]>
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 <linux/list_lru.h> | |
21 | #include "internal.h" | |
22 | ||
23 | /* | |
24 | * Inode locking rules: | |
25 | * | |
26 | * inode->i_lock protects: | |
27 | * inode->i_state, inode->i_hash, __iget() | |
28 | * Inode LRU list locks protect: | |
29 | * inode->i_sb->s_inode_lru, inode->i_lru | |
30 | * inode_sb_list_lock protects: | |
31 | * sb->s_inodes, inode->i_sb_list | |
32 | * bdi->wb.list_lock protects: | |
33 | * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list | |
34 | * inode_hash_lock protects: | |
35 | * inode_hashtable, inode->i_hash | |
36 | * | |
37 | * Lock ordering: | |
38 | * | |
39 | * inode_sb_list_lock | |
40 | * inode->i_lock | |
41 | * Inode LRU list locks | |
42 | * | |
43 | * bdi->wb.list_lock | |
44 | * inode->i_lock | |
45 | * | |
46 | * inode_hash_lock | |
47 | * inode_sb_list_lock | |
48 | * inode->i_lock | |
49 | * | |
50 | * iunique_lock | |
51 | * inode_hash_lock | |
52 | */ | |
53 | ||
54 | static unsigned int i_hash_mask __read_mostly; | |
55 | static unsigned int i_hash_shift __read_mostly; | |
56 | static struct hlist_head *inode_hashtable __read_mostly; | |
57 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock); | |
58 | ||
59 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock); | |
60 | ||
61 | /* | |
62 | * Empty aops. Can be used for the cases where the user does not | |
63 | * define any of the address_space operations. | |
64 | */ | |
65 | const struct address_space_operations empty_aops = { | |
66 | }; | |
67 | EXPORT_SYMBOL(empty_aops); | |
68 | ||
69 | /* | |
70 | * Statistics gathering.. | |
71 | */ | |
72 | struct inodes_stat_t inodes_stat; | |
73 | ||
74 | static DEFINE_PER_CPU(unsigned long, nr_inodes); | |
75 | static DEFINE_PER_CPU(unsigned long, nr_unused); | |
76 | ||
77 | static struct kmem_cache *inode_cachep __read_mostly; | |
78 | ||
79 | static long get_nr_inodes(void) | |
80 | { | |
81 | int i; | |
82 | long sum = 0; | |
83 | for_each_possible_cpu(i) | |
84 | sum += per_cpu(nr_inodes, i); | |
85 | return sum < 0 ? 0 : sum; | |
86 | } | |
87 | ||
88 | static inline long get_nr_inodes_unused(void) | |
89 | { | |
90 | int i; | |
91 | long sum = 0; | |
92 | for_each_possible_cpu(i) | |
93 | sum += per_cpu(nr_unused, i); | |
94 | return sum < 0 ? 0 : sum; | |
95 | } | |
96 | ||
97 | long get_nr_dirty_inodes(void) | |
98 | { | |
99 | /* not actually dirty inodes, but a wild approximation */ | |
100 | long nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); | |
101 | return nr_dirty > 0 ? nr_dirty : 0; | |
102 | } | |
103 | ||
104 | /* | |
105 | * Handle nr_inode sysctl | |
106 | */ | |
107 | #ifdef CONFIG_SYSCTL | |
108 | int proc_nr_inodes(struct ctl_table *table, int write, | |
109 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
110 | { | |
111 | inodes_stat.nr_inodes = get_nr_inodes(); | |
112 | inodes_stat.nr_unused = get_nr_inodes_unused(); | |
113 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); | |
114 | } | |
115 | #endif | |
116 | ||
117 | static int no_open(struct inode *inode, struct file *file) | |
118 | { | |
119 | return -ENXIO; | |
120 | } | |
121 | ||
122 | /** | |
123 | * inode_init_always - perform inode structure intialisation | |
124 | * @sb: superblock inode belongs to | |
125 | * @inode: inode to initialise | |
126 | * | |
127 | * These are initializations that need to be done on every inode | |
128 | * allocation as the fields are not initialised by slab allocation. | |
129 | */ | |
130 | int inode_init_always(struct super_block *sb, struct inode *inode) | |
131 | { | |
132 | static const struct inode_operations empty_iops; | |
133 | static const struct file_operations no_open_fops = {.open = no_open}; | |
134 | struct address_space *const mapping = &inode->i_data; | |
135 | ||
136 | inode->i_sb = sb; | |
137 | inode->i_blkbits = sb->s_blocksize_bits; | |
138 | inode->i_flags = 0; | |
139 | atomic_set(&inode->i_count, 1); | |
140 | inode->i_op = &empty_iops; | |
141 | inode->i_fop = &no_open_fops; | |
142 | inode->__i_nlink = 1; | |
143 | inode->i_opflags = 0; | |
144 | i_uid_write(inode, 0); | |
145 | i_gid_write(inode, 0); | |
146 | atomic_set(&inode->i_writecount, 0); | |
147 | inode->i_size = 0; | |
148 | inode->i_blocks = 0; | |
149 | inode->i_bytes = 0; | |
150 | inode->i_generation = 0; | |
151 | inode->i_pipe = NULL; | |
152 | inode->i_bdev = NULL; | |
153 | inode->i_cdev = NULL; | |
154 | inode->i_rdev = 0; | |
155 | inode->dirtied_when = 0; | |
156 | ||
157 | if (security_inode_alloc(inode)) | |
158 | goto out; | |
159 | spin_lock_init(&inode->i_lock); | |
160 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
161 | ||
162 | mutex_init(&inode->i_mutex); | |
163 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
164 | ||
165 | atomic_set(&inode->i_dio_count, 0); | |
166 | ||
167 | mapping->a_ops = &empty_aops; | |
168 | mapping->host = inode; | |
169 | mapping->flags = 0; | |
170 | atomic_set(&mapping->i_mmap_writable, 0); | |
171 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); | |
172 | mapping->private_data = NULL; | |
173 | mapping->backing_dev_info = &default_backing_dev_info; | |
174 | mapping->writeback_index = 0; | |
175 | ||
176 | /* | |
177 | * If the block_device provides a backing_dev_info for client | |
178 | * inodes then use that. Otherwise the inode share the bdev's | |
179 | * backing_dev_info. | |
180 | */ | |
181 | if (sb->s_bdev) { | |
182 | struct backing_dev_info *bdi; | |
183 | ||
184 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; | |
185 | mapping->backing_dev_info = bdi; | |
186 | } | |
187 | inode->i_private = NULL; | |
188 | inode->i_mapping = mapping; | |
189 | INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */ | |
190 | #ifdef CONFIG_FS_POSIX_ACL | |
191 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
192 | #endif | |
193 | ||
194 | #ifdef CONFIG_FSNOTIFY | |
195 | inode->i_fsnotify_mask = 0; | |
196 | #endif | |
197 | inode->i_flctx = NULL; | |
198 | this_cpu_inc(nr_inodes); | |
199 | ||
200 | return 0; | |
201 | out: | |
202 | return -ENOMEM; | |
203 | } | |
204 | EXPORT_SYMBOL(inode_init_always); | |
205 | ||
206 | static struct inode *alloc_inode(struct super_block *sb) | |
207 | { | |
208 | struct inode *inode; | |
209 | ||
210 | if (sb->s_op->alloc_inode) | |
211 | inode = sb->s_op->alloc_inode(sb); | |
212 | else | |
213 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
214 | ||
215 | if (!inode) | |
216 | return NULL; | |
217 | ||
218 | if (unlikely(inode_init_always(sb, inode))) { | |
219 | if (inode->i_sb->s_op->destroy_inode) | |
220 | inode->i_sb->s_op->destroy_inode(inode); | |
221 | else | |
222 | kmem_cache_free(inode_cachep, inode); | |
223 | return NULL; | |
224 | } | |
225 | ||
226 | return inode; | |
227 | } | |
228 | ||
229 | void free_inode_nonrcu(struct inode *inode) | |
230 | { | |
231 | kmem_cache_free(inode_cachep, inode); | |
232 | } | |
233 | EXPORT_SYMBOL(free_inode_nonrcu); | |
234 | ||
235 | void __destroy_inode(struct inode *inode) | |
236 | { | |
237 | BUG_ON(inode_has_buffers(inode)); | |
238 | security_inode_free(inode); | |
239 | fsnotify_inode_delete(inode); | |
240 | locks_free_lock_context(inode->i_flctx); | |
241 | if (!inode->i_nlink) { | |
242 | WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0); | |
243 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
244 | } | |
245 | ||
246 | #ifdef CONFIG_FS_POSIX_ACL | |
247 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) | |
248 | posix_acl_release(inode->i_acl); | |
249 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) | |
250 | posix_acl_release(inode->i_default_acl); | |
251 | #endif | |
252 | this_cpu_dec(nr_inodes); | |
253 | } | |
254 | EXPORT_SYMBOL(__destroy_inode); | |
255 | ||
256 | static void i_callback(struct rcu_head *head) | |
257 | { | |
258 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
259 | kmem_cache_free(inode_cachep, inode); | |
260 | } | |
261 | ||
262 | static void destroy_inode(struct inode *inode) | |
263 | { | |
264 | BUG_ON(!list_empty(&inode->i_lru)); | |
265 | __destroy_inode(inode); | |
266 | if (inode->i_sb->s_op->destroy_inode) | |
267 | inode->i_sb->s_op->destroy_inode(inode); | |
268 | else | |
269 | call_rcu(&inode->i_rcu, i_callback); | |
270 | } | |
271 | ||
272 | /** | |
273 | * drop_nlink - directly drop an inode's link count | |
