Root/
Source at commit be977234bfb4a6dca8a39e7c52165e4cd536ad71 created 12 years 9 months ago. By Lars-Peter Clausen, jz4740: Fix compile error | |
---|---|
1 | /* |
2 | * linux/fs/inode.c |
3 | * |
4 | * (C) 1997 Linus Torvalds |
5 | */ |
6 | |
7 | #include <linux/fs.h> |
8 | #include <linux/mm.h> |
9 | #include <linux/dcache.h> |
10 | #include <linux/init.h> |
11 | #include <linux/slab.h> |
12 | #include <linux/writeback.h> |
13 | #include <linux/module.h> |
14 | #include <linux/backing-dev.h> |
15 | #include <linux/wait.h> |
16 | #include <linux/rwsem.h> |
17 | #include <linux/hash.h> |
18 | #include <linux/swap.h> |
19 | #include <linux/security.h> |
20 | #include <linux/pagemap.h> |
21 | #include <linux/cdev.h> |
22 | #include <linux/bootmem.h> |
23 | #include <linux/fsnotify.h> |
24 | #include <linux/mount.h> |
25 | #include <linux/async.h> |
26 | #include <linux/posix_acl.h> |
27 | #include <linux/ima.h> |
28 | #include <linux/cred.h> |
29 | #include "internal.h" |
30 | |
31 | /* |
32 | * inode locking rules. |
33 | * |
34 | * inode->i_lock protects: |
35 | * inode->i_state, inode->i_hash, __iget() |
36 | * inode_lru_lock protects: |
37 | * inode_lru, inode->i_lru |
38 | * inode_sb_list_lock protects: |
39 | * sb->s_inodes, inode->i_sb_list |
40 | * inode_wb_list_lock protects: |
41 | * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list |
42 | * inode_hash_lock protects: |
43 | * inode_hashtable, inode->i_hash |
44 | * |
45 | * Lock ordering: |
46 | * |
47 | * inode_sb_list_lock |
48 | * inode->i_lock |
49 | * inode_lru_lock |
50 | * |
51 | * inode_wb_list_lock |
52 | * inode->i_lock |
53 | * |
54 | * inode_hash_lock |
55 | * inode_sb_list_lock |
56 | * inode->i_lock |
57 | * |
58 | * iunique_lock |
59 | * inode_hash_lock |
60 | */ |
61 | |
62 | /* |
63 | * This is needed for the following functions: |
64 | * - inode_has_buffers |
65 | * - invalidate_bdev |
66 | * |
67 | * FIXME: remove all knowledge of the buffer layer from this file |
68 | */ |
69 | #include <linux/buffer_head.h> |
70 | |
71 | /* |
72 | * New inode.c implementation. |
73 | * |
74 | * This implementation has the basic premise of trying |
75 | * to be extremely low-overhead and SMP-safe, yet be |
76 | * simple enough to be "obviously correct". |
77 | * |
78 | * Famous last words. |
79 | */ |
80 | |
81 | /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */ |
82 | |
83 | /* #define INODE_PARANOIA 1 */ |
84 | /* #define INODE_DEBUG 1 */ |
85 | |
86 | /* |
87 | * Inode lookup is no longer as critical as it used to be: |
88 | * most of the lookups are going to be through the dcache. |
89 | */ |
90 | #define I_HASHBITS i_hash_shift |
91 | #define I_HASHMASK i_hash_mask |
92 | |
93 | static unsigned int i_hash_mask __read_mostly; |
94 | static unsigned int i_hash_shift __read_mostly; |
95 | static struct hlist_head *inode_hashtable __read_mostly; |
96 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock); |
97 | |
98 | /* |
99 | * Each inode can be on two separate lists. One is |
100 | * the hash list of the inode, used for lookups. The |
101 | * other linked list is the "type" list: |
102 | * "in_use" - valid inode, i_count > 0, i_nlink > 0 |
103 | * "dirty" - as "in_use" but also dirty |
104 | * "unused" - valid inode, i_count = 0 |
105 | * |
106 | * A "dirty" list is maintained for each super block, |
107 | * allowing for low-overhead inode sync() operations. |
108 | */ |
109 | |
110 | static LIST_HEAD(inode_lru); |
111 | static DEFINE_SPINLOCK(inode_lru_lock); |
112 | |
113 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock); |
114 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_wb_list_lock); |
115 | |
116 | /* |
117 | * iprune_sem provides exclusion between the icache shrinking and the |
118 | * umount path. |
119 | * |
120 | * We don't actually need it to protect anything in the umount path, |
121 | * but only need to cycle through it to make sure any inode that |
122 | * prune_icache took off the LRU list has been fully torn down by the |
123 | * time we are past evict_inodes. |
124 | */ |
125 | static DECLARE_RWSEM(iprune_sem); |
126 | |
127 | /* |
128 | * Empty aops. Can be used for the cases where the user does not |
129 | * define any of the address_space operations. |
130 | */ |
131 | const struct address_space_operations empty_aops = { |
132 | }; |
133 | EXPORT_SYMBOL(empty_aops); |
134 | |
135 | /* |
136 | * Statistics gathering.. |
137 | */ |
138 | struct inodes_stat_t inodes_stat; |
139 | |
140 | static DEFINE_PER_CPU(unsigned int, nr_inodes); |
141 | |
142 | static struct kmem_cache *inode_cachep __read_mostly; |
143 | |
144 | static int get_nr_inodes(void) |
145 | { |
146 | int i; |
147 | int sum = 0; |
148 | for_each_possible_cpu(i) |
149 | sum += per_cpu(nr_inodes, i); |
150 | return sum < 0 ? 0 : sum; |
151 | } |
152 | |
153 | static inline int get_nr_inodes_unused(void) |
154 | { |
155 | return inodes_stat.nr_unused; |
156 | } |
157 | |
158 | int get_nr_dirty_inodes(void) |
159 | { |
160 | /* not actually dirty inodes, but a wild approximation */ |
161 | int nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); |
162 | return nr_dirty > 0 ? nr_dirty : 0; |
163 | } |
164 | |
165 | /* |
166 | * Handle nr_inode sysctl |
167 | */ |
168 | #ifdef CONFIG_SYSCTL |
169 | int proc_nr_inodes(ctl_table *table, int write, |
170 | void __user *buffer, size_t *lenp, loff_t *ppos) |
171 | { |
172 | inodes_stat.nr_inodes = get_nr_inodes(); |
173 | return proc_dointvec(table, write, buffer, lenp, ppos); |
174 | } |
175 | #endif |
176 | |
177 | /** |
178 | * inode_init_always - perform inode structure intialisation |
179 | * @sb: superblock inode belongs to |
180 | * @inode: inode to initialise |
181 | * |
182 | * These are initializations that need to be done on every inode |
183 | * allocation as the fields are not initialised by slab allocation. |
184 | */ |
185 | int inode_init_always(struct super_block *sb, struct inode *inode) |
186 | { |
187 | static const struct inode_operations empty_iops; |
188 | static const struct file_operations empty_fops; |
189 | struct address_space *const mapping = &inode->i_data; |
190 | |
191 | inode->i_sb = sb; |
192 | inode->i_blkbits = sb->s_blocksize_bits; |
193 | inode->i_flags = 0; |
194 | atomic_set(&inode->i_count, 1); |
195 | inode->i_op = &empty_iops; |
196 | inode->i_fop = &empty_fops; |
197 | inode->i_nlink = 1; |
198 | inode->i_uid = 0; |
199 | inode->i_gid = 0; |
200 | atomic_set(&inode->i_writecount, 0); |
201 | inode->i_size = 0; |
202 | inode->i_blocks = 0; |
203 | inode->i_bytes = 0; |
204 | inode->i_generation = 0; |
205 | #ifdef CONFIG_QUOTA |
206 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
207 | #endif |
208 | inode->i_pipe = NULL; |
209 | inode->i_bdev = NULL; |
210 | inode->i_cdev = NULL; |
211 | inode->i_rdev = 0; |
212 | inode->dirtied_when = 0; |
213 | |
214 | if (security_inode_alloc(inode)) |
215 | goto out; |
216 | spin_lock_init(&inode->i_lock); |
217 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); |
218 | |
219 | mutex_init(&inode->i_mutex); |
220 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); |
