]>
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 <trace/events/writeback.h> | |
22 | #include "internal.h" | |
23 | ||
24 | /* | |
25 | * Inode locking rules: | |
26 | * | |
27 | * inode->i_lock protects: | |
28 | * inode->i_state, inode->i_hash, __iget() | |
29 | * Inode LRU list locks protect: | |
30 | * inode->i_sb->s_inode_lru, inode->i_lru | |
31 | * inode->i_sb->s_inode_list_lock protects: | |
32 | * inode->i_sb->s_inodes, inode->i_sb_list | |
33 | * bdi->wb.list_lock protects: | |
34 | * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list | |
35 | * inode_hash_lock protects: | |
36 | * inode_hashtable, inode->i_hash | |
37 | * | |
38 | * Lock ordering: | |
39 | * | |
40 | * inode->i_sb->s_inode_list_lock | |
41 | * inode->i_lock | |
42 | * Inode LRU list locks | |
43 | * | |
44 | * bdi->wb.list_lock | |
45 | * inode->i_lock | |
46 | * | |
47 | * inode_hash_lock | |
48 | * inode->i_sb->s_inode_list_lock | |
49 | * inode->i_lock | |
50 | * | |
51 | * iunique_lock | |
52 | * inode_hash_lock | |
53 | */ | |
54 | ||
55 | static unsigned int i_hash_mask __read_mostly; | |
56 | static unsigned int i_hash_shift __read_mostly; | |
57 | static struct hlist_head *inode_hashtable __read_mostly; | |
58 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_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 long, nr_inodes); | |
74 | static DEFINE_PER_CPU(unsigned long, nr_unused); | |
75 | ||
76 | static struct kmem_cache *inode_cachep __read_mostly; | |
77 | ||
78 | static long get_nr_inodes(void) | |
79 | { | |
80 | int i; | |
81 | long 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 long get_nr_inodes_unused(void) | |
88 | { | |
89 | int i; | |
90 | long sum = 0; | |
91 | for_each_possible_cpu(i) | |
92 | sum += per_cpu(nr_unused, i); | |
93 | return sum < 0 ? 0 : sum; | |
94 | } | |
95 | ||
96 | long get_nr_dirty_inodes(void) | |
97 | { | |
98 | /* not actually dirty inodes, but a wild approximation */ | |
99 | long 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(struct 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_doulongvec_minmax(table, write, buffer, lenp, ppos); | |
113 | } | |
114 | #endif | |
115 | ||
116 | static int no_open(struct inode *inode, struct file *file) | |
117 | { | |
118 | return -ENXIO; | |
119 | } | |
120 | ||
121 | /** | |
122 | * inode_init_always - perform inode structure intialisation | |
123 | * @sb: superblock inode belongs to | |
124 | * @inode: inode to initialise | |
125 | * | |
126 | * These are initializations that need to be done on every inode | |
127 | * allocation as the fields are not initialised by slab allocation. | |
128 | */ | |
129 | int inode_init_always(struct super_block *sb, struct inode *inode) | |
130 | { | |
131 | static const struct inode_operations empty_iops; | |
132 | static const struct file_operations no_open_fops = {.open = no_open}; | |
133 | struct address_space *const mapping = &inode->i_data; | |
134 | ||
135 | inode->i_sb = sb; | |
136 | inode->i_blkbits = sb->s_blocksize_bits; | |
137 | inode->i_flags = 0; | |
138 | atomic_set(&inode->i_count, 1); | |
139 | inode->i_op = &empty_iops; | |
140 | inode->i_fop = &no_open_fops; | |
141 | inode->__i_nlink = 1; | |
142 | inode->i_opflags = 0; | |
143 | if (sb->s_xattr) | |
144 | inode->i_opflags |= IOP_XATTR; | |
145 | i_uid_write(inode, 0); | |
146 | i_gid_write(inode, 0); | |
147 | atomic_set(&inode->i_writecount, 0); | |
148 | inode->i_size = 0; | |
149 | inode->i_blocks = 0; | |
150 | inode->i_bytes = 0; | |
151 | inode->i_generation = 0; | |
152 | inode->i_pipe = NULL; | |
153 | inode->i_bdev = NULL; | |
154 | inode->i_cdev = NULL; | |
155 | inode->i_link = NULL; | |
156 | inode->i_dir_seq = 0; | |
157 | inode->i_rdev = 0; | |
158 | inode->dirtied_when = 0; | |
159 | ||
160 | #ifdef CONFIG_CGROUP_WRITEBACK | |
161 | inode->i_wb_frn_winner = 0; | |
162 | inode->i_wb_frn_avg_time = 0; | |
163 | inode->i_wb_frn_history = 0; | |
164 | #endif | |
165 | ||
166 | if (security_inode_alloc(inode)) | |
167 | goto out; | |
168 | spin_lock_init(&inode->i_lock); | |
169 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
170 | ||
171 | init_rwsem(&inode->i_rwsem); | |
172 | lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key); | |
173 | ||
174 | atomic_set(&inode->i_dio_count, 0); | |
175 | ||
176 | mapping->a_ops = &empty_aops; | |
177 | mapping->host = inode; | |
178 | mapping->flags = 0; | |
179 | atomic_set(&mapping->i_mmap_writable, 0); | |
180 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); | |
181 | mapping->private_data = NULL; | |
182 | mapping->writeback_index = 0; | |
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 | inode->i_flctx = NULL; | |
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 | inode_detach_wb(inode); | |
235 | security_inode_free(inode); | |
236 | fsnotify_inode_delete(inode); | |
237 | locks_free_lock_context(inode); | |
238 | if (!inode->i_nlink) { | |
239 | WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0); | |
240 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
241 | } | |
242 | ||
243 | #ifdef CONFIG_FS_POSIX_ACL | |
244 | if (inode->i_acl && !is_uncached_acl(inode->i_acl)) | |
245 | posix_acl_release(inode->i_acl); | |
246 | if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl)) | |