274 | * @inode: inode | |
275 | * | |
276 | * This is a low-level filesystem helper to replace any | |
277 | * direct filesystem manipulation of i_nlink. In cases | |
278 | * where we are attempting to track writes to the | |
279 | * filesystem, a decrement to zero means an imminent | |
280 | * write when the file is truncated and actually unlinked | |
281 | * on the filesystem. | |
282 | */ | |
283 | void drop_nlink(struct inode *inode) | |
284 | { | |
285 | WARN_ON(inode->i_nlink == 0); | |
286 | inode->__i_nlink--; | |
287 | if (!inode->i_nlink) | |
288 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
289 | } | |
290 | EXPORT_SYMBOL(drop_nlink); | |
291 | ||
292 | /** | |
293 | * clear_nlink - directly zero an inode's link count | |
294 | * @inode: inode | |
295 | * | |
296 | * This is a low-level filesystem helper to replace any | |
297 | * direct filesystem manipulation of i_nlink. See | |
298 | * drop_nlink() for why we care about i_nlink hitting zero. | |
299 | */ | |
300 | void clear_nlink(struct inode *inode) | |
301 | { | |
302 | if (inode->i_nlink) { | |
303 | inode->__i_nlink = 0; | |
304 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
305 | } | |
306 | } | |
307 | EXPORT_SYMBOL(clear_nlink); | |
308 | ||
309 | /** | |
310 | * set_nlink - directly set an inode's link count | |
311 | * @inode: inode | |
312 | * @nlink: new nlink (should be non-zero) | |
313 | * | |
314 | * This is a low-level filesystem helper to replace any | |
315 | * direct filesystem manipulation of i_nlink. | |
316 | */ | |
317 | void set_nlink(struct inode *inode, unsigned int nlink) | |
318 | { | |
319 | if (!nlink) { | |
320 | clear_nlink(inode); | |
321 | } else { | |
322 | /* Yes, some filesystems do change nlink from zero to one */ | |
323 | if (inode->i_nlink == 0) | |
324 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
325 | ||
326 | inode->__i_nlink = nlink; | |
327 | } | |
328 | } | |
329 | EXPORT_SYMBOL(set_nlink); | |
330 | ||
331 | /** | |
332 | * inc_nlink - directly increment an inode's link count | |
333 | * @inode: inode | |
334 | * | |
335 | * This is a low-level filesystem helper to replace any | |
336 | * direct filesystem manipulation of i_nlink. Currently, | |
337 | * it is only here for parity with dec_nlink(). | |
338 | */ | |
339 | void inc_nlink(struct inode *inode) | |
340 | { | |
341 | if (unlikely(inode->i_nlink == 0)) { | |
342 | WARN_ON(!(inode->i_state & I_LINKABLE)); | |
343 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
344 | } | |
345 | ||
346 | inode->__i_nlink++; | |
347 | } | |
348 | EXPORT_SYMBOL(inc_nlink); | |
349 | ||
350 | void address_space_init_once(struct address_space *mapping) | |
351 | { | |
352 | memset(mapping, 0, sizeof(*mapping)); | |
353 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC); | |
354 | spin_lock_init(&mapping->tree_lock); | |
355 | init_rwsem(&mapping->i_mmap_rwsem); | |
356 | INIT_LIST_HEAD(&mapping->private_list); | |
357 | spin_lock_init(&mapping->private_lock); | |
358 | mapping->i_mmap = RB_ROOT; | |
359 | } | |
360 | EXPORT_SYMBOL(address_space_init_once); | |
361 | ||
362 | /* | |
363 | * These are initializations that only need to be done | |
364 | * once, because the fields are idempotent across use | |
365 | * of the inode, so let the slab aware of that. | |
366 | */ | |
367 | void inode_init_once(struct inode *inode) | |
368 | { | |
369 | memset(inode, 0, sizeof(*inode)); | |
370 | INIT_HLIST_NODE(&inode->i_hash); | |
371 | INIT_LIST_HEAD(&inode->i_devices); | |
372 | INIT_LIST_HEAD(&inode->i_wb_list); | |
373 | INIT_LIST_HEAD(&inode->i_lru); | |
374 | address_space_init_once(&inode->i_data); | |
375 | i_size_ordered_init(inode); | |
376 | #ifdef CONFIG_FSNOTIFY | |
377 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); | |
378 | #endif | |
379 | } | |
380 | EXPORT_SYMBOL(inode_init_once); | |
381 | ||
382 | static void init_once(void *foo) | |
383 | { | |
384 | struct inode *inode = (struct inode *) foo; | |
385 | ||
386 | inode_init_once(inode); | |
387 | } | |
388 | ||
389 | /* | |
390 | * inode->i_lock must be held | |
391 | */ | |
392 | void __iget(struct inode *inode) | |
393 | { | |
394 | atomic_inc(&inode->i_count); | |
395 | } | |
396 | ||
397 | /* | |
398 | * get additional reference to inode; caller must already hold one. | |
399 | */ | |
400 | void ihold(struct inode *inode) | |
401 | { | |
402 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); | |
403 | } | |
404 | EXPORT_SYMBOL(ihold); | |
405 | ||
406 | static void inode_lru_list_add(struct inode *inode) | |
407 | { | |
408 | if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru)) | |
409 | this_cpu_inc(nr_unused); | |
410 | } | |
411 | ||
412 | /* | |
413 | * Add inode to LRU if needed (inode is unused and clean). | |
414 | * | |
415 | * Needs inode->i_lock held. | |
416 | */ | |
417 | void inode_add_lru(struct inode *inode) | |
418 | { | |
419 | if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) && | |
420 | !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE) | |
421 | inode_lru_list_add(inode); | |
422 | } | |
423 | ||
424 | ||
425 | static void inode_lru_list_del(struct inode *inode) | |
426 | { | |
427 | ||
428 | if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru)) | |
429 | this_cpu_dec(nr_unused); | |
430 | } | |
431 | ||
432 | /** | |
433 | * inode_sb_list_add - add inode to the superblock list of inodes | |
434 | * @inode: inode to add | |
435 | */ | |
436 | void inode_sb_list_add(struct inode *inode) | |
437 | { | |
438 | spin_lock(&inode_sb_list_lock); | |
439 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); | |
440 | spin_unlock(&inode_sb_list_lock); | |
441 | } | |
442 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
443 | ||
444 | static inline void inode_sb_list_del(struct inode *inode) | |
445 | { | |
446 | if (!list_empty(&inode->i_sb_list)) { | |
447 | spin_lock(&inode_sb_list_lock); | |
448 | list_del_init(&inode->i_sb_list); | |
449 | spin_unlock(&inode_sb_list_lock); | |
450 | } | |
451 | } | |
452 | ||
453 | static unsigned long hash(struct super_block *sb, unsigned long hashval) | |
454 | { | |
455 | unsigned long tmp; | |
456 | ||
457 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
458 | L1_CACHE_BYTES; | |
459 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift); | |
460 | return tmp & i_hash_mask; | |
461 | } | |
462 | ||
463 | /** | |
464 | * __insert_inode_hash - hash an inode | |
465 | * @inode: unhashed inode | |
466 | * @hashval: unsigned long value used to locate this object in the | |
467 | * inode_hashtable. | |
468 | * | |
469 | * Add an inode to the inode hash for this superblock. | |
470 | */ | |
471 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
472 | { | |
473 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); | |
474 | ||
475 | spin_lock(&inode_hash_lock); | |
476 | spin_lock(&inode->i_lock); | |
477 | hlist_add_head(&inode->i_hash, b); | |
478 | spin_unlock(&inode->i_lock); | |
479 | spin_unlock(&inode_hash_lock); | |
480 | } | |
481 | EXPORT_SYMBOL(__insert_inode_hash); | |
482 | ||
483 | /** | |
484 | * __remove_inode_hash - remove an inode from the hash | |
485 | * @inode: inode to unhash | |
486 | * | |
487 | * Remove an inode from the superblock. | |
488 | */ | |
489 | void __remove_inode_hash(struct inode *inode) | |
490 | { | |
491 | spin_lock(&inode_hash_lock); | |
492 | spin_lock(&inode->i_lock); | |
493 | hlist_del_init(&inode->i_hash); | |
494 | spin_unlock(&inode->i_lock); | |
495 | spin_unlock(&inode_hash_lock); | |
496 | } | |
497 | EXPORT_SYMBOL(__remove_inode_hash); | |
498 | ||
499 | void clear_inode(struct inode *inode) | |
500 | { | |
501 | might_sleep(); | |
502 | /* | |
503 | * We have to cycle tree_lock here because reclaim can be still in the | |
504 | * process of removing the last page (in __delete_from_page_cache()) | |
505 | * and we must not free mapping under it. | |
506 | */ | |
507 | spin_lock_irq(&inode->i_data.tree_lock); | |
508 | BUG_ON(inode->i_data.nrpages); | |
509 | BUG_ON(inode->i_data.nrshadows); | |
510 | spin_unlock_irq(&inode->i_data.tree_lock); | |
511 | BUG_ON(!list_empty(&inode->i_data.private_list)); | |
512 | BUG_ON(!(inode->i_state & I_FREEING)); | |
513 | BUG_ON(inode->i_state & I_CLEAR); | |
514 | /* don't need i_lock here, no concurrent mods to i_state */ | |
515 | inode->i_state = I_FREEING | I_CLEAR; | |
516 | } | |
