Root/
1 | /* |
2 | * (C) 1997 Linus Torvalds |
3 | * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation) |
4 | */ |
5 | #include <linux/fs.h> |
6 | #include <linux/mm.h> |
7 | #include <linux/dcache.h> |
8 | #include <linux/init.h> |
9 | #include <linux/slab.h> |
10 | #include <linux/writeback.h> |
11 | #include <linux/module.h> |
12 | #include <linux/backing-dev.h> |
13 | #include <linux/wait.h> |
14 | #include <linux/rwsem.h> |
15 | #include <linux/hash.h> |
16 | #include <linux/swap.h> |
17 | #include <linux/security.h> |
18 | #include <linux/pagemap.h> |
19 | #include <linux/cdev.h> |
20 | #include <linux/bootmem.h> |
21 | #include <linux/fsnotify.h> |
22 | #include <linux/mount.h> |
23 | #include <linux/async.h> |
24 | #include <linux/posix_acl.h> |
25 | #include <linux/prefetch.h> |
26 | #include <linux/ima.h> |
27 | #include <linux/cred.h> |
28 | #include <linux/buffer_head.h> /* for inode_has_buffers */ |
29 | #include <linux/ratelimit.h> |
30 | #include "internal.h" |
31 | |
32 | /* |
33 | * Inode locking rules: |
34 | * |
35 | * inode->i_lock protects: |
36 | * inode->i_state, inode->i_hash, __iget() |
37 | * inode->i_sb->s_inode_lru_lock protects: |
38 | * inode->i_sb->s_inode_lru, inode->i_lru |
39 | * inode_sb_list_lock protects: |
40 | * sb->s_inodes, inode->i_sb_list |
41 | * bdi->wb.list_lock protects: |
42 | * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list |
43 | * inode_hash_lock protects: |
44 | * inode_hashtable, inode->i_hash |
45 | * |
46 | * Lock ordering: |
47 | * |
48 | * inode_sb_list_lock |
49 | * inode->i_lock |
50 | * inode->i_sb->s_inode_lru_lock |
51 | * |
52 | * bdi->wb.list_lock |
53 | * inode->i_lock |
54 | * |
55 | * inode_hash_lock |
56 | * inode_sb_list_lock |
57 | * inode->i_lock |
58 | * |
59 | * iunique_lock |
60 | * inode_hash_lock |
61 | */ |
62 | |
63 | static unsigned int i_hash_mask __read_mostly; |
64 | static unsigned int i_hash_shift __read_mostly; |
65 | static struct hlist_head *inode_hashtable __read_mostly; |
66 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock); |
67 | |
68 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock); |
69 | |
70 | /* |
71 | * Empty aops. Can be used for the cases where the user does not |
72 | * define any of the address_space operations. |
73 | */ |
74 | const struct address_space_operations empty_aops = { |
75 | }; |
76 | EXPORT_SYMBOL(empty_aops); |
77 | |
78 | /* |
79 | * Statistics gathering.. |
80 | */ |
81 | struct inodes_stat_t inodes_stat; |
82 | |
83 | static DEFINE_PER_CPU(unsigned int, nr_inodes); |
84 | static DEFINE_PER_CPU(unsigned int, nr_unused); |
85 | |
86 | static struct kmem_cache *inode_cachep __read_mostly; |
87 | |
88 | static int get_nr_inodes(void) |
89 | { |
90 | int i; |
91 | int sum = 0; |
92 | for_each_possible_cpu(i) |
93 | sum += per_cpu(nr_inodes, i); |
94 | return sum < 0 ? 0 : sum; |
95 | } |
96 | |
97 | static inline int get_nr_inodes_unused(void) |
98 | { |
99 | int i; |
100 | int sum = 0; |
101 | for_each_possible_cpu(i) |
102 | sum += per_cpu(nr_unused, i); |
103 | return sum < 0 ? 0 : sum; |
104 | } |
105 | |
106 | int get_nr_dirty_inodes(void) |
107 | { |
108 | /* not actually dirty inodes, but a wild approximation */ |
109 | int nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); |
110 | return nr_dirty > 0 ? nr_dirty : 0; |
111 | } |
112 | |
113 | /* |
114 | * Handle nr_inode sysctl |
115 | */ |
116 | #ifdef CONFIG_SYSCTL |
117 | int proc_nr_inodes(ctl_table *table, int write, |
118 | void __user *buffer, size_t *lenp, loff_t *ppos) |
119 | { |
120 | inodes_stat.nr_inodes = get_nr_inodes(); |
121 | inodes_stat.nr_unused = get_nr_inodes_unused(); |
122 | return proc_dointvec(table, write, buffer, lenp, ppos); |
123 | } |
124 | #endif |
125 | |
126 | /** |
127 | * inode_init_always - perform inode structure intialisation |
128 | * @sb: superblock inode belongs to |
129 | * @inode: inode to initialise |
130 | * |
131 | * These are initializations that need to be done on every inode |
132 | * allocation as the fields are not initialised by slab allocation. |
133 | */ |
134 | int inode_init_always(struct super_block *sb, struct inode *inode) |
135 | { |
136 | static const struct inode_operations empty_iops; |
137 | static const struct file_operations empty_fops; |
138 | struct address_space *const mapping = &inode->i_data; |
139 | |
140 | inode->i_sb = sb; |
141 | inode->i_blkbits = sb->s_blocksize_bits; |
142 | inode->i_flags = 0; |
143 | atomic_set(&inode->i_count, 1); |
144 | inode->i_op = &empty_iops; |
145 | inode->i_fop = &empty_fops; |
146 | inode->__i_nlink = 1; |
147 | inode->i_opflags = 0; |
148 | inode->i_uid = 0; |
149 | inode->i_gid = 0; |
150 | atomic_set(&inode->i_writecount, 0); |
151 | inode->i_size = 0; |
152 | inode->i_blocks = 0; |
153 | inode->i_bytes = 0; |
154 | inode->i_generation = 0; |
155 | #ifdef CONFIG_QUOTA |
156 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
157 | #endif |
158 | inode->i_pipe = NULL; |
159 | inode->i_bdev = NULL; |
160 | inode->i_cdev = NULL; |
161 | inode->i_rdev = 0; |
162 | inode->dirtied_when = 0; |
163 | |
164 | if (security_inode_alloc(inode)) |
165 | goto out; |
166 | spin_lock_init(&inode->i_lock); |
167 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); |
168 | |
169 | mutex_init(&inode->i_mutex); |
170 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); |
171 | |
172 | atomic_set(&inode->i_dio_count, 0); |
173 | |
174 | mapping->a_ops = &empty_aops; |
175 | mapping->host = inode; |
176 | mapping->flags = 0; |
177 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
178 | mapping->assoc_mapping = NULL; |
179 | mapping->backing_dev_info = &default_backing_dev_info; |
180 | mapping->writeback_index = 0; |
181 | |
182 | /* |
183 | * If the block_device provides a backing_dev_info for client |
184 | * inodes then use that. Otherwise the inode share the bdev's |
185 | * backing_dev_info. |
186 | */ |
187 | if (sb->s_bdev) { |
188 | struct backing_dev_info *bdi; |
189 | |
190 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; |
191 | mapping->backing_dev_info = bdi; |
192 | } |
193 | inode->i_private = NULL; |
194 | inode->i_mapping = mapping; |
195 | INIT_LIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */ |
196 | #ifdef CONFIG_FS_POSIX_ACL |
197 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; |
198 | #endif |
199 | |
200 | #ifdef CONFIG_FSNOTIFY |
201 | inode->i_fsnotify_mask = 0; |
202 | #endif |
203 | |
204 | this_cpu_inc(nr_inodes); |
205 | |
206 | return 0; |
207 | out: |
208 | return -ENOMEM; |
209 | } |
210 | EXPORT_SYMBOL(inode_init_always); |
211 | |
212 | static struct inode *alloc_inode(struct super_block *sb) |
213 | { |
214 | struct inode *inode; |
215 | |
216 | if (sb->s_op->alloc_inode) |
217 | inode = sb->s_op->alloc_inode(sb); |
218 | else |
219 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); |
220 | |
221 | if (!inode) |
222 | return NULL; |
223 | |
224 | if (unlikely(inode_init_always(sb, inode))) { |
225 | if (inode->i_sb->s_op->destroy_inode) |
226 | inode->i_sb->s_op->destroy_inode(inode); |
227 | else |
228 | kmem_cache_free(inode_cachep, inode); |
229 | return NULL; |
230 | } |
231 | |
232 | return inode; |
233 | } |
234 | |
235 | void free_inode_nonrcu(struct inode *inode) |
