]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * (C) 1997 Linus Torvalds | |
3 | * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation) | |
4 | */ | |
5 | #include <linux/export.h> | |
6 | #include <linux/fs.h> | |
7 | #include <linux/mm.h> | |
8 | #include <linux/backing-dev.h> | |
9 | #include <linux/hash.h> | |
10 | #include <linux/swap.h> | |
11 | #include <linux/security.h> | |
12 | #include <linux/cdev.h> | |
13 | #include <linux/bootmem.h> | |
14 | #include <linux/fsnotify.h> | |
15 | #include <linux/mount.h> | |
16 | #include <linux/posix_acl.h> | |
17 | #include <linux/prefetch.h> | |
18 | #include <linux/buffer_head.h> /* for inode_has_buffers */ | |
19 | #include <linux/ratelimit.h> | |
20 | #include "internal.h" | |
21 | ||
22 | /* | |
23 | * Inode locking rules: | |
24 | * | |
25 | * inode->i_lock protects: | |
26 | * inode->i_state, inode->i_hash, __iget() | |
27 | * inode->i_sb->s_inode_lru_lock protects: | |
28 | * inode->i_sb->s_inode_lru, inode->i_lru | |
29 | * inode_sb_list_lock protects: | |
30 | * sb->s_inodes, inode->i_sb_list | |
31 | * bdi->wb.list_lock protects: | |
32 | * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list | |
33 | * inode_hash_lock protects: | |
34 | * inode_hashtable, inode->i_hash | |
35 | * | |
36 | * Lock ordering: | |
37 | * | |
38 | * inode_sb_list_lock | |
39 | * inode->i_lock | |
40 | * inode->i_sb->s_inode_lru_lock | |
41 | * | |
42 | * bdi->wb.list_lock | |
43 | * inode->i_lock | |
44 | * | |
45 | * inode_hash_lock | |
46 | * inode_sb_list_lock | |
47 | * inode->i_lock | |
48 | * | |
49 | * iunique_lock | |
50 | * inode_hash_lock | |
51 | */ | |
52 | ||
53 | static unsigned int i_hash_mask __read_mostly; | |
54 | static unsigned int i_hash_shift __read_mostly; | |
55 | static struct hlist_head *inode_hashtable __read_mostly; | |
56 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock); | |
57 | ||
58 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock); | |
59 | ||
60 | /* | |
61 | * Empty aops. Can be used for the cases where the user does not | |
62 | * define any of the address_space operations. | |
63 | */ | |
64 | const struct address_space_operations empty_aops = { | |
65 | }; | |
66 | EXPORT_SYMBOL(empty_aops); | |
67 | ||
68 | /* | |
69 | * Statistics gathering.. | |
70 | */ | |
71 | struct inodes_stat_t inodes_stat; | |
72 | ||
73 | static DEFINE_PER_CPU(unsigned int, nr_inodes); | |
74 | static DEFINE_PER_CPU(unsigned int, nr_unused); | |
75 | ||
76 | static struct kmem_cache *inode_cachep __read_mostly; | |
77 | ||
78 | static int get_nr_inodes(void) | |
79 | { | |
80 | int i; | |
81 | int sum = 0; | |
82 | for_each_possible_cpu(i) | |
83 | sum += per_cpu(nr_inodes, i); | |
84 | return sum < 0 ? 0 : sum; | |
85 | } | |
86 | ||
87 | static inline int get_nr_inodes_unused(void) | |
88 | { | |
89 | int i; | |
90 | int sum = 0; | |
91 | for_each_possible_cpu(i) | |
92 | sum += per_cpu(nr_unused, i); | |
93 | return sum < 0 ? 0 : sum; | |
94 | } | |
95 | ||
96 | int get_nr_dirty_inodes(void) | |
97 | { | |
98 | /* not actually dirty inodes, but a wild approximation */ | |
99 | int nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); | |
100 | return nr_dirty > 0 ? nr_dirty : 0; | |
101 | } | |
102 | ||
103 | /* | |
104 | * Handle nr_inode sysctl | |
105 | */ | |
106 | #ifdef CONFIG_SYSCTL | |
107 | int proc_nr_inodes(ctl_table *table, int write, | |
108 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
109 | { | |
110 | inodes_stat.nr_inodes = get_nr_inodes(); | |
111 | inodes_stat.nr_unused = get_nr_inodes_unused(); | |
112 | return proc_dointvec(table, write, buffer, lenp, ppos); | |
113 | } | |
114 | #endif | |
115 | ||
116 | /** | |
117 | * inode_init_always - perform inode structure intialisation | |
118 | * @sb: superblock inode belongs to | |
119 | * @inode: inode to initialise | |
120 | * | |
121 | * These are initializations that need to be done on every inode | |
122 | * allocation as the fields are not initialised by slab allocation. | |
123 | */ | |
124 | int inode_init_always(struct super_block *sb, struct inode *inode) | |
125 | { | |
126 | static const struct inode_operations empty_iops; | |
127 | static const struct file_operations empty_fops; | |
128 | struct address_space *const mapping = &inode->i_data; | |
129 | ||
130 | inode->i_sb = sb; | |
131 | inode->i_blkbits = sb->s_blocksize_bits; | |
132 | inode->i_flags = 0; | |
133 | atomic_set(&inode->i_count, 1); | |
134 | inode->i_op = &empty_iops; | |
135 | inode->i_fop = &empty_fops; | |
136 | inode->__i_nlink = 1; | |
137 | inode->i_opflags = 0; | |
138 | i_uid_write(inode, 0); | |
139 | i_gid_write(inode, 0); | |
140 | atomic_set(&inode->i_writecount, 0); | |
141 | inode->i_size = 0; | |
142 | inode->i_blocks = 0; | |
143 | inode->i_bytes = 0; | |
144 | inode->i_generation = 0; | |
145 | #ifdef CONFIG_QUOTA | |
146 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); | |
147 | #endif | |
148 | inode->i_pipe = NULL; | |
149 | inode->i_bdev = NULL; | |
150 | inode->i_cdev = NULL; | |
151 | inode->i_rdev = 0; | |
152 | inode->dirtied_when = 0; | |
153 | ||
154 | if (security_inode_alloc(inode)) | |
155 | goto out; | |
156 | spin_lock_init(&inode->i_lock); | |
157 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
158 | ||
159 | mutex_init(&inode->i_mutex); | |
160 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
161 | ||
162 | atomic_set(&inode->i_dio_count, 0); | |
163 | ||
164 | mapping->a_ops = &empty_aops; | |
165 | mapping->host = inode; | |
166 | mapping->flags = 0; | |
167 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); | |
168 | mapping->assoc_mapping = NULL; | |
169 | mapping->backing_dev_info = &default_backing_dev_info; | |
170 | mapping->writeback_index = 0; | |
171 | ||
172 | /* | |
173 | * If the block_device provides a backing_dev_info for client | |
174 | * inodes then use that. Otherwise the inode share the bdev's | |
175 | * backing_dev_info. | |
176 | */ | |
177 | if (sb->s_bdev) { | |
178 | struct backing_dev_info *bdi; | |
179 | ||
180 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; | |
181 | mapping->backing_dev_info = bdi; | |
182 | } | |
183 | inode->i_private = NULL; | |
184 | inode->i_mapping = mapping; | |
185 | INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */ | |
186 | #ifdef CONFIG_FS_POSIX_ACL | |
187 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
188 | #endif | |
189 | ||
190 | #ifdef CONFIG_FSNOTIFY | |
191 | inode->i_fsnotify_mask = 0; | |
192 | #endif | |
193 | ||
194 | this_cpu_inc(nr_inodes); | |
195 | ||
196 | return 0; | |
197 | out: | |
198 | return -ENOMEM; | |
199 | } | |
200 | EXPORT_SYMBOL(inode_init_always); | |
201 | ||
202 | static struct inode *alloc_inode(struct super_block *sb) | |
203 | { | |
204 | struct inode *inode; | |
205 | ||
206 | if (sb->s_op->alloc_inode) | |
207 | inode = sb->s_op->alloc_inode(sb); | |
208 | else | |
209 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
210 | ||
211 | if (!inode) | |
212 | return NULL; | |
213 | ||
214 | if (unlikely(inode_init_always(sb, inode))) { | |
215 | if (inode->i_sb->s_op->destroy_inode) | |
216 | inode->i_sb->s_op->destroy_inode(inode); | |
217 | else | |
218 | kmem_cache_free(inode_cachep, inode); | |
219 | return NULL; | |
220 | } | |
221 | ||
222 | return inode; | |
223 | } | |
224 | ||
225 | void free_inode_nonrcu(struct inode *inode) | |
226 | { | |
227 | kmem_cache_free(inode_cachep, inode); | |
228 | } | |
229 | EXPORT_SYMBOL(free_inode_nonrcu); | |
230 | ||
231 | void __destroy_inode(struct inode *inode) | |
232 | { | |
233 | BUG_ON(inode_has_buffers(inode)); | |
234 | security_inode_free(inode); | |
235 | fsnotify_inode_delete(inode); | |
236 | if (!inode->i_nlink) { | |
237 | WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0); | |
238 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
239 | } | |
240 | ||
241 | #ifdef CONFIG_FS_POSIX_ACL | |
242 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) | |
243 | posix_acl_release(inode->i_acl); | |
244 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) | |
245 | posix_acl_release(inode->i_default_acl); | |
246 | #endif | |
247 | this_cpu_dec(nr_inodes); | |
248 | } | |
249 | EXPORT_SYMBOL(__destroy_inode); | |
250 | ||