517 | EXPORT_SYMBOL(clear_inode); | |
518 | ||
519 | /* | |
520 | * Free the inode passed in, removing it from the lists it is still connected | |
521 | * to. We remove any pages still attached to the inode and wait for any IO that | |
522 | * is still in progress before finally destroying the inode. | |
523 | * | |
524 | * An inode must already be marked I_FREEING so that we avoid the inode being | |
525 | * moved back onto lists if we race with other code that manipulates the lists | |
526 | * (e.g. writeback_single_inode). The caller is responsible for setting this. | |
527 | * | |
528 | * An inode must already be removed from the LRU list before being evicted from | |
529 | * the cache. This should occur atomically with setting the I_FREEING state | |
530 | * flag, so no inodes here should ever be on the LRU when being evicted. | |
531 | */ | |
532 | static void evict(struct inode *inode) | |
533 | { | |
534 | const struct super_operations *op = inode->i_sb->s_op; | |
535 | ||
536 | BUG_ON(!(inode->i_state & I_FREEING)); | |
537 | BUG_ON(!list_empty(&inode->i_lru)); | |
538 | ||
539 | if (!list_empty(&inode->i_wb_list)) | |
540 | inode_wb_list_del(inode); | |
541 | ||
542 | inode_sb_list_del(inode); | |
543 | ||
544 | /* | |
545 | * Wait for flusher thread to be done with the inode so that filesystem | |
546 | * does not start destroying it while writeback is still running. Since | |
547 | * the inode has I_FREEING set, flusher thread won't start new work on | |
548 | * the inode. We just have to wait for running writeback to finish. | |
549 | */ | |
550 | inode_wait_for_writeback(inode); | |
551 | ||
552 | if (op->evict_inode) { | |
553 | op->evict_inode(inode); | |
554 | } else { | |
555 | truncate_inode_pages_final(&inode->i_data); | |
556 | clear_inode(inode); | |
557 | } | |
558 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) | |
559 | bd_forget(inode); | |
560 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
561 | cd_forget(inode); | |
562 | ||
563 | remove_inode_hash(inode); | |
564 | ||
565 | spin_lock(&inode->i_lock); | |
566 | wake_up_bit(&inode->i_state, __I_NEW); | |
567 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); | |
568 | spin_unlock(&inode->i_lock); | |
569 | ||
570 | destroy_inode(inode); | |
571 | } | |
572 | ||
573 | /* | |
574 | * dispose_list - dispose of the contents of a local list | |
575 | * @head: the head of the list to free | |
576 | * | |
577 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
578 | * need to worry about list corruption and SMP locks. | |
579 | */ | |
580 | static void dispose_list(struct list_head *head) | |
581 | { | |
582 | while (!list_empty(head)) { | |
583 | struct inode *inode; | |
584 | ||
585 | inode = list_first_entry(head, struct inode, i_lru); | |
586 | list_del_init(&inode->i_lru); | |
587 | ||
588 | evict(inode); | |
589 | } | |
590 | } | |
591 | ||
592 | /** | |
593 | * evict_inodes - evict all evictable inodes for a superblock | |
594 | * @sb: superblock to operate on | |
595 | * | |
596 | * Make sure that no inodes with zero refcount are retained. This is | |
597 | * called by superblock shutdown after having MS_ACTIVE flag removed, | |
598 | * so any inode reaching zero refcount during or after that call will | |
599 | * be immediately evicted. | |
600 | */ | |
601 | void evict_inodes(struct super_block *sb) | |
602 | { | |
603 | struct inode *inode, *next; | |
604 | LIST_HEAD(dispose); | |
605 | ||
606 | spin_lock(&inode_sb_list_lock); | |
607 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { | |
608 | if (atomic_read(&inode->i_count)) | |
609 | continue; | |
610 | ||
611 | spin_lock(&inode->i_lock); | |
612 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
613 | spin_unlock(&inode->i_lock); | |
614 | continue; | |
615 | } | |
616 | ||
617 | inode->i_state |= I_FREEING; | |
618 | inode_lru_list_del(inode); | |
619 | spin_unlock(&inode->i_lock); | |
620 | list_add(&inode->i_lru, &dispose); | |
621 | } | |
622 | spin_unlock(&inode_sb_list_lock); | |
623 | ||
624 | dispose_list(&dispose); | |
625 | } | |
626 | ||
627 | /** | |
628 | * invalidate_inodes - attempt to free all inodes on a superblock | |
629 | * @sb: superblock to operate on | |
630 | * @kill_dirty: flag to guide handling of dirty inodes | |
631 | * | |
632 | * Attempts to free all inodes for a given superblock. If there were any | |
633 | * busy inodes return a non-zero value, else zero. | |
634 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat | |
635 | * them as busy. | |
636 | */ | |
637 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) | |
638 | { | |
639 | int busy = 0; | |
640 | struct inode *inode, *next; | |
641 | LIST_HEAD(dispose); | |
642 | ||
643 | spin_lock(&inode_sb_list_lock); | |
644 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { | |
645 | spin_lock(&inode->i_lock); | |
646 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
647 | spin_unlock(&inode->i_lock); | |
648 | continue; | |
649 | } | |
650 | if (inode->i_state & I_DIRTY && !kill_dirty) { | |
651 | spin_unlock(&inode->i_lock); | |
652 | busy = 1; | |
653 | continue; | |
654 | } | |
655 | if (atomic_read(&inode->i_count)) { | |
656 | spin_unlock(&inode->i_lock); | |
657 | busy = 1; | |
658 | continue; | |
659 | } | |
660 | ||
661 | inode->i_state |= I_FREEING; | |
662 | inode_lru_list_del(inode); | |
663 | spin_unlock(&inode->i_lock); | |
664 | list_add(&inode->i_lru, &dispose); | |
665 | } | |
666 | spin_unlock(&inode_sb_list_lock); | |
667 | ||
668 | dispose_list(&dispose); | |
669 | ||
670 | return busy; | |
671 | } | |
672 | ||
673 | /* | |
674 | * Isolate the inode from the LRU in preparation for freeing it. | |
675 | * | |
676 | * Any inodes which are pinned purely because of attached pagecache have their | |
677 | * pagecache removed. If the inode has metadata buffers attached to | |
678 | * mapping->private_list then try to remove them. | |
679 | * | |
680 | * If the inode has the I_REFERENCED flag set, then it means that it has been | |
681 | * used recently - the flag is set in iput_final(). When we encounter such an | |
682 | * inode, clear the flag and move it to the back of the LRU so it gets another | |
683 | * pass through the LRU before it gets reclaimed. This is necessary because of | |
684 | * the fact we are doing lazy LRU updates to minimise lock contention so the | |
685 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes | |
686 | * with this flag set because they are the inodes that are out of order. | |
687 | */ | |
688 | static enum lru_status | |
689 | inode_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) | |
690 | { | |
691 | struct list_head *freeable = arg; | |
692 | struct inode *inode = container_of(item, struct inode, i_lru); | |
693 | ||
694 | /* | |
695 | * we are inverting the lru lock/inode->i_lock here, so use a trylock. | |
696 | * If we fail to get the lock, just skip it. | |
697 | */ | |
698 | if (!spin_trylock(&inode->i_lock)) | |
699 | return LRU_SKIP; | |
700 | ||
701 | /* | |
702 | * Referenced or dirty inodes are still in use. Give them another pass | |
703 | * through the LRU as we canot reclaim them now. | |
704 | */ | |
705 | if (atomic_read(&inode->i_count) || | |
706 | (inode->i_state & ~I_REFERENCED)) { | |
707 | list_del_init(&inode->i_lru); | |
708 | spin_unlock(&inode->i_lock); | |
709 | this_cpu_dec(nr_unused); | |
710 | return LRU_REMOVED; | |
711 | } | |
712 | ||
713 | /* recently referenced inodes get one more pass */ | |
714 | if (inode->i_state & I_REFERENCED) { | |
715 | inode->i_state &= ~I_REFERENCED; | |
716 | spin_unlock(&inode->i_lock); | |
717 | return LRU_ROTATE; | |
718 | } | |
719 | ||
720 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
721 | __iget(inode); | |
722 | spin_unlock(&inode->i_lock); | |
723 | spin_unlock(lru_lock); | |
724 | if (remove_inode_buffers(inode)) { | |
725 | unsigned long reap; | |
726 | reap = invalidate_mapping_pages(&inode->i_data, 0, -1); | |
727 | if (current_is_kswapd()) | |
728 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
729 | else | |
730 | __count_vm_events(PGINODESTEAL, reap); | |
731 | if (current->reclaim_state) | |
732 | current->reclaim_state->reclaimed_slab += reap; | |
733 | } | |
734 | iput(inode); | |
735 | spin_lock(lru_lock); | |
736 | return LRU_RETRY; | |
737 | } | |
738 | ||
739 | WARN_ON(inode->i_state & I_NEW); | |