236 | { |
237 | kmem_cache_free(inode_cachep, inode); |
238 | } |
239 | EXPORT_SYMBOL(free_inode_nonrcu); |
240 | |
241 | void __destroy_inode(struct inode *inode) |
242 | { |
243 | BUG_ON(inode_has_buffers(inode)); |
244 | security_inode_free(inode); |
245 | fsnotify_inode_delete(inode); |
246 | if (!inode->i_nlink) { |
247 | WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0); |
248 | atomic_long_dec(&inode->i_sb->s_remove_count); |
249 | } |
250 | |
251 | #ifdef CONFIG_FS_POSIX_ACL |
252 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) |
253 | posix_acl_release(inode->i_acl); |
254 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) |
255 | posix_acl_release(inode->i_default_acl); |
256 | #endif |
257 | this_cpu_dec(nr_inodes); |
258 | } |
259 | EXPORT_SYMBOL(__destroy_inode); |
260 | |
261 | static void i_callback(struct rcu_head *head) |
262 | { |
263 | struct inode *inode = container_of(head, struct inode, i_rcu); |
264 | kmem_cache_free(inode_cachep, inode); |
265 | } |
266 | |
267 | static void destroy_inode(struct inode *inode) |
268 | { |
269 | BUG_ON(!list_empty(&inode->i_lru)); |
270 | __destroy_inode(inode); |
271 | if (inode->i_sb->s_op->destroy_inode) |
272 | inode->i_sb->s_op->destroy_inode(inode); |
273 | else |
274 | call_rcu(&inode->i_rcu, i_callback); |
275 | } |
276 | |
277 | /** |
278 | * drop_nlink - directly drop an inode's link count |
279 | * @inode: inode |
280 | * |
281 | * This is a low-level filesystem helper to replace any |
282 | * direct filesystem manipulation of i_nlink. In cases |
283 | * where we are attempting to track writes to the |
284 | * filesystem, a decrement to zero means an imminent |
285 | * write when the file is truncated and actually unlinked |
286 | * on the filesystem. |
287 | */ |
288 | void drop_nlink(struct inode *inode) |
289 | { |
290 | WARN_ON(inode->i_nlink == 0); |
291 | inode->__i_nlink--; |
292 | if (!inode->i_nlink) |
293 | atomic_long_inc(&inode->i_sb->s_remove_count); |
294 | } |
295 | EXPORT_SYMBOL(drop_nlink); |
296 | |
297 | /** |
298 | * clear_nlink - directly zero an inode's link count |
299 | * @inode: inode |
300 | * |
301 | * This is a low-level filesystem helper to replace any |
302 | * direct filesystem manipulation of i_nlink. See |
303 | * drop_nlink() for why we care about i_nlink hitting zero. |
304 | */ |
305 | void clear_nlink(struct inode *inode) |
306 | { |
307 | if (inode->i_nlink) { |
308 | inode->__i_nlink = 0; |
309 | atomic_long_inc(&inode->i_sb->s_remove_count); |
310 | } |
311 | } |
312 | EXPORT_SYMBOL(clear_nlink); |
313 | |
314 | /** |
315 | * set_nlink - directly set an inode's link count |
316 | * @inode: inode |
317 | * @nlink: new nlink (should be non-zero) |
318 | * |
319 | * This is a low-level filesystem helper to replace any |
320 | * direct filesystem manipulation of i_nlink. |
321 | */ |
322 | void set_nlink(struct inode *inode, unsigned int nlink) |
323 | { |
324 | if (!nlink) { |
325 | clear_nlink(inode); |
326 | } else { |
327 | /* Yes, some filesystems do change nlink from zero to one */ |
328 | if (inode->i_nlink == 0) |
329 | atomic_long_dec(&inode->i_sb->s_remove_count); |
330 | |
331 | inode->__i_nlink = nlink; |
332 | } |
333 | } |
334 | EXPORT_SYMBOL(set_nlink); |
335 | |
336 | /** |
337 | * inc_nlink - directly increment an inode's link count |
338 | * @inode: inode |
339 | * |
340 | * This is a low-level filesystem helper to replace any |
341 | * direct filesystem manipulation of i_nlink. Currently, |
342 | * it is only here for parity with dec_nlink(). |
343 | */ |
344 | void inc_nlink(struct inode *inode) |
345 | { |
346 | if (WARN_ON(inode->i_nlink == 0)) |
347 | atomic_long_dec(&inode->i_sb->s_remove_count); |
348 | |
349 | inode->__i_nlink++; |
350 | } |
351 | EXPORT_SYMBOL(inc_nlink); |
352 | |
353 | void address_space_init_once(struct address_space *mapping) |
354 | { |
355 | memset(mapping, 0, sizeof(*mapping)); |
356 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC); |
357 | spin_lock_init(&mapping->tree_lock); |
358 | mutex_init(&mapping->i_mmap_mutex); |
359 | INIT_LIST_HEAD(&mapping->private_list); |
360 | spin_lock_init(&mapping->private_lock); |
361 | INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap); |
362 | INIT_LIST_HEAD(&mapping->i_mmap_nonlinear); |
363 | } |
364 | EXPORT_SYMBOL(address_space_init_once); |
365 | |
366 | /* |
367 | * These are initializations that only need to be done |
368 | * once, because the fields are idempotent across use |
369 | * of the inode, so let the slab aware of that. |
370 | */ |
371 | void inode_init_once(struct inode *inode) |
372 | { |
373 | memset(inode, 0, sizeof(*inode)); |
374 | INIT_HLIST_NODE(&inode->i_hash); |
375 | INIT_LIST_HEAD(&inode->i_devices); |
376 | INIT_LIST_HEAD(&inode->i_wb_list); |
377 | INIT_LIST_HEAD(&inode->i_lru); |
378 | address_space_init_once(&inode->i_data); |
379 | i_size_ordered_init(inode); |
380 | #ifdef CONFIG_FSNOTIFY |
381 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); |
382 | #endif |
383 | } |
384 | EXPORT_SYMBOL(inode_init_once); |
385 | |
386 | static void init_once(void *foo) |
387 | { |
388 | struct inode *inode = (struct inode *) foo; |
389 | |
390 | inode_init_once(inode); |
391 | } |
392 | |
393 | /* |
394 | * inode->i_lock must be held |
395 | */ |
396 | void __iget(struct inode *inode) |
397 | { |
398 | atomic_inc(&inode->i_count); |
399 | } |
400 | |
401 | /* |
402 | * get additional reference to inode; caller must already hold one. |
403 | */ |
404 | void ihold(struct inode *inode) |
405 | { |
406 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); |
407 | } |
408 | EXPORT_SYMBOL(ihold); |
409 | |
410 | static void inode_lru_list_add(struct inode *inode) |
411 | { |
412 | spin_lock(&inode->i_sb->s_inode_lru_lock); |
413 | if (list_empty(&inode->i_lru)) { |
414 | list_add(&inode->i_lru, &inode->i_sb->s_inode_lru); |
415 | inode->i_sb->s_nr_inodes_unused++; |
416 | this_cpu_inc(nr_unused); |
417 | } |
418 | spin_unlock(&inode->i_sb->s_inode_lru_lock); |
419 | } |
420 | |
421 | static void inode_lru_list_del(struct inode *inode) |
422 | { |
423 | spin_lock(&inode->i_sb->s_inode_lru_lock); |
424 | if (!list_empty(&inode->i_lru)) { |
425 | list_del_init(&inode->i_lru); |
426 | inode->i_sb->s_nr_inodes_unused--; |
427 | this_cpu_dec(nr_unused); |
428 | } |
429 | spin_unlock(&inode->i_sb->s_inode_lru_lock); |
430 | } |
431 | |
432 | /** |
433 | * inode_sb_list_add - add inode to the superblock list of inodes |
434 | * @inode: inode to add |
435 | */ |
436 | void inode_sb_list_add(struct inode *inode) |
437 | { |
438 | spin_lock(&inode_sb_list_lock); |
439 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); |
440 | spin_unlock(&inode_sb_list_lock); |
441 | } |
442 | EXPORT_SYMBOL_GPL(inode_sb_list_add); |
443 | |
444 | static inline void inode_sb_list_del(struct inode *inode) |
445 | { |
446 | if (!list_empty(&inode->i_sb_list)) { |
447 | spin_lock(&inode_sb_list_lock); |
448 | list_del_init(&inode->i_sb_list); |
449 | spin_unlock(&inode_sb_list_lock); |
450 | } |
451 | } |
452 | |
453 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
454 | { |
455 | unsigned long tmp; |
456 | |
457 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / |
458 | L1_CACHE_BYTES; |