247 | posix_acl_release(inode->i_default_acl); | |
248 | #endif | |
249 | this_cpu_dec(nr_inodes); | |
250 | } | |
251 | EXPORT_SYMBOL(__destroy_inode); | |
252 | ||
253 | static void i_callback(struct rcu_head *head) | |
254 | { | |
255 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
256 | kmem_cache_free(inode_cachep, inode); | |
257 | } | |
258 | ||
259 | static void destroy_inode(struct inode *inode) | |
260 | { | |
261 | BUG_ON(!list_empty(&inode->i_lru)); | |
262 | __destroy_inode(inode); | |
263 | if (inode->i_sb->s_op->destroy_inode) | |
264 | inode->i_sb->s_op->destroy_inode(inode); | |
265 | else | |
266 | call_rcu(&inode->i_rcu, i_callback); | |
267 | } | |
268 | ||
269 | /** | |
270 | * drop_nlink - directly drop an inode's link count | |
271 | * @inode: inode | |
272 | * | |
273 | * This is a low-level filesystem helper to replace any | |
274 | * direct filesystem manipulation of i_nlink. In cases | |
275 | * where we are attempting to track writes to the | |
276 | * filesystem, a decrement to zero means an imminent | |
277 | * write when the file is truncated and actually unlinked | |
278 | * on the filesystem. | |
279 | */ | |
280 | void drop_nlink(struct inode *inode) | |
281 | { | |
282 | WARN_ON(inode->i_nlink == 0); | |
283 | inode->__i_nlink--; | |
284 | if (!inode->i_nlink) | |
285 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
286 | } | |
287 | EXPORT_SYMBOL(drop_nlink); | |
288 | ||
289 | /** | |
290 | * clear_nlink - directly zero an inode's link count | |
291 | * @inode: inode | |
292 | * | |
293 | * This is a low-level filesystem helper to replace any | |
294 | * direct filesystem manipulation of i_nlink. See | |
295 | * drop_nlink() for why we care about i_nlink hitting zero. | |
296 | */ | |
297 | void clear_nlink(struct inode *inode) | |
298 | { | |
299 | if (inode->i_nlink) { | |
300 | inode->__i_nlink = 0; | |
301 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
302 | } | |
303 | } | |
304 | EXPORT_SYMBOL(clear_nlink); | |
305 | ||
306 | /** | |
307 | * set_nlink - directly set an inode's link count | |
308 | * @inode: inode | |
309 | * @nlink: new nlink (should be non-zero) | |
310 | * | |
311 | * This is a low-level filesystem helper to replace any | |
312 | * direct filesystem manipulation of i_nlink. | |
313 | */ | |
314 | void set_nlink(struct inode *inode, unsigned int nlink) | |
315 | { | |
316 | if (!nlink) { | |
317 | clear_nlink(inode); | |
318 | } else { | |
319 | /* Yes, some filesystems do change nlink from zero to one */ | |
320 | if (inode->i_nlink == 0) | |
321 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
322 | ||
323 | inode->__i_nlink = nlink; | |
324 | } | |
325 | } | |
326 | EXPORT_SYMBOL(set_nlink); | |
327 | ||
328 | /** | |
329 | * inc_nlink - directly increment an inode's link count | |
330 | * @inode: inode | |
331 | * | |
332 | * This is a low-level filesystem helper to replace any | |
333 | * direct filesystem manipulation of i_nlink. Currently, | |
334 | * it is only here for parity with dec_nlink(). | |
335 | */ | |
336 | void inc_nlink(struct inode *inode) | |
337 | { | |
338 | if (unlikely(inode->i_nlink == 0)) { | |
339 | WARN_ON(!(inode->i_state & I_LINKABLE)); | |
340 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
341 | } | |
342 | ||
343 | inode->__i_nlink++; | |
344 | } | |
345 | EXPORT_SYMBOL(inc_nlink); | |
346 | ||
347 | void address_space_init_once(struct address_space *mapping) | |
348 | { | |
349 | memset(mapping, 0, sizeof(*mapping)); | |
350 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC | __GFP_ACCOUNT); | |
351 | spin_lock_init(&mapping->tree_lock); | |
352 | init_rwsem(&mapping->i_mmap_rwsem); | |
353 | INIT_LIST_HEAD(&mapping->private_list); | |
354 | spin_lock_init(&mapping->private_lock); | |
355 | mapping->i_mmap = RB_ROOT; | |
356 | } | |
357 | EXPORT_SYMBOL(address_space_init_once); | |
358 | ||
359 | /* | |
360 | * These are initializations that only need to be done | |
361 | * once, because the fields are idempotent across use | |
362 | * of the inode, so let the slab aware of that. | |
363 | */ | |
364 | void inode_init_once(struct inode *inode) | |
365 | { | |
366 | memset(inode, 0, sizeof(*inode)); | |
367 | INIT_HLIST_NODE(&inode->i_hash); | |
368 | INIT_LIST_HEAD(&inode->i_devices); | |
369 | INIT_LIST_HEAD(&inode->i_io_list); | |
370 | INIT_LIST_HEAD(&inode->i_wb_list); | |
371 | INIT_LIST_HEAD(&inode->i_lru); | |
372 | address_space_init_once(&inode->i_data); | |
373 | i_size_ordered_init(inode); | |
374 | #ifdef CONFIG_FSNOTIFY | |
375 | INIT_HLIST_HEAD(&inode->i_fsnotify_marks); | |
376 | #endif | |
377 | } | |
378 | EXPORT_SYMBOL(inode_init_once); | |
379 | ||
380 | static void init_once(void *foo) | |
381 | { | |
382 | struct inode *inode = (struct inode *) foo; | |
383 | ||
384 | inode_init_once(inode); | |
385 | } | |
386 | ||
387 | /* | |
388 | * inode->i_lock must be held | |
389 | */ | |
390 | void __iget(struct inode *inode) | |
391 | { | |
392 | atomic_inc(&inode->i_count); | |
393 | } | |
394 | ||
395 | /* | |
396 | * get additional reference to inode; caller must already hold one. | |
397 | */ | |
398 | void ihold(struct inode *inode) | |
399 | { | |
400 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); | |
401 | } | |
402 | EXPORT_SYMBOL(ihold); | |