740 | inode->i_state |= I_FREEING; | |
741 | list_move(&inode->i_lru, freeable); | |
742 | spin_unlock(&inode->i_lock); | |
743 | ||
744 | this_cpu_dec(nr_unused); | |
745 | return LRU_REMOVED; | |
746 | } | |
747 | ||
748 | /* | |
749 | * Walk the superblock inode LRU for freeable inodes and attempt to free them. | |
750 | * This is called from the superblock shrinker function with a number of inodes | |
751 | * to trim from the LRU. Inodes to be freed are moved to a temporary list and | |
752 | * then are freed outside inode_lock by dispose_list(). | |
753 | */ | |
754 | long prune_icache_sb(struct super_block *sb, struct shrink_control *sc) | |
755 | { | |
756 | LIST_HEAD(freeable); | |
757 | long freed; | |
758 | ||
759 | freed = list_lru_shrink_walk(&sb->s_inode_lru, sc, | |
760 | inode_lru_isolate, &freeable); | |
761 | dispose_list(&freeable); | |
762 | return freed; | |
763 | } | |
764 | ||
765 | static void __wait_on_freeing_inode(struct inode *inode); | |
766 | /* | |
767 | * Called with the inode lock held. | |
768 | */ | |
769 | static struct inode *find_inode(struct super_block *sb, | |
770 | struct hlist_head *head, | |
771 | int (*test)(struct inode *, void *), | |
772 | void *data) | |
773 | { | |
774 | struct inode *inode = NULL; | |
775 | ||
776 | repeat: | |
777 | hlist_for_each_entry(inode, head, i_hash) { | |
778 | if (inode->i_sb != sb) | |
779 | continue; | |
780 | if (!test(inode, data)) | |
781 | continue; | |
782 | spin_lock(&inode->i_lock); | |
783 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { | |
784 | __wait_on_freeing_inode(inode); | |
785 | goto repeat; | |
786 | } | |
787 | __iget(inode); | |
788 | spin_unlock(&inode->i_lock); | |
789 | return inode; | |
790 | } | |
791 | return NULL; | |
792 | } | |
793 | ||
794 | /* | |
795 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
796 | * iget_locked for details. | |
797 | */ | |
798 | static struct inode *find_inode_fast(struct super_block *sb, | |
799 | struct hlist_head *head, unsigned long ino) | |
800 | { | |
801 | struct inode *inode = NULL; | |
802 | ||
803 | repeat: | |
804 | hlist_for_each_entry(inode, head, i_hash) { | |
805 | if (inode->i_ino != ino) | |
806 | continue; | |
807 | if (inode->i_sb != sb) | |
808 | continue; | |
809 | spin_lock(&inode->i_lock); | |
810 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { | |
811 | __wait_on_freeing_inode(inode); | |
812 | goto repeat; | |
813 | } | |
814 | __iget(inode); | |
815 | spin_unlock(&inode->i_lock); | |
816 | return inode; | |
817 | } | |
818 | return NULL; | |
819 | } | |
820 | ||
821 | /* | |
822 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
823 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
824 | * to renew the exhausted range. | |
825 | * | |
826 | * This does not significantly increase overflow rate because every CPU can | |
827 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
828 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
829 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
830 | * overflow rate by 2x, which does not seem too significant. | |
831 | * | |
832 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
833 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
834 | * here to attempt to avoid that. | |
835 | */ | |
836 | #define LAST_INO_BATCH 1024 | |
837 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
838 | ||
839 | unsigned int get_next_ino(void) | |
840 | { | |
841 | unsigned int *p = &get_cpu_var(last_ino); | |
842 | unsigned int res = *p; | |
843 | ||
844 | #ifdef CONFIG_SMP | |
845 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
846 | static atomic_t shared_last_ino; | |
847 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
848 | ||
849 | res = next - LAST_INO_BATCH; | |
850 | } | |
851 | #endif | |
852 | ||
853 | *p = ++res; | |
854 | put_cpu_var(last_ino); | |
855 | return res; | |
856 | } | |
857 | EXPORT_SYMBOL(get_next_ino); | |
858 | ||
859 | /** | |
860 | * new_inode_pseudo - obtain an inode | |
861 | * @sb: superblock | |
862 | * | |
863 | * Allocates a new inode for given superblock. | |
864 | * Inode wont be chained in superblock s_inodes list | |
865 | * This means : | |
866 | * - fs can't be unmount | |
867 | * - quotas, fsnotify, writeback can't work | |
868 | */ | |
869 | struct inode *new_inode_pseudo(struct super_block *sb) | |
870 | { | |
871 | struct inode *inode = alloc_inode(sb); | |
872 | ||
873 | if (inode) { | |
874 | spin_lock(&inode->i_lock); | |
875 | inode->i_state = 0; | |
876 | spin_unlock(&inode->i_lock); | |
877 | INIT_LIST_HEAD(&inode->i_sb_list); | |
878 | } | |
879 | return inode; | |
880 | } | |
881 | ||
882 | /** | |
883 | * new_inode - obtain an inode | |
884 | * @sb: superblock | |
885 | * | |
886 | * Allocates a new inode for given superblock. The default gfp_mask | |
887 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. | |
888 | * If HIGHMEM pages are unsuitable or it is known that pages allocated | |
889 | * for the page cache are not reclaimable or migratable, | |
890 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
891 | * newly created inode's mapping | |
892 | * | |
893 | */ | |
894 | struct inode *new_inode(struct super_block *sb) | |
895 | { | |
896 | struct inode *inode; | |
897 | ||
898 | spin_lock_prefetch(&inode_sb_list_lock); | |
899 | ||
900 | inode = new_inode_pseudo(sb); | |
901 | if (inode) | |
902 | inode_sb_list_add(inode); | |
903 | return inode; | |
904 | } | |
905 | EXPORT_SYMBOL(new_inode); | |
906 | ||
907 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
908 | void lockdep_annotate_inode_mutex_key(struct inode *inode) | |
909 | { | |
910 | if (S_ISDIR(inode->i_mode)) { | |
911 | struct file_system_type *type = inode->i_sb->s_type; | |
912 | ||
913 | /* Set new key only if filesystem hasn't already changed it */ | |
914 | if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) { | |
915 | /* | |
916 | * ensure nobody is actually holding i_mutex | |
917 | */ | |
918 | mutex_destroy(&inode->i_mutex); | |
919 | mutex_init(&inode->i_mutex); | |
920 | lockdep_set_class(&inode->i_mutex, | |
921 | &type->i_mutex_dir_key); | |
922 | } | |
923 | } | |
924 | } | |
925 | EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key); | |
926 | #endif | |
927 | ||
928 | /** | |
929 | * unlock_new_inode - clear the I_NEW state and wake up any waiters | |
930 | * @inode: new inode to unlock | |
931 | * | |
932 | * Called when the inode is fully initialised to clear the new state of the | |
933 | * inode and wake up anyone waiting for the inode to finish initialisation. | |
934 | */ | |
935 | void unlock_new_inode(struct inode *inode) | |
936 | { | |
937 | lockdep_annotate_inode_mutex_key(inode); | |
938 | spin_lock(&inode->i_lock); | |
939 | WARN_ON(!(inode->i_state & I_NEW)); | |
940 | inode->i_state &= ~I_NEW; | |
941 | smp_mb(); | |
942 | wake_up_bit(&inode->i_state, __I_NEW); | |
943 | spin_unlock(&inode->i_lock); | |
944 | } | |
945 | EXPORT_SYMBOL(unlock_new_inode); | |
946 | ||
947 | /** | |
948 | * lock_two_nondirectories - take two i_mutexes on non-directory objects | |
949 | * | |
950 | * Lock any non-NULL argument that is not a directory. | |
951 | * Zero, one or two objects may be locked by this function. | |
952 | * | |
953 | * @inode1: first inode to lock | |
954 | * @inode2: second inode to lock | |
955 | */ | |
956 | void lock_two_nondirectories(struct inode *inode1, struct inode *inode2) | |
957 | { | |
958 | if (inode1 > inode2) | |
959 | swap(inode1, inode2); | |
960 | ||
961 | if (inode1 && !S_ISDIR(inode1->i_mode)) | |
962 | mutex_lock(&inode1->i_mutex); | |
963 | if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1) | |
964 | mutex_lock_nested(&inode2->i_mutex, I_MUTEX_NONDIR2); | |
965 | } | |
966 | EXPORT_SYMBOL(lock_two_nondirectories); | |
967 | ||
968 | /** | |
969 | * unlock_two_nondirectories - release locks from lock_two_nondirectories() | |
970 | * @inode1: first inode to unlock | |
971 | * @inode2: second inode to unlock | |
972 | */ | |
973 | void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2) | |
974 | { | |
975 | if (inode1 && !S_ISDIR(inode1->i_mode)) | |
976 | mutex_unlock(&inode1->i_mutex); | |
977 | if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1) | |
978 | mutex_unlock(&inode2->i_mutex); | |
979 | } | |
980 | EXPORT_SYMBOL(unlock_two_nondirectories); | |
981 | ||
982 | /** | |