221 | |
222 | init_rwsem(&inode->i_alloc_sem); |
223 | lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key); |
224 | |
225 | mapping->a_ops = &empty_aops; |
226 | mapping->host = inode; |
227 | mapping->flags = 0; |
228 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
229 | mapping->assoc_mapping = NULL; |
230 | mapping->backing_dev_info = &default_backing_dev_info; |
231 | mapping->writeback_index = 0; |
232 | |
233 | /* |
234 | * If the block_device provides a backing_dev_info for client |
235 | * inodes then use that. Otherwise the inode share the bdev's |
236 | * backing_dev_info. |
237 | */ |
238 | if (sb->s_bdev) { |
239 | struct backing_dev_info *bdi; |
240 | |
241 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; |
242 | mapping->backing_dev_info = bdi; |
243 | } |
244 | inode->i_private = NULL; |
245 | inode->i_mapping = mapping; |
246 | #ifdef CONFIG_FS_POSIX_ACL |
247 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; |
248 | #endif |
249 | |
250 | #ifdef CONFIG_FSNOTIFY |
251 | inode->i_fsnotify_mask = 0; |
252 | #endif |
253 | |
254 | this_cpu_inc(nr_inodes); |
255 | |
256 | return 0; |
257 | out: |
258 | return -ENOMEM; |
259 | } |
260 | EXPORT_SYMBOL(inode_init_always); |
261 | |
262 | static struct inode *alloc_inode(struct super_block *sb) |
263 | { |
264 | struct inode *inode; |
265 | |
266 | if (sb->s_op->alloc_inode) |
267 | inode = sb->s_op->alloc_inode(sb); |
268 | else |
269 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); |
270 | |
271 | if (!inode) |
272 | return NULL; |
273 | |
274 | if (unlikely(inode_init_always(sb, inode))) { |
275 | if (inode->i_sb->s_op->destroy_inode) |
276 | inode->i_sb->s_op->destroy_inode(inode); |
277 | else |
278 | kmem_cache_free(inode_cachep, inode); |
279 | return NULL; |
280 | } |
281 | |
282 | return inode; |
283 | } |
284 | |
285 | void free_inode_nonrcu(struct inode *inode) |
286 | { |
287 | kmem_cache_free(inode_cachep, inode); |
288 | } |
289 | EXPORT_SYMBOL(free_inode_nonrcu); |
290 | |
291 | void __destroy_inode(struct inode *inode) |
292 | { |
293 | BUG_ON(inode_has_buffers(inode)); |
294 | security_inode_free(inode); |
295 | fsnotify_inode_delete(inode); |
296 | #ifdef CONFIG_FS_POSIX_ACL |
297 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) |
298 | posix_acl_release(inode->i_acl); |
299 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) |
300 | posix_acl_release(inode->i_default_acl); |
301 | #endif |
302 | this_cpu_dec(nr_inodes); |
303 | } |
304 | EXPORT_SYMBOL(__destroy_inode); |
305 | |
306 | static void i_callback(struct rcu_head *head) |
307 | { |
308 | struct inode *inode = container_of(head, struct inode, i_rcu); |
309 | INIT_LIST_HEAD(&inode->i_dentry); |
310 | kmem_cache_free(inode_cachep, inode); |
311 | } |
312 | |
313 | static void destroy_inode(struct inode *inode) |
314 | { |
315 | BUG_ON(!list_empty(&inode->i_lru)); |
316 | __destroy_inode(inode); |
317 | if (inode->i_sb->s_op->destroy_inode) |
318 | inode->i_sb->s_op->destroy_inode(inode); |
319 | else |
320 | call_rcu(&inode->i_rcu, i_callback); |
321 | } |
322 | |
323 | void address_space_init_once(struct address_space *mapping) |
324 | { |
325 | memset(mapping, 0, sizeof(*mapping)); |
326 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC); |
327 | spin_lock_init(&mapping->tree_lock); |
328 | spin_lock_init(&mapping->i_mmap_lock); |
329 | INIT_LIST_HEAD(&mapping->private_list); |
330 | spin_lock_init(&mapping->private_lock); |
331 | INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap); |
332 | INIT_LIST_HEAD(&mapping->i_mmap_nonlinear); |
333 | mutex_init(&mapping->unmap_mutex); |
334 | } |
335 | EXPORT_SYMBOL(address_space_init_once); |
336 | |
337 | /* |
338 | * These are initializations that only need to be done |
339 | * once, because the fields are idempotent across use |
340 | * of the inode, so let the slab aware of that. |
341 | */ |
342 | void inode_init_once(struct inode *inode) |
343 | { |
344 | memset(inode, 0, sizeof(*inode)); |
345 | INIT_HLIST_NODE(&inode->i_hash); |
346 | INIT_LIST_HEAD(&inode->i_dentry); |
347 | INIT_LIST_HEAD(&inode->i_devices); |
348 | INIT_LIST_HEAD(&inode->i_wb_list); |
349 | INIT_LIST_HEAD(&inode->i_lru); |
350 | address_space_init_once(&inode->i_data); |
351 | i_size_ordered_init(inode); |
352 | #ifdef CONFIG_FSNOTIFY |
353 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); |
354 | #endif |
355 | } |
356 | EXPORT_SYMBOL(inode_init_once); |
357 | |
358 | static void init_once(void *foo) |
359 | { |
360 | struct inode *inode = (struct inode *) foo; |
361 | |
362 | inode_init_once(inode); |
363 | } |
364 | |
365 | /* |
366 | * inode->i_lock must be held |
367 | */ |
368 | void __iget(struct inode *inode) |
369 | { |
370 | atomic_inc(&inode->i_count); |
371 | } |
372 | |
373 | /* |
374 | * get additional reference to inode; caller must already hold one. |
375 | */ |
376 | void ihold(struct inode *inode) |
377 | { |
378 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); |
379 | } |
380 | EXPORT_SYMBOL(ihold); |
381 | |
382 | static void inode_lru_list_add(struct inode *inode) |
383 | { |
384 | spin_lock(&inode_lru_lock); |
385 | if (list_empty(&inode->i_lru)) { |
386 | list_add(&inode->i_lru, &inode_lru); |
387 | inodes_stat.nr_unused++; |
388 | } |
389 | spin_unlock(&inode_lru_lock); |
390 | } |
391 | |
392 | static void inode_lru_list_del(struct inode *inode) |
393 | { |
394 | spin_lock(&inode_lru_lock); |
395 | if (!list_empty(&inode->i_lru)) { |
396 | list_del_init(&inode->i_lru); |
397 | inodes_stat.nr_unused--; |
398 | } |
399 | spin_unlock(&inode_lru_lock); |
400 | } |
401 | |
402 | /** |
403 | * inode_sb_list_add - add inode to the superblock list of inodes |
404 | * @inode: inode to add |
405 | */ |
406 | void inode_sb_list_add(struct inode *inode) |
407 | { |
408 | spin_lock(&inode_sb_list_lock); |
409 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); |
410 | spin_unlock(&inode_sb_list_lock); |
411 | } |
412 | EXPORT_SYMBOL_GPL(inode_sb_list_add); |
413 | |
414 | static inline void inode_sb_list_del(struct inode *inode) |
415 | { |
416 | spin_lock(&inode_sb_list_lock); |
417 | list_del_init(&inode->i_sb_list); |
418 | spin_unlock(&inode_sb_list_lock); |
419 | } |
420 | |
421 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
422 | { |
423 | unsigned long tmp; |
424 | |
425 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / |
426 | L1_CACHE_BYTES; |
427 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS); |
428 | return tmp & I_HASHMASK; |
429 | } |
430 | |
431 | /** |
432 | * __insert_inode_hash - hash an inode |
433 | * @inode: unhashed inode |
434 | * @hashval: unsigned long value used to locate this object in the |
435 | * inode_hashtable. |
436 | * |
437 | * Add an inode to the inode hash for this superblock. |
438 | */ |
439 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) |
440 | { |
441 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); |
442 | |
443 | spin_lock(&inode_hash_lock); |
444 | spin_lock(&inode->i_lock); |
445 | hlist_add_head(&inode->i_hash, b); |
446 | spin_unlock(&inode->i_lock); |