251 | static void i_callback(struct rcu_head *head) | |
252 | { | |
253 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
254 | kmem_cache_free(inode_cachep, inode); | |
255 | } | |
256 | ||
257 | static void destroy_inode(struct inode *inode) | |
258 | { | |
259 | BUG_ON(!list_empty(&inode->i_lru)); | |
260 | __destroy_inode(inode); | |
261 | if (inode->i_sb->s_op->destroy_inode) | |
262 | inode->i_sb->s_op->destroy_inode(inode); | |
263 | else | |
264 | call_rcu(&inode->i_rcu, i_callback); | |
265 | } | |
266 | ||
267 | /** | |
268 | * drop_nlink - directly drop an inode's link count | |
269 | * @inode: inode | |
270 | * | |
271 | * This is a low-level filesystem helper to replace any | |
272 | * direct filesystem manipulation of i_nlink. In cases | |
273 | * where we are attempting to track writes to the | |
274 | * filesystem, a decrement to zero means an imminent | |
275 | * write when the file is truncated and actually unlinked | |
276 | * on the filesystem. | |
277 | */ | |
278 | void drop_nlink(struct inode *inode) | |
279 | { | |
280 | WARN_ON(inode->i_nlink == 0); | |
281 | inode->__i_nlink--; | |
282 | if (!inode->i_nlink) | |
283 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
284 | } | |
285 | EXPORT_SYMBOL(drop_nlink); | |
286 | ||
287 | /** | |
288 | * clear_nlink - directly zero an inode's link count | |
289 | * @inode: inode | |
290 | * | |
291 | * This is a low-level filesystem helper to replace any | |
292 | * direct filesystem manipulation of i_nlink. See | |
293 | * drop_nlink() for why we care about i_nlink hitting zero. | |
294 | */ | |
295 | void clear_nlink(struct inode *inode) | |
296 | { | |
297 | if (inode->i_nlink) { | |
298 | inode->__i_nlink = 0; | |
299 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
300 | } | |
301 | } | |
302 | EXPORT_SYMBOL(clear_nlink); | |
303 | ||
304 | /** | |
305 | * set_nlink - directly set an inode's link count | |
306 | * @inode: inode | |
307 | * @nlink: new nlink (should be non-zero) | |
308 | * | |
309 | * This is a low-level filesystem helper to replace any | |
310 | * direct filesystem manipulation of i_nlink. | |
311 | */ | |
312 | void set_nlink(struct inode *inode, unsigned int nlink) | |
313 | { | |
314 | if (!nlink) { | |
315 | clear_nlink(inode); | |
316 | } else { | |
317 | /* Yes, some filesystems do change nlink from zero to one */ | |
318 | if (inode->i_nlink == 0) | |
319 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
320 | ||
321 | inode->__i_nlink = nlink; | |
322 | } | |
323 | } | |
324 | EXPORT_SYMBOL(set_nlink); | |
325 | ||
326 | /** | |
327 | * inc_nlink - directly increment an inode's link count | |
328 | * @inode: inode | |
329 | * | |
330 | * This is a low-level filesystem helper to replace any | |
331 | * direct filesystem manipulation of i_nlink. Currently, | |
332 | * it is only here for parity with dec_nlink(). | |
333 | */ | |
334 | void inc_nlink(struct inode *inode) | |
335 | { | |
336 | if (WARN_ON(inode->i_nlink == 0)) | |
337 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
338 | ||
339 | inode->__i_nlink++; | |
340 | } | |
341 | EXPORT_SYMBOL(inc_nlink); | |
342 | ||
343 | void address_space_init_once(struct address_space *mapping) | |
344 | { | |
345 | memset(mapping, 0, sizeof(*mapping)); | |
346 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC); | |
347 | spin_lock_init(&mapping->tree_lock); | |
348 | mutex_init(&mapping->i_mmap_mutex); | |
349 | INIT_LIST_HEAD(&mapping->private_list); | |
350 | spin_lock_init(&mapping->private_lock); | |
351 | mapping->i_mmap = RB_ROOT; | |
352 | INIT_LIST_HEAD(&mapping->i_mmap_nonlinear); | |
353 | } | |
354 | EXPORT_SYMBOL(address_space_init_once); | |
355 | ||
356 | /* | |
357 | * These are initializations that only need to be done | |
358 | * once, because the fields are idempotent across use | |
359 | * of the inode, so let the slab aware of that. | |
360 | */ | |
361 | void inode_init_once(struct inode *inode) | |
362 | { | |
363 | memset(inode, 0, sizeof(*inode)); | |
364 | INIT_HLIST_NODE(&inode->i_hash); | |
365 | INIT_LIST_HEAD(&inode->i_devices); | |
366 | INIT_LIST_HEAD(&inode->i_wb_list); | |
367 | INIT_LIST_HEAD(&inode->i_lru); | |
368 | address_space_init_once(&inode->i_data); | |
369 | i_size_ordered_init(inode); | |
370 | #ifdef CONFIG_FSNOTIFY | |
371 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); | |
372 | #endif | |
373 | } | |
374 | EXPORT_SYMBOL(inode_init_once); | |
375 | ||
376 | static void init_once(void *foo) | |
377 | { | |
378 | struct inode *inode = (struct inode *) foo; | |
379 | ||
380 | inode_init_once(inode); | |
381 | } | |
382 | ||
383 | /* | |
384 | * inode->i_lock must be held | |
385 | */ | |
386 | void __iget(struct inode *inode) | |
387 | { | |
388 | atomic_inc(&inode->i_count); | |
389 | } | |
390 | ||
391 | /* | |
392 | * get additional reference to inode; caller must already hold one. | |
393 | */ | |
394 | void ihold(struct inode *inode) | |
395 | { | |
396 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); | |
397 | } | |
398 | EXPORT_SYMBOL(ihold); | |
399 | ||
400 | static void inode_lru_list_add(struct inode *inode) | |
401 | { | |
402 | spin_lock(&inode->i_sb->s_inode_lru_lock); | |
403 | if (list_empty(&inode->i_lru)) { | |
404 | list_add(&inode->i_lru, &inode->i_sb->s_inode_lru); | |
405 | inode->i_sb->s_nr_inodes_unused++; | |
406 | this_cpu_inc(nr_unused); | |
407 | } | |
408 | spin_unlock(&inode->i_sb->s_inode_lru_lock); | |
409 | } | |
410 | ||
411 | /* | |
412 | * Add inode to LRU if needed (inode is unused and clean). | |
413 | * | |
414 | * Needs inode->i_lock held. | |
415 | */ | |
416 | void inode_add_lru(struct inode *inode) | |
417 | { | |
418 | if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) && | |
419 | !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE) | |
420 | inode_lru_list_add(inode); | |
421 | } | |
422 | ||
423 | ||
424 | static void inode_lru_list_del(struct inode *inode) | |
425 | { | |
426 | spin_lock(&inode->i_sb->s_inode_lru_lock); | |
427 | if (!list_empty(&inode->i_lru)) { | |
428 | list_del_init(&inode->i_lru); | |
429 | inode->i_sb->s_nr_inodes_unused--; | |
430 | this_cpu_dec(nr_unused); | |
431 | } | |
432 | spin_unlock(&inode->i_sb->s_inode_lru_lock); | |
433 | } | |
434 | ||
435 | /** | |
436 | * inode_sb_list_add - add inode to the superblock list of inodes | |
437 | * @inode: inode to add | |
438 | */ | |
439 | void inode_sb_list_add(struct inode *inode) | |
440 | { | |
441 | spin_lock(&inode_sb_list_lock); | |
442 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); | |
443 | spin_unlock(&inode_sb_list_lock); | |
444 | } | |
445 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
446 | ||
447 | static inline void inode_sb_list_del(struct inode *inode) | |
448 | { | |
449 | if (!list_empty(&inode->i_sb_list)) { | |
450 | spin_lock(&inode_sb_list_lock); | |
451 | list_del_init(&inode->i_sb_list); | |
452 | spin_unlock(&inode_sb_list_lock); | |
453 | } | |
454 | } | |
455 | ||
456 | static unsigned long hash(struct super_block *sb, unsigned long hashval) | |
457 | { | |
458 | unsigned long tmp; | |
459 | ||
460 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
461 | L1_CACHE_BYTES; | |
462 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift); | |
463 | return tmp & i_hash_mask; | |
464 | } | |
465 | ||
466 | /** | |
467 | * __insert_inode_hash - hash an inode | |
468 | * @inode: unhashed inode | |
469 | * @hashval: unsigned long value used to locate this object in the | |
470 | * inode_hashtable. | |
471 | * | |
472 | * Add an inode to the inode hash for this superblock. | |
473 | */ | |
474 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
475 | { | |
476 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); | |
477 | ||
478 | spin_lock(&inode_hash_lock); | |
479 | spin_lock(&inode->i_lock); | |
480 | hlist_add_head(&inode->i_hash, b); | |
481 | spin_unlock(&inode->i_lock); | |
482 | spin_unlock(&inode_hash_lock); | |
483 | } | |
484 | EXPORT_SYMBOL(__insert_inode_hash); | |
485 | ||
486 | /** | |
487 | * __remove_inode_hash - remove an inode from the hash | |
488 | * @inode: inode to unhash | |
489 | * | |
490 | * Remove an inode from the superblock. | |
491 | */ | |
492 | void __remove_inode_hash(struct inode *inode) | |