983 | * iget5_locked - obtain an inode from a mounted file system | |
984 | * @sb: super block of file system | |
985 | * @hashval: hash value (usually inode number) to get | |
986 | * @test: callback used for comparisons between inodes | |
987 | * @set: callback used to initialize a new struct inode | |
988 | * @data: opaque data pointer to pass to @test and @set | |
989 | * | |
990 | * Search for the inode specified by @hashval and @data in the inode cache, | |
991 | * and if present it is return it with an increased reference count. This is | |
992 | * a generalized version of iget_locked() for file systems where the inode | |
993 | * number is not sufficient for unique identification of an inode. | |
994 | * | |
995 | * If the inode is not in cache, allocate a new inode and return it locked, | |
996 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
997 | * before unlocking it via unlock_new_inode(). | |
998 | * | |
999 | * Note both @test and @set are called with the inode_hash_lock held, so can't | |
1000 | * sleep. | |
1001 | */ | |
1002 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
1003 | int (*test)(struct inode *, void *), | |
1004 | int (*set)(struct inode *, void *), void *data) | |
1005 | { | |
1006 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1007 | struct inode *inode; | |
1008 | ||
1009 | spin_lock(&inode_hash_lock); | |
1010 | inode = find_inode(sb, head, test, data); | |
1011 | spin_unlock(&inode_hash_lock); | |
1012 | ||
1013 | if (inode) { | |
1014 | wait_on_inode(inode); | |
1015 | return inode; | |
1016 | } | |
1017 | ||
1018 | inode = alloc_inode(sb); | |
1019 | if (inode) { | |
1020 | struct inode *old; | |
1021 | ||
1022 | spin_lock(&inode_hash_lock); | |
1023 | /* We released the lock, so.. */ | |
1024 | old = find_inode(sb, head, test, data); | |
1025 | if (!old) { | |
1026 | if (set(inode, data)) | |
1027 | goto set_failed; | |
1028 | ||
1029 | spin_lock(&inode->i_lock); | |
1030 | inode->i_state = I_NEW; | |
1031 | hlist_add_head(&inode->i_hash, head); | |
1032 | spin_unlock(&inode->i_lock); | |
1033 | inode_sb_list_add(inode); | |
1034 | spin_unlock(&inode_hash_lock); | |
1035 | ||
1036 | /* Return the locked inode with I_NEW set, the | |
1037 | * caller is responsible for filling in the contents | |
1038 | */ | |
1039 | return inode; | |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * Uhhuh, somebody else created the same inode under | |
1044 | * us. Use the old inode instead of the one we just | |
1045 | * allocated. | |
1046 | */ | |
1047 | spin_unlock(&inode_hash_lock); | |
1048 | destroy_inode(inode); | |
1049 | inode = old; | |
1050 | wait_on_inode(inode); | |
1051 | } | |
1052 | return inode; | |
1053 | ||
1054 | set_failed: | |
1055 | spin_unlock(&inode_hash_lock); | |
1056 | destroy_inode(inode); | |
1057 | return NULL; | |
1058 | } | |
1059 | EXPORT_SYMBOL(iget5_locked); | |
1060 | ||
1061 | /** | |
1062 | * iget_locked - obtain an inode from a mounted file system | |
1063 | * @sb: super block of file system | |
1064 | * @ino: inode number to get | |
1065 | * | |
1066 | * Search for the inode specified by @ino in the inode cache and if present | |
1067 | * return it with an increased reference count. This is for file systems | |
1068 | * where the inode number is sufficient for unique identification of an inode. | |
1069 | * | |
1070 | * If the inode is not in cache, allocate a new inode and return it locked, | |
1071 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
1072 | * before unlocking it via unlock_new_inode(). | |
1073 | */ | |
1074 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) | |
1075 | { | |
1076 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1077 | struct inode *inode; | |
1078 | ||
1079 | spin_lock(&inode_hash_lock); | |
1080 | inode = find_inode_fast(sb, head, ino); | |
1081 | spin_unlock(&inode_hash_lock); | |
1082 | if (inode) { | |
1083 | wait_on_inode(inode); | |
1084 | return inode; | |
1085 | } | |
1086 | ||
1087 | inode = alloc_inode(sb); | |
1088 | if (inode) { | |
1089 | struct inode *old; | |
1090 | ||
1091 | spin_lock(&inode_hash_lock); | |
1092 | /* We released the lock, so.. */ | |
1093 | old = find_inode_fast(sb, head, ino); | |
1094 | if (!old) { | |
1095 | inode->i_ino = ino; | |
1096 | spin_lock(&inode->i_lock); | |
1097 | inode->i_state = I_NEW; | |
1098 | hlist_add_head(&inode->i_hash, head); | |
1099 | spin_unlock(&inode->i_lock); | |
1100 | inode_sb_list_add(inode); | |
1101 | spin_unlock(&inode_hash_lock); | |
1102 | ||
1103 | /* Return the locked inode with I_NEW set, the | |
1104 | * caller is responsible for filling in the contents | |
1105 | */ | |
1106 | return inode; | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * Uhhuh, somebody else created the same inode under | |
1111 | * us. Use the old inode instead of the one we just | |
1112 | * allocated. | |
1113 | */ | |
1114 | spin_unlock(&inode_hash_lock); | |
1115 | destroy_inode(inode); | |
1116 | inode = old; | |
1117 | wait_on_inode(inode); | |
1118 | } | |
1119 | return inode; | |
1120 | } | |
1121 | EXPORT_SYMBOL(iget_locked); | |
1122 | ||
1123 | /* | |
1124 | * search the inode cache for a matching inode number. | |
1125 | * If we find one, then the inode number we are trying to | |
1126 | * allocate is not unique and so we should not use it. | |
1127 | * | |
1128 | * Returns 1 if the inode number is unique, 0 if it is not. | |
1129 | */ | |
1130 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
1131 | { | |
1132 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
1133 | struct inode *inode; | |
1134 | ||
1135 | spin_lock(&inode_hash_lock); | |
1136 | hlist_for_each_entry(inode, b, i_hash) { | |
1137 | if (inode->i_ino == ino && inode->i_sb == sb) { | |
1138 | spin_unlock(&inode_hash_lock); | |
1139 | return 0; | |
1140 | } | |
1141 | } | |
1142 | spin_unlock(&inode_hash_lock); | |
1143 | ||
1144 | return 1; | |
1145 | } | |
1146 | ||
1147 | /** | |
1148 | * iunique - get a unique inode number | |
1149 | * @sb: superblock | |
1150 | * @max_reserved: highest reserved inode number | |
1151 | * | |
1152 | * Obtain an inode number that is unique on the system for a given | |
1153 | * superblock. This is used by file systems that have no natural | |
1154 | * permanent inode numbering system. An inode number is returned that | |
1155 | * is higher than the reserved limit but unique. | |
1156 | * | |
1157 | * BUGS: | |
1158 | * With a large number of inodes live on the file system this function | |
1159 | * currently becomes quite slow. | |
1160 | */ | |
1161 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
1162 | { | |
1163 | /* | |
1164 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
1165 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
1166 | * here to attempt to avoid that. | |
1167 | */ | |
1168 | static DEFINE_SPINLOCK(iunique_lock); | |
1169 | static unsigned int counter; | |
1170 | ino_t res; | |
1171 | ||
1172 | spin_lock(&iunique_lock); | |
1173 | do { | |
1174 | if (counter <= max_reserved) | |
1175 | counter = max_reserved + 1; | |
1176 | res = counter++; | |
1177 | } while (!test_inode_iunique(sb, res)); | |
1178 | spin_unlock(&iunique_lock); | |
1179 | ||
1180 | return res; | |
1181 | } | |
1182 | EXPORT_SYMBOL(iunique); | |
1183 | ||
1184 | struct inode *igrab(struct inode *inode) | |
1185 | { | |
1186 | spin_lock(&inode->i_lock); | |
1187 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { | |
1188 | __iget(inode); | |
1189 | spin_unlock(&inode->i_lock); | |
1190 | } else { | |
1191 | spin_unlock(&inode->i_lock); | |
1192 | /* | |
1193 | * Handle the case where s_op->clear_inode is not been | |
1194 | * called yet, and somebody is calling igrab | |
1195 | * while the inode is getting freed. | |
1196 | */ | |
1197 | inode = NULL; | |
1198 | } | |
1199 | return inode; | |
1200 | } | |
1201 | EXPORT_SYMBOL(igrab); | |
1202 | ||
1203 | /** | |
1204 | * ilookup5_nowait - search for an inode in the inode cache | |
1205 | * @sb: super block of file system to search | |
1206 | * @hashval: hash value (usually inode number) to search for | |
1207 | * @test: callback used for comparisons between inodes | |
1208 | * @data: opaque data pointer to pass to @test | |
1209 | * | |
1210 | * Search for the inode specified by @hashval and @data in the inode cache. | |
1211 | * If the inode is in the cache, the inode is returned with an incremented | |