447 | spin_unlock(&inode_hash_lock); |
448 | } |
449 | EXPORT_SYMBOL(__insert_inode_hash); |
450 | |
451 | /** |
452 | * remove_inode_hash - remove an inode from the hash |
453 | * @inode: inode to unhash |
454 | * |
455 | * Remove an inode from the superblock. |
456 | */ |
457 | void remove_inode_hash(struct inode *inode) |
458 | { |
459 | spin_lock(&inode_hash_lock); |
460 | spin_lock(&inode->i_lock); |
461 | hlist_del_init(&inode->i_hash); |
462 | spin_unlock(&inode->i_lock); |
463 | spin_unlock(&inode_hash_lock); |
464 | } |
465 | EXPORT_SYMBOL(remove_inode_hash); |
466 | |
467 | void end_writeback(struct inode *inode) |
468 | { |
469 | might_sleep(); |
470 | BUG_ON(inode->i_data.nrpages); |
471 | BUG_ON(!list_empty(&inode->i_data.private_list)); |
472 | BUG_ON(!(inode->i_state & I_FREEING)); |
473 | BUG_ON(inode->i_state & I_CLEAR); |
474 | inode_sync_wait(inode); |
475 | /* don't need i_lock here, no concurrent mods to i_state */ |
476 | inode->i_state = I_FREEING | I_CLEAR; |
477 | } |
478 | EXPORT_SYMBOL(end_writeback); |
479 | |
480 | /* |
481 | * Free the inode passed in, removing it from the lists it is still connected |
482 | * to. We remove any pages still attached to the inode and wait for any IO that |
483 | * is still in progress before finally destroying the inode. |
484 | * |
485 | * An inode must already be marked I_FREEING so that we avoid the inode being |
486 | * moved back onto lists if we race with other code that manipulates the lists |
487 | * (e.g. writeback_single_inode). The caller is responsible for setting this. |
488 | * |
489 | * An inode must already be removed from the LRU list before being evicted from |
490 | * the cache. This should occur atomically with setting the I_FREEING state |
491 | * flag, so no inodes here should ever be on the LRU when being evicted. |
492 | */ |
493 | static void evict(struct inode *inode) |
494 | { |
495 | const struct super_operations *op = inode->i_sb->s_op; |
496 | |
497 | BUG_ON(!(inode->i_state & I_FREEING)); |
498 | BUG_ON(!list_empty(&inode->i_lru)); |
499 | |
500 | inode_wb_list_del(inode); |
501 | inode_sb_list_del(inode); |
502 | |
503 | if (op->evict_inode) { |
504 | op->evict_inode(inode); |
505 | } else { |
506 | if (inode->i_data.nrpages) |
507 | truncate_inode_pages(&inode->i_data, 0); |
508 | end_writeback(inode); |
509 | } |
510 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
511 | bd_forget(inode); |
512 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) |
513 | cd_forget(inode); |
514 | |
515 | remove_inode_hash(inode); |
516 | |
517 | spin_lock(&inode->i_lock); |
518 | wake_up_bit(&inode->i_state, __I_NEW); |
519 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); |
520 | spin_unlock(&inode->i_lock); |
521 | |
522 | destroy_inode(inode); |
523 | } |
524 | |
525 | /* |
526 | * dispose_list - dispose of the contents of a local list |
527 | * @head: the head of the list to free |
528 | * |
529 | * Dispose-list gets a local list with local inodes in it, so it doesn't |
530 | * need to worry about list corruption and SMP locks. |
531 | */ |
532 | static void dispose_list(struct list_head *head) |
533 | { |
534 | while (!list_empty(head)) { |
535 | struct inode *inode; |
536 | |
537 | inode = list_first_entry(head, struct inode, i_lru); |
538 | list_del_init(&inode->i_lru); |
539 | |
540 | evict(inode); |
541 | } |
542 | } |
543 | |
544 | /** |
545 | * evict_inodes - evict all evictable inodes for a superblock |
546 | * @sb: superblock to operate on |
547 | * |
548 | * Make sure that no inodes with zero refcount are retained. This is |
549 | * called by superblock shutdown after having MS_ACTIVE flag removed, |
550 | * so any inode reaching zero refcount during or after that call will |
551 | * be immediately evicted. |
552 | */ |
553 | void evict_inodes(struct super_block *sb) |
554 | { |
555 | struct inode *inode, *next; |
556 | LIST_HEAD(dispose); |
557 | |
558 | spin_lock(&inode_sb_list_lock); |
559 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
560 | if (atomic_read(&inode->i_count)) |
561 | continue; |
562 | |
563 | spin_lock(&inode->i_lock); |
564 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
565 | spin_unlock(&inode->i_lock); |
566 | continue; |
567 | } |
568 | |
569 | inode->i_state |= I_FREEING; |
570 | inode_lru_list_del(inode); |
571 | spin_unlock(&inode->i_lock); |
572 | list_add(&inode->i_lru, &dispose); |
573 | } |
574 | spin_unlock(&inode_sb_list_lock); |
575 | |
576 | dispose_list(&dispose); |
577 | |
578 | /* |
579 | * Cycle through iprune_sem to make sure any inode that prune_icache |
580 | * moved off the list before we took the lock has been fully torn |
581 | * down. |
582 | */ |
583 | down_write(&iprune_sem); |
584 | up_write(&iprune_sem); |
585 | } |
586 | |
587 | /** |
588 | * invalidate_inodes - attempt to free all inodes on a superblock |
589 | * @sb: superblock to operate on |
590 | * @kill_dirty: flag to guide handling of dirty inodes |
591 | * |
592 | * Attempts to free all inodes for a given superblock. If there were any |
593 | * busy inodes return a non-zero value, else zero. |
594 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat |
595 | * them as busy. |
596 | */ |
597 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) |
598 | { |
599 | int busy = 0; |
600 | struct inode *inode, *next; |
601 | LIST_HEAD(dispose); |
602 | |
603 | spin_lock(&inode_sb_list_lock); |
604 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
605 | spin_lock(&inode->i_lock); |
606 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
607 | spin_unlock(&inode->i_lock); |
608 | continue; |
609 | } |
610 | if (inode->i_state & I_DIRTY && !kill_dirty) { |
611 | spin_unlock(&inode->i_lock); |
612 | busy = 1; |
613 | continue; |
614 | } |
615 | if (atomic_read(&inode->i_count)) { |
616 | spin_unlock(&inode->i_lock); |
617 | busy = 1; |
618 | continue; |
619 | } |
620 | |
621 | inode->i_state |= I_FREEING; |
622 | inode_lru_list_del(inode); |
623 | spin_unlock(&inode->i_lock); |
624 | list_add(&inode->i_lru, &dispose); |
625 | } |
626 | spin_unlock(&inode_sb_list_lock); |
627 | |
628 | dispose_list(&dispose); |
629 | |
630 | return busy; |
631 | } |
632 | |
633 | static int can_unuse(struct inode *inode) |
634 | { |
635 | if (inode->i_state & ~I_REFERENCED) |
636 | return 0; |
637 | if (inode_has_buffers(inode)) |
638 | return 0; |
639 | if (atomic_read(&inode->i_count)) |
640 | return 0; |
641 | if (inode->i_data.nrpages) |
642 | return 0; |
643 | return 1; |
644 | } |
645 | |
646 | /* |
647 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to a |
648 | * temporary list and then are freed outside inode_lru_lock by dispose_list(). |
649 | * |
650 | * Any inodes which are pinned purely because of attached pagecache have their |
651 | * pagecache removed. If the inode has metadata buffers attached to |
652 | * mapping->private_list then try to remove them. |
653 | * |
654 | * If the inode has the I_REFERENCED flag set, then it means that it has been |
655 | * used recently - the flag is set in iput_final(). When we encounter such an |