403 | ||
404 | static void inode_lru_list_add(struct inode *inode) | |
405 | { | |
406 | if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru)) | |
407 | this_cpu_inc(nr_unused); | |
408 | } | |
409 | ||
410 | /* | |
411 | * Add inode to LRU if needed (inode is unused and clean). | |
412 | * | |
413 | * Needs inode->i_lock held. | |
414 | */ | |
415 | void inode_add_lru(struct inode *inode) | |
416 | { | |
417 | if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC | | |
418 | 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 | ||
427 | if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru)) | |
428 | this_cpu_dec(nr_unused); | |
429 | } | |
430 | ||
431 | /** | |
432 | * inode_sb_list_add - add inode to the superblock list of inodes | |
433 | * @inode: inode to add | |
434 | */ | |
435 | void inode_sb_list_add(struct inode *inode) | |
436 | { | |
437 | spin_lock(&inode->i_sb->s_inode_list_lock); | |
438 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); | |
439 | spin_unlock(&inode->i_sb->s_inode_list_lock); | |
440 | } | |
441 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
442 | ||
443 | static inline void inode_sb_list_del(struct inode *inode) | |
444 | { | |
445 | if (!list_empty(&inode->i_sb_list)) { | |
446 | spin_lock(&inode->i_sb->s_inode_list_lock); | |
447 | list_del_init(&inode->i_sb_list); | |
448 | spin_unlock(&inode->i_sb->s_inode_list_lock); | |
449 | } | |
450 | } | |
451 | ||
452 | static unsigned long hash(struct super_block *sb, unsigned long hashval) | |
453 | { | |
454 | unsigned long tmp; | |
455 | ||
456 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
457 | L1_CACHE_BYTES; | |
458 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift); | |
459 | return tmp & i_hash_mask; | |
460 | } | |
461 | ||
462 | /** | |
463 | * __insert_inode_hash - hash an inode | |
464 | * @inode: unhashed inode | |
465 | * @hashval: unsigned long value used to locate this object in the | |
466 | * inode_hashtable. | |
467 | * | |
468 | * Add an inode to the inode hash for this superblock. | |
469 | */ | |
470 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
471 | { | |
472 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); | |
473 | ||
474 | spin_lock(&inode_hash_lock); | |
475 | spin_lock(&inode->i_lock); | |
476 | hlist_add_head(&inode->i_hash, b); | |
477 | spin_unlock(&inode->i_lock); | |
478 | spin_unlock(&inode_hash_lock); | |
479 | } | |
480 | EXPORT_SYMBOL(__insert_inode_hash); | |
481 | ||
482 | /** | |
483 | * __remove_inode_hash - remove an inode from the hash | |
484 | * @inode: inode to unhash | |
485 | * | |
486 | * Remove an inode from the superblock. | |
487 | */ | |
488 | void __remove_inode_hash(struct inode *inode) | |
489 | { | |
490 | spin_lock(&inode_hash_lock); | |
491 | spin_lock(&inode->i_lock); | |
492 | hlist_del_init(&inode->i_hash); | |
493 | spin_unlock(&inode->i_lock); | |
494 | spin_unlock(&inode_hash_lock); | |
495 | } | |
496 | EXPORT_SYMBOL(__remove_inode_hash); | |
497 | ||
498 | void clear_inode(struct inode *inode) | |
499 | { | |
500 | might_sleep(); | |
501 | /* | |
502 | * We have to cycle tree_lock here because reclaim can be still in the | |
503 | * process of removing the last page (in __delete_from_page_cache()) | |
504 | * and we must not free mapping under it. | |
505 | */ | |
506 | spin_lock_irq(&inode->i_data.tree_lock); | |
507 | BUG_ON(inode->i_data.nrpages); | |
508 | BUG_ON(inode->i_data.nrexceptional); | |
509 | spin_unlock_irq(&inode->i_data.tree_lock); | |
510 | BUG_ON(!list_empty(&inode->i_data.private_list)); | |
511 | BUG_ON(!(inode->i_state & I_FREEING)); | |
512 | BUG_ON(inode->i_state & I_CLEAR); | |
513 | BUG_ON(!list_empty(&inode->i_wb_list)); | |
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_io_list)) | |
540 | inode_io_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 | cond_resched(); | |
590 | } | |
591 | } | |
592 | ||
593 | /** | |
594 | * evict_inodes - evict all evictable inodes for a superblock | |
595 | * @sb: superblock to operate on | |
596 | * | |
597 | * Make sure that no inodes with zero refcount are retained. This is | |
598 | * called by superblock shutdown after having MS_ACTIVE flag removed, | |
599 | * so any inode reaching zero refcount during or after that call will | |
600 | * be immediately evicted. | |
601 | */ | |
602 | void evict_inodes(struct super_block *sb) | |
603 | { | |
604 | struct inode *inode, *next; | |
605 | LIST_HEAD(dispose); | |
606 | ||
607 | again: | |
608 | spin_lock(&sb->s_inode_list_lock); | |
609 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { | |
610 | if (atomic_read(&inode->i_count)) | |
611 | continue; | |
612 | ||
613 | spin_lock(&inode->i_lock); | |
614 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
615 | spin_unlock(&inode->i_lock); | |
616 | continue; | |
617 | } | |
618 | ||
619 | inode->i_state |= I_FREEING; | |
620 | inode_lru_list_del(inode); | |
621 | spin_unlock(&inode->i_lock); | |
622 | list_add(&inode->i_lru, &dispose); | |
623 | ||
624 | /* | |
625 | * We can have a ton of inodes to evict at unmount time given | |
626 | * enough memory, check to see if we need to go to sleep for a | |
627 | * bit so we don't livelock. | |
628 | */ | |
629 | if (need_resched()) { | |
630 | spin_unlock(&sb->s_inode_list_lock); | |