493 | { | |
494 | spin_lock(&inode_hash_lock); | |
495 | spin_lock(&inode->i_lock); | |
496 | hlist_del_init(&inode->i_hash); | |
497 | spin_unlock(&inode->i_lock); | |
498 | spin_unlock(&inode_hash_lock); | |
499 | } | |
500 | EXPORT_SYMBOL(__remove_inode_hash); | |
501 | ||
502 | void clear_inode(struct inode *inode) | |
503 | { | |
504 | might_sleep(); | |
505 | /* | |
506 | * We have to cycle tree_lock here because reclaim can be still in the | |
507 | * process of removing the last page (in __delete_from_page_cache()) | |
508 | * and we must not free mapping under it. | |
509 | */ | |
510 | spin_lock_irq(&inode->i_data.tree_lock); | |
511 | BUG_ON(inode->i_data.nrpages); | |
512 | spin_unlock_irq(&inode->i_data.tree_lock); | |
513 | BUG_ON(!list_empty(&inode->i_data.private_list)); | |
514 | BUG_ON(!(inode->i_state & I_FREEING)); | |
515 | BUG_ON(inode->i_state & I_CLEAR); | |
516 | /* don't need i_lock here, no concurrent mods to i_state */ | |
517 | inode->i_state = I_FREEING | I_CLEAR; | |
518 | } | |
519 | EXPORT_SYMBOL(clear_inode); | |
520 | ||
521 | /* | |
522 | * Free the inode passed in, removing it from the lists it is still connected | |
523 | * to. We remove any pages still attached to the inode and wait for any IO that | |
524 | * is still in progress before finally destroying the inode. | |
525 | * | |
526 | * An inode must already be marked I_FREEING so that we avoid the inode being | |
527 | * moved back onto lists if we race with other code that manipulates the lists | |
528 | * (e.g. writeback_single_inode). The caller is responsible for setting this. | |
529 | * | |
530 | * An inode must already be removed from the LRU list before being evicted from | |
531 | * the cache. This should occur atomically with setting the I_FREEING state | |
532 | * flag, so no inodes here should ever be on the LRU when being evicted. | |
533 | */ | |
534 | static void evict(struct inode *inode) | |
535 | { | |
536 | const struct super_operations *op = inode->i_sb->s_op; | |
537 | ||
538 | BUG_ON(!(inode->i_state & I_FREEING)); | |
539 | BUG_ON(!list_empty(&inode->i_lru)); | |
540 | ||
541 | if (!list_empty(&inode->i_wb_list)) | |
542 | inode_wb_list_del(inode); | |
543 | ||
544 | inode_sb_list_del(inode); | |
545 | ||
546 | /* | |
547 | * Wait for flusher thread to be done with the inode so that filesystem | |
548 | * does not start destroying it while writeback is still running. Since | |
549 | * the inode has I_FREEING set, flusher thread won't start new work on | |
550 | * the inode. We just have to wait for running writeback to finish. | |
551 | */ | |
552 | inode_wait_for_writeback(inode); | |
553 | ||
554 | if (op->evict_inode) { | |
555 | op->evict_inode(inode); | |
556 | } else { | |
557 | if (inode->i_data.nrpages) | |
558 | truncate_inode_pages(&inode->i_data, 0); | |
559 | clear_inode(inode); | |
560 | } | |
561 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) | |
562 | bd_forget(inode); | |
563 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
564 | cd_forget(inode); | |
565 | ||
566 | remove_inode_hash(inode); | |
567 | ||
568 | spin_lock(&inode->i_lock); | |
569 | wake_up_bit(&inode->i_state, __I_NEW); | |
570 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); | |
571 | spin_unlock(&inode->i_lock); | |
572 | ||
573 | destroy_inode(inode); | |
574 | } | |
575 | ||
576 | /* | |
577 | * dispose_list - dispose of the contents of a local list | |
578 | * @head: the head of the list to free | |
579 | * | |
580 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
581 | * need to worry about list corruption and SMP locks. | |
582 | */ | |
583 | static void dispose_list(struct list_head *head) | |
584 | { | |
585 | while (!list_empty(head)) { | |
586 | struct inode *inode; | |
587 | ||
588 | inode = list_first_entry(head, struct inode, i_lru); | |
589 | list_del_init(&inode->i_lru); | |
590 | ||
591 | evict(inode); | |
592 | } | |
593 | } | |
594 | ||
595 | /** | |
596 | * evict_inodes - evict all evictable inodes for a superblock | |
597 | * @sb: superblock to operate on | |
598 | * | |
599 | * Make sure that no inodes with zero refcount are retained. This is | |
600 | * called by superblock shutdown after having MS_ACTIVE flag removed, | |
601 | * so any inode reaching zero refcount during or after that call will | |
602 | * be immediately evicted. | |
603 | */ | |
604 | void evict_inodes(struct super_block *sb) | |
605 | { | |
606 | struct inode *inode, *next; | |
607 | LIST_HEAD(dispose); | |
608 | ||
609 | spin_lock(&inode_sb_list_lock); | |
610 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { | |
611 | if (atomic_read(&inode->i_count)) | |
612 | continue; | |
613 | ||
614 | spin_lock(&inode->i_lock); | |
615 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
616 | spin_unlock(&inode->i_lock); | |
617 | continue; | |
618 | } | |
619 | ||
620 | inode->i_state |= I_FREEING; | |
621 | inode_lru_list_del(inode); | |
622 | spin_unlock(&inode->i_lock); | |
623 | list_add(&inode->i_lru, &dispose); | |
624 | } | |
625 | spin_unlock(&inode_sb_list_lock); | |
626 | ||
627 | dispose_list(&dispose); | |
628 | } | |
629 | ||
630 | /** | |
631 | * invalidate_inodes - attempt to free all inodes on a superblock | |
632 | * @sb: superblock to operate on | |
633 | * @kill_dirty: flag to guide handling of dirty inodes | |
634 | * | |
635 | * Attempts to free all inodes for a given superblock. If there were any | |
636 | * busy inodes return a non-zero value, else zero. | |
637 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat | |
638 | * them as busy. | |
639 | */ | |
640 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) | |
641 | { | |
642 | int busy = 0; | |
643 | struct inode *inode, *next; | |
644 | LIST_HEAD(dispose); | |
645 | ||
646 | spin_lock(&inode_sb_list_lock); | |
647 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { | |
648 | spin_lock(&inode->i_lock); | |
649 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
650 | spin_unlock(&inode->i_lock); | |
651 | continue; | |
652 | } | |
653 | if (inode->i_state & I_DIRTY && !kill_dirty) { | |
654 | spin_unlock(&inode->i_lock); | |
655 | busy = 1; | |
656 | continue; | |
657 | } | |
658 | if (atomic_read(&inode->i_count)) { | |
659 | spin_unlock(&inode->i_lock); | |
660 | busy = 1; | |
661 | continue; | |
662 | } | |
663 | ||
664 | inode->i_state |= I_FREEING; | |
665 | inode_lru_list_del(inode); | |
666 | spin_unlock(&inode->i_lock); | |
667 | list_add(&inode->i_lru, &dispose); | |
668 | } | |
669 | spin_unlock(&inode_sb_list_lock); | |
670 | ||
671 | dispose_list(&dispose); | |
672 | ||
673 | return busy; | |
674 | } | |
675 | ||
676 | static int can_unuse(struct inode *inode) | |
677 | { | |
678 | if (inode->i_state & ~I_REFERENCED) | |
679 | return 0; | |
680 | if (inode_has_buffers(inode)) | |
681 | return 0; | |
682 | if (atomic_read(&inode->i_count)) | |
683 | return 0; | |
684 | if (inode->i_data.nrpages) | |
685 | return 0; | |
686 | return 1; | |
687 | } | |
688 | ||
689 | /* | |
690 | * Walk the superblock inode LRU for freeable inodes and attempt to free them. | |
691 | * This is called from the superblock shrinker function with a number of inodes | |
692 | * to trim from the LRU. Inodes to be freed are moved to a temporary list and | |
693 | * then are freed outside inode_lock by dispose_list(). | |
694 | * | |
695 | * Any inodes which are pinned purely because of attached pagecache have their | |
696 | * pagecache removed. If the inode has metadata buffers attached to | |
697 | * mapping->private_list then try to remove them. | |
698 | * | |
699 | * If the inode has the I_REFERENCED flag set, then it means that it has been | |
700 | * used recently - the flag is set in iput_final(). When we encounter such an | |
701 | * inode, clear the flag and move it to the back of the LRU so it gets another | |
702 | * pass through the LRU before it gets reclaimed. This is necessary because of | |
703 | * the fact we are doing lazy LRU updates to minimise lock contention so the | |
704 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes | |
705 | * with this flag set because they are the inodes that are out of order. | |
706 | */ | |
707 | void prune_icache_sb(struct super_block *sb, int nr_to_scan) | |