656 | * inode, clear the flag and move it to the back of the LRU so it gets another |
657 | * pass through the LRU before it gets reclaimed. This is necessary because of |
658 | * the fact we are doing lazy LRU updates to minimise lock contention so the |
659 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes |
660 | * with this flag set because they are the inodes that are out of order. |
661 | */ |
662 | static void prune_icache(int nr_to_scan) |
663 | { |
664 | LIST_HEAD(freeable); |
665 | int nr_scanned; |
666 | unsigned long reap = 0; |
667 | |
668 | down_read(&iprune_sem); |
669 | spin_lock(&inode_lru_lock); |
670 | for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) { |
671 | struct inode *inode; |
672 | |
673 | if (list_empty(&inode_lru)) |
674 | break; |
675 | |
676 | inode = list_entry(inode_lru.prev, struct inode, i_lru); |
677 | |
678 | /* |
679 | * we are inverting the inode_lru_lock/inode->i_lock here, |
680 | * so use a trylock. If we fail to get the lock, just move the |
681 | * inode to the back of the list so we don't spin on it. |
682 | */ |
683 | if (!spin_trylock(&inode->i_lock)) { |
684 | list_move(&inode->i_lru, &inode_lru); |
685 | continue; |
686 | } |
687 | |
688 | /* |
689 | * Referenced or dirty inodes are still in use. Give them |
690 | * another pass through the LRU as we canot reclaim them now. |
691 | */ |
692 | if (atomic_read(&inode->i_count) || |
693 | (inode->i_state & ~I_REFERENCED)) { |
694 | list_del_init(&inode->i_lru); |
695 | spin_unlock(&inode->i_lock); |
696 | inodes_stat.nr_unused--; |
697 | continue; |
698 | } |
699 | |
700 | /* recently referenced inodes get one more pass */ |
701 | if (inode->i_state & I_REFERENCED) { |
702 | inode->i_state &= ~I_REFERENCED; |
703 | list_move(&inode->i_lru, &inode_lru); |
704 | spin_unlock(&inode->i_lock); |
705 | continue; |
706 | } |
707 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { |
708 | __iget(inode); |
709 | spin_unlock(&inode->i_lock); |
710 | spin_unlock(&inode_lru_lock); |
711 | if (remove_inode_buffers(inode)) |
712 | reap += invalidate_mapping_pages(&inode->i_data, |
713 | 0, -1); |
714 | iput(inode); |
715 | spin_lock(&inode_lru_lock); |
716 | |
717 | if (inode != list_entry(inode_lru.next, |
718 | struct inode, i_lru)) |
719 | continue; /* wrong inode or list_empty */ |
720 | /* avoid lock inversions with trylock */ |
721 | if (!spin_trylock(&inode->i_lock)) |
722 | continue; |
723 | if (!can_unuse(inode)) { |
724 | spin_unlock(&inode->i_lock); |
725 | continue; |
726 | } |
727 | } |
728 | WARN_ON(inode->i_state & I_NEW); |
729 | inode->i_state |= I_FREEING; |
730 | spin_unlock(&inode->i_lock); |
731 | |
732 | list_move(&inode->i_lru, &freeable); |
733 | inodes_stat.nr_unused--; |
734 | } |
735 | if (current_is_kswapd()) |
736 | __count_vm_events(KSWAPD_INODESTEAL, reap); |
737 | else |
738 | __count_vm_events(PGINODESTEAL, reap); |
739 | spin_unlock(&inode_lru_lock); |
740 | |
741 | dispose_list(&freeable); |
742 | up_read(&iprune_sem); |
743 | } |
744 | |
745 | /* |
746 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, |
747 | * "unused" means that no dentries are referring to the inodes: the files are |
748 | * not open and the dcache references to those inodes have already been |
749 | * reclaimed. |
750 | * |
751 | * This function is passed the number of inodes to scan, and it returns the |
752 | * total number of remaining possibly-reclaimable inodes. |
753 | */ |
754 | static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
755 | { |
756 | if (nr) { |
757 | /* |
758 | * Nasty deadlock avoidance. We may hold various FS locks, |
759 | * and we don't want to recurse into the FS that called us |
760 | * in clear_inode() and friends.. |
761 | */ |
762 | if (!(gfp_mask & __GFP_FS)) |
763 | return -1; |
764 | prune_icache(nr); |
765 | } |
766 | return (get_nr_inodes_unused() / 100) * sysctl_vfs_cache_pressure; |
767 | } |
768 | |
769 | static struct shrinker icache_shrinker = { |
770 | .shrink = shrink_icache_memory, |
771 | .seeks = DEFAULT_SEEKS, |
772 | }; |
773 | |
774 | static void __wait_on_freeing_inode(struct inode *inode); |
775 | /* |
776 | * Called with the inode lock held. |
777 | */ |
778 | static struct inode *find_inode(struct super_block *sb, |
779 | struct hlist_head *head, |
780 | int (*test)(struct inode *, void *), |
781 | void *data) |
782 | { |
783 | struct hlist_node *node; |
784 | struct inode *inode = NULL; |
785 | |
786 | repeat: |
787 | hlist_for_each_entry(inode, node, head, i_hash) { |
788 | spin_lock(&inode->i_lock); |
789 | if (inode->i_sb != sb) { |
790 | spin_unlock(&inode->i_lock); |
791 | continue; |
792 | } |
793 | if (!test(inode, data)) { |
794 | spin_unlock(&inode->i_lock); |
795 | continue; |
796 | } |
797 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
798 | __wait_on_freeing_inode(inode); |
799 | goto repeat; |
800 | } |
801 | __iget(inode); |
802 | spin_unlock(&inode->i_lock); |
803 | return inode; |
804 | } |
805 | return NULL; |
806 | } |
807 | |
808 | /* |
809 | * find_inode_fast is the fast path version of find_inode, see the comment at |
810 | * iget_locked for details. |
811 | */ |
812 | static struct inode *find_inode_fast(struct super_block *sb, |
813 | struct hlist_head *head, unsigned long ino) |
814 | { |
815 | struct hlist_node *node; |
816 | struct inode *inode = NULL; |
817 | |
818 | repeat: |
819 | hlist_for_each_entry(inode, node, head, i_hash) { |
820 | spin_lock(&inode->i_lock); |
821 | if (inode->i_ino != ino) { |
822 | spin_unlock(&inode->i_lock); |
823 | continue; |
824 | } |
825 | if (inode->i_sb != sb) { |
826 | spin_unlock(&inode->i_lock); |
827 | continue; |
828 | } |
829 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
830 | __wait_on_freeing_inode(inode); |
831 | goto repeat; |
832 | } |
833 | __iget(inode); |
834 | spin_unlock(&inode->i_lock); |
835 | return inode; |
836 | } |
837 | return NULL; |
838 | } |
839 | |
840 | /* |
841 | * Each cpu owns a range of LAST_INO_BATCH numbers. |
842 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, |
843 | * to renew the exhausted range. |
844 | * |
845 | * This does not significantly increase overflow rate because every CPU can |
846 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is |
847 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the |
848 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase |
849 | * overflow rate by 2x, which does not seem too significant. |
850 | * |
851 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW |
852 | * error if st_ino won't fit in target struct field. Use 32bit counter |
853 | * here to attempt to avoid that. |
854 | */ |
855 | #define LAST_INO_BATCH 1024 |
856 | static DEFINE_PER_CPU(unsigned int, last_ino); |
857 | |
858 | unsigned int get_next_ino(void) |
859 | { |
860 | unsigned int *p = &get_cpu_var(last_ino); |
861 | unsigned int res = *p; |
862 | |
863 | #ifdef CONFIG_SMP |
864 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { |
865 | static atomic_t shared_last_ino; |
866 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); |
867 | |
868 | res = next - LAST_INO_BATCH; |
869 | } |
870 | #endif |