631 | cond_resched(); | |
632 | dispose_list(&dispose); | |
633 | goto again; | |
634 | } | |
635 | } | |
636 | spin_unlock(&sb->s_inode_list_lock); | |
637 | ||
638 | dispose_list(&dispose); | |
639 | } | |
640 | ||
641 | /** | |
642 | * invalidate_inodes - attempt to free all inodes on a superblock | |
643 | * @sb: superblock to operate on | |
644 | * @kill_dirty: flag to guide handling of dirty inodes | |
645 | * | |
646 | * Attempts to free all inodes for a given superblock. If there were any | |
647 | * busy inodes return a non-zero value, else zero. | |
648 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat | |
649 | * them as busy. | |
650 | */ | |
651 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) | |
652 | { | |
653 | int busy = 0; | |
654 | struct inode *inode, *next; | |
655 | LIST_HEAD(dispose); | |
656 | ||
657 | spin_lock(&sb->s_inode_list_lock); | |
658 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { | |
659 | spin_lock(&inode->i_lock); | |
660 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
661 | spin_unlock(&inode->i_lock); | |
662 | continue; | |
663 | } | |
664 | if (inode->i_state & I_DIRTY_ALL && !kill_dirty) { | |
665 | spin_unlock(&inode->i_lock); | |
666 | busy = 1; | |
667 | continue; | |
668 | } | |
669 | if (atomic_read(&inode->i_count)) { | |
670 | spin_unlock(&inode->i_lock); | |
671 | busy = 1; | |
672 | continue; | |
673 | } | |
674 | ||
675 | inode->i_state |= I_FREEING; | |
676 | inode_lru_list_del(inode); | |
677 | spin_unlock(&inode->i_lock); | |
678 | list_add(&inode->i_lru, &dispose); | |
679 | } | |
680 | spin_unlock(&sb->s_inode_list_lock); | |
681 | ||
682 | dispose_list(&dispose); | |
683 | ||
684 | return busy; | |
685 | } | |
686 | ||
687 | /* | |
688 | * Isolate the inode from the LRU in preparation for freeing it. | |
689 | * | |
690 | * Any inodes which are pinned purely because of attached pagecache have their | |
691 | * pagecache removed. If the inode has metadata buffers attached to | |
692 | * mapping->private_list then try to remove them. | |
693 | * | |
694 | * If the inode has the I_REFERENCED flag set, then it means that it has been | |
695 | * used recently - the flag is set in iput_final(). When we encounter such an | |
696 | * inode, clear the flag and move it to the back of the LRU so it gets another | |
697 | * pass through the LRU before it gets reclaimed. This is necessary because of | |
698 | * the fact we are doing lazy LRU updates to minimise lock contention so the | |
699 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes | |
700 | * with this flag set because they are the inodes that are out of order. | |
701 | */ | |
702 | static enum lru_status inode_lru_isolate(struct list_head *item, | |
703 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
704 | { | |
705 | struct list_head *freeable = arg; | |
706 | struct inode *inode = container_of(item, struct inode, i_lru); | |
707 | ||
708 | /* | |
709 | * we are inverting the lru lock/inode->i_lock here, so use a trylock. | |
710 | * If we fail to get the lock, just skip it. | |
711 | */ | |
712 | if (!spin_trylock(&inode->i_lock)) | |
713 | return LRU_SKIP; | |
714 | ||
715 | /* | |
716 | * Referenced or dirty inodes are still in use. Give them another pass | |
717 | * through the LRU as we canot reclaim them now. | |
718 | */ | |
719 | if (atomic_read(&inode->i_count) || | |
720 | (inode->i_state & ~I_REFERENCED)) { | |
721 | list_lru_isolate(lru, &inode->i_lru); | |
722 | spin_unlock(&inode->i_lock); | |
723 | this_cpu_dec(nr_unused); | |
724 | return LRU_REMOVED; | |
725 | } | |
726 | ||
727 | /* recently referenced inodes get one more pass */ | |
728 | if (inode->i_state & I_REFERENCED) { | |
729 | inode->i_state &= ~I_REFERENCED; | |
730 | spin_unlock(&inode->i_lock); | |
731 | return LRU_ROTATE; | |
732 | } | |
733 | ||
734 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
735 | __iget(inode); | |
736 | spin_unlock(&inode->i_lock); | |
737 | spin_unlock(lru_lock); | |
738 | if (remove_inode_buffers(inode)) { | |
739 | unsigned long reap; | |
740 | reap = invalidate_mapping_pages(&inode->i_data, 0, -1); | |
741 | if (current_is_kswapd()) | |
742 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
743 | else | |
744 | __count_vm_events(PGINODESTEAL, reap); | |
745 | if (current->reclaim_state) | |
746 | current->reclaim_state->reclaimed_slab += reap; | |
747 | } | |
748 | iput(inode); | |
749 | spin_lock(lru_lock); | |
750 | return LRU_RETRY; | |
751 | } | |
752 | ||
753 | WARN_ON(inode->i_state & I_NEW); | |
754 | inode->i_state |= I_FREEING; | |
755 | list_lru_isolate_move(lru, &inode->i_lru, freeable); | |
756 | spin_unlock(&inode->i_lock); | |
757 | ||
758 | this_cpu_dec(nr_unused); | |
759 | return LRU_REMOVED; | |
760 | } | |
761 | ||
762 | /* | |
763 | * Walk the superblock inode LRU for freeable inodes and attempt to free them. | |
764 | * This is called from the superblock shrinker function with a number of inodes | |
765 | * to trim from the LRU. Inodes to be freed are moved to a temporary list and | |
766 | * then are freed outside inode_lock by dispose_list(). | |
767 | */ | |
768 | long prune_icache_sb(struct super_block *sb, struct shrink_control *sc) | |
769 | { | |
770 | LIST_HEAD(freeable); | |