708 | { | |
709 | LIST_HEAD(freeable); | |
710 | int nr_scanned; | |
711 | unsigned long reap = 0; | |
712 | ||
713 | spin_lock(&sb->s_inode_lru_lock); | |
714 | for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) { | |
715 | struct inode *inode; | |
716 | ||
717 | if (list_empty(&sb->s_inode_lru)) | |
718 | break; | |
719 | ||
720 | inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru); | |
721 | ||
722 | /* | |
723 | * we are inverting the sb->s_inode_lru_lock/inode->i_lock here, | |
724 | * so use a trylock. If we fail to get the lock, just move the | |
725 | * inode to the back of the list so we don't spin on it. | |
726 | */ | |
727 | if (!spin_trylock(&inode->i_lock)) { | |
728 | list_move_tail(&inode->i_lru, &sb->s_inode_lru); | |
729 | continue; | |
730 | } | |
731 | ||
732 | /* | |
733 | * Referenced or dirty inodes are still in use. Give them | |
734 | * another pass through the LRU as we canot reclaim them now. | |
735 | */ | |
736 | if (atomic_read(&inode->i_count) || | |
737 | (inode->i_state & ~I_REFERENCED)) { | |
738 | list_del_init(&inode->i_lru); | |
739 | spin_unlock(&inode->i_lock); | |
740 | sb->s_nr_inodes_unused--; | |
741 | this_cpu_dec(nr_unused); | |
742 | continue; | |
743 | } | |
744 | ||
745 | /* recently referenced inodes get one more pass */ | |
746 | if (inode->i_state & I_REFERENCED) { | |
747 | inode->i_state &= ~I_REFERENCED; | |
748 | list_move(&inode->i_lru, &sb->s_inode_lru); | |
749 | spin_unlock(&inode->i_lock); | |
750 | continue; | |
751 | } | |
752 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
753 | __iget(inode); | |
754 | spin_unlock(&inode->i_lock); | |
755 | spin_unlock(&sb->s_inode_lru_lock); | |
756 | if (remove_inode_buffers(inode)) | |
757 | reap += invalidate_mapping_pages(&inode->i_data, | |
758 | 0, -1); | |
759 | iput(inode); | |
760 | spin_lock(&sb->s_inode_lru_lock); | |
761 | ||
762 | if (inode != list_entry(sb->s_inode_lru.next, | |
763 | struct inode, i_lru)) | |
764 | continue; /* wrong inode or list_empty */ | |
765 | /* avoid lock inversions with trylock */ | |
766 | if (!spin_trylock(&inode->i_lock)) | |
767 | continue; | |
768 | if (!can_unuse(inode)) { | |
769 | spin_unlock(&inode->i_lock); | |
770 | continue; | |
771 | } | |
772 | } | |
773 | WARN_ON(inode->i_state & I_NEW); | |
774 | inode->i_state |= I_FREEING; | |
775 | spin_unlock(&inode->i_lock); | |
776 | ||
777 | list_move(&inode->i_lru, &freeable); | |
778 | sb->s_nr_inodes_unused--; | |
779 | this_cpu_dec(nr_unused); | |
780 | } | |
781 | if (current_is_kswapd()) | |
782 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
783 | else | |
784 | __count_vm_events(PGINODESTEAL, reap); | |
785 | spin_unlock(&sb->s_inode_lru_lock); | |
786 | if (current->reclaim_state) | |
787 | current->reclaim_state->reclaimed_slab += reap; | |
788 | ||
789 | dispose_list(&freeable); | |
790 | } | |
791 | ||
792 | static void __wait_on_freeing_inode(struct inode *inode); | |
793 | /* | |
794 | * Called with the inode lock held. | |
795 | */ | |
796 | static struct inode *find_inode(struct super_block *sb, | |
797 | struct hlist_head *head, | |
798 | int (*test)(struct inode *, void *), | |
799 | void *data) | |
800 | { | |
801 | struct hlist_node *node; | |
802 | struct inode *inode = NULL; | |
803 | ||
804 | repeat: | |
805 | hlist_for_each_entry(inode, node, head, i_hash) { | |
806 | spin_lock(&inode->i_lock); | |
807 | if (inode->i_sb != sb) { | |
808 | spin_unlock(&inode->i_lock); | |
809 | continue; | |
810 | } | |
811 | if (!test(inode, data)) { | |
812 | spin_unlock(&inode->i_lock); | |
813 | continue; | |
814 | } | |
815 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { | |
816 | __wait_on_freeing_inode(inode); | |
817 | goto repeat; | |
818 | } | |
819 | __iget(inode); | |
820 | spin_unlock(&inode->i_lock); | |
821 | return inode; | |
822 | } | |
823 | return NULL; | |
824 | } | |
825 | ||
826 | /* | |
827 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
828 | * iget_locked for details. | |
829 | */ | |
830 | static struct inode *find_inode_fast(struct super_block *sb, | |
831 | struct hlist_head *head, unsigned long ino) | |
832 | { | |
833 | struct hlist_node *node; | |
834 | struct inode *inode = NULL; | |
835 | ||
836 | repeat: | |
837 | hlist_for_each_entry(inode, node, head, i_hash) { | |
838 | spin_lock(&inode->i_lock); | |
839 | if (inode->i_ino != ino) { | |
840 | spin_unlock(&inode->i_lock); | |
841 | continue; | |
842 | } | |
843 | if (inode->i_sb != sb) { | |
844 | spin_unlock(&inode->i_lock); | |
845 | continue; | |
846 | } | |
847 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { | |
848 | __wait_on_freeing_inode(inode); | |
849 | goto repeat; | |
850 | } | |
851 | __iget(inode); | |
852 | spin_unlock(&inode->i_lock); | |
853 | return inode; | |
854 | } | |
855 | return NULL; | |
856 | } | |
857 | ||
858 | /* | |
859 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
860 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
861 | * to renew the exhausted range. | |
862 | * | |
863 | * This does not significantly increase overflow rate because every CPU can | |
864 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
865 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
866 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
867 | * overflow rate by 2x, which does not seem too significant. | |
868 | * | |
869 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
870 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
871 | * here to attempt to avoid that. | |
872 | */ | |
873 | #define LAST_INO_BATCH 1024 | |
874 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
875 | ||
876 | unsigned int get_next_ino(void) | |
877 | { | |
878 | unsigned int *p = &get_cpu_var(last_ino); | |
879 | unsigned int res = *p; | |
880 | ||
881 | #ifdef CONFIG_SMP | |
882 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
883 | static atomic_t shared_last_ino; | |
884 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
885 | ||
886 | res = next - LAST_INO_BATCH; | |
887 | } | |
888 | #endif | |
889 | ||
890 | *p = ++res; | |
891 | put_cpu_var(last_ino); | |
892 | return res; | |
893 | } | |
894 | EXPORT_SYMBOL(get_next_ino); | |
895 | ||
896 | /** | |
897 | * new_inode_pseudo - obtain an inode | |
898 | * @sb: superblock | |
899 | * | |
900 | * Allocates a new inode for given superblock. | |
901 | * Inode wont be chained in superblock s_inodes list | |
902 | * This means : | |
903 | * - fs can't be unmount | |
904 | * - quotas, fsnotify, writeback can't work | |
905 | */ | |
906 | struct inode *new_inode_pseudo(struct super_block *sb) | |
907 | { | |
908 | struct inode *inode = alloc_inode(sb); | |
909 | ||
910 | if (inode) { | |
911 | spin_lock(&inode->i_lock); | |
912 | inode->i_state = 0; | |
913 | spin_unlock(&inode->i_lock); | |
914 | INIT_LIST_HEAD(&inode->i_sb_list); | |
915 | } | |
916 | return inode; | |
917 | } | |
918 | ||
919 | /** | |
920 | * new_inode - obtain an inode | |
921 | * @sb: superblock | |
922 | * | |
923 | * Allocates a new inode for given superblock. The default gfp_mask | |
924 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. | |
925 | * If HIGHMEM pages are unsuitable or it is known that pages allocated | |
926 | * for the page cache are not reclaimable or migratable, | |
927 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
928 | * newly created inode's mapping | |
929 | * | |
930 | */ | |
931 | struct inode *new_inode(struct super_block *sb) | |
932 | { | |
933 | struct inode *inode; | |
934 | ||
935 | spin_lock_prefetch(&inode_sb_list_lock); | |
936 | ||
937 | inode = new_inode_pseudo(sb); | |
938 | if (inode) | |
939 | inode_sb_list_add(inode); | |
940 | return inode; | |
941 | } | |
942 | EXPORT_SYMBOL(new_inode); | |
943 | ||
944 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
945 | void lockdep_annotate_inode_mutex_key(struct inode *inode) | |
946 | { | |
947 | if (S_ISDIR(inode->i_mode)) { | |
948 | struct file_system_type *type = inode->i_sb->s_type; | |
949 | ||
950 | /* Set new key only if filesystem hasn't already changed it */ | |
951 | if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) { | |
952 | /* | |
953 | * ensure nobody is actually holding i_mutex | |
954 | */ | |