871 | |
872 | *p = ++res; |
873 | put_cpu_var(last_ino); |
874 | return res; |
875 | } |
876 | EXPORT_SYMBOL(get_next_ino); |
877 | |
878 | /** |
879 | * new_inode - obtain an inode |
880 | * @sb: superblock |
881 | * |
882 | * Allocates a new inode for given superblock. The default gfp_mask |
883 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
884 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
885 | * for the page cache are not reclaimable or migratable, |
886 | * mapping_set_gfp_mask() must be called with suitable flags on the |
887 | * newly created inode's mapping |
888 | * |
889 | */ |
890 | struct inode *new_inode(struct super_block *sb) |
891 | { |
892 | struct inode *inode; |
893 | |
894 | spin_lock_prefetch(&inode_sb_list_lock); |
895 | |
896 | inode = alloc_inode(sb); |
897 | if (inode) { |
898 | spin_lock(&inode->i_lock); |
899 | inode->i_state = 0; |
900 | spin_unlock(&inode->i_lock); |
901 | inode_sb_list_add(inode); |
902 | } |
903 | return inode; |
904 | } |
905 | EXPORT_SYMBOL(new_inode); |
906 | |
907 | /** |
908 | * unlock_new_inode - clear the I_NEW state and wake up any waiters |
909 | * @inode: new inode to unlock |
910 | * |
911 | * Called when the inode is fully initialised to clear the new state of the |
912 | * inode and wake up anyone waiting for the inode to finish initialisation. |
913 | */ |
914 | void unlock_new_inode(struct inode *inode) |
915 | { |
916 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
917 | if (S_ISDIR(inode->i_mode)) { |
918 | struct file_system_type *type = inode->i_sb->s_type; |
919 | |
920 | /* Set new key only if filesystem hasn't already changed it */ |
921 | if (!lockdep_match_class(&inode->i_mutex, |
922 | &type->i_mutex_key)) { |
923 | /* |
924 | * ensure nobody is actually holding i_mutex |
925 | */ |
926 | mutex_destroy(&inode->i_mutex); |
927 | mutex_init(&inode->i_mutex); |
928 | lockdep_set_class(&inode->i_mutex, |
929 | &type->i_mutex_dir_key); |
930 | } |
931 | } |
932 | #endif |
933 | spin_lock(&inode->i_lock); |
934 | WARN_ON(!(inode->i_state & I_NEW)); |
935 | inode->i_state &= ~I_NEW; |
936 | wake_up_bit(&inode->i_state, __I_NEW); |
937 | spin_unlock(&inode->i_lock); |
938 | } |
939 | EXPORT_SYMBOL(unlock_new_inode); |
940 | |
941 | /** |
942 | * iget5_locked - obtain an inode from a mounted file system |
943 | * @sb: super block of file system |
944 | * @hashval: hash value (usually inode number) to get |
945 | * @test: callback used for comparisons between inodes |
946 | * @set: callback used to initialize a new struct inode |
947 | * @data: opaque data pointer to pass to @test and @set |
948 | * |
949 | * Search for the inode specified by @hashval and @data in the inode cache, |
950 | * and if present it is return it with an increased reference count. This is |
951 | * a generalized version of iget_locked() for file systems where the inode |
952 | * number is not sufficient for unique identification of an inode. |
953 | * |
954 | * If the inode is not in cache, allocate a new inode and return it locked, |
955 | * hashed, and with the I_NEW flag set. The file system gets to fill it in |
956 | * before unlocking it via unlock_new_inode(). |
957 | * |
958 | * Note both @test and @set are called with the inode_hash_lock held, so can't |
959 | * sleep. |
960 | */ |
961 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, |
962 | int (*test)(struct inode *, void *), |
963 | int (*set)(struct inode *, void *), void *data) |
964 | { |
965 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
966 | struct inode *inode; |
967 | |
968 | spin_lock(&inode_hash_lock); |
969 | inode = find_inode(sb, head, test, data); |
970 | spin_unlock(&inode_hash_lock); |
971 | |
972 | if (inode) { |
973 | wait_on_inode(inode); |
974 | return inode; |
975 | } |
976 | |
977 | inode = alloc_inode(sb); |
978 | if (inode) { |
979 | struct inode *old; |
980 | |
981 | spin_lock(&inode_hash_lock); |
982 | /* We released the lock, so.. */ |
983 | old = find_inode(sb, head, test, data); |
984 | if (!old) { |
985 | if (set(inode, data)) |
986 | goto set_failed; |
987 | |
988 | spin_lock(&inode->i_lock); |
989 | inode->i_state = I_NEW; |
990 | hlist_add_head(&inode->i_hash, head); |
991 | spin_unlock(&inode->i_lock); |
992 | inode_sb_list_add(inode); |
993 | spin_unlock(&inode_hash_lock); |
994 | |
995 | /* Return the locked inode with I_NEW set, the |
996 | * caller is responsible for filling in the contents |
997 | */ |
998 | return inode; |
999 | } |
1000 | |
1001 | /* |
1002 | * Uhhuh, somebody else created the same inode under |
1003 | * us. Use the old inode instead of the one we just |
1004 | * allocated. |
1005 | */ |
1006 | spin_unlock(&inode_hash_lock); |
1007 | destroy_inode(inode); |
1008 | inode = old; |
1009 | wait_on_inode(inode); |
1010 | } |
1011 | return inode; |
1012 | |
1013 | set_failed: |
1014 | spin_unlock(&inode_hash_lock); |
1015 | destroy_inode(inode); |
1016 | return NULL; |
1017 | } |
1018 | EXPORT_SYMBOL(iget5_locked); |
1019 | |
1020 | /** |
1021 | * iget_locked - obtain an inode from a mounted file system |
1022 | * @sb: super block of file system |
1023 | * @ino: inode number to get |
1024 | * |
1025 | * Search for the inode specified by @ino in the inode cache and if present |
1026 | * return it with an increased reference count. This is for file systems |
1027 | * where the inode number is sufficient for unique identification of an inode. |
1028 | * |
1029 | * If the inode is not in cache, allocate a new inode and return it locked, |
1030 | * hashed, and with the I_NEW flag set. The file system gets to fill it in |
1031 | * before unlocking it via unlock_new_inode(). |
1032 | */ |
1033 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) |
1034 | { |
1035 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
1036 | struct inode *inode; |
1037 | |
1038 | spin_lock(&inode_hash_lock); |
1039 | inode = find_inode_fast(sb, head, ino); |
1040 | spin_unlock(&inode_hash_lock); |
1041 | if (inode) { |
1042 | wait_on_inode(inode); |
1043 | return inode; |
1044 | } |
1045 | |
1046 | inode = alloc_inode(sb); |
1047 | if (inode) { |
1048 | struct inode *old; |
1049 | |
1050 | spin_lock(&inode_hash_lock); |
1051 | /* We released the lock, so.. */ |
1052 | old = find_inode_fast(sb, head, ino); |
1053 | if (!old) { |
1054 | inode->i_ino = ino; |
1055 | spin_lock(&inode->i_lock); |
1056 | inode->i_state = I_NEW; |
1057 | hlist_add_head(&inode->i_hash, head); |
1058 | spin_unlock(&inode->i_lock); |
1059 | inode_sb_list_add(inode); |
1060 | spin_unlock(&inode_hash_lock); |
1061 | |
1062 | /* Return the locked inode with I_NEW set, the |
1063 | * caller is responsible for filling in the contents |
1064 | */ |
1065 | return inode; |
1066 | } |
1067 | |
1068 | /* |
1069 | * Uhhuh, somebody else created the same inode under |
1070 | * us. Use the old inode instead of the one we just |
1071 | * allocated. |
1072 | */ |
1073 | spin_unlock(&inode_hash_lock); |
1074 | destroy_inode(inode); |
1075 | inode = old; |
1076 | wait_on_inode(inode); |
1077 | } |
1078 | return inode; |
1079 | } |
1080 | EXPORT_SYMBOL(iget_locked); |
