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