955 | mutex_destroy(&inode->i_mutex); | |
956 | mutex_init(&inode->i_mutex); | |
957 | lockdep_set_class(&inode->i_mutex, | |
958 | &type->i_mutex_dir_key); | |
959 | } | |
960 | } | |
961 | } | |
962 | EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key); | |
963 | #endif | |
964 | ||
965 | /** | |
966 | * unlock_new_inode - clear the I_NEW state and wake up any waiters | |
967 | * @inode: new inode to unlock | |
968 | * | |
969 | * Called when the inode is fully initialised to clear the new state of the | |
970 | * inode and wake up anyone waiting for the inode to finish initialisation. | |
971 | */ | |
972 | void unlock_new_inode(struct inode *inode) | |
973 | { | |
974 | lockdep_annotate_inode_mutex_key(inode); | |
975 | spin_lock(&inode->i_lock); | |
976 | WARN_ON(!(inode->i_state & I_NEW)); | |
977 | inode->i_state &= ~I_NEW; | |
978 | smp_mb(); | |
979 | wake_up_bit(&inode->i_state, __I_NEW); | |
980 | spin_unlock(&inode->i_lock); | |
981 | } | |
982 | EXPORT_SYMBOL(unlock_new_inode); | |
983 | ||
984 | /** | |
985 | * iget5_locked - obtain an inode from a mounted file system | |
986 | * @sb: super block of file system | |
987 | * @hashval: hash value (usually inode number) to get | |
988 | * @test: callback used for comparisons between inodes | |
989 | * @set: callback used to initialize a new struct inode | |
990 | * @data: opaque data pointer to pass to @test and @set | |
991 | * | |
992 | * Search for the inode specified by @hashval and @data in the inode cache, | |
993 | * and if present it is return it with an increased reference count. This is | |
994 | * a generalized version of iget_locked() for file systems where the inode | |
995 | * number is not sufficient for unique identification of an inode. | |
996 | * | |
997 | * If the inode is not in cache, allocate a new inode and return it locked, | |
998 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
999 | * before unlocking it via unlock_new_inode(). | |
1000 | * | |
1001 | * Note both @test and @set are called with the inode_hash_lock held, so can't | |
1002 | * sleep. | |
1003 | */ | |
1004 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
1005 | int (*test)(struct inode *, void *), | |
1006 | int (*set)(struct inode *, void *), void *data) | |
1007 | { | |
1008 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1009 | struct inode *inode; | |
1010 | ||
1011 | spin_lock(&inode_hash_lock); | |
1012 | inode = find_inode(sb, head, test, data); | |
1013 | spin_unlock(&inode_hash_lock); | |
1014 | ||
1015 | if (inode) { | |
1016 | wait_on_inode(inode); | |
1017 | return inode; | |
1018 | } | |
1019 | ||
1020 | inode = alloc_inode(sb); | |
1021 | if (inode) { | |
1022 | struct inode *old; | |
1023 | ||
1024 | spin_lock(&inode_hash_lock); | |
1025 | /* We released the lock, so.. */ | |
1026 | old = find_inode(sb, head, test, data); | |
1027 | if (!old) { | |
1028 | if (set(inode, data)) | |
1029 | goto set_failed; | |
1030 | ||
1031 | spin_lock(&inode->i_lock); | |
1032 | inode->i_state = I_NEW; | |
1033 | hlist_add_head(&inode->i_hash, head); | |
1034 | spin_unlock(&inode->i_lock); | |
1035 | inode_sb_list_add(inode); | |
1036 | spin_unlock(&inode_hash_lock); | |
1037 | ||
1038 | /* Return the locked inode with I_NEW set, the | |
1039 | * caller is responsible for filling in the contents | |
1040 | */ | |
1041 | return inode; | |
1042 | } | |
1043 | ||
1044 | /* | |
1045 | * Uhhuh, somebody else created the same inode under | |
1046 | * us. Use the old inode instead of the one we just | |
1047 | * allocated. | |
1048 | */ | |
1049 | spin_unlock(&inode_hash_lock); | |
1050 | destroy_inode(inode); | |
1051 | inode = old; | |
1052 | wait_on_inode(inode); | |
1053 | } | |
1054 | return inode; | |
1055 | ||
1056 | set_failed: | |
1057 | spin_unlock(&inode_hash_lock); | |
1058 | destroy_inode(inode); | |
1059 | return NULL; | |
1060 | } | |
1061 | EXPORT_SYMBOL(iget5_locked); | |
1062 | ||
1063 | /** | |
1064 | * iget_locked - obtain an inode from a mounted file system | |
1065 | * @sb: super block of file system | |
1066 | * @ino: inode number to get | |
1067 | * | |
1068 | * Search for the inode specified by @ino in the inode cache and if present | |
1069 | * return it with an increased reference count. This is for file systems | |
1070 | * where the inode number is sufficient for unique identification of an inode. | |
1071 | * | |
1072 | * If the inode is not in cache, allocate a new inode and return it locked, | |
1073 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
1074 | * before unlocking it via unlock_new_inode(). | |
1075 | */ | |
1076 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) | |
1077 | { | |
1078 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1079 | struct inode *inode; | |
1080 | ||
1081 | spin_lock(&inode_hash_lock); | |
1082 | inode = find_inode_fast(sb, head, ino); | |
1083 | spin_unlock(&inode_hash_lock); | |
1084 | if (inode) { | |
1085 | wait_on_inode(inode); | |
1086 | return inode; | |
1087 | } | |
1088 | ||
1089 | inode = alloc_inode(sb); | |
1090 | if (inode) { | |
1091 | struct inode *old; | |
1092 | ||
1093 | spin_lock(&inode_hash_lock); | |
1094 | /* We released the lock, so.. */ | |
1095 | old = find_inode_fast(sb, head, ino); | |
1096 | if (!old) { | |
1097 | inode->i_ino = ino; | |
1098 | spin_lock(&inode->i_lock); | |
1099 | inode->i_state = I_NEW; | |
1100 | hlist_add_head(&inode->i_hash, head); | |
1101 | spin_unlock(&inode->i_lock); | |
1102 | inode_sb_list_add(inode); | |
1103 | spin_unlock(&inode_hash_lock); | |
1104 | ||
1105 | /* Return the locked inode with I_NEW set, the | |
1106 | * caller is responsible for filling in the contents | |
1107 | */ | |
1108 | return inode; | |
1109 | } | |
1110 | ||
1111 | /* | |
1112 | * Uhhuh, somebody else created the same inode under | |
1113 | * us. Use the old inode instead of the one we just | |
1114 | * allocated. | |
1115 | */ | |
1116 | spin_unlock(&inode_hash_lock); | |
1117 | destroy_inode(inode); | |
1118 | inode = old; | |
1119 | wait_on_inode(inode); | |
1120 | } | |
1121 | return inode; | |
1122 | } | |
1123 | EXPORT_SYMBOL(iget_locked); | |
1124 | ||
1125 | /* | |
1126 | * search the inode cache for a matching inode number. | |
1127 | * If we find one, then the inode number we are trying to | |
1128 | * allocate is not unique and so we should not use it. | |
1129 | * | |
1130 | * Returns 1 if the inode number is unique, 0 if it is not. | |
1131 | */ | |
1132 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
1133 | { | |
1134 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
1135 | struct hlist_node *node; | |
1136 | struct inode *inode; | |
1137 | ||
1138 | spin_lock(&inode_hash_lock); | |
1139 | hlist_for_each_entry(inode, node, b, i_hash) { | |
1140 | if (inode->i_ino == ino && inode->i_sb == sb) { | |
1141 | spin_unlock(&inode_hash_lock); | |
1142 | return 0; | |
1143 | } | |
1144 | } | |
1145 | spin_unlock(&inode_hash_lock); | |
1146 | ||
1147 | return 1; | |
1148 | } | |
1149 | ||
1150 | /** | |
1151 | * iunique - get a unique inode number | |
1152 | * @sb: superblock | |
1153 | * @max_reserved: highest reserved inode number | |
1154 | * | |
1155 | * Obtain an inode number that is unique on the system for a given | |
1156 | * superblock. This is used by file systems that have no natural | |
1157 | * permanent inode numbering system. An inode number is returned that | |
1158 | * is higher than the reserved limit but unique. | |
1159 | * | |
1160 | * BUGS: | |
1161 | * With a large number of inodes live on the file system this function | |
1162 | * currently becomes quite slow. | |
1163 | */ | |
1164 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
1165 | { | |
1166 | /* | |
1167 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
1168 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
1169 | * here to attempt to avoid that. | |
1170 | */ | |
1171 | static DEFINE_SPINLOCK(iunique_lock); | |
1172 | static unsigned int counter; | |
1173 | ino_t res; | |
1174 | ||
1175 | spin_lock(&iunique_lock); | |
1176 | do { | |
1177 | if (counter <= max_reserved) | |
1178 | counter = max_reserved + 1; | |
1179 | res = counter++; | |
1180 | } while (!test_inode_iunique(sb, res)); | |
1181 | spin_unlock(&iunique_lock); | |