1081 | |
1082 | /* |
1083 | * search the inode cache for a matching inode number. |
1084 | * If we find one, then the inode number we are trying to |
1085 | * allocate is not unique and so we should not use it. |
1086 | * |
1087 | * Returns 1 if the inode number is unique, 0 if it is not. |
1088 | */ |
1089 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) |
1090 | { |
1091 | struct hlist_head *b = inode_hashtable + hash(sb, ino); |
1092 | struct hlist_node *node; |
1093 | struct inode *inode; |
1094 | |
1095 | spin_lock(&inode_hash_lock); |
1096 | hlist_for_each_entry(inode, node, b, i_hash) { |
1097 | if (inode->i_ino == ino && inode->i_sb == sb) { |
1098 | spin_unlock(&inode_hash_lock); |
1099 | return 0; |
1100 | } |
1101 | } |
1102 | spin_unlock(&inode_hash_lock); |
1103 | |
1104 | return 1; |
1105 | } |
1106 | |
1107 | /** |
1108 | * iunique - get a unique inode number |
1109 | * @sb: superblock |
1110 | * @max_reserved: highest reserved inode number |
1111 | * |
1112 | * Obtain an inode number that is unique on the system for a given |
1113 | * superblock. This is used by file systems that have no natural |
1114 | * permanent inode numbering system. An inode number is returned that |
1115 | * is higher than the reserved limit but unique. |
1116 | * |
1117 | * BUGS: |
1118 | * With a large number of inodes live on the file system this function |
1119 | * currently becomes quite slow. |
1120 | */ |
1121 | ino_t iunique(struct super_block *sb, ino_t max_reserved) |
1122 | { |
1123 | /* |
1124 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW |
1125 | * error if st_ino won't fit in target struct field. Use 32bit counter |
1126 | * here to attempt to avoid that. |
1127 | */ |
1128 | static DEFINE_SPINLOCK(iunique_lock); |
1129 | static unsigned int counter; |
1130 | ino_t res; |
1131 | |
1132 | spin_lock(&iunique_lock); |
1133 | do { |
1134 | if (counter <= max_reserved) |
1135 | counter = max_reserved + 1; |
1136 | res = counter++; |
1137 | } while (!test_inode_iunique(sb, res)); |
1138 | spin_unlock(&iunique_lock); |
1139 | |
1140 | return res; |
1141 | } |
1142 | EXPORT_SYMBOL(iunique); |
1143 | |
1144 | struct inode *igrab(struct inode *inode) |
1145 | { |
1146 | spin_lock(&inode->i_lock); |
1147 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { |
1148 | __iget(inode); |
1149 | spin_unlock(&inode->i_lock); |
1150 | } else { |
1151 | spin_unlock(&inode->i_lock); |
1152 | /* |
1153 | * Handle the case where s_op->clear_inode is not been |
1154 | * called yet, and somebody is calling igrab |
1155 | * while the inode is getting freed. |
1156 | */ |
1157 | inode = NULL; |
1158 | } |
1159 | return inode; |
1160 | } |
1161 | EXPORT_SYMBOL(igrab); |
1162 | |
1163 | /** |
1164 | * ilookup5_nowait - search for an inode in the inode cache |
1165 | * @sb: super block of file system to search |
1166 | * @hashval: hash value (usually inode number) to search for |
1167 | * @test: callback used for comparisons between inodes |
1168 | * @data: opaque data pointer to pass to @test |
1169 | * |
1170 | * Search for the inode specified by @hashval and @data in the inode cache. |
1171 | * If the inode is in the cache, the inode is returned with an incremented |
1172 | * reference count. |
1173 | * |
1174 | * Note: I_NEW is not waited upon so you have to be very careful what you do |
1175 | * with the returned inode. You probably should be using ilookup5() instead. |
1176 | * |
1177 | * Note2: @test is called with the inode_hash_lock held, so can't sleep. |
1178 | */ |
1179 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, |
1180 | int (*test)(struct inode *, void *), void *data) |
1181 | { |
1182 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1183 | struct inode *inode; |
1184 | |
1185 | spin_lock(&inode_hash_lock); |
1186 | inode = find_inode(sb, head, test, data); |
1187 | spin_unlock(&inode_hash_lock); |
1188 | |
1189 | return inode; |
1190 | } |
1191 | EXPORT_SYMBOL(ilookup5_nowait); |
1192 | |
1193 | /** |
1194 | * ilookup5 - search for an inode in the inode cache |
1195 | * @sb: super block of file system to search |
1196 | * @hashval: hash value (usually inode number) to search for |
1197 | * @test: callback used for comparisons between inodes |
1198 | * @data: opaque data pointer to pass to @test |
1199 | * |
1200 | * Search for the inode specified by @hashval and @data in the inode cache, |
1201 | * and if the inode is in the cache, return the inode with an incremented |
1202 | * reference count. Waits on I_NEW before returning the inode. |
1203 | * returned with an incremented reference count. |
1204 | * |
1205 | * This is a generalized version of ilookup() for file systems where the |
1206 | * inode number is not sufficient for unique identification of an inode. |
1207 | * |
1208 | * Note: @test is called with the inode_hash_lock held, so can't sleep. |
1209 | */ |
1210 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, |
1211 | int (*test)(struct inode *, void *), void *data) |
1212 | { |
1213 | struct inode *inode = ilookup5_nowait(sb, hashval, test, data); |
1214 | |
1215 | if (inode) |
1216 | wait_on_inode(inode); |
1217 | return inode; |
1218 | } |
1219 | EXPORT_SYMBOL(ilookup5); |
1220 | |
1221 | /** |
1222 | * ilookup - search for an inode in the inode cache |
1223 | * @sb: super block of file system to search |
1224 | * @ino: inode number to search for |
1225 | * |
1226 | * Search for the inode @ino in the inode cache, and if the inode is in the |
1227 | * cache, the inode is returned with an incremented reference count. |
1228 | */ |
1229 | struct inode *ilookup(struct super_block *sb, unsigned long ino) |
1230 | { |
1231 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
1232 | struct inode *inode; |
1233 | |
1234 | spin_lock(&inode_hash_lock); |
1235 | inode = find_inode_fast(sb, head, ino); |
1236 | spin_unlock(&inode_hash_lock); |
1237 | |
1238 | if (inode) |
1239 | wait_on_inode(inode); |
1240 | return inode; |
1241 | } |
1242 | EXPORT_SYMBOL(ilookup); |
1243 | |
1244 | int insert_inode_locked(struct inode *inode) |
1245 | { |
1246 | struct super_block *sb = inode->i_sb; |
1247 | ino_t ino = inode->i_ino; |
1248 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
1249 | |
1250 | while (1) { |
1251 | struct hlist_node *node; |
1252 | struct inode *old = NULL; |
1253 | spin_lock(&inode_hash_lock); |
1254 | hlist_for_each_entry(old, node, head, i_hash) { |
1255 | if (old->i_ino != ino) |
1256 | continue; |
1257 | if (old->i_sb != sb) |
1258 | continue; |
1259 | spin_lock(&old->i_lock); |
1260 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { |
1261 | spin_unlock(&old->i_lock); |
1262 | continue; |
1263 | } |
1264 | break; |
1265 | } |
1266 | if (likely(!node)) { |
1267 | spin_lock(&inode->i_lock); |
1268 | inode->i_state |= I_NEW; |
1269 | hlist_add_head(&inode->i_hash, head); |
1270 | spin_unlock(&inode->i_lock); |
1271 | spin_unlock(&inode_hash_lock); |
1272 | return 0; |
1273 | } |
1274 | __iget(old); |
1275 | spin_unlock(&old->i_lock); |
1276 | spin_unlock(&inode_hash_lock); |
1277 | wait_on_inode(old); |
1278 | if (unlikely(!inode_unhashed(old))) { |