771 | long freed; | |
772 | ||
773 | freed = list_lru_shrink_walk(&sb->s_inode_lru, sc, | |
774 | inode_lru_isolate, &freeable); | |
775 | dispose_list(&freeable); | |
776 | return freed; | |
777 | } | |
778 | ||
779 | static void __wait_on_freeing_inode(struct inode *inode); | |
780 | /* | |
781 | * Called with the inode lock held. | |
782 | */ | |
783 | static struct inode *find_inode(struct super_block *sb, | |
784 | struct hlist_head *head, | |
785 | int (*test)(struct inode *, void *), | |
786 | void *data) | |
787 | { | |
788 | struct inode *inode = NULL; | |
789 | ||
790 | repeat: | |
791 | hlist_for_each_entry(inode, head, i_hash) { | |
792 | if (inode->i_sb != sb) | |
793 | continue; | |
794 | if (!test(inode, data)) | |
795 | continue; | |
796 | spin_lock(&inode->i_lock); | |
797 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { | |
798 | __wait_on_freeing_inode(inode); | |
799 | goto repeat; | |
800 | } | |
801 | __iget(inode); | |
802 | spin_unlock(&inode->i_lock); | |
803 | return inode; | |
804 | } | |
805 | return NULL; | |
806 | } | |
807 | ||
808 | /* | |
809 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
810 | * iget_locked for details. | |
811 | */ | |
812 | static struct inode *find_inode_fast(struct super_block *sb, | |
813 | struct hlist_head *head, unsigned long ino) | |
814 | { | |
815 | struct inode *inode = NULL; | |
816 | ||
817 | repeat: | |
818 | hlist_for_each_entry(inode, head, i_hash) { | |
819 | if (inode->i_ino != ino) | |
820 | continue; | |
821 | if (inode->i_sb != sb) | |
822 | continue; | |
823 | spin_lock(&inode->i_lock); | |
824 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { | |
825 | __wait_on_freeing_inode(inode); | |
826 | goto repeat; | |
827 | } | |
828 | __iget(inode); | |
829 | spin_unlock(&inode->i_lock); | |
830 | return inode; | |
831 | } | |
832 | return NULL; | |
833 | } | |
834 | ||
835 | /* | |
836 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
837 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
838 | * to renew the exhausted range. | |
839 | * | |
840 | * This does not significantly increase overflow rate because every CPU can | |
841 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
842 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
843 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
844 | * overflow rate by 2x, which does not seem too significant. | |
845 | * | |
846 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
847 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
848 | * here to attempt to avoid that. | |
849 | */ | |
850 | #define LAST_INO_BATCH 1024 | |
851 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
852 | ||
853 | unsigned int get_next_ino(void) | |
854 | { | |
855 | unsigned int *p = &get_cpu_var(last_ino); | |
856 | unsigned int res = *p; | |
857 | ||
858 | #ifdef CONFIG_SMP | |
859 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
860 | static atomic_t shared_last_ino; | |
861 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
862 | ||
863 | res = next - LAST_INO_BATCH; | |
864 | } | |
865 | #endif | |
866 | ||
867 | res++; | |
868 | /* get_next_ino should not provide a 0 inode number */ | |
869 | if (unlikely(!res)) | |
870 | res++; | |
871 | *p = res; | |
872 | put_cpu_var(last_ino); | |
873 | return res; | |
874 | } | |
875 | EXPORT_SYMBOL(get_next_ino); | |
876 | ||
877 | /** | |
878 | * new_inode_pseudo - obtain an inode | |
879 | * @sb: superblock | |
880 | * | |
881 | * Allocates a new inode for given superblock. | |
882 | * Inode wont be chained in superblock s_inodes list | |
883 | * This means : | |
884 | * - fs can't be unmount | |
885 | * - quotas, fsnotify, writeback can't work | |
886 | */ | |
887 | struct inode *new_inode_pseudo(struct super_block *sb) | |
888 | { | |
889 | struct inode *inode = alloc_inode(sb); | |
890 | ||
891 | if (inode) { | |
892 | spin_lock(&inode->i_lock); | |
893 | inode->i_state = 0; | |
894 | spin_unlock(&inode->i_lock); | |
895 | INIT_LIST_HEAD(&inode->i_sb_list); | |
896 | } | |
897 | return inode; | |
898 | } | |
899 | ||
900 | /** | |
901 | * new_inode - obtain an inode | |
902 | * @sb: superblock | |
903 | * | |
904 | * Allocates a new inode for given superblock. The default gfp_mask | |
905 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. | |
906 | * If HIGHMEM pages are unsuitable or it is known that pages allocated | |
907 | * for the page cache are not reclaimable or migratable, | |
908 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
909 | * newly created inode's mapping | |
910 | * | |
911 | */ | |
912 | struct inode *new_inode(struct super_block *sb) | |
913 | { | |
914 | struct inode *inode; | |
915 | ||
916 | spin_lock_prefetch(&sb->s_inode_list_lock); | |
917 | ||
918 | inode = new_inode_pseudo(sb); | |
919 | if (inode) | |
920 | inode_sb_list_add(inode); | |
921 | return inode; | |
922 | } | |
923 | EXPORT_SYMBOL(new_inode); | |
924 | ||
925 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
926 | void lockdep_annotate_inode_mutex_key(struct inode *inode) | |
927 | { | |
928 | if (S_ISDIR(inode->i_mode)) { | |
929 | struct file_system_type *type = inode->i_sb->s_type; | |
930 | ||
931 | /* Set new key only if filesystem hasn't already changed it */ | |
932 | if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) { | |