1182 | ||
1183 | return res; | |
1184 | } | |
1185 | EXPORT_SYMBOL(iunique); | |
1186 | ||
1187 | struct inode *igrab(struct inode *inode) | |
1188 | { | |
1189 | spin_lock(&inode->i_lock); | |
1190 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { | |
1191 | __iget(inode); | |
1192 | spin_unlock(&inode->i_lock); | |
1193 | } else { | |
1194 | spin_unlock(&inode->i_lock); | |
1195 | /* | |
1196 | * Handle the case where s_op->clear_inode is not been | |
1197 | * called yet, and somebody is calling igrab | |
1198 | * while the inode is getting freed. | |
1199 | */ | |
1200 | inode = NULL; | |
1201 | } | |
1202 | return inode; | |
1203 | } | |
1204 | EXPORT_SYMBOL(igrab); | |
1205 | ||
1206 | /** | |
1207 | * ilookup5_nowait - search for an inode in the inode cache | |
1208 | * @sb: super block of file system to search | |
1209 | * @hashval: hash value (usually inode number) to search for | |
1210 | * @test: callback used for comparisons between inodes | |
1211 | * @data: opaque data pointer to pass to @test | |
1212 | * | |
1213 | * Search for the inode specified by @hashval and @data in the inode cache. | |
1214 | * If the inode is in the cache, the inode is returned with an incremented | |
1215 | * reference count. | |
1216 | * | |
1217 | * Note: I_NEW is not waited upon so you have to be very careful what you do | |
1218 | * with the returned inode. You probably should be using ilookup5() instead. | |
1219 | * | |
1220 | * Note2: @test is called with the inode_hash_lock held, so can't sleep. | |
1221 | */ | |
1222 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
1223 | int (*test)(struct inode *, void *), void *data) | |
1224 | { | |
1225 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1226 | struct inode *inode; | |
1227 | ||
1228 | spin_lock(&inode_hash_lock); | |
1229 | inode = find_inode(sb, head, test, data); | |
1230 | spin_unlock(&inode_hash_lock); | |
1231 | ||
1232 | return inode; | |
1233 | } | |
1234 | EXPORT_SYMBOL(ilookup5_nowait); | |
1235 | ||
1236 | /** | |
1237 | * ilookup5 - search for an inode in the inode cache | |
1238 | * @sb: super block of file system to search | |
1239 | * @hashval: hash value (usually inode number) to search for | |
1240 | * @test: callback used for comparisons between inodes | |
1241 | * @data: opaque data pointer to pass to @test | |
1242 | * | |
1243 | * Search for the inode specified by @hashval and @data in the inode cache, | |
1244 | * and if the inode is in the cache, return the inode with an incremented | |
1245 | * reference count. Waits on I_NEW before returning the inode. | |
1246 | * returned with an incremented reference count. | |
1247 | * | |
1248 | * This is a generalized version of ilookup() for file systems where the | |
1249 | * inode number is not sufficient for unique identification of an inode. | |
1250 | * | |
1251 | * Note: @test is called with the inode_hash_lock held, so can't sleep. | |
1252 | */ | |
1253 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1254 | int (*test)(struct inode *, void *), void *data) | |
1255 | { | |
1256 | struct inode *inode = ilookup5_nowait(sb, hashval, test, data); | |
1257 | ||
1258 | if (inode) | |
1259 | wait_on_inode(inode); | |
1260 | return inode; | |
1261 | } | |
1262 | EXPORT_SYMBOL(ilookup5); | |
1263 | ||
1264 | /** | |
1265 | * ilookup - search for an inode in the inode cache | |
1266 | * @sb: super block of file system to search | |
1267 | * @ino: inode number to search for | |
1268 | * | |
1269 | * Search for the inode @ino in the inode cache, and if the inode is in the | |
1270 | * cache, the inode is returned with an incremented reference count. | |
1271 | */ | |
1272 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1273 | { | |
1274 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1275 | struct inode *inode; | |
1276 | ||
1277 | spin_lock(&inode_hash_lock); | |
1278 | inode = find_inode_fast(sb, head, ino); | |
1279 | spin_unlock(&inode_hash_lock); | |
1280 | ||
1281 | if (inode) | |
1282 | wait_on_inode(inode); | |
1283 | return inode; | |
1284 | } | |
1285 | EXPORT_SYMBOL(ilookup); | |
1286 | ||
1287 | int insert_inode_locked(struct inode *inode) | |
1288 | { | |
1289 | struct super_block *sb = inode->i_sb; | |
1290 | ino_t ino = inode->i_ino; | |
1291 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1292 | ||
1293 | while (1) { | |
1294 | struct hlist_node *node; | |
1295 | struct inode *old = NULL; | |
1296 | spin_lock(&inode_hash_lock); | |
1297 | hlist_for_each_entry(old, node, head, i_hash) { | |
1298 | if (old->i_ino != ino) | |
1299 | continue; | |
1300 | if (old->i_sb != sb) | |
1301 | continue; | |
1302 | spin_lock(&old->i_lock); | |
1303 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1304 | spin_unlock(&old->i_lock); | |
1305 | continue; | |
1306 | } | |
1307 | break; | |
1308 | } | |
1309 | if (likely(!node)) { | |
1310 | spin_lock(&inode->i_lock); | |
1311 | inode->i_state |= I_NEW; | |
1312 | hlist_add_head(&inode->i_hash, head); | |
1313 | spin_unlock(&inode->i_lock); | |
1314 | spin_unlock(&inode_hash_lock); | |
1315 | return 0; | |
1316 | } | |
1317 | __iget(old); | |
1318 | spin_unlock(&old->i_lock); | |
1319 | spin_unlock(&inode_hash_lock); | |
1320 | wait_on_inode(old); | |
1321 | if (unlikely(!inode_unhashed(old))) { | |
1322 | iput(old); | |
1323 | return -EBUSY; | |
1324 | } | |
1325 | iput(old); | |
1326 | } | |
1327 | } | |
1328 | EXPORT_SYMBOL(insert_inode_locked); | |
1329 | ||
1330 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1331 | int (*test)(struct inode *, void *), void *data) | |
1332 | { | |
1333 | struct super_block *sb = inode->i_sb; | |
1334 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1335 | ||
1336 | while (1) { | |
1337 | struct hlist_node *node; | |
1338 | struct inode *old = NULL; | |
1339 | ||
1340 | spin_lock(&inode_hash_lock); | |
1341 | hlist_for_each_entry(old, node, head, i_hash) { | |
1342 | if (old->i_sb != sb) | |
1343 | continue; | |
1344 | if (!test(old, data)) | |
1345 | continue; | |
1346 | spin_lock(&old->i_lock); | |
1347 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1348 | spin_unlock(&old->i_lock); | |
1349 | continue; | |
1350 | } | |
1351 | break; | |
1352 | } | |
1353 | if (likely(!node)) { | |
1354 | spin_lock(&inode->i_lock); | |
1355 | inode->i_state |= I_NEW; | |
1356 | hlist_add_head(&inode->i_hash, head); | |
1357 | spin_unlock(&inode->i_lock); | |
1358 | spin_unlock(&inode_hash_lock); | |
1359 | return 0; | |
1360 | } | |
1361 | __iget(old); | |
1362 | spin_unlock(&old->i_lock); | |
1363 | spin_unlock(&inode_hash_lock); | |
1364 | wait_on_inode(old); | |
1365 | if (unlikely(!inode_unhashed(old))) { | |
1366 | iput(old); | |
1367 | return -EBUSY; | |
1368 | } | |
1369 | iput(old); | |
1370 | } | |
1371 | } | |
1372 | EXPORT_SYMBOL(insert_inode_locked4); | |
1373 | ||
1374 | ||
1375 | int generic_delete_inode(struct inode *inode) | |
1376 | { | |
1377 | return 1; | |
1378 | } | |
1379 | EXPORT_SYMBOL(generic_delete_inode); | |
1380 | ||
1381 | /* | |
1382 | * Called when we're dropping the last reference | |
1383 | * to an inode. | |
1384 | * | |
1385 | * Call the FS "drop_inode()" function, defaulting to | |
1386 | * the legacy UNIX filesystem behaviour. If it tells | |
1387 | * us to evict inode, do so. Otherwise, retain inode | |
1388 | * in cache if fs is alive, sync and evict if fs is | |
1389 | * shutting down. | |
1390 | */ | |
1391 | static void iput_final(struct inode *inode) | |
1392 | { | |
1393 | struct super_block *sb = inode->i_sb; | |
1394 | const struct super_operations *op = inode->i_sb->s_op; | |
1395 | int drop; | |
1396 | ||
1397 | WARN_ON(inode->i_state & I_NEW); | |
1398 | ||
1399 | if (op->drop_inode) | |
1400 | drop = op->drop_inode(inode); | |
1401 | else | |
1402 | drop = generic_drop_inode(inode); | |
1403 | ||
1404 | if (!drop && (sb->s_flags & MS_ACTIVE)) { | |
1405 | inode->i_state |= I_REFERENCED; | |
1406 | inode_add_lru(inode); | |
1407 | spin_unlock(&inode->i_lock); | |
1408 | return; | |
1409 | } | |
1410 | ||
1411 | if (!drop) { | |
1412 | inode->i_state |= I_WILL_FREE; | |
1413 | spin_unlock(&inode->i_lock); | |
1414 | write_inode_now(inode, 1); | |
1415 | spin_lock(&inode->i_lock); | |
1416 | WARN_ON(inode->i_state & I_NEW); | |