1279 | iput(old); |
1280 | return -EBUSY; |
1281 | } |
1282 | iput(old); |
1283 | } |
1284 | } |
1285 | EXPORT_SYMBOL(insert_inode_locked); |
1286 | |
1287 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, |
1288 | int (*test)(struct inode *, void *), void *data) |
1289 | { |
1290 | struct super_block *sb = inode->i_sb; |
1291 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1292 | |
1293 | while (1) { |
1294 | struct hlist_node *node; |
1295 | struct inode *old = NULL; |
1296 | |
1297 | spin_lock(&inode_hash_lock); |
1298 | hlist_for_each_entry(old, node, head, i_hash) { |
1299 | if (old->i_sb != sb) |
1300 | continue; |
1301 | if (!test(old, data)) |
1302 | continue; |
1303 | spin_lock(&old->i_lock); |
1304 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { |
1305 | spin_unlock(&old->i_lock); |
1306 | continue; |
1307 | } |
1308 | break; |
1309 | } |
1310 | if (likely(!node)) { |
1311 | spin_lock(&inode->i_lock); |
1312 | inode->i_state |= I_NEW; |
1313 | hlist_add_head(&inode->i_hash, head); |
1314 | spin_unlock(&inode->i_lock); |
1315 | spin_unlock(&inode_hash_lock); |
1316 | return 0; |
1317 | } |
1318 | __iget(old); |
1319 | spin_unlock(&old->i_lock); |
1320 | spin_unlock(&inode_hash_lock); |
1321 | wait_on_inode(old); |
1322 | if (unlikely(!inode_unhashed(old))) { |
1323 | iput(old); |
1324 | return -EBUSY; |
1325 | } |
1326 | iput(old); |
1327 | } |
1328 | } |
1329 | EXPORT_SYMBOL(insert_inode_locked4); |
1330 | |
1331 | |
1332 | int generic_delete_inode(struct inode *inode) |
1333 | { |
1334 | return 1; |
1335 | } |
1336 | EXPORT_SYMBOL(generic_delete_inode); |
1337 | |
1338 | /* |
1339 | * Normal UNIX filesystem behaviour: delete the |
1340 | * inode when the usage count drops to zero, and |
1341 | * i_nlink is zero. |
1342 | */ |
1343 | int generic_drop_inode(struct inode *inode) |
1344 | { |
1345 | return !inode->i_nlink || inode_unhashed(inode); |
1346 | } |
1347 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1348 | |
1349 | /* |
1350 | * Called when we're dropping the last reference |
1351 | * to an inode. |
1352 | * |
1353 | * Call the FS "drop_inode()" function, defaulting to |
1354 | * the legacy UNIX filesystem behaviour. If it tells |
1355 | * us to evict inode, do so. Otherwise, retain inode |
1356 | * in cache if fs is alive, sync and evict if fs is |
1357 | * shutting down. |
1358 | */ |
1359 | static void iput_final(struct inode *inode) |
1360 | { |
1361 | struct super_block *sb = inode->i_sb; |
1362 | const struct super_operations *op = inode->i_sb->s_op; |
1363 | int drop; |
1364 | |
1365 | WARN_ON(inode->i_state & I_NEW); |
1366 | |
1367 | if (op && op->drop_inode) |
1368 | drop = op->drop_inode(inode); |
1369 | else |
1370 | drop = generic_drop_inode(inode); |
1371 | |
1372 | if (!drop && (sb->s_flags & MS_ACTIVE)) { |
1373 | inode->i_state |= I_REFERENCED; |
1374 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) |
1375 | inode_lru_list_add(inode); |
1376 | spin_unlock(&inode->i_lock); |
1377 | return; |
1378 | } |
1379 | |
1380 | if (!drop) { |
1381 | inode->i_state |= I_WILL_FREE; |
1382 | spin_unlock(&inode->i_lock); |
1383 | write_inode_now(inode, 1); |
1384 | spin_lock(&inode->i_lock); |
1385 | WARN_ON(inode->i_state & I_NEW); |
1386 | inode->i_state &= ~I_WILL_FREE; |
1387 | } |
1388 | |
1389 | inode->i_state |= I_FREEING; |
1390 | inode_lru_list_del(inode); |
1391 | spin_unlock(&inode->i_lock); |
1392 | |
1393 | evict(inode); |
1394 | } |
1395 | |
1396 | /** |
1397 | * iput - put an inode |
1398 | * @inode: inode to put |
1399 | * |
1400 | * Puts an inode, dropping its usage count. If the inode use count hits |
1401 | * zero, the inode is then freed and may also be destroyed. |
1402 | * |
1403 | * Consequently, iput() can sleep. |
1404 | */ |
1405 | void iput(struct inode *inode) |
1406 | { |
1407 | if (inode) { |
1408 | BUG_ON(inode->i_state & I_CLEAR); |
1409 | |
1410 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) |
1411 | iput_final(inode); |
1412 | } |
1413 | } |
1414 | EXPORT_SYMBOL(iput); |
1415 | |
1416 | /** |
1417 | * bmap - find a block number in a file |
1418 | * @inode: inode of file |
1419 | * @block: block to find |
1420 | * |
1421 | * Returns the block number on the device holding the inode that |
1422 | * is the disk block number for the block of the file requested. |
1423 | * That is, asked for block 4 of inode 1 the function will return the |
1424 | * disk block relative to the disk start that holds that block of the |
1425 | * file. |
1426 | */ |
1427 | sector_t bmap(struct inode *inode, sector_t block) |
1428 | { |
1429 | sector_t res = 0; |
1430 | if (inode->i_mapping->a_ops->bmap) |
1431 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); |
1432 | return res; |
1433 | } |
1434 | EXPORT_SYMBOL(bmap); |
1435 | |
1436 | /* |
1437 | * With relative atime, only update atime if the previous atime is |
1438 | * earlier than either the ctime or mtime or if at least a day has |
1439 | * passed since the last atime update. |
1440 | */ |
1441 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, |
1442 | struct timespec now) |
1443 | { |
1444 | |
1445 | if (!(mnt->mnt_flags & MNT_RELATIME)) |
1446 | return 1; |
1447 | /* |
1448 | * Is mtime younger than atime? If yes, update atime: |
1449 | */ |
1450 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) |
1451 | return 1; |
1452 | /* |
1453 | * Is ctime younger than atime? If yes, update atime: |
1454 | */ |
1455 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) |
1456 | return 1; |
1457 | |
1458 | /* |
1459 | * Is the previous atime value older than a day? If yes, |
1460 | * update atime: |
1461 | */ |
1462 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) |
1463 | return 1; |
1464 | /* |
1465 | * Good, we can skip the atime update: |
1466 | */ |
1467 | return 0; |
1468 | } |
1469 | |
1470 | /** |
1471 | * touch_atime - update the access time |
1472 | * @mnt: mount the inode is accessed on |
1473 | * @dentry: dentry accessed |
1474 | * |
1475 | * Update the accessed time on an inode and mark it for writeback. |
1476 | * This function automatically handles read only file systems and media, |
1477 | * as well as the "noatime" flag and inode specific "noatime" markers. |
1478 | */ |
1479 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1480 | { |
1481 | struct inode *inode = dentry->d_inode; |
1482 | struct timespec now; |
1483 | |
1484 | if (inode->i_flags & S_NOATIME) |
1485 | return; |
1486 | if (IS_NOATIME(inode)) |
1487 | return; |
1488 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
1489 | return; |
1490 | |
1491 | if (mnt->mnt_flags & MNT_NOATIME) |
1492 | return; |
1493 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
1494 | return; |
1495 | |
1496 | now = current_fs_time(inode->i_sb); |
1497 | |
1498 | if (!relatime_need_update(mnt, inode, now)) |
1499 | return; |
1500 | |
1501 | if (timespec_equal(&inode->i_atime, &now)) |
1502 | return; |
1503 | |
1504 | if (mnt_want_write(mnt)) |
1505 | return; |
1506 | |
1507 | inode->i_atime = now; |
1508 | mark_inode_dirty_sync(inode); |
1509 | mnt_drop_write(mnt); |