459 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift); |
460 | return tmp & i_hash_mask; |
461 | } |
462 | |
463 | /** |
464 | * __insert_inode_hash - hash an inode |
465 | * @inode: unhashed inode |
466 | * @hashval: unsigned long value used to locate this object in the |
467 | * inode_hashtable. |
468 | * |
469 | * Add an inode to the inode hash for this superblock. |
470 | */ |
471 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) |
472 | { |
473 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); |
474 | |
475 | spin_lock(&inode_hash_lock); |
476 | spin_lock(&inode->i_lock); |
477 | hlist_add_head(&inode->i_hash, b); |
478 | spin_unlock(&inode->i_lock); |
479 | spin_unlock(&inode_hash_lock); |
480 | } |
481 | EXPORT_SYMBOL(__insert_inode_hash); |
482 | |
483 | /** |
484 | * __remove_inode_hash - remove an inode from the hash |
485 | * @inode: inode to unhash |
486 | * |
487 | * Remove an inode from the superblock. |
488 | */ |
489 | void __remove_inode_hash(struct inode *inode) |
490 | { |
491 | spin_lock(&inode_hash_lock); |
492 | spin_lock(&inode->i_lock); |
493 | hlist_del_init(&inode->i_hash); |
494 | spin_unlock(&inode->i_lock); |
495 | spin_unlock(&inode_hash_lock); |
496 | } |
497 | EXPORT_SYMBOL(__remove_inode_hash); |
498 | |
499 | void end_writeback(struct inode *inode) |
500 | { |
501 | might_sleep(); |
502 | /* |
503 | * We have to cycle tree_lock here because reclaim can be still in the |
504 | * process of removing the last page (in __delete_from_page_cache()) |
505 | * and we must not free mapping under it. |
506 | */ |
507 | spin_lock_irq(&inode->i_data.tree_lock); |
508 | BUG_ON(inode->i_data.nrpages); |
509 | spin_unlock_irq(&inode->i_data.tree_lock); |
510 | BUG_ON(!list_empty(&inode->i_data.private_list)); |
511 | BUG_ON(!(inode->i_state & I_FREEING)); |
512 | BUG_ON(inode->i_state & I_CLEAR); |
513 | inode_sync_wait(inode); |
514 | /* don't need i_lock here, no concurrent mods to i_state */ |
515 | inode->i_state = I_FREEING | I_CLEAR; |
516 | } |
517 | EXPORT_SYMBOL(end_writeback); |
518 | |
519 | /* |
520 | * Free the inode passed in, removing it from the lists it is still connected |
521 | * to. We remove any pages still attached to the inode and wait for any IO that |
522 | * is still in progress before finally destroying the inode. |
523 | * |
524 | * An inode must already be marked I_FREEING so that we avoid the inode being |
525 | * moved back onto lists if we race with other code that manipulates the lists |
526 | * (e.g. writeback_single_inode). The caller is responsible for setting this. |
527 | * |
528 | * An inode must already be removed from the LRU list before being evicted from |
529 | * the cache. This should occur atomically with setting the I_FREEING state |
530 | * flag, so no inodes here should ever be on the LRU when being evicted. |
531 | */ |
532 | static void evict(struct inode *inode) |
533 | { |
534 | const struct super_operations *op = inode->i_sb->s_op; |
535 | |
536 | BUG_ON(!(inode->i_state & I_FREEING)); |
537 | BUG_ON(!list_empty(&inode->i_lru)); |
538 | |
539 | if (!list_empty(&inode->i_wb_list)) |
540 | inode_wb_list_del(inode); |
541 | |
542 | inode_sb_list_del(inode); |
543 | |
544 | if (op->evict_inode) { |
545 | op->evict_inode(inode); |
546 | } else { |
547 | if (inode->i_data.nrpages) |
548 | truncate_inode_pages(&inode->i_data, 0); |
549 | end_writeback(inode); |
550 | } |
551 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
552 | bd_forget(inode); |
553 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) |
554 | cd_forget(inode); |
555 | |
556 | remove_inode_hash(inode); |
557 | |
558 | spin_lock(&inode->i_lock); |
559 | wake_up_bit(&inode->i_state, __I_NEW); |
560 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); |
561 | spin_unlock(&inode->i_lock); |
562 | |
563 | destroy_inode(inode); |
564 | } |
565 | |
566 | /* |
567 | * dispose_list - dispose of the contents of a local list |
568 | * @head: the head of the list to free |
569 | * |
570 | * Dispose-list gets a local list with local inodes in it, so it doesn't |
571 | * need to worry about list corruption and SMP locks. |
572 | */ |
573 | static void dispose_list(struct list_head *head) |
574 | { |
575 | while (!list_empty(head)) { |
576 | struct inode *inode; |
577 | |
578 | inode = list_first_entry(head, struct inode, i_lru); |
579 | list_del_init(&inode->i_lru); |
580 | |
581 | evict(inode); |
582 | } |
583 | } |
584 | |
585 | /** |
586 | * evict_inodes - evict all evictable inodes for a superblock |
587 | * @sb: superblock to operate on |
588 | * |
589 | * Make sure that no inodes with zero refcount are retained. This is |
590 | * called by superblock shutdown after having MS_ACTIVE flag removed, |
591 | * so any inode reaching zero refcount during or after that call will |
592 | * be immediately evicted. |
593 | */ |
594 | void evict_inodes(struct super_block *sb) |
595 | { |
596 | struct inode *inode, *next; |
597 | LIST_HEAD(dispose); |
598 | |
599 | spin_lock(&inode_sb_list_lock); |
600 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
601 | if (atomic_read(&inode->i_count)) |
602 | continue; |
603 | |
604 | spin_lock(&inode->i_lock); |
605 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
606 | spin_unlock(&inode->i_lock); |
607 | continue; |
608 | } |
609 | |
610 | inode->i_state |= I_FREEING; |
611 | inode_lru_list_del(inode); |
612 | spin_unlock(&inode->i_lock); |
613 | list_add(&inode->i_lru, &dispose); |
614 | } |
615 | spin_unlock(&inode_sb_list_lock); |
616 | |
617 | dispose_list(&dispose); |
618 | } |
619 | |
620 | /** |
621 | * invalidate_inodes - attempt to free all inodes on a superblock |
622 | * @sb: superblock to operate on |
623 | * @kill_dirty: flag to guide handling of dirty inodes |
624 | * |
625 | * Attempts to free all inodes for a given superblock. If there were any |
626 | * busy inodes return a non-zero value, else zero. |
627 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat |
628 | * them as busy. |
629 | */ |
630 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) |
631 | { |
632 | int busy = 0; |
633 | struct inode *inode, *next; |
634 | LIST_HEAD(dispose); |
635 | |
636 | spin_lock(&inode_sb_list_lock); |
637 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
638 | spin_lock(&inode->i_lock); |
639 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
640 | spin_unlock(&inode->i_lock); |
641 | continue; |
642 | } |
643 | if (inode->i_state & I_DIRTY && !kill_dirty) { |
644 | spin_unlock(&inode->i_lock); |
645 | busy = 1; |
646 | continue; |
647 | } |
648 | if (atomic_read(&inode->i_count)) { |
649 | spin_unlock(&inode->i_lock); |
650 | busy = 1; |
651 | continue; |
652 | } |
653 | |
654 | inode->i_state |= I_FREEING; |
655 | inode_lru_list_del(inode); |
656 | spin_unlock(&inode->i_lock); |
657 | list_add(&inode->i_lru, &dispose); |
658 | } |
659 | spin_unlock(&inode_sb_list_lock); |
660 | |
661 | dispose_list(&dispose); |
662 | |
663 | return busy; |
664 | } |
665 | |
666 | static int can_unuse(struct inode *inode) |
667 | { |
668 | if (inode->i_state & ~I_REFERENCED) |
669 | return 0; |
670 | if (inode_has_buffers(inode)) |
671 | return 0; |
672 | if (atomic_read(&inode->i_count)) |
673 | return 0; |
674 | if (inode->i_data.nrpages) |