1417 | inode->i_state &= ~I_WILL_FREE; | |
1418 | } | |
1419 | ||
1420 | inode->i_state |= I_FREEING; | |
1421 | if (!list_empty(&inode->i_lru)) | |
1422 | inode_lru_list_del(inode); | |
1423 | spin_unlock(&inode->i_lock); | |
1424 | ||
1425 | evict(inode); | |
1426 | } | |
1427 | ||
1428 | /** | |
1429 | * iput - put an inode | |
1430 | * @inode: inode to put | |
1431 | * | |
1432 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1433 | * zero, the inode is then freed and may also be destroyed. | |
1434 | * | |
1435 | * Consequently, iput() can sleep. | |
1436 | */ | |
1437 | void iput(struct inode *inode) | |
1438 | { | |
1439 | if (inode) { | |
1440 | BUG_ON(inode->i_state & I_CLEAR); | |
1441 | ||
1442 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) | |
1443 | iput_final(inode); | |
1444 | } | |
1445 | } | |
1446 | EXPORT_SYMBOL(iput); | |
1447 | ||
1448 | /** | |
1449 | * bmap - find a block number in a file | |
1450 | * @inode: inode of file | |
1451 | * @block: block to find | |
1452 | * | |
1453 | * Returns the block number on the device holding the inode that | |
1454 | * is the disk block number for the block of the file requested. | |
1455 | * That is, asked for block 4 of inode 1 the function will return the | |
1456 | * disk block relative to the disk start that holds that block of the | |
1457 | * file. | |
1458 | */ | |
1459 | sector_t bmap(struct inode *inode, sector_t block) | |
1460 | { | |
1461 | sector_t res = 0; | |
1462 | if (inode->i_mapping->a_ops->bmap) | |
1463 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1464 | return res; | |
1465 | } | |
1466 | EXPORT_SYMBOL(bmap); | |
1467 | ||
1468 | /* | |
1469 | * With relative atime, only update atime if the previous atime is | |
1470 | * earlier than either the ctime or mtime or if at least a day has | |
1471 | * passed since the last atime update. | |
1472 | */ | |
1473 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1474 | struct timespec now) | |
1475 | { | |
1476 | ||
1477 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1478 | return 1; | |
1479 | /* | |
1480 | * Is mtime younger than atime? If yes, update atime: | |
1481 | */ | |
1482 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1483 | return 1; | |
1484 | /* | |
1485 | * Is ctime younger than atime? If yes, update atime: | |
1486 | */ | |
1487 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1488 | return 1; | |
1489 | ||
1490 | /* | |
1491 | * Is the previous atime value older than a day? If yes, | |
1492 | * update atime: | |
1493 | */ | |
1494 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1495 | return 1; | |
1496 | /* | |
1497 | * Good, we can skip the atime update: | |
1498 | */ | |
1499 | return 0; | |
1500 | } | |
1501 | ||
1502 | /* | |
1503 | * This does the actual work of updating an inodes time or version. Must have | |
1504 | * had called mnt_want_write() before calling this. | |
1505 | */ | |
1506 | static int update_time(struct inode *inode, struct timespec *time, int flags) | |
1507 | { | |
1508 | if (inode->i_op->update_time) | |
1509 | return inode->i_op->update_time(inode, time, flags); | |
1510 | ||
1511 | if (flags & S_ATIME) | |
1512 | inode->i_atime = *time; | |
1513 | if (flags & S_VERSION) | |
1514 | inode_inc_iversion(inode); | |
1515 | if (flags & S_CTIME) | |
1516 | inode->i_ctime = *time; | |
1517 | if (flags & S_MTIME) | |
1518 | inode->i_mtime = *time; | |
1519 | mark_inode_dirty_sync(inode); | |
1520 | return 0; | |
1521 | } | |
1522 | ||
1523 | /** | |
1524 | * touch_atime - update the access time | |
1525 | * @path: the &struct path to update | |
1526 | * | |
1527 | * Update the accessed time on an inode and mark it for writeback. | |
1528 | * This function automatically handles read only file systems and media, | |
1529 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1530 | */ | |
1531 | void touch_atime(struct path *path) | |
1532 | { | |
1533 | struct vfsmount *mnt = path->mnt; | |
1534 | struct inode *inode = path->dentry->d_inode; | |
1535 | struct timespec now; | |
1536 | ||
1537 | if (inode->i_flags & S_NOATIME) | |
1538 | return; | |
1539 | if (IS_NOATIME(inode)) | |
1540 | return; | |
1541 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1542 | return; | |
1543 | ||
1544 | if (mnt->mnt_flags & MNT_NOATIME) | |
1545 | return; | |
1546 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1547 | return; | |
1548 | ||
1549 | now = current_fs_time(inode->i_sb); | |
1550 | ||
1551 | if (!relatime_need_update(mnt, inode, now)) | |
1552 | return; | |
1553 | ||
1554 | if (timespec_equal(&inode->i_atime, &now)) | |
1555 | return; | |
1556 | ||
1557 | if (!sb_start_write_trylock(inode->i_sb)) | |
1558 | return; | |
1559 | ||
1560 | if (__mnt_want_write(mnt)) | |
1561 | goto skip_update; | |
1562 | /* | |
1563 | * File systems can error out when updating inodes if they need to | |
1564 | * allocate new space to modify an inode (such is the case for | |
1565 | * Btrfs), but since we touch atime while walking down the path we | |
1566 | * really don't care if we failed to update the atime of the file, | |
1567 | * so just ignore the return value. | |
1568 | * We may also fail on filesystems that have the ability to make parts | |
1569 | * of the fs read only, e.g. subvolumes in Btrfs. | |
1570 | */ | |
1571 | update_time(inode, &now, S_ATIME); | |
1572 | __mnt_drop_write(mnt); | |
1573 | skip_update: | |
1574 | sb_end_write(inode->i_sb); | |
1575 | } | |
1576 | EXPORT_SYMBOL(touch_atime); | |
1577 | ||
1578 | /* | |
1579 | * The logic we want is | |
1580 | * | |
1581 | * if suid or (sgid and xgrp) | |
1582 | * remove privs | |
1583 | */ | |
1584 | int should_remove_suid(struct dentry *dentry) | |
1585 | { | |
1586 | umode_t mode = dentry->d_inode->i_mode; | |
1587 | int kill = 0; | |
1588 | ||
1589 | /* suid always must be killed */ | |
1590 | if (unlikely(mode & S_ISUID)) | |
1591 | kill = ATTR_KILL_SUID; | |
1592 | ||
1593 | /* | |
1594 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1595 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1596 | */ | |
1597 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1598 | kill |= ATTR_KILL_SGID; | |
1599 | ||
1600 | if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) | |
1601 | return kill; | |
1602 | ||
1603 | return 0; | |
1604 | } | |
1605 | EXPORT_SYMBOL(should_remove_suid); | |
1606 | ||
1607 | static int __remove_suid(struct dentry *dentry, int kill) | |
1608 | { | |
1609 | struct iattr newattrs; | |
1610 | ||
1611 | newattrs.ia_valid = ATTR_FORCE | kill; | |
1612 | return notify_change(dentry, &newattrs); | |
1613 | } | |
1614 | ||
1615 | int file_remove_suid(struct file *file) | |
1616 | { | |
1617 | struct dentry *dentry = file->f_path.dentry; | |
1618 | struct inode *inode = dentry->d_inode; | |
1619 | int killsuid; | |
1620 | int killpriv; | |
1621 | int error = 0; | |
1622 | ||
1623 | /* Fast path for nothing security related */ | |
1624 | if (IS_NOSEC(inode)) | |
1625 | return 0; | |
1626 | ||
1627 | killsuid = should_remove_suid(dentry); | |
1628 | killpriv = security_inode_need_killpriv(dentry); | |
1629 | ||
1630 | if (killpriv < 0) | |
1631 | return killpriv; | |
1632 | if (killpriv) | |
1633 | error = security_inode_killpriv(dentry); | |
1634 | if (!error && killsuid) | |
1635 | error = __remove_suid(dentry, killsuid); | |
1636 | if (!error && (inode->i_sb->s_flags & MS_NOSEC)) | |
1637 | inode->i_flags |= S_NOSEC; | |
1638 | ||
1639 | return error; | |
1640 | } | |
1641 | EXPORT_SYMBOL(file_remove_suid); | |
1642 | ||
1643 | /** | |
1644 | * file_update_time - update mtime and ctime time | |
1645 | * @file: file accessed | |
1646 | * | |
1647 | * Update the mtime and ctime members of an inode and mark the inode | |
1648 | * for writeback. Note that this function is meant exclusively for | |
1649 | * usage in the file write path of filesystems, and filesystems may | |
1650 | * choose to explicitly ignore update via this function with the | |
1651 | * S_NOCMTIME inode flag, e.g. for network filesystem where these | |
1652 | * timestamps are handled by the server. This can return an error for | |
1653 | * file systems who need to allocate space in order to update an inode. | |
1654 | */ | |
1655 | ||