1510 | } |
1511 | EXPORT_SYMBOL(touch_atime); |
1512 | |
1513 | /** |
1514 | * file_update_time - update mtime and ctime time |
1515 | * @file: file accessed |
1516 | * |
1517 | * Update the mtime and ctime members of an inode and mark the inode |
1518 | * for writeback. Note that this function is meant exclusively for |
1519 | * usage in the file write path of filesystems, and filesystems may |
1520 | * choose to explicitly ignore update via this function with the |
1521 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
1522 | * timestamps are handled by the server. |
1523 | */ |
1524 | |
1525 | void file_update_time(struct file *file) |
1526 | { |
1527 | struct inode *inode = file->f_path.dentry->d_inode; |
1528 | struct timespec now; |
1529 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; |
1530 | |
1531 | /* First try to exhaust all avenues to not sync */ |
1532 | if (IS_NOCMTIME(inode)) |
1533 | return; |
1534 | |
1535 | now = current_fs_time(inode->i_sb); |
1536 | if (!timespec_equal(&inode->i_mtime, &now)) |
1537 | sync_it = S_MTIME; |
1538 | |
1539 | if (!timespec_equal(&inode->i_ctime, &now)) |
1540 | sync_it |= S_CTIME; |
1541 | |
1542 | if (IS_I_VERSION(inode)) |
1543 | sync_it |= S_VERSION; |
1544 | |
1545 | if (!sync_it) |
1546 | return; |
1547 | |
1548 | /* Finally allowed to write? Takes lock. */ |
1549 | if (mnt_want_write_file(file)) |
1550 | return; |
1551 | |
1552 | /* Only change inode inside the lock region */ |
1553 | if (sync_it & S_VERSION) |
1554 | inode_inc_iversion(inode); |
1555 | if (sync_it & S_CTIME) |
1556 | inode->i_ctime = now; |
1557 | if (sync_it & S_MTIME) |
1558 | inode->i_mtime = now; |
1559 | mark_inode_dirty_sync(inode); |
1560 | mnt_drop_write(file->f_path.mnt); |
1561 | } |
1562 | EXPORT_SYMBOL(file_update_time); |
1563 | |
1564 | int inode_needs_sync(struct inode *inode) |
1565 | { |
1566 | if (IS_SYNC(inode)) |
1567 | return 1; |
1568 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) |
1569 | return 1; |
1570 | return 0; |
1571 | } |
1572 | EXPORT_SYMBOL(inode_needs_sync); |
1573 | |
1574 | int inode_wait(void *word) |
1575 | { |
1576 | schedule(); |
1577 | return 0; |
1578 | } |
1579 | EXPORT_SYMBOL(inode_wait); |
1580 | |
1581 | /* |
1582 | * If we try to find an inode in the inode hash while it is being |
1583 | * deleted, we have to wait until the filesystem completes its |
1584 | * deletion before reporting that it isn't found. This function waits |
1585 | * until the deletion _might_ have completed. Callers are responsible |
1586 | * to recheck inode state. |
1587 | * |
1588 | * It doesn't matter if I_NEW is not set initially, a call to |
1589 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list |
1590 | * will DTRT. |
1591 | */ |
1592 | static void __wait_on_freeing_inode(struct inode *inode) |
1593 | { |
1594 | wait_queue_head_t *wq; |
1595 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1596 | wq = bit_waitqueue(&inode->i_state, __I_NEW); |
1597 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
1598 | spin_unlock(&inode->i_lock); |
1599 | spin_unlock(&inode_hash_lock); |
1600 | schedule(); |
1601 | finish_wait(wq, &wait.wait); |
1602 | spin_lock(&inode_hash_lock); |
1603 | } |
1604 | |
1605 | static __initdata unsigned long ihash_entries; |
1606 | static int __init set_ihash_entries(char *str) |
1607 | { |
1608 | if (!str) |
1609 | return 0; |
1610 | ihash_entries = simple_strtoul(str, &str, 0); |
1611 | return 1; |
1612 | } |
1613 | __setup("ihash_entries=", set_ihash_entries); |
1614 | |
1615 | /* |
1616 | * Initialize the waitqueues and inode hash table. |
1617 | */ |
1618 | void __init inode_init_early(void) |
1619 | { |
1620 | int loop; |
1621 | |
1622 | /* If hashes are distributed across NUMA nodes, defer |
1623 | * hash allocation until vmalloc space is available. |
1624 | */ |
1625 | if (hashdist) |
1626 | return; |
1627 | |
1628 | inode_hashtable = |
1629 | alloc_large_system_hash("Inode-cache", |
1630 | sizeof(struct hlist_head), |
1631 | ihash_entries, |
1632 | 14, |
1633 | HASH_EARLY, |
1634 | &i_hash_shift, |
1635 | &i_hash_mask, |
1636 | 0); |
1637 | |
1638 | for (loop = 0; loop < (1 << i_hash_shift); loop++) |
1639 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1640 | } |
1641 | |
1642 | void __init inode_init(void) |
1643 | { |
1644 | int loop; |
1645 | |
1646 | /* inode slab cache */ |
1647 | inode_cachep = kmem_cache_create("inode_cache", |
1648 | sizeof(struct inode), |
1649 | 0, |
1650 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| |
1651 | SLAB_MEM_SPREAD), |
1652 | init_once); |
1653 | register_shrinker(&icache_shrinker); |
1654 | |
1655 | /* Hash may have been set up in inode_init_early */ |
1656 | if (!hashdist) |
1657 | return; |
1658 | |
1659 | inode_hashtable = |
1660 | alloc_large_system_hash("Inode-cache", |
1661 | sizeof(struct hlist_head), |
1662 | ihash_entries, |
1663 | 14, |
1664 | 0, |
1665 | &i_hash_shift, |
1666 | &i_hash_mask, |
1667 | 0); |
1668 | |
1669 | for (loop = 0; loop < (1 << i_hash_shift); loop++) |
1670 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1671 | } |
1672 | |
1673 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) |
1674 | { |
1675 | inode->i_mode = mode; |
1676 | if (S_ISCHR(mode)) { |
1677 | inode->i_fop = &def_chr_fops; |
1678 | inode->i_rdev = rdev; |
1679 | } else if (S_ISBLK(mode)) { |
1680 | inode->i_fop = &def_blk_fops; |
1681 | inode->i_rdev = rdev; |
1682 | } else if (S_ISFIFO(mode)) |
1683 | inode->i_fop = &def_fifo_fops; |
1684 | else if (S_ISSOCK(mode)) |
1685 | inode->i_fop = &bad_sock_fops; |
1686 | else |
1687 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1688 | " inode %s:%lu\n", mode, inode->i_sb->s_id, |
1689 | inode->i_ino); |
1690 | } |
1691 | EXPORT_SYMBOL(init_special_inode); |
1692 | |
1693 | /** |
1694 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards |
1695 | * @inode: New inode |
1696 | * @dir: Directory inode |
1697 | * @mode: mode of the new inode |
1698 | */ |
1699 | void inode_init_owner(struct inode *inode, const struct inode *dir, |
1700 | mode_t mode) |
1701 | { |
1702 | inode->i_uid = current_fsuid(); |
1703 | if (dir && dir->i_mode & S_ISGID) { |
1704 | inode->i_gid = dir->i_gid; |
1705 | if (S_ISDIR(mode)) |
1706 | mode |= S_ISGID; |
1707 | } else |
1708 | inode->i_gid = current_fsgid(); |
1709 | inode->i_mode = mode; |
1710 | } |
1711 | EXPORT_SYMBOL(inode_init_owner); |
1712 | |
1713 | /** |
1714 | * inode_owner_or_capable - check current task permissions to inode |
1715 | * @inode: inode being checked |
1716 | * |
1717 | * Return true if current either has CAP_FOWNER to the inode, or |
1718 | * owns the file. |
1719 | */ |
1720 | bool inode_owner_or_capable(const struct inode *inode) |
1721 | { |
1722 | struct user_namespace *ns = inode_userns(inode); |
1723 | |
1724 | if (current_user_ns() == ns && current_fsuid() == inode->i_uid) |
1725 | return true; |
1726 | if (ns_capable(ns, CAP_FOWNER)) |
1727 | return true; |
1728 | return false; |
1729 | } |
1730 | EXPORT_SYMBOL(inode_owner_or_capable); |
1731 |
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