933 | /* | |
934 | * ensure nobody is actually holding i_mutex | |
935 | */ | |
936 | // mutex_destroy(&inode->i_mutex); | |
937 | init_rwsem(&inode->i_rwsem); | |
938 | lockdep_set_class(&inode->i_rwsem, | |
939 | &type->i_mutex_dir_key); | |
940 | } | |
941 | } | |
942 | } | |
943 | EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key); | |
944 | #endif | |
945 | ||
946 | /** | |
947 | * unlock_new_inode - clear the I_NEW state and wake up any waiters | |
948 | * @inode: new inode to unlock | |
949 | * | |
950 | * Called when the inode is fully initialised to clear the new state of the | |
951 | * inode and wake up anyone waiting for the inode to finish initialisation. | |
952 | */ | |
953 | void unlock_new_inode(struct inode *inode) | |
954 | { | |
955 | lockdep_annotate_inode_mutex_key(inode); | |
956 | spin_lock(&inode->i_lock); | |
957 | WARN_ON(!(inode->i_state & I_NEW)); | |
958 | inode->i_state &= ~I_NEW; | |
959 | smp_mb(); | |
960 | wake_up_bit(&inode->i_state, __I_NEW); | |
961 | spin_unlock(&inode->i_lock); | |
962 | } | |
963 | EXPORT_SYMBOL(unlock_new_inode); | |
964 | ||
965 | /** | |
966 | * lock_two_nondirectories - take two i_mutexes on non-directory objects | |
967 | * | |
968 | * Lock any non-NULL argument that is not a directory. | |
969 | * Zero, one or two objects may be locked by this function. | |
970 | * | |
971 | * @inode1: first inode to lock | |
972 | * @inode2: second inode to lock | |
973 | */ | |
974 | void lock_two_nondirectories(struct inode *inode1, struct inode *inode2) | |
975 | { | |
976 | if (inode1 > inode2) | |
977 | swap(inode1, inode2); | |
978 | ||
979 | if (inode1 && !S_ISDIR(inode1->i_mode)) | |
980 | inode_lock(inode1); | |
981 | if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1) | |
982 | inode_lock_nested(inode2, I_MUTEX_NONDIR2); | |
983 | } | |
984 | EXPORT_SYMBOL(lock_two_nondirectories); | |
985 | ||
986 | /** | |
987 | * unlock_two_nondirectories - release locks from lock_two_nondirectories() | |
988 | * @inode1: first inode to unlock | |
989 | * @inode2: second inode to unlock | |
990 | */ | |
991 | void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2) | |
992 | { | |
993 | if (inode1 && !S_ISDIR(inode1->i_mode)) | |
994 | inode_unlock(inode1); | |
995 | if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1) | |
996 | inode_unlock(inode2); | |
997 | } | |
998 | EXPORT_SYMBOL(unlock_two_nondirectories); | |
999 | ||
1000 | /** | |
1001 | * iget5_locked - obtain an inode from a mounted file system | |
1002 | * @sb: super block of file system | |
1003 | * @hashval: hash value (usually inode number) to get | |
1004 | * @test: callback used for comparisons between inodes | |
1005 | * @set: callback used to initialize a new struct inode | |
1006 | * @data: opaque data pointer to pass to @test and @set | |
1007 | * | |
1008 | * Search for the inode specified by @hashval and @data in the inode cache, | |
1009 | * and if present it is return it with an increased reference count. This is | |
1010 | * a generalized version of iget_locked() for file systems where the inode | |
1011 | * number is not sufficient for unique identification of an inode. | |
1012 | * | |
1013 | * If the inode is not in cache, allocate a new inode and return it locked, | |
1014 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
1015 | * before unlocking it via unlock_new_inode(). | |
1016 | * | |
1017 | * Note both @test and @set are called with the inode_hash_lock held, so can't | |
1018 | * sleep. | |
1019 | */ | |
1020 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
1021 | int (*test)(struct inode *, void *), | |
1022 | int (*set)(struct inode *, void *), void *data) | |
1023 | { | |
1024 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1025 | struct inode *inode; | |
1026 | again: | |
1027 | spin_lock(&inode_hash_lock); | |
1028 | inode = find_inode(sb, head, test, data); | |
1029 | spin_unlock(&inode_hash_lock); | |
1030 | ||
1031 | if (inode) { | |
1032 | wait_on_inode(inode); | |
1033 | if (unlikely(inode_unhashed(inode))) { | |
1034 | iput(inode); | |
1035 | goto again; | |
1036 | } | |
1037 | return inode; | |
1038 | } | |
1039 | ||
1040 | inode = alloc_inode(sb); | |
1041 | if (inode) { | |
1042 | struct inode *old; | |
1043 | ||
1044 | spin_lock(&inode_hash_lock); | |
1045 | /* We released the lock, so.. */ | |
1046 | old = find_inode(sb, head, test, data); | |
1047 | if (!old) { | |
1048 | if (set(inode, data)) | |
1049 | goto set_failed; | |
1050 | ||
1051 | spin_lock(&inode->i_lock); | |
1052 | inode->i_state = I_NEW; | |
1053 | hlist_add_head(&inode->i_hash, head); | |
1054 | spin_unlock(&inode->i_lock); | |
1055 | inode_sb_list_add(inode); | |
1056 | spin_unlock(&inode_hash_lock); | |
1057 | ||
1058 | /* Return the locked inode with I_NEW set, the | |
1059 | * caller is responsible for filling in the contents | |
1060 | */ | |
1061 | return inode; | |
1062 | } | |
1063 | ||
1064 | /* | |
1065 | * Uhhuh, somebody else created the same inode under | |
1066 | * us. Use the old inode instead of the one we just | |
1067 | * allocated. | |
1068 | */ | |
1069 | spin_unlock(&inode_hash_lock); | |
1070 | destroy_inode(inode); | |
1071 | inode = old; | |
1072 | wait_on_inode(inode); | |