1212 | * reference count. | |
1213 | * | |
1214 | * Note: I_NEW is not waited upon so you have to be very careful what you do | |
1215 | * with the returned inode. You probably should be using ilookup5() instead. | |
1216 | * | |
1217 | * Note2: @test is called with the inode_hash_lock held, so can't sleep. | |
1218 | */ | |
1219 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
1220 | int (*test)(struct inode *, void *), void *data) | |
1221 | { | |
1222 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1223 | struct inode *inode; | |
1224 | ||
1225 | spin_lock(&inode_hash_lock); | |
1226 | inode = find_inode(sb, head, test, data); | |
1227 | spin_unlock(&inode_hash_lock); | |
1228 | ||
1229 | return inode; | |
1230 | } | |
1231 | EXPORT_SYMBOL(ilookup5_nowait); | |
1232 | ||
1233 | /** | |
1234 | * ilookup5 - search for an inode in the inode cache | |
1235 | * @sb: super block of file system to search | |
1236 | * @hashval: hash value (usually inode number) to search for | |
1237 | * @test: callback used for comparisons between inodes | |
1238 | * @data: opaque data pointer to pass to @test | |
1239 | * | |
1240 | * Search for the inode specified by @hashval and @data in the inode cache, | |
1241 | * and if the inode is in the cache, return the inode with an incremented | |
1242 | * reference count. Waits on I_NEW before returning the inode. | |
1243 | * returned with an incremented reference count. | |
1244 | * | |
1245 | * This is a generalized version of ilookup() for file systems where the | |
1246 | * inode number is not sufficient for unique identification of an inode. | |
1247 | * | |
1248 | * Note: @test is called with the inode_hash_lock held, so can't sleep. | |
1249 | */ | |
1250 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1251 | int (*test)(struct inode *, void *), void *data) | |
1252 | { | |
1253 | struct inode *inode = ilookup5_nowait(sb, hashval, test, data); | |
1254 | ||
1255 | if (inode) | |
1256 | wait_on_inode(inode); | |
1257 | return inode; | |
1258 | } | |
1259 | EXPORT_SYMBOL(ilookup5); | |
1260 | ||
1261 | /** | |
1262 | * ilookup - search for an inode in the inode cache | |
1263 | * @sb: super block of file system to search | |
1264 | * @ino: inode number to search for | |
1265 | * | |
1266 | * Search for the inode @ino in the inode cache, and if the inode is in the | |
1267 | * cache, the inode is returned with an incremented reference count. | |
1268 | */ | |
1269 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1270 | { | |
1271 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1272 | struct inode *inode; | |
1273 | ||
1274 | spin_lock(&inode_hash_lock); | |
1275 | inode = find_inode_fast(sb, head, ino); | |
1276 | spin_unlock(&inode_hash_lock); | |
1277 | ||
1278 | if (inode) | |
1279 | wait_on_inode(inode); | |
1280 | return inode; | |
1281 | } | |
1282 | EXPORT_SYMBOL(ilookup); | |
1283 | ||
1284 | int insert_inode_locked(struct inode *inode) | |
1285 | { | |
1286 | struct super_block *sb = inode->i_sb; | |
1287 | ino_t ino = inode->i_ino; | |
1288 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1289 | ||
1290 | while (1) { | |
1291 | struct inode *old = NULL; | |
1292 | spin_lock(&inode_hash_lock); | |
1293 | hlist_for_each_entry(old, head, i_hash) { | |
1294 | if (old->i_ino != ino) | |
1295 | continue; | |
1296 | if (old->i_sb != sb) | |
1297 | continue; | |
1298 | spin_lock(&old->i_lock); | |
1299 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1300 | spin_unlock(&old->i_lock); | |
1301 | continue; | |
1302 | } | |
1303 | break; | |
1304 | } | |
1305 | if (likely(!old)) { | |
1306 | spin_lock(&inode->i_lock); | |
1307 | inode->i_state |= I_NEW; | |
1308 | hlist_add_head(&inode->i_hash, head); | |
1309 | spin_unlock(&inode->i_lock); | |
1310 | spin_unlock(&inode_hash_lock); | |
1311 | return 0; | |
1312 | } | |
1313 | __iget(old); | |
1314 | spin_unlock(&old->i_lock); | |
1315 | spin_unlock(&inode_hash_lock); | |
1316 | wait_on_inode(old); | |
1317 | if (unlikely(!inode_unhashed(old))) { | |
1318 | iput(old); | |
1319 | return -EBUSY; | |
1320 | } | |
1321 | iput(old); | |
1322 | } | |
1323 | } | |
1324 | EXPORT_SYMBOL(insert_inode_locked); | |
1325 | ||
1326 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1327 | int (*test)(struct inode *, void *), void *data) | |
1328 | { | |
1329 | struct super_block *sb = inode->i_sb; | |
1330 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1331 | ||
1332 | while (1) { | |
1333 | struct inode *old = NULL; | |
1334 | ||
1335 | spin_lock(&inode_hash_lock); | |
1336 | hlist_for_each_entry(old, head, i_hash) { | |
1337 | if (old->i_sb != sb) | |
1338 | continue; | |
1339 | if (!test(old, data)) | |
1340 | continue; | |
1341 | spin_lock(&old->i_lock); | |
1342 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1343 | spin_unlock(&old->i_lock); | |
1344 | continue; | |
1345 | } | |
1346 | break; | |
1347 | } | |
1348 | if (likely(!old)) { | |
1349 | spin_lock(&inode->i_lock); | |
1350 | inode->i_state |= I_NEW; | |
1351 | hlist_add_head(&inode->i_hash, head); | |
1352 | spin_unlock(&inode->i_lock); | |
1353 | spin_unlock(&inode_hash_lock); | |
1354 | return 0; | |
1355 | } | |
1356 | __iget(old); | |
1357 | spin_unlock(&old->i_lock); | |
1358 | spin_unlock(&inode_hash_lock); | |
1359 | wait_on_inode(old); | |
1360 | if (unlikely(!inode_unhashed(old))) { | |
1361 | iput(old); | |
1362 | return -EBUSY; | |
1363 | } | |
1364 | iput(old); | |
1365 | } | |
1366 | } | |
1367 | EXPORT_SYMBOL(insert_inode_locked4); | |
1368 | ||
1369 | ||
1370 | int generic_delete_inode(struct inode *inode) | |
1371 | { | |
1372 | return 1; | |
1373 | } | |
1374 | EXPORT_SYMBOL(generic_delete_inode); | |
1375 | ||
1376 | /* | |
1377 | * Called when we're dropping the last reference | |
1378 | * to an inode. | |
1379 | * | |
1380 | * Call the FS "drop_inode()" function, defaulting to | |
1381 | * the legacy UNIX filesystem behaviour. If it tells | |
1382 | * us to evict inode, do so. Otherwise, retain inode | |
1383 | * in cache if fs is alive, sync and evict if fs is | |
1384 | * shutting down. | |
1385 | */ | |
1386 | static void iput_final(struct inode *inode) | |
1387 | { | |
1388 | struct super_block *sb = inode->i_sb; | |
1389 | const struct super_operations *op = inode->i_sb->s_op; | |
1390 | int drop; | |
1391 | ||
1392 | WARN_ON(inode->i_state & I_NEW); | |
1393 | ||
1394 | if (op->drop_inode) | |
1395 | drop = op->drop_inode(inode); | |
1396 | else | |
1397 | drop = generic_drop_inode(inode); | |
1398 | ||
1399 | if (!drop && (sb->s_flags & MS_ACTIVE)) { | |
1400 | inode->i_state |= I_REFERENCED; | |
1401 | inode_add_lru(inode); | |
1402 | spin_unlock(&inode->i_lock); | |
1403 | return; | |
1404 | } | |
1405 | ||
1406 | if (!drop) { | |
1407 | inode->i_state |= I_WILL_FREE; | |
1408 | spin_unlock(&inode->i_lock); | |
1409 | write_inode_now(inode, 1); | |
1410 | spin_lock(&inode->i_lock); | |
1411 | WARN_ON(inode->i_state & I_NEW); | |
1412 | inode->i_state &= ~I_WILL_FREE; | |
1413 | } | |
1414 | ||
1415 | inode->i_state |= I_FREEING; | |
1416 | if (!list_empty(&inode->i_lru)) | |
1417 | inode_lru_list_del(inode); | |
1418 | spin_unlock(&inode->i_lock); | |
1419 | ||
1420 | evict(inode); | |
1421 | } | |
1422 | ||
1423 | /** | |
1424 | * iput - put an inode | |
1425 | * @inode: inode to put | |
1426 | * | |
1427 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1428 | * zero, the inode is then freed and may also be destroyed. | |
1429 | * | |
1430 | * Consequently, iput() can sleep. | |
1431 | */ | |
1432 | void iput(struct inode *inode) | |
1433 | { | |
1434 | if (inode) { | |
1435 | BUG_ON(inode->i_state & I_CLEAR); | |
1436 | ||
1437 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) | |
1438 | iput_final(inode); | |
1439 | } | |
1440 | } | |
1441 | EXPORT_SYMBOL(iput); | |
1442 | ||
1443 | /** | |
1444 | * bmap - find a block number in a file | |
1445 | * @inode: inode of file | |
1446 | * @block: block to find | |
1447 | * | |
1448 | * Returns the block number on the device holding the inode that | |
1449 | * is the disk block number for the block of the file requested. | |
1450 | * That is, asked for block 4 of inode 1 the function will return the | |
1451 | * disk block relative to the disk start that holds that block of the | |
1452 | * file. | |
1453 | */ | |