675 | return 0; |
676 | return 1; |
677 | } |
678 | |
679 | /* |
680 | * Walk the superblock inode LRU for freeable inodes and attempt to free them. |
681 | * This is called from the superblock shrinker function with a number of inodes |
682 | * to trim from the LRU. Inodes to be freed are moved to a temporary list and |
683 | * then are freed outside inode_lock by dispose_list(). |
684 | * |
685 | * Any inodes which are pinned purely because of attached pagecache have their |
686 | * pagecache removed. If the inode has metadata buffers attached to |
687 | * mapping->private_list then try to remove them. |
688 | * |
689 | * If the inode has the I_REFERENCED flag set, then it means that it has been |
690 | * used recently - the flag is set in iput_final(). When we encounter such an |
691 | * inode, clear the flag and move it to the back of the LRU so it gets another |
692 | * pass through the LRU before it gets reclaimed. This is necessary because of |
693 | * the fact we are doing lazy LRU updates to minimise lock contention so the |
694 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes |
695 | * with this flag set because they are the inodes that are out of order. |
696 | */ |
697 | void prune_icache_sb(struct super_block *sb, int nr_to_scan) |
698 | { |
699 | LIST_HEAD(freeable); |
700 | int nr_scanned; |
701 | unsigned long reap = 0; |
702 | |
703 | spin_lock(&sb->s_inode_lru_lock); |
704 | for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) { |
705 | struct inode *inode; |
706 | |
707 | if (list_empty(&sb->s_inode_lru)) |
708 | break; |
709 | |
710 | inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru); |
711 | |
712 | /* |
713 | * we are inverting the sb->s_inode_lru_lock/inode->i_lock here, |
714 | * so use a trylock. If we fail to get the lock, just move the |
715 | * inode to the back of the list so we don't spin on it. |
716 | */ |
717 | if (!spin_trylock(&inode->i_lock)) { |
718 | list_move_tail(&inode->i_lru, &sb->s_inode_lru); |
719 | continue; |
720 | } |
721 | |
722 | /* |
723 | * Referenced or dirty inodes are still in use. Give them |
724 | * another pass through the LRU as we canot reclaim them now. |
725 | */ |
726 | if (atomic_read(&inode->i_count) || |
727 | (inode->i_state & ~I_REFERENCED)) { |
728 | list_del_init(&inode->i_lru); |
729 | spin_unlock(&inode->i_lock); |
730 | sb->s_nr_inodes_unused--; |
731 | this_cpu_dec(nr_unused); |
732 | continue; |
733 | } |
734 | |
735 | /* recently referenced inodes get one more pass */ |
736 | if (inode->i_state & I_REFERENCED) { |
737 | inode->i_state &= ~I_REFERENCED; |
738 | list_move(&inode->i_lru, &sb->s_inode_lru); |
739 | spin_unlock(&inode->i_lock); |
740 | continue; |
741 | } |
742 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { |
743 | __iget(inode); |
744 | spin_unlock(&inode->i_lock); |
745 | spin_unlock(&sb->s_inode_lru_lock); |
746 | if (remove_inode_buffers(inode)) |
747 | reap += invalidate_mapping_pages(&inode->i_data, |
748 | 0, -1); |
749 | iput(inode); |
750 | spin_lock(&sb->s_inode_lru_lock); |
751 | |
752 | if (inode != list_entry(sb->s_inode_lru.next, |
753 | struct inode, i_lru)) |
754 | continue; /* wrong inode or list_empty */ |
755 | /* avoid lock inversions with trylock */ |
756 | if (!spin_trylock(&inode->i_lock)) |
757 | continue; |
758 | if (!can_unuse(inode)) { |
759 | spin_unlock(&inode->i_lock); |
760 | continue; |
761 | } |
762 | } |
763 | WARN_ON(inode->i_state & I_NEW); |
764 | inode->i_state |= I_FREEING; |
765 | spin_unlock(&inode->i_lock); |
766 | |
767 | list_move(&inode->i_lru, &freeable); |
768 | sb->s_nr_inodes_unused--; |
769 | this_cpu_dec(nr_unused); |
770 | } |
771 | if (current_is_kswapd()) |
772 | __count_vm_events(KSWAPD_INODESTEAL, reap); |
773 | else |
774 | __count_vm_events(PGINODESTEAL, reap); |
775 | spin_unlock(&sb->s_inode_lru_lock); |
776 | if (current->reclaim_state) |
777 | current->reclaim_state->reclaimed_slab += reap; |
778 | |
779 | dispose_list(&freeable); |
780 | } |
781 | |
782 | static void __wait_on_freeing_inode(struct inode *inode); |
783 | /* |
784 | * Called with the inode lock held. |
785 | */ |
786 | static struct inode *find_inode(struct super_block *sb, |
787 | struct hlist_head *head, |
788 | int (*test)(struct inode *, void *), |
789 | void *data) |
790 | { |
791 | struct hlist_node *node; |
792 | struct inode *inode = NULL; |
793 | |
794 | repeat: |
795 | hlist_for_each_entry(inode, node, head, i_hash) { |
796 | spin_lock(&inode->i_lock); |
797 | if (inode->i_sb != sb) { |
798 | spin_unlock(&inode->i_lock); |
799 | continue; |
800 | } |
801 | if (!test(inode, data)) { |
802 | spin_unlock(&inode->i_lock); |
803 | continue; |
804 | } |
805 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
806 | __wait_on_freeing_inode(inode); |
807 | goto repeat; |
808 | } |
809 | __iget(inode); |
810 | spin_unlock(&inode->i_lock); |
811 | return inode; |
812 | } |
813 | return NULL; |
814 | } |
815 | |
816 | /* |
817 | * find_inode_fast is the fast path version of find_inode, see the comment at |
818 | * iget_locked for details. |
819 | */ |
820 | static struct inode *find_inode_fast(struct super_block *sb, |
821 | struct hlist_head *head, unsigned long ino) |
822 | { |
823 | struct hlist_node *node; |
824 | struct inode *inode = NULL; |
825 | |
826 | repeat: |
827 | hlist_for_each_entry(inode, node, head, i_hash) { |
828 | spin_lock(&inode->i_lock); |
829 | if (inode->i_ino != ino) { |
830 | spin_unlock(&inode->i_lock); |
831 | continue; |
832 | } |
833 | if (inode->i_sb != sb) { |
834 | spin_unlock(&inode->i_lock); |
835 | continue; |
836 | } |
837 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
838 | __wait_on_freeing_inode(inode); |
839 | goto repeat; |
840 | } |
841 | __iget(inode); |
842 | spin_unlock(&inode->i_lock); |
843 | return inode; |
844 | } |
845 | return NULL; |
846 | } |
847 | |
848 | /* |
849 | * Each cpu owns a range of LAST_INO_BATCH numbers. |
850 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, |
851 | * to renew the exhausted range. |
852 | * |
853 | * This does not significantly increase overflow rate because every CPU can |
854 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is |
855 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the |
856 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase |
857 | * overflow rate by 2x, which does not seem too significant. |
858 | * |
859 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW |
860 | * error if st_ino won't fit in target struct field. Use 32bit counter |
861 | * here to attempt to avoid that. |
862 | */ |
863 | #define LAST_INO_BATCH 1024 |
864 | static DEFINE_PER_CPU(unsigned int, last_ino); |
865 | |
866 | unsigned int get_next_ino(void) |
867 | { |
868 | unsigned int *p = &get_cpu_var(last_ino); |
869 | unsigned int res = *p; |
870 | |
871 | #ifdef CONFIG_SMP |
872 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { |
873 | static atomic_t shared_last_ino; |
874 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); |
875 | |
876 | res = next - LAST_INO_BATCH; |
877 | } |
878 | #endif |
879 | |
880 | *p = ++res; |
881 | put_cpu_var(last_ino); |
882 | return res; |
883 | } |
884 | EXPORT_SYMBOL(get_next_ino); |
885 | |
886 | /** |
887 | * new_inode_pseudo - obtain an inode |