1656 | int file_update_time(struct file *file) | |
1657 | { | |
1658 | struct inode *inode = file->f_path.dentry->d_inode; | |
1659 | struct timespec now; | |
1660 | int sync_it = 0; | |
1661 | int ret; | |
1662 | ||
1663 | /* First try to exhaust all avenues to not sync */ | |
1664 | if (IS_NOCMTIME(inode)) | |
1665 | return 0; | |
1666 | ||
1667 | now = current_fs_time(inode->i_sb); | |
1668 | if (!timespec_equal(&inode->i_mtime, &now)) | |
1669 | sync_it = S_MTIME; | |
1670 | ||
1671 | if (!timespec_equal(&inode->i_ctime, &now)) | |
1672 | sync_it |= S_CTIME; | |
1673 | ||
1674 | if (IS_I_VERSION(inode)) | |
1675 | sync_it |= S_VERSION; | |
1676 | ||
1677 | if (!sync_it) | |
1678 | return 0; | |
1679 | ||
1680 | /* Finally allowed to write? Takes lock. */ | |
1681 | if (__mnt_want_write_file(file)) | |
1682 | return 0; | |
1683 | ||
1684 | ret = update_time(inode, &now, sync_it); | |
1685 | __mnt_drop_write_file(file); | |
1686 | ||
1687 | return ret; | |
1688 | } | |
1689 | EXPORT_SYMBOL(file_update_time); | |
1690 | ||
1691 | int inode_needs_sync(struct inode *inode) | |
1692 | { | |
1693 | if (IS_SYNC(inode)) | |
1694 | return 1; | |
1695 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1696 | return 1; | |
1697 | return 0; | |
1698 | } | |
1699 | EXPORT_SYMBOL(inode_needs_sync); | |
1700 | ||
1701 | int inode_wait(void *word) | |
1702 | { | |
1703 | schedule(); | |
1704 | return 0; | |
1705 | } | |
1706 | EXPORT_SYMBOL(inode_wait); | |
1707 | ||
1708 | /* | |
1709 | * If we try to find an inode in the inode hash while it is being | |
1710 | * deleted, we have to wait until the filesystem completes its | |
1711 | * deletion before reporting that it isn't found. This function waits | |
1712 | * until the deletion _might_ have completed. Callers are responsible | |
1713 | * to recheck inode state. | |
1714 | * | |
1715 | * It doesn't matter if I_NEW is not set initially, a call to | |
1716 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list | |
1717 | * will DTRT. | |
1718 | */ | |
1719 | static void __wait_on_freeing_inode(struct inode *inode) | |
1720 | { | |
1721 | wait_queue_head_t *wq; | |
1722 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); | |
1723 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1724 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
1725 | spin_unlock(&inode->i_lock); | |
1726 | spin_unlock(&inode_hash_lock); | |
1727 | schedule(); | |
1728 | finish_wait(wq, &wait.wait); | |
1729 | spin_lock(&inode_hash_lock); | |
1730 | } | |
1731 | ||
1732 | static __initdata unsigned long ihash_entries; | |
1733 | static int __init set_ihash_entries(char *str) | |
1734 | { | |
1735 | if (!str) | |
1736 | return 0; | |
1737 | ihash_entries = simple_strtoul(str, &str, 0); | |
1738 | return 1; | |
1739 | } | |
1740 | __setup("ihash_entries=", set_ihash_entries); | |
1741 | ||
1742 | /* | |
1743 | * Initialize the waitqueues and inode hash table. | |
1744 | */ | |
1745 | void __init inode_init_early(void) | |
1746 | { | |
1747 | unsigned int loop; | |
1748 | ||
1749 | /* If hashes are distributed across NUMA nodes, defer | |
1750 | * hash allocation until vmalloc space is available. | |
1751 | */ | |
1752 | if (hashdist) | |
1753 | return; | |
1754 | ||
1755 | inode_hashtable = | |
1756 | alloc_large_system_hash("Inode-cache", | |
1757 | sizeof(struct hlist_head), | |
1758 | ihash_entries, | |
1759 | 14, | |
1760 | HASH_EARLY, | |
1761 | &i_hash_shift, | |
1762 | &i_hash_mask, | |
1763 | 0, | |
1764 | 0); | |
1765 | ||
1766 | for (loop = 0; loop < (1U << i_hash_shift); loop++) | |
1767 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1768 | } | |
1769 | ||
1770 | void __init inode_init(void) | |
1771 | { | |
1772 | unsigned int loop; | |
1773 | ||
1774 | /* inode slab cache */ | |
1775 | inode_cachep = kmem_cache_create("inode_cache", | |
1776 | sizeof(struct inode), | |
1777 | 0, | |
1778 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1779 | SLAB_MEM_SPREAD), | |
1780 | init_once); | |
1781 | ||
1782 | /* Hash may have been set up in inode_init_early */ | |
1783 | if (!hashdist) | |
1784 | return; | |
1785 | ||
1786 | inode_hashtable = | |
1787 | alloc_large_system_hash("Inode-cache", | |
1788 | sizeof(struct hlist_head), | |
1789 | ihash_entries, | |
1790 | 14, | |
1791 | 0, | |
1792 | &i_hash_shift, | |
1793 | &i_hash_mask, | |
1794 | 0, | |
1795 | 0); | |
1796 | ||
1797 | for (loop = 0; loop < (1U << i_hash_shift); loop++) | |
1798 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1799 | } | |
1800 | ||
1801 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1802 | { | |
1803 | inode->i_mode = mode; | |
1804 | if (S_ISCHR(mode)) { | |
1805 | inode->i_fop = &def_chr_fops; | |
1806 | inode->i_rdev = rdev; | |
1807 | } else if (S_ISBLK(mode)) { | |
1808 | inode->i_fop = &def_blk_fops; | |
1809 | inode->i_rdev = rdev; | |
1810 | } else if (S_ISFIFO(mode)) | |
1811 | inode->i_fop = &def_fifo_fops; | |
1812 | else if (S_ISSOCK(mode)) | |
1813 | inode->i_fop = &bad_sock_fops; | |
1814 | else | |
1815 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" | |
1816 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1817 | inode->i_ino); | |
1818 | } | |
1819 | EXPORT_SYMBOL(init_special_inode); | |
1820 | ||
1821 | /** | |
1822 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards | |
1823 | * @inode: New inode | |
1824 | * @dir: Directory inode | |
1825 | * @mode: mode of the new inode | |
1826 | */ | |
1827 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
1828 | umode_t mode) | |
1829 | { | |
1830 | inode->i_uid = current_fsuid(); | |
1831 | if (dir && dir->i_mode & S_ISGID) { | |
1832 | inode->i_gid = dir->i_gid; | |
1833 | if (S_ISDIR(mode)) | |
1834 | mode |= S_ISGID; | |
1835 | } else | |
1836 | inode->i_gid = current_fsgid(); | |
1837 | inode->i_mode = mode; | |
1838 | } | |
1839 | EXPORT_SYMBOL(inode_init_owner); | |
1840 | ||
1841 | /** | |
1842 | * inode_owner_or_capable - check current task permissions to inode | |
1843 | * @inode: inode being checked | |
1844 | * | |
1845 | * Return true if current either has CAP_FOWNER to the inode, or | |
1846 | * owns the file. | |
1847 | */ | |
1848 | bool inode_owner_or_capable(const struct inode *inode) | |
1849 | { | |
1850 | if (uid_eq(current_fsuid(), inode->i_uid)) | |
1851 | return true; | |
1852 | if (inode_capable(inode, CAP_FOWNER)) | |
1853 | return true; | |
1854 | return false; | |
1855 | } | |
1856 | EXPORT_SYMBOL(inode_owner_or_capable); | |
1857 | ||
1858 | /* | |
1859 | * Direct i/o helper functions | |
1860 | */ | |
1861 | static void __inode_dio_wait(struct inode *inode) | |
1862 | { | |
1863 | wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP); | |
1864 | DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP); | |
1865 | ||
1866 | do { | |
1867 | prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE); | |
1868 | if (atomic_read(&inode->i_dio_count)) | |
1869 | schedule(); | |
1870 | } while (atomic_read(&inode->i_dio_count)); | |
1871 | finish_wait(wq, &q.wait); | |
1872 | } | |
1873 | ||
1874 | /** | |
1875 | * inode_dio_wait - wait for outstanding DIO requests to finish | |
1876 | * @inode: inode to wait for | |
1877 | * | |
1878 | * Waits for all pending direct I/O requests to finish so that we can | |
1879 | * proceed with a truncate or equivalent operation. | |
1880 | * | |
1881 | * Must be called under a lock that serializes taking new references | |
1882 | * to i_dio_count, usually by inode->i_mutex. | |
1883 | */ | |
1884 | void inode_dio_wait(struct inode *inode) | |
1885 | { | |
1886 | if (atomic_read(&inode->i_dio_count)) | |
1887 | __inode_dio_wait(inode); | |
1888 | } | |
1889 | EXPORT_SYMBOL(inode_dio_wait); | |
1890 | ||
1891 | /* | |
1892 | * inode_dio_done - signal finish of a direct I/O requests | |
1893 | * @inode: inode the direct I/O happens on | |
1894 | * | |
1895 | * This is called once we've finished processing a direct I/O request, | |
1896 | * and is used to wake up callers waiting for direct I/O to be quiesced. | |
1897 | */ | |
1898 | void inode_dio_done(struct inode *inode) | |
1899 | { | |
1900 | if (atomic_dec_and_test(&inode->i_dio_count)) | |
1901 | wake_up_bit(&inode->i_state, __I_DIO_WAKEUP); | |
1902 | } | |
1903 | EXPORT_SYMBOL(inode_dio_done); |