1454 | sector_t bmap(struct inode *inode, sector_t block) | |
1455 | { | |
1456 | sector_t res = 0; | |
1457 | if (inode->i_mapping->a_ops->bmap) | |
1458 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1459 | return res; | |
1460 | } | |
1461 | EXPORT_SYMBOL(bmap); | |
1462 | ||
1463 | /* | |
1464 | * With relative atime, only update atime if the previous atime is | |
1465 | * earlier than either the ctime or mtime or if at least a day has | |
1466 | * passed since the last atime update. | |
1467 | */ | |
1468 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1469 | struct timespec now) | |
1470 | { | |
1471 | ||
1472 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1473 | return 1; | |
1474 | /* | |
1475 | * Is mtime younger than atime? If yes, update atime: | |
1476 | */ | |
1477 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1478 | return 1; | |
1479 | /* | |
1480 | * Is ctime younger than atime? If yes, update atime: | |
1481 | */ | |
1482 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1483 | return 1; | |
1484 | ||
1485 | /* | |
1486 | * Is the previous atime value older than a day? If yes, | |
1487 | * update atime: | |
1488 | */ | |
1489 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1490 | return 1; | |
1491 | /* | |
1492 | * Good, we can skip the atime update: | |
1493 | */ | |
1494 | return 0; | |
1495 | } | |
1496 | ||
1497 | /* | |
1498 | * This does the actual work of updating an inodes time or version. Must have | |
1499 | * had called mnt_want_write() before calling this. | |
1500 | */ | |
1501 | static int update_time(struct inode *inode, struct timespec *time, int flags) | |
1502 | { | |
1503 | if (inode->i_op->update_time) | |
1504 | return inode->i_op->update_time(inode, time, flags); | |
1505 | ||
1506 | if (flags & S_ATIME) | |
1507 | inode->i_atime = *time; | |
1508 | if (flags & S_VERSION) | |
1509 | inode_inc_iversion(inode); | |
1510 | if (flags & S_CTIME) | |
1511 | inode->i_ctime = *time; | |
1512 | if (flags & S_MTIME) | |
1513 | inode->i_mtime = *time; | |
1514 | mark_inode_dirty_sync(inode); | |
1515 | return 0; | |
1516 | } | |
1517 | ||
1518 | /** | |
1519 | * touch_atime - update the access time | |
1520 | * @path: the &struct path to update | |
1521 | * | |
1522 | * Update the accessed time on an inode and mark it for writeback. | |
1523 | * This function automatically handles read only file systems and media, | |
1524 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1525 | */ | |
1526 | void touch_atime(const struct path *path) | |
1527 | { | |
1528 | struct vfsmount *mnt = path->mnt; | |
1529 | struct inode *inode = path->dentry->d_inode; | |
1530 | struct timespec now; | |
1531 | ||
1532 | if (inode->i_flags & S_NOATIME) | |
1533 | return; | |
1534 | if (IS_NOATIME(inode)) | |
1535 | return; | |
1536 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1537 | return; | |
1538 | ||
1539 | if (mnt->mnt_flags & MNT_NOATIME) | |
1540 | return; | |
1541 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1542 | return; | |
1543 | ||
1544 | now = current_fs_time(inode->i_sb); | |
1545 | ||
1546 | if (!relatime_need_update(mnt, inode, now)) | |
1547 | return; | |
1548 | ||
1549 | if (timespec_equal(&inode->i_atime, &now)) | |
1550 | return; | |
1551 | ||
1552 | if (!sb_start_write_trylock(inode->i_sb)) | |
1553 | return; | |
1554 | ||
1555 | if (__mnt_want_write(mnt)) | |
1556 | goto skip_update; | |
1557 | /* | |
1558 | * File systems can error out when updating inodes if they need to | |
1559 | * allocate new space to modify an inode (such is the case for | |
1560 | * Btrfs), but since we touch atime while walking down the path we | |
1561 | * really don't care if we failed to update the atime of the file, | |
1562 | * so just ignore the return value. | |
1563 | * We may also fail on filesystems that have the ability to make parts | |
1564 | * of the fs read only, e.g. subvolumes in Btrfs. | |
1565 | */ | |
1566 | update_time(inode, &now, S_ATIME); | |
1567 | __mnt_drop_write(mnt); | |
1568 | skip_update: | |
1569 | sb_end_write(inode->i_sb); | |
1570 | } | |
1571 | EXPORT_SYMBOL(touch_atime); | |
1572 | ||
1573 | /* | |
1574 | * The logic we want is | |
1575 | * | |
1576 | * if suid or (sgid and xgrp) | |
1577 | * remove privs | |
1578 | */ | |
1579 | int should_remove_suid(struct dentry *dentry) | |
1580 | { | |
1581 | umode_t mode = dentry->d_inode->i_mode; | |
1582 | int kill = 0; | |
1583 | ||
1584 | /* suid always must be killed */ | |
1585 | if (unlikely(mode & S_ISUID)) | |
1586 | kill = ATTR_KILL_SUID; | |
1587 | ||
1588 | /* | |
1589 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1590 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1591 | */ | |
1592 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1593 | kill |= ATTR_KILL_SGID; | |
1594 | ||
1595 | if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) | |
1596 | return kill; | |
1597 | ||
1598 | return 0; | |
1599 | } | |
1600 | EXPORT_SYMBOL(should_remove_suid); | |
1601 | ||
1602 | static int __remove_suid(struct dentry *dentry, int kill) | |
1603 | { | |
1604 | struct iattr newattrs; | |
1605 | ||
1606 | newattrs.ia_valid = ATTR_FORCE | kill; | |
1607 | /* | |
1608 | * Note we call this on write, so notify_change will not | |
1609 | * encounter any conflicting delegations: | |
1610 | */ | |
1611 | return notify_change(dentry, &newattrs, NULL); | |
1612 | } | |
1613 | ||
1614 | int file_remove_suid(struct file *file) | |
1615 | { | |
1616 | struct dentry *dentry = file->f_path.dentry; | |
1617 | struct inode *inode = dentry->d_inode; | |
1618 | int killsuid; | |
1619 | int killpriv; | |
1620 | int error = 0; | |
1621 | ||
1622 | /* Fast path for nothing security related */ | |
1623 | if (IS_NOSEC(inode)) | |
1624 | return 0; | |
1625 | ||
1626 | killsuid = should_remove_suid(dentry); | |
1627 | killpriv = security_inode_need_killpriv(dentry); | |
1628 | ||
1629 | if (killpriv < 0) | |
1630 | return killpriv; | |
1631 | if (killpriv) | |
1632 | error = security_inode_killpriv(dentry); | |
1633 | if (!error && killsuid) | |
1634 | error = __remove_suid(dentry, killsuid); | |
1635 | if (!error && (inode->i_sb->s_flags & MS_NOSEC)) | |
1636 | inode->i_flags |= S_NOSEC; | |
1637 | ||
1638 | return error; | |
1639 | } | |
1640 | EXPORT_SYMBOL(file_remove_suid); | |
1641 | ||
1642 | /** | |
1643 | * file_update_time - update mtime and ctime time | |
1644 | * @file: file accessed | |
1645 | * | |
1646 | * Update the mtime and ctime members of an inode and mark the inode | |
1647 | * for writeback. Note that this function is meant exclusively for | |
1648 | * usage in the file write path of filesystems, and filesystems may | |
1649 | * choose to explicitly ignore update via this function with the | |
1650 | * S_NOCMTIME inode flag, e.g. for network filesystem where these | |
1651 | * timestamps are handled by the server. This can return an error for | |
1652 | * file systems who need to allocate space in order to update an inode. | |
1653 | */ | |
1654 | ||
1655 | int file_update_time(struct file *file) | |
1656 | { | |
1657 | struct inode *inode = file_inode(file); | |
1658 | struct timespec now; | |
1659 | int sync_it = 0; | |
1660 | int ret; | |
1661 | ||
1662 | /* First try to exhaust all avenues to not sync */ | |
1663 | if (IS_NOCMTIME(inode)) | |
1664 | return 0; | |
1665 | ||
1666 | now = current_fs_time(inode->i_sb); | |
1667 | if (!timespec_equal(&inode->i_mtime, &now)) | |
1668 | sync_it = S_MTIME; | |
1669 | ||
1670 | if (!timespec_equal(&inode->i_ctime, &now)) | |
1671 | sync_it |= S_CTIME; | |
1672 | ||
1673 | if (IS_I_VERSION(inode)) | |
1674 | sync_it |= S_VERSION; | |
1675 | ||
1676 | if (!sync_it) | |
1677 | return 0; | |
1678 | ||
1679 | /* Finally allowed to write? Takes lock. */ | |
1680 | if (__mnt_want_write_file(file)) | |
1681 | return 0; | |
1682 | ||
1683 | ret = update_time(inode, &now, sync_it); | |
1684 | __mnt_drop_write_file(file); | |
1685 | ||
1686 | return ret; | |
1687 | } | |
1688 | EXPORT_SYMBOL(file_update_time); | |
1689 | ||
1690 | int inode_needs_sync(struct inode *inode) | |
1691 | { | |
1692 | if (IS_SYNC(inode)) | |
1693 | return 1; | |
1694 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1695 | return 1; | |
1696 | return 0; | |
1697 | } | |
1698 | EXPORT_SYMBOL(inode_needs_sync); | |
1699 | ||
1700 | /* | |
1701 | * If we try to find an inode in the inode hash while it is being | |