888 | * @sb: superblock |
889 | * |
890 | * Allocates a new inode for given superblock. |
891 | * Inode wont be chained in superblock s_inodes list |
892 | * This means : |
893 | * - fs can't be unmount |
894 | * - quotas, fsnotify, writeback can't work |
895 | */ |
896 | struct inode *new_inode_pseudo(struct super_block *sb) |
897 | { |
898 | struct inode *inode = alloc_inode(sb); |
899 | |
900 | if (inode) { |
901 | spin_lock(&inode->i_lock); |
902 | inode->i_state = 0; |
903 | spin_unlock(&inode->i_lock); |
904 | INIT_LIST_HEAD(&inode->i_sb_list); |
905 | } |
906 | return inode; |
907 | } |
908 | |
909 | /** |
910 | * new_inode - obtain an inode |
911 | * @sb: superblock |
912 | * |
913 | * Allocates a new inode for given superblock. The default gfp_mask |
914 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
915 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
916 | * for the page cache are not reclaimable or migratable, |
917 | * mapping_set_gfp_mask() must be called with suitable flags on the |
918 | * newly created inode's mapping |
919 | * |
920 | */ |
921 | struct inode *new_inode(struct super_block *sb) |
922 | { |
923 | struct inode *inode; |
924 | |
925 | spin_lock_prefetch(&inode_sb_list_lock); |
926 | |
927 | inode = new_inode_pseudo(sb); |
928 | if (inode) |
929 | inode_sb_list_add(inode); |
930 | return inode; |
931 | } |
932 | EXPORT_SYMBOL(new_inode); |
933 | |
934 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
935 | void lockdep_annotate_inode_mutex_key(struct inode *inode) |
936 | { |
937 | if (S_ISDIR(inode->i_mode)) { |
938 | struct file_system_type *type = inode->i_sb->s_type; |
939 | |
940 | /* Set new key only if filesystem hasn't already changed it */ |
941 | if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) { |
942 | /* |
943 | * ensure nobody is actually holding i_mutex |
944 | */ |
945 | mutex_destroy(&inode->i_mutex); |
946 | mutex_init(&inode->i_mutex); |
947 | lockdep_set_class(&inode->i_mutex, |
948 | &type->i_mutex_dir_key); |
949 | } |
950 | } |
951 | } |
952 | EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key); |
953 | #endif |
954 | |
955 | /** |
956 | * unlock_new_inode - clear the I_NEW state and wake up any waiters |
957 | * @inode: new inode to unlock |
958 | * |
959 | * Called when the inode is fully initialised to clear the new state of the |
960 | * inode and wake up anyone waiting for the inode to finish initialisation. |
961 | */ |
962 | void unlock_new_inode(struct inode *inode) |
963 | { |
964 | lockdep_annotate_inode_mutex_key(inode); |
965 | spin_lock(&inode->i_lock); |
966 | WARN_ON(!(inode->i_state & I_NEW)); |
967 | inode->i_state &= ~I_NEW; |
968 | smp_mb(); |
969 | wake_up_bit(&inode->i_state, __I_NEW); |
970 | spin_unlock(&inode->i_lock); |
971 | } |
972 | EXPORT_SYMBOL(unlock_new_inode); |
973 | |
974 | /** |
975 | * iget5_locked - obtain an inode from a mounted file system |
976 | * @sb: super block of file system |
977 | * @hashval: hash value (usually inode number) to get |
978 | * @test: callback used for comparisons between inodes |
979 | * @set: callback used to initialize a new struct inode |
980 | * @data: opaque data pointer to pass to @test and @set |
981 | * |
982 | * Search for the inode specified by @hashval and @data in the inode cache, |
983 | * and if present it is return it with an increased reference count. This is |
984 | * a generalized version of iget_locked() for file systems where the inode |
985 | * number is not sufficient for unique identification of an inode. |
986 | * |
987 | * If the inode is not in cache, allocate a new inode and return it locked, |
988 | * hashed, and with the I_NEW flag set. The file system gets to fill it in |
989 | * before unlocking it via unlock_new_inode(). |
990 | * |
991 | * Note both @test and @set are called with the inode_hash_lock held, so can't |
992 | * sleep. |
993 | */ |
994 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, |
995 | int (*test)(struct inode *, void *), |
996 | int (*set)(struct inode *, void *), void *data) |
997 | { |
998 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
999 | struct inode *inode; |
1000 | |
1001 | spin_lock(&inode_hash_lock); |
1002 | inode = find_inode(sb, head, test, data); |
1003 | spin_unlock(&inode_hash_lock); |
1004 | |
1005 | if (inode) { |
1006 | wait_on_inode(inode); |
1007 | return inode; |
1008 | } |
1009 | |
1010 | inode = alloc_inode(sb); |
1011 | if (inode) { |
1012 | struct inode *old; |
1013 | |
1014 | spin_lock(&inode_hash_lock); |
1015 | /* We released the lock, so.. */ |
1016 | old = find_inode(sb, head, test, data); |
1017 | if (!old) { |
1018 | if (set(inode, data)) |
1019 | goto set_failed; |
1020 | |
1021 | spin_lock(&inode->i_lock); |
1022 | inode->i_state = I_NEW; |
1023 | hlist_add_head(&inode->i_hash, head); |
1024 | spin_unlock(&inode->i_lock); |
1025 | inode_sb_list_add(inode); |
1026 | spin_unlock(&inode_hash_lock); |
1027 | |
1028 | /* Return the locked inode with I_NEW set, the |
1029 | * caller is responsible for filling in the contents |
1030 | */ |
1031 | return inode; |
1032 | } |
1033 | |
1034 | /* |
1035 | * Uhhuh, somebody else created the same inode under |
1036 | * us. Use the old inode instead of the one we just |
1037 | * allocated. |
1038 | */ |
1039 | spin_unlock(&inode_hash_lock); |
1040 | destroy_inode(inode); |
1041 | inode = old; |
1042 | wait_on_inode(inode); |
1043 | } |
1044 | return inode; |
1045 | |
1046 | set_failed: |
1047 | spin_unlock(&inode_hash_lock); |
1048 | destroy_inode(inode); |
1049 | return NULL; |
1050 | } |
1051 | EXPORT_SYMBOL(iget5_locked); |
1052 | |
1053 | /** |
1054 | * iget_locked - obtain an inode from a mounted file system |
1055 | * @sb: super block of file system |
1056 | * @ino: inode number to get |
1057 | * |
1058 | * Search for the inode specified by @ino in the inode cache and if present |
1059 | * return it with an increased reference count. This is for file systems |
1060 | * where the inode number is sufficient for unique identification of an inode. |
1061 | * |
1062 | * If the inode is not in cache, allocate a new inode and return it locked, |
1063 | * hashed, and with the I_NEW flag set. The file system gets to fill it in |
1064 | * before unlocking it via unlock_new_inode(). |
1065 | */ |
1066 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) |
1067 | { |
1068 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
1069 | struct inode *inode; |
1070 | |
1071 | spin_lock(&inode_hash_lock); |
1072 | inode = find_inode_fast(sb, head, ino); |
1073 | spin_unlock(&inode_hash_lock); |
1074 | if (inode) { |
1075 | wait_on_inode(inode); |
1076 | return inode; |
1077 | } |
1078 | |
1079 | inode = alloc_inode(sb); |
1080 | if (inode) { |
1081 | struct inode *old; |
1082 | |
1083 | spin_lock(&inode_hash_lock); |
1084 | /* We released the lock, so.. */ |
1085 | old = find_inode_fast(sb, head, ino); |
1086 | if (!old) { |
1087 | inode->i_ino = ino; |
1088 | spin_lock(&inode->i_lock); |
1089 | inode->i_state = I_NEW; |
1090 | hlist_add_head(&inode->i_hash, head); |
1091 | spin_unlock(&inode->i_lock); |
1092 | inode_sb_list_add(inode); |
1093 | spin_unlock(&inode_hash_lock); |
1094 | |
1095 | /* Return the locked inode with I_NEW set, the |