1073 | if (unlikely(inode_unhashed(inode))) { | |
1074 | iput(inode); | |
1075 | goto again; | |
1076 | } | |
1077 | } | |
1078 | return inode; | |
1079 | ||
1080 | set_failed: | |
1081 | spin_unlock(&inode_hash_lock); | |
1082 | destroy_inode(inode); | |
1083 | return NULL; | |
1084 | } | |
1085 | EXPORT_SYMBOL(iget5_locked); | |
1086 | ||
1087 | /** | |
1088 | * iget_locked - obtain an inode from a mounted file system | |
1089 | * @sb: super block of file system | |
1090 | * @ino: inode number to get | |
1091 | * | |
1092 | * Search for the inode specified by @ino in the inode cache and if present | |
1093 | * return it with an increased reference count. This is for file systems | |
1094 | * where the inode number is sufficient for unique identification of an inode. | |
1095 | * | |
1096 | * If the inode is not in cache, allocate a new inode and return it locked, | |
1097 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
1098 | * before unlocking it via unlock_new_inode(). | |
1099 | */ | |
1100 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) | |
1101 | { | |
1102 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1103 | struct inode *inode; | |
1104 | again: | |
1105 | spin_lock(&inode_hash_lock); | |
1106 | inode = find_inode_fast(sb, head, ino); | |
1107 | spin_unlock(&inode_hash_lock); | |
1108 | if (inode) { | |
1109 | wait_on_inode(inode); | |
1110 | if (unlikely(inode_unhashed(inode))) { | |
1111 | iput(inode); | |
1112 | goto again; | |
1113 | } | |
1114 | return inode; | |
1115 | } | |
1116 | ||
1117 | inode = alloc_inode(sb); | |
1118 | if (inode) { | |
1119 | struct inode *old; | |
1120 | ||
1121 | spin_lock(&inode_hash_lock); | |
1122 | /* We released the lock, so.. */ | |
1123 | old = find_inode_fast(sb, head, ino); | |
1124 | if (!old) { | |
1125 | inode->i_ino = ino; | |
1126 | spin_lock(&inode->i_lock); | |
1127 | inode->i_state = I_NEW; | |
1128 | hlist_add_head(&inode->i_hash, head); | |
1129 | spin_unlock(&inode->i_lock); | |
1130 | inode_sb_list_add(inode); | |
1131 | spin_unlock(&inode_hash_lock); | |
1132 | ||
1133 | /* Return the locked inode with I_NEW set, the | |
1134 | * caller is responsible for filling in the contents | |
1135 | */ | |
1136 | return inode; | |
1137 | } | |
1138 | ||
1139 | /* | |
1140 | * Uhhuh, somebody else created the same inode under | |
1141 | * us. Use the old inode instead of the one we just | |
1142 | * allocated. | |
1143 | */ | |
1144 | spin_unlock(&inode_hash_lock); | |
1145 | destroy_inode(inode); | |
1146 | inode = old; | |
1147 | wait_on_inode(inode); | |
1148 | if (unlikely(inode_unhashed(inode))) { | |
1149 | iput(inode); | |
1150 | goto again; | |
1151 | } | |
1152 | } | |
1153 | return inode; | |
1154 | } | |
1155 | EXPORT_SYMBOL(iget_locked); | |
1156 | ||
1157 | /* | |
1158 | * search the inode cache for a matching inode number. | |
1159 | * If we find one, then the inode number we are trying to | |
1160 | * allocate is not unique and so we should not use it. | |
1161 | * | |
1162 | * Returns 1 if the inode number is unique, 0 if it is not. | |
1163 | */ | |
1164 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
1165 | { | |
1166 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
1167 | struct inode *inode; | |
1168 | ||
1169 | spin_lock(&inode_hash_lock); | |
1170 | hlist_for_each_entry(inode, b, i_hash) { | |
1171 | if (inode->i_ino == ino && inode->i_sb == sb) { | |
1172 | spin_unlock(&inode_hash_lock); | |
1173 | return 0; | |
1174 | } | |
1175 | } | |
1176 | spin_unlock(&inode_hash_lock); | |
1177 | ||
1178 | return 1; | |
1179 | } | |
1180 | ||
1181 | /** | |
1182 | * iunique - get a unique inode number | |
1183 | * @sb: superblock | |
1184 | * @max_reserved: highest reserved inode number | |
1185 | * | |
1186 | * Obtain an inode number that is unique on the system for a given | |
1187 | * superblock. This is used by file systems that have no natural | |
1188 | * permanent inode numbering system. An inode number is returned that | |
1189 | * is higher than the reserved limit but unique. | |
1190 | * | |
1191 | * BUGS: | |
1192 | * With a large number of inodes live on the file system this function | |
1193 | * currently becomes quite slow. | |
1194 | */ | |
1195 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
1196 | { | |
1197 | /* | |
1198 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
1199 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
1200 | * here to attempt to avoid that. | |
1201 | */ | |
1202 | static DEFINE_SPINLOCK(iunique_lock); | |
1203 | static unsigned int counter; | |
1204 | ino_t res; | |
1205 | ||
1206 | spin_lock(&iunique_lock); | |
1207 | do { | |
1208 | if (counter <= max_reserved) | |
1209 | counter = max_reserved + 1; | |
1210 | res = counter++; | |
1211 | } while (!test_inode_iunique(sb, res)); | |
1212 | spin_unlock(&iunique_lock); | |
1213 | ||
1214 | return res; | |
1215 | } | |
1216 | EXPORT_SYMBOL(iunique); | |
1217 | ||
1218 | struct inode *igrab(struct inode *inode) | |
1219 | { | |
1220 | spin_lock(&inode->i_lock); | |
1221 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { | |
1222 | __iget(inode); | |
1223 | spin_unlock(&inode->i_lock); | |
1224 | } else { | |
1225 | spin_unlock(&inode->i_lock); | |
1226 | /* | |
1227 | * Handle the case where s_op->clear_inode is not been | |