1702 | * deleted, we have to wait until the filesystem completes its | |
1703 | * deletion before reporting that it isn't found. This function waits | |
1704 | * until the deletion _might_ have completed. Callers are responsible | |
1705 | * to recheck inode state. | |
1706 | * | |
1707 | * It doesn't matter if I_NEW is not set initially, a call to | |
1708 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list | |
1709 | * will DTRT. | |
1710 | */ | |
1711 | static void __wait_on_freeing_inode(struct inode *inode) | |
1712 | { | |
1713 | wait_queue_head_t *wq; | |
1714 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); | |
1715 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1716 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
1717 | spin_unlock(&inode->i_lock); | |
1718 | spin_unlock(&inode_hash_lock); | |
1719 | schedule(); | |
1720 | finish_wait(wq, &wait.wait); | |
1721 | spin_lock(&inode_hash_lock); | |
1722 | } | |
1723 | ||
1724 | static __initdata unsigned long ihash_entries; | |
1725 | static int __init set_ihash_entries(char *str) | |
1726 | { | |
1727 | if (!str) | |
1728 | return 0; | |
1729 | ihash_entries = simple_strtoul(str, &str, 0); | |
1730 | return 1; | |
1731 | } | |
1732 | __setup("ihash_entries=", set_ihash_entries); | |
1733 | ||
1734 | /* | |
1735 | * Initialize the waitqueues and inode hash table. | |
1736 | */ | |
1737 | void __init inode_init_early(void) | |
1738 | { | |
1739 | unsigned int loop; | |
1740 | ||
1741 | /* If hashes are distributed across NUMA nodes, defer | |
1742 | * hash allocation until vmalloc space is available. | |
1743 | */ | |
1744 | if (hashdist) | |
1745 | return; | |
1746 | ||
1747 | inode_hashtable = | |
1748 | alloc_large_system_hash("Inode-cache", | |
1749 | sizeof(struct hlist_head), | |
1750 | ihash_entries, | |
1751 | 14, | |
1752 | HASH_EARLY, | |
1753 | &i_hash_shift, | |
1754 | &i_hash_mask, | |
1755 | 0, | |
1756 | 0); | |
1757 | ||
1758 | for (loop = 0; loop < (1U << i_hash_shift); loop++) | |
1759 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1760 | } | |
1761 | ||
1762 | void __init inode_init(void) | |
1763 | { | |
1764 | unsigned int loop; | |
1765 | ||
1766 | /* inode slab cache */ | |
1767 | inode_cachep = kmem_cache_create("inode_cache", | |
1768 | sizeof(struct inode), | |
1769 | 0, | |
1770 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1771 | SLAB_MEM_SPREAD), | |
1772 | init_once); | |
1773 | ||
1774 | /* Hash may have been set up in inode_init_early */ | |
1775 | if (!hashdist) | |
1776 | return; | |
1777 | ||
1778 | inode_hashtable = | |
1779 | alloc_large_system_hash("Inode-cache", | |
1780 | sizeof(struct hlist_head), | |
1781 | ihash_entries, | |
1782 | 14, | |
1783 | 0, | |
1784 | &i_hash_shift, | |
1785 | &i_hash_mask, | |
1786 | 0, | |
1787 | 0); | |
1788 | ||
1789 | for (loop = 0; loop < (1U << i_hash_shift); loop++) | |
1790 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1791 | } | |
1792 | ||
1793 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1794 | { | |
1795 | inode->i_mode = mode; | |
1796 | if (S_ISCHR(mode)) { | |
1797 | inode->i_fop = &def_chr_fops; | |
1798 | inode->i_rdev = rdev; | |
1799 | } else if (S_ISBLK(mode)) { | |
1800 | inode->i_fop = &def_blk_fops; | |
1801 | inode->i_rdev = rdev; | |
1802 | } else if (S_ISFIFO(mode)) | |
1803 | inode->i_fop = &pipefifo_fops; | |
1804 | else if (S_ISSOCK(mode)) | |
1805 | ; /* leave it no_open_fops */ | |
1806 | else | |
1807 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" | |
1808 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1809 | inode->i_ino); | |
1810 | } | |
1811 | EXPORT_SYMBOL(init_special_inode); | |
1812 | ||
1813 | /** | |
1814 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards | |
1815 | * @inode: New inode | |
1816 | * @dir: Directory inode | |
1817 | * @mode: mode of the new inode | |
1818 | */ | |
1819 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
1820 | umode_t mode) | |
1821 | { | |
1822 | inode->i_uid = current_fsuid(); | |
1823 | if (dir && dir->i_mode & S_ISGID) { | |
1824 | inode->i_gid = dir->i_gid; | |
1825 | if (S_ISDIR(mode)) | |
1826 | mode |= S_ISGID; | |
1827 | } else | |
1828 | inode->i_gid = current_fsgid(); | |
1829 | inode->i_mode = mode; | |
1830 | } | |
1831 | EXPORT_SYMBOL(inode_init_owner); | |
1832 | ||
1833 | /** | |
1834 | * inode_owner_or_capable - check current task permissions to inode | |
1835 | * @inode: inode being checked | |
1836 | * | |
1837 | * Return true if current either has CAP_FOWNER in a namespace with the | |
1838 | * inode owner uid mapped, or owns the file. | |
1839 | */ | |
1840 | bool inode_owner_or_capable(const struct inode *inode) | |
1841 | { | |
1842 | struct user_namespace *ns; | |
1843 | ||
1844 | if (uid_eq(current_fsuid(), inode->i_uid)) | |
1845 | return true; | |
1846 | ||
1847 | ns = current_user_ns(); | |
1848 | if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid)) | |
1849 | return true; | |
1850 | return false; | |
1851 | } | |
1852 | EXPORT_SYMBOL(inode_owner_or_capable); | |
1853 | ||
1854 | /* | |
1855 | * Direct i/o helper functions | |
1856 | */ | |
1857 | static void __inode_dio_wait(struct inode *inode) | |
1858 | { | |
1859 | wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP); | |
1860 | DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP); | |
1861 | ||
1862 | do { | |
1863 | prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE); | |
1864 | if (atomic_read(&inode->i_dio_count)) | |
1865 | schedule(); | |
1866 | } while (atomic_read(&inode->i_dio_count)); | |
1867 | finish_wait(wq, &q.wait); | |
1868 | } | |
1869 | ||
1870 | /** | |
1871 | * inode_dio_wait - wait for outstanding DIO requests to finish | |
1872 | * @inode: inode to wait for | |
1873 | * | |
1874 | * Waits for all pending direct I/O requests to finish so that we can | |
1875 | * proceed with a truncate or equivalent operation. | |
1876 | * | |
1877 | * Must be called under a lock that serializes taking new references | |
1878 | * to i_dio_count, usually by inode->i_mutex. | |
1879 | */ | |
1880 | void inode_dio_wait(struct inode *inode) | |
1881 | { | |
1882 | if (atomic_read(&inode->i_dio_count)) | |
1883 | __inode_dio_wait(inode); | |
1884 | } | |
1885 | EXPORT_SYMBOL(inode_dio_wait); | |
1886 | ||
1887 | /* | |
1888 | * inode_dio_done - signal finish of a direct I/O requests | |
1889 | * @inode: inode the direct I/O happens on | |
1890 | * | |
1891 | * This is called once we've finished processing a direct I/O request, | |
1892 | * and is used to wake up callers waiting for direct I/O to be quiesced. | |
1893 | */ | |
1894 | void inode_dio_done(struct inode *inode) | |
1895 | { | |
1896 | if (atomic_dec_and_test(&inode->i_dio_count)) | |
1897 | wake_up_bit(&inode->i_state, __I_DIO_WAKEUP); | |
1898 | } | |
1899 | EXPORT_SYMBOL(inode_dio_done); | |
1900 | ||
1901 | /* | |
1902 | * inode_set_flags - atomically set some inode flags | |
1903 | * | |
1904 | * Note: the caller should be holding i_mutex, or else be sure that | |
1905 | * they have exclusive access to the inode structure (i.e., while the | |
1906 | * inode is being instantiated). The reason for the cmpxchg() loop | |
1907 | * --- which wouldn't be necessary if all code paths which modify | |
1908 | * i_flags actually followed this rule, is that there is at least one | |
1909 | * code path which doesn't today --- for example, | |
1910 | * __generic_file_aio_write() calls file_remove_suid() without holding | |
1911 | * i_mutex --- so we use cmpxchg() out of an abundance of caution. | |
1912 | * | |
1913 | * In the long run, i_mutex is overkill, and we should probably look | |
1914 | * at using the i_lock spinlock to protect i_flags, and then make sure | |
1915 | * it is so documented in include/linux/fs.h and that all code follows | |
1916 | * the locking convention!! | |
1917 | */ | |
1918 | void inode_set_flags(struct inode *inode, unsigned int flags, | |
1919 | unsigned int mask) | |
1920 | { | |
1921 | unsigned int old_flags, new_flags; | |
1922 | ||
1923 | WARN_ON_ONCE(flags & ~mask); | |
1924 | do { | |
1925 | old_flags = ACCESS_ONCE(inode->i_flags); | |
1926 | new_flags = (old_flags & ~mask) | flags; | |
1927 | } while (unlikely(cmpxchg(&inode->i_flags, old_flags, | |
1928 | new_flags) != old_flags)); | |
1929 | } | |
1930 | EXPORT_SYMBOL(inode_set_flags); |