1096 | * caller is responsible for filling in the contents |
1097 | */ |
1098 | return inode; |
1099 | } |
1100 | |
1101 | /* |
1102 | * Uhhuh, somebody else created the same inode under |
1103 | * us. Use the old inode instead of the one we just |
1104 | * allocated. |
1105 | */ |
1106 | spin_unlock(&inode_hash_lock); |
1107 | destroy_inode(inode); |
1108 | inode = old; |
1109 | wait_on_inode(inode); |
1110 | } |
1111 | return inode; |
1112 | } |
1113 | EXPORT_SYMBOL(iget_locked); |
1114 | |
1115 | /* |
1116 | * search the inode cache for a matching inode number. |
1117 | * If we find one, then the inode number we are trying to |
1118 | * allocate is not unique and so we should not use it. |
1119 | * |
1120 | * Returns 1 if the inode number is unique, 0 if it is not. |
1121 | */ |
1122 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) |
1123 | { |
1124 | struct hlist_head *b = inode_hashtable + hash(sb, ino); |
1125 | struct hlist_node *node; |
1126 | struct inode *inode; |
1127 | |
1128 | spin_lock(&inode_hash_lock); |
1129 | hlist_for_each_entry(inode, node, b, i_hash) { |
1130 | if (inode->i_ino == ino && inode->i_sb == sb) { |
1131 | spin_unlock(&inode_hash_lock); |
1132 | return 0; |
1133 | } |
1134 | } |
1135 | spin_unlock(&inode_hash_lock); |
1136 | |
1137 | return 1; |
1138 | } |
1139 | |
1140 | /** |
1141 | * iunique - get a unique inode number |
1142 | * @sb: superblock |
1143 | * @max_reserved: highest reserved inode number |
1144 | * |
1145 | * Obtain an inode number that is unique on the system for a given |
1146 | * superblock. This is used by file systems that have no natural |
1147 | * permanent inode numbering system. An inode number is returned that |
1148 | * is higher than the reserved limit but unique. |
1149 | * |
1150 | * BUGS: |
1151 | * With a large number of inodes live on the file system this function |
1152 | * currently becomes quite slow. |
1153 | */ |
1154 | ino_t iunique(struct super_block *sb, ino_t max_reserved) |
1155 | { |
1156 | /* |
1157 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW |
1158 | * error if st_ino won't fit in target struct field. Use 32bit counter |
1159 | * here to attempt to avoid that. |
1160 | */ |
1161 | static DEFINE_SPINLOCK(iunique_lock); |
1162 | static unsigned int counter; |
1163 | ino_t res; |
1164 | |
1165 | spin_lock(&iunique_lock); |
1166 | do { |
1167 | if (counter <= max_reserved) |
1168 | counter = max_reserved + 1; |
1169 | res = counter++; |
1170 | } while (!test_inode_iunique(sb, res)); |
1171 | spin_unlock(&iunique_lock); |
1172 | |
1173 | return res; |
1174 | } |
1175 | EXPORT_SYMBOL(iunique); |
1176 | |
1177 | struct inode *igrab(struct inode *inode) |
1178 | { |
1179 | spin_lock(&inode->i_lock); |
1180 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { |
1181 | __iget(inode); |
1182 | spin_unlock(&inode->i_lock); |
1183 | } else { |
1184 | spin_unlock(&inode->i_lock); |
1185 | /* |
1186 | * Handle the case where s_op->clear_inode is not been |
1187 | * called yet, and somebody is calling igrab |
1188 | * while the inode is getting freed. |
1189 | */ |
1190 | inode = NULL; |
1191 | } |
1192 | return inode; |
1193 | } |
1194 | EXPORT_SYMBOL(igrab); |
1195 | |
1196 | /** |
1197 | * ilookup5_nowait - search for an inode in the inode cache |
1198 | * @sb: super block of file system to search |
1199 | * @hashval: hash value (usually inode number) to search for |
1200 | * @test: callback used for comparisons between inodes |
1201 | * @data: opaque data pointer to pass to @test |
1202 | * |
1203 | * Search for the inode specified by @hashval and @data in the inode cache. |
1204 | * If the inode is in the cache, the inode is returned with an incremented |
1205 | * reference count. |
1206 | * |
1207 | * Note: I_NEW is not waited upon so you have to be very careful what you do |
1208 | * with the returned inode. You probably should be using ilookup5() instead. |
1209 | * |
1210 | * Note2: @test is called with the inode_hash_lock held, so can't sleep. |
1211 | */ |
1212 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, |
1213 | int (*test)(struct inode *, void *), void *data) |
1214 | { |
1215 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1216 | struct inode *inode; |
1217 | |
1218 | spin_lock(&inode_hash_lock); |
1219 | inode = find_inode(sb, head, test, data); |
1220 | spin_unlock(&inode_hash_lock); |
1221 | |
1222 | return inode; |
1223 | } |
1224 | EXPORT_SYMBOL(ilookup5_nowait); |
1225 | |
1226 | /** |
1227 | * ilookup5 - search for an inode in the inode cache |
1228 | * @sb: super block of file system to search |
1229 | * @hashval: hash value (usually inode number) to search for |
1230 | * @test: callback used for comparisons between inodes |
1231 | * @data: opaque data pointer to pass to @test |
1232 | * |
1233 | * Search for the inode specified by @hashval and @data in the inode cache, |
1234 | * and if the inode is in the cache, return the inode with an incremented |
1235 | * reference count. Waits on I_NEW before returning the inode. |
1236 | * returned with an incremented reference count. |
1237 | * |
1238 | * This is a generalized version of ilookup() for file systems where the |
1239 | * inode number is not sufficient for unique identification of an inode. |
1240 | * |
1241 | * Note: @test is called with the inode_hash_lock held, so can't sleep. |
1242 | */ |
1243 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, |
1244 | int (*test)(struct inode *, void *), void *data) |
1245 | { |
1246 | struct inode *inode = ilookup5_nowait(sb, hashval, test, data); |
1247 | |
1248 | if (inode) |
1249 | wait_on_inode(inode); |
1250 | return inode; |
1251 | } |
1252 | EXPORT_SYMBOL(ilookup5); |
1253 | |
1254 | /** |
1255 | * ilookup - search for an inode in the inode cache |
1256 | * @sb: super block of file system to search |
1257 | * @ino: inode number to search for |
1258 | * |
1259 | * Search for the inode @ino in the inode cache, and if the inode is in the |
1260 | * cache, the inode is returned with an incremented reference count. |
1261 | */ |
1262 | struct inode *ilookup(struct super_block *sb, unsigned long ino) |
1263 | { |
1264 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
1265 | struct inode *inode; |
1266 | |
1267 | spin_lock(&inode_hash_lock); |
1268 | inode = find_inode_fast(sb, head, ino); |
1269 | spin_unlock(&inode_hash_lock); |
1270 | |
1271 | if (inode) |
1272 | wait_on_inode(inode); |
1273 | return inode; |
1274 | } |
1275 | EXPORT_SYMBOL(ilookup); |
1276 | |
1277 | int insert_inode_locked(struct inode *inode) |
1278 | { |
1279 | struct super_block *sb = inode->i_sb; |
1280 | ino_t ino = inode->i_ino; |
1281 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
1282 | |
1283 | while (1) { |
1284 | struct hlist_node *node; |
1285 | struct inode *old = NULL; |
1286 | spin_lock(&inode_hash_lock); |
1287 | hlist_for_each_entry(old, node, head, i_hash) { |
1288 | if (old->i_ino != ino) |
1289 | continue; |
1290 | if (old->i_sb != sb) |
1291 | continue; |
1292 | spin_lock(&old->i_lock); |
1293 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { |
1294 | spin_unlock(&old->i_lock); |
1295 | continue; |
1296 | } |
1297 | break; |
1298 | } |
1299 | if (likely(!node)) { |
1300 | spin_lock(&inode->i_lock); |
1301 | inode->i_state |= I_NEW; |
1302 | hlist_add_head(&inode->i_hash, head); |
1303 | spin_unlock(&inode->i_lock); |
1304 | spin_unlock(&inode_hash_lock); |
1305 | return 0; |
1306 | } |
1307 | __iget(old); |
1308 | spin_unlock(&old->i_lock); |
1309 | spin_unlock(&inode_hash_lock); |
1310 | wait_on_inode(old); |
1311 | if (unlikely(!inode_unhashed(old))) { |
1312 | iput(old); |
1313 | return -EBUSY; |
1314 | } |
1315 | iput(old); |
1316 | } |
1317 | } |
1318 | EXPORT_SYMBOL(insert_inode_locked); |
1319 | |
1320 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, |
1321 | int (*test)(struct inode *, void *), void *data) |
1322 | { |
1323 | struct super_block *sb = inode->i_sb; |
1324 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1325 | |
1326 | while (1) { |
1327 | struct hlist_node *node; |
1328 | struct inode *old = NULL; |
1329 | |
1330 | spin_lock(&inode_hash_lock); |
1331 | hlist_for_each_entry(old, node, head, i_hash) { |
1332 | if (old->i_sb != sb) |
1333 | continue; |
1334 | if (!test(old, data)) |
1335 | continue; |
1336 | spin_lock(&old->i_lock); |
1337 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { |
1338 | spin_unlock(&old->i_lock); |
1339 | continue; |
1340 | } |
1341 | break; |
1342 | } |
1343 | if (likely(!node)) { |
1344 | spin_lock(&inode->i_lock); |
1345 | inode->i_state |= I_NEW; |
1346 | hlist_add_head(&inode->i_hash, head); |
1347 | spin_unlock(&inode->i_lock); |
1348 | spin_unlock(&inode_hash_lock); |
1349 | return 0; |
1350 | } |
1351 | __iget(old); |
1352 | spin_unlock(&old->i_lock); |
1353 | spin_unlock(&inode_hash_lock); |
1354 | wait_on_inode(old); |
1355 | if (unlikely(!inode_unhashed(old))) { |
1356 | iput(old); |
1357 | return -EBUSY; |
1358 | } |
1359 | iput(old); |
1360 | } |
1361 | } |
1362 | EXPORT_SYMBOL(insert_inode_locked4); |
1363 | |
1364 | |
1365 | int generic_delete_inode(struct inode *inode) |
1366 | { |
1367 | return 1; |
1368 | } |
1369 | EXPORT_SYMBOL(generic_delete_inode); |
1370 | |
1371 | /* |
1372 | * Normal UNIX filesystem behaviour: delete the |
1373 | * inode when the usage count drops to zero, and |
1374 | * i_nlink is zero. |
1375 | */ |
1376 | int generic_drop_inode(struct inode *inode) |
1377 | { |
1378 | return !inode->i_nlink || inode_unhashed(inode); |
1379 | } |
1380 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1381 | |
1382 | /* |
1383 | * Called when we're dropping the last reference |
1384 | * to an inode. |
1385 | * |
1386 | * Call the FS "drop_inode()" function, defaulting to |
1387 | * the legacy UNIX filesystem behaviour. If it tells |
1388 | * us to evict inode, do so. Otherwise, retain inode |
1389 | * in cache if fs is alive, sync and evict if fs is |
1390 | * shutting down. |
1391 | */ |
1392 | static void iput_final(struct inode *inode) |
1393 | { |
1394 | struct super_block *sb = inode->i_sb; |
1395 | const struct super_operations *op = inode->i_sb->s_op; |
1396 | int drop; |
1397 | |
1398 | WARN_ON(inode->i_state & I_NEW); |
1399 | |
1400 | if (op->drop_inode) |
1401 | drop = op->drop_inode(inode); |
1402 | else |
1403 | drop = generic_drop_inode(inode); |
1404 | |
1405 | if (!drop && (sb->s_flags & MS_ACTIVE)) { |
1406 | inode->i_state |= I_REFERENCED; |
1407 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) |
1408 | inode_lru_list_add(inode); |
1409 | spin_unlock(&inode->i_lock); |
1410 | return; |
1411 | } |
1412 | |
1413 | if (!drop) { |
1414 | inode->i_state |= I_WILL_FREE; |
1415 | spin_unlock(&inode->i_lock); |
1416 | write_inode_now(inode, 1); |
1417 | spin_lock(&inode->i_lock); |
1418 | WARN_ON(inode->i_state & I_NEW); |
1419 | inode->i_state &= ~I_WILL_FREE; |
1420 | } |
1421 | |
1422 | inode->i_state |= I_FREEING; |
1423 | if (!list_empty(&inode->i_lru)) |
1424 | inode_lru_list_del(inode); |
1425 | spin_unlock(&inode->i_lock); |
1426 | |
1427 | evict(inode); |
1428 | } |
1429 | |
1430 | /** |
1431 | * iput - put an inode |
1432 | * @inode: inode to put |
1433 | * |
1434 | * Puts an inode, dropping its usage count. If the inode use count hits |
1435 | * zero, the inode is then freed and may also be destroyed. |
1436 | * |
1437 | * Consequently, iput() can sleep. |
1438 | */ |
1439 | void iput(struct inode *inode) |
1440 | { |
1441 | if (inode) { |
1442 | BUG_ON(inode->i_state & I_CLEAR); |
1443 | |
1444 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) |
1445 | iput_final(inode); |
1446 | } |
1447 | } |
1448 | EXPORT_SYMBOL(iput); |
1449 | |
1450 | /** |
1451 | * bmap - find a block number in a file |
1452 | * @inode: inode of file |
1453 | * @block: block to find |
1454 | * |
1455 | * Returns the block number on the device holding the inode that |
1456 | * is the disk block number for the block of the file requested. |
1457 | * That is, asked for block 4 of inode 1 the function will return the |
1458 | * disk block relative to the disk start that holds that block of the |
1459 | * file. |
1460 | */ |
1461 | sector_t bmap(struct inode *inode, sector_t block) |
1462 | { |
1463 | sector_t res = 0; |
1464 | if (inode->i_mapping->a_ops->bmap) |
1465 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); |
1466 | return res; |
1467 | } |
1468 | EXPORT_SYMBOL(bmap); |
1469 | |
1470 | /* |
1471 | * With relative atime, only update atime if the previous atime is |
1472 | * earlier than either the ctime or mtime or if at least a day has |
1473 | * passed since the last atime update. |
1474 | */ |
1475 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, |
1476 | struct timespec now) |
1477 | { |
1478 | |
1479 | if (!(mnt->mnt_flags & MNT_RELATIME)) |
1480 | return 1; |
1481 | /* |
1482 | * Is mtime younger than atime? If yes, update atime: |
1483 | */ |
1484 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) |
1485 | return 1; |
1486 | /* |
1487 | * Is ctime younger than atime? If yes, update atime: |
1488 | */ |
1489 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) |
1490 | return 1; |
1491 | |
1492 | /* |
1493 | * Is the previous atime value older than a day? If yes, |
1494 | * update atime: |
1495 | */ |
1496 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) |
1497 | return 1; |
1498 | /* |
1499 | * Good, we can skip the atime update: |
1500 | */ |
1501 | return 0; |
1502 | } |
1503 | |
1504 | /** |
1505 | * touch_atime - update the access time |
1506 | * @mnt: mount the inode is accessed on |
1507 | * @dentry: dentry accessed |
1508 | * |
1509 | * Update the accessed time on an inode and mark it for writeback. |
1510 | * This function automatically handles read only file systems and media, |
1511 | * as well as the "noatime" flag and inode specific "noatime" markers. |
1512 | */ |
1513 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1514 | { |
1515 | struct inode *inode = dentry->d_inode; |
1516 | struct timespec now; |
1517 | |
1518 | if (inode->i_flags & S_NOATIME) |
1519 | return; |
1520 | if (IS_NOATIME(inode)) |
1521 | return; |
1522 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
1523 | return; |
1524 | |
1525 | if (mnt->mnt_flags & MNT_NOATIME) |
1526 | return; |
1527 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
1528 | return; |
1529 | |
1530 | now = current_fs_time(inode->i_sb); |
1531 | |
1532 | if (!relatime_need_update(mnt, inode, now)) |
1533 | return; |
1534 | |
1535 | if (timespec_equal(&inode->i_atime, &now)) |
1536 | return; |
1537 | |
1538 | if (mnt_want_write(mnt)) |
1539 | return; |
1540 | |
1541 | inode->i_atime = now; |
1542 | mark_inode_dirty_sync(inode); |
1543 | mnt_drop_write(mnt); |
1544 | } |
1545 | EXPORT_SYMBOL(touch_atime); |
1546 | |
1547 | /** |
1548 | * file_update_time - update mtime and ctime time |
1549 | * @file: file accessed |
1550 | * |
1551 | * Update the mtime and ctime members of an inode and mark the inode |
1552 | * for writeback. Note that this function is meant exclusively for |
1553 | * usage in the file write path of filesystems, and filesystems may |
1554 | * choose to explicitly ignore update via this function with the |
1555 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
1556 | * timestamps are handled by the server. |
1557 | */ |
1558 | |
1559 | void file_update_time(struct file *file) |
1560 | { |
1561 | struct inode *inode = file->f_path.dentry->d_inode; |
1562 | struct timespec now; |
1563 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; |
1564 | |
1565 | /* First try to exhaust all avenues to not sync */ |
1566 | if (IS_NOCMTIME(inode)) |
1567 | return; |
1568 | |
1569 | now = current_fs_time(inode->i_sb); |
1570 | if (!timespec_equal(&inode->i_mtime, &now)) |
1571 | sync_it = S_MTIME; |
1572 | |
1573 | if (!timespec_equal(&inode->i_ctime, &now)) |
1574 | sync_it |= S_CTIME; |
1575 | |
1576 | if (IS_I_VERSION(inode)) |
1577 | sync_it |= S_VERSION; |
1578 | |
1579 | if (!sync_it) |
1580 | return; |
1581 | |
1582 | /* Finally allowed to write? Takes lock. */ |
1583 | if (mnt_want_write_file(file)) |
1584 | return; |
1585 | |
1586 | /* Only change inode inside the lock region */ |
1587 | if (sync_it & S_VERSION) |
1588 | inode_inc_iversion(inode); |
1589 | if (sync_it & S_CTIME) |
1590 | inode->i_ctime = now; |
1591 | if (sync_it & S_MTIME) |
1592 | inode->i_mtime = now; |
1593 | mark_inode_dirty_sync(inode); |
1594 | mnt_drop_write_file(file); |
1595 | } |
1596 | EXPORT_SYMBOL(file_update_time); |
1597 | |
1598 | int inode_needs_sync(struct inode *inode) |
1599 | { |
1600 | if (IS_SYNC(inode)) |
1601 | return 1; |
1602 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) |
1603 | return 1; |
1604 | return 0; |
1605 | } |
1606 | EXPORT_SYMBOL(inode_needs_sync); |
1607 | |
1608 | int inode_wait(void *word) |
1609 | { |
1610 | schedule(); |
1611 | return 0; |
1612 | } |
1613 | EXPORT_SYMBOL(inode_wait); |
1614 | |
1615 | /* |
1616 | * If we try to find an inode in the inode hash while it is being |
1617 | * deleted, we have to wait until the filesystem completes its |
1618 | * deletion before reporting that it isn't found. This function waits |
1619 | * until the deletion _might_ have completed. Callers are responsible |
1620 | * to recheck inode state. |
1621 | * |
1622 | * It doesn't matter if I_NEW is not set initially, a call to |
1623 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list |
1624 | * will DTRT. |
1625 | */ |
1626 | static void __wait_on_freeing_inode(struct inode *inode) |
1627 | { |
1628 | wait_queue_head_t *wq; |
1629 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1630 | wq = bit_waitqueue(&inode->i_state, __I_NEW); |
1631 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
1632 | spin_unlock(&inode->i_lock); |
1633 | spin_unlock(&inode_hash_lock); |
1634 | schedule(); |
1635 | finish_wait(wq, &wait.wait); |
1636 | spin_lock(&inode_hash_lock); |
1637 | } |
1638 | |
1639 | static __initdata unsigned long ihash_entries; |
1640 | static int __init set_ihash_entries(char *str) |
1641 | { |
1642 | if (!str) |
1643 | return 0; |
1644 | ihash_entries = simple_strtoul(str, &str, 0); |
1645 | return 1; |
1646 | } |
1647 | __setup("ihash_entries=", set_ihash_entries); |
1648 | |
1649 | /* |
1650 | * Initialize the waitqueues and inode hash table. |
1651 | */ |
1652 | void __init inode_init_early(void) |
1653 | { |
1654 | unsigned int loop; |
1655 | |
1656 | /* If hashes are distributed across NUMA nodes, defer |
1657 | * hash allocation until vmalloc space is available. |
1658 | */ |
1659 | if (hashdist) |
1660 | return; |
1661 | |
1662 | inode_hashtable = |
1663 | alloc_large_system_hash("Inode-cache", |
1664 | sizeof(struct hlist_head), |
1665 | ihash_entries, |
1666 | 14, |
1667 | HASH_EARLY, |
1668 | &i_hash_shift, |
1669 | &i_hash_mask, |
1670 | 0); |
1671 | |
1672 | for (loop = 0; loop < (1U << i_hash_shift); loop++) |
1673 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1674 | } |
1675 | |
1676 | void __init inode_init(void) |
1677 | { |
1678 | unsigned int loop; |
1679 | |
1680 | /* inode slab cache */ |
1681 | inode_cachep = kmem_cache_create("inode_cache", |
1682 | sizeof(struct inode), |
1683 | 0, |
1684 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| |
1685 | SLAB_MEM_SPREAD), |
1686 | init_once); |
1687 | |
1688 | /* Hash may have been set up in inode_init_early */ |
1689 | if (!hashdist) |
1690 | return; |
1691 | |
1692 | inode_hashtable = |
1693 | alloc_large_system_hash("Inode-cache", |
1694 | sizeof(struct hlist_head), |
1695 | ihash_entries, |
1696 | 14, |
1697 | 0, |
1698 | &i_hash_shift, |
1699 | &i_hash_mask, |
1700 | 0); |
1701 | |
1702 | for (loop = 0; loop < (1U << i_hash_shift); loop++) |
1703 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1704 | } |
1705 | |
1706 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) |
1707 | { |
1708 | inode->i_mode = mode; |
1709 | if (S_ISCHR(mode)) { |
1710 | inode->i_fop = &def_chr_fops; |
1711 | inode->i_rdev = rdev; |
1712 | } else if (S_ISBLK(mode)) { |
1713 | inode->i_fop = &def_blk_fops; |
1714 | inode->i_rdev = rdev; |
1715 | } else if (S_ISFIFO(mode)) |
1716 | inode->i_fop = &def_fifo_fops; |
1717 | else if (S_ISSOCK(mode)) |
1718 | inode->i_fop = &bad_sock_fops; |
1719 | else |
1720 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1721 | " inode %s:%lu\n", mode, inode->i_sb->s_id, |
1722 | inode->i_ino); |
1723 | } |
1724 | EXPORT_SYMBOL(init_special_inode); |
1725 | |
1726 | /** |
1727 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards |
1728 | * @inode: New inode |
1729 | * @dir: Directory inode |
1730 | * @mode: mode of the new inode |
1731 | */ |
1732 | void inode_init_owner(struct inode *inode, const struct inode *dir, |
1733 | umode_t mode) |
1734 | { |
1735 | inode->i_uid = current_fsuid(); |
1736 | if (dir && dir->i_mode & S_ISGID) { |
1737 | inode->i_gid = dir->i_gid; |
1738 | if (S_ISDIR(mode)) |
1739 | mode |= S_ISGID; |
1740 | } else |
1741 | inode->i_gid = current_fsgid(); |
1742 | inode->i_mode = mode; |
1743 | } |
1744 | EXPORT_SYMBOL(inode_init_owner); |
1745 | |
1746 | /** |
1747 | * inode_owner_or_capable - check current task permissions to inode |
1748 | * @inode: inode being checked |
1749 | * |
1750 | * Return true if current either has CAP_FOWNER to the inode, or |
1751 | * owns the file. |
1752 | */ |
1753 | bool inode_owner_or_capable(const struct inode *inode) |
1754 | { |
1755 | struct user_namespace *ns = inode_userns(inode); |
1756 | |
1757 | if (current_user_ns() == ns && current_fsuid() == inode->i_uid) |
1758 | return true; |
1759 | if (ns_capable(ns, CAP_FOWNER)) |
1760 | return true; |
1761 | return false; |
1762 | } |
1763 | EXPORT_SYMBOL(inode_owner_or_capable); |
1764 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9