1228 | * called yet, and somebody is calling igrab | |
1229 | * while the inode is getting freed. | |
1230 | */ | |
1231 | inode = NULL; | |
1232 | } | |
1233 | return inode; | |
1234 | } | |
1235 | EXPORT_SYMBOL(igrab); | |
1236 | ||
1237 | /** | |
1238 | * ilookup5_nowait - search for an inode in the inode cache | |
1239 | * @sb: super block of file system to search | |
1240 | * @hashval: hash value (usually inode number) to search for | |
1241 | * @test: callback used for comparisons between inodes | |
1242 | * @data: opaque data pointer to pass to @test | |
1243 | * | |
1244 | * Search for the inode specified by @hashval and @data in the inode cache. | |
1245 | * If the inode is in the cache, the inode is returned with an incremented | |
1246 | * reference count. | |
1247 | * | |
1248 | * Note: I_NEW is not waited upon so you have to be very careful what you do | |
1249 | * with the returned inode. You probably should be using ilookup5() instead. | |
1250 | * | |
1251 | * Note2: @test is called with the inode_hash_lock held, so can't sleep. | |
1252 | */ | |
1253 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
1254 | int (*test)(struct inode *, void *), void *data) | |
1255 | { | |
1256 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1257 | struct inode *inode; | |
1258 | ||
1259 | spin_lock(&inode_hash_lock); | |
1260 | inode = find_inode(sb, head, test, data); | |
1261 | spin_unlock(&inode_hash_lock); | |
1262 | ||
1263 | return inode; | |
1264 | } | |
1265 | EXPORT_SYMBOL(ilookup5_nowait); | |
1266 | ||
1267 | /** | |
1268 | * ilookup5 - search for an inode in the inode cache | |
1269 | * @sb: super block of file system to search | |
1270 | * @hashval: hash value (usually inode number) to search for | |
1271 | * @test: callback used for comparisons between inodes | |
1272 | * @data: opaque data pointer to pass to @test | |
1273 | * | |
1274 | * Search for the inode specified by @hashval and @data in the inode cache, | |
1275 | * and if the inode is in the cache, return the inode with an incremented | |
1276 | * reference count. Waits on I_NEW before returning the inode. | |
1277 | * returned with an incremented reference count. | |
1278 | * | |
1279 | * This is a generalized version of ilookup() for file systems where the | |
1280 | * inode number is not sufficient for unique identification of an inode. | |
1281 | * | |
1282 | * Note: @test is called with the inode_hash_lock held, so can't sleep. | |
1283 | */ | |
1284 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1285 | int (*test)(struct inode *, void *), void *data) | |
1286 | { | |
1287 | struct inode *inode; | |
1288 | again: | |
1289 | inode = ilookup5_nowait(sb, hashval, test, data); | |
1290 | if (inode) { | |
1291 | wait_on_inode(inode); | |
1292 | if (unlikely(inode_unhashed(inode))) { | |
1293 | iput(inode); | |
1294 | goto again; | |
1295 | } | |
1296 | } | |
1297 | return inode; | |
1298 | } | |
1299 | EXPORT_SYMBOL(ilookup5); | |
1300 | ||
1301 | /** | |
1302 | * ilookup - search for an inode in the inode cache | |
1303 | * @sb: super block of file system to search | |
1304 | * @ino: inode number to search for | |
1305 | * | |
1306 | * Search for the inode @ino in the inode cache, and if the inode is in the | |
1307 | * cache, the inode is returned with an incremented reference count. | |
1308 | */ | |
1309 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1310 | { | |
1311 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1312 | struct inode *inode; | |
1313 | again: | |
1314 | spin_lock(&inode_hash_lock); | |
1315 | inode = find_inode_fast(sb, head, ino); | |
1316 | spin_unlock(&inode_hash_lock); | |
1317 | ||
1318 | if (inode) { | |
1319 | wait_on_inode(inode); | |
1320 | if (unlikely(inode_unhashed(inode))) { | |
1321 | iput(inode); | |
1322 | goto again; | |
1323 | } | |
1324 | } | |
1325 | return inode; | |
1326 | } | |
1327 | EXPORT_SYMBOL(ilookup); | |
1328 | ||
1329 | /** | |
1330 | * find_inode_nowait - find an inode in the inode cache | |
1331 | * @sb: super block of file system to search | |
1332 | * @hashval: hash value (usually inode number) to search for | |
1333 | * @match: callback used for comparisons between inodes | |
1334 | * @data: opaque data pointer to pass to @match | |
1335 | * | |
1336 | * Search for the inode specified by @hashval and @data in the inode | |
1337 | * cache, where the helper function @match will return 0 if the inode | |
1338 | * does not match, 1 if the inode does match, and -1 if the search | |
1339 | * should be stopped. The @match function must be responsible for | |
1340 | * taking the i_lock spin_lock and checking i_state for an inode being | |
1341 | * freed or being initialized, and incrementing the reference count | |
1342 | * before returning 1. It also must not sleep, since it is called with | |
1343 | * the inode_hash_lock spinlock held. | |
1344 | * | |
1345 | * This is a even more generalized version of ilookup5() when the | |
1346 | * function must never block --- find_inode() can block in | |
1347 | * __wait_on_freeing_inode() --- or when the caller can not increment | |
1348 | * the reference count because the resulting iput() might cause an | |
1349 | * inode eviction. The tradeoff is that the @match funtion must be | |
1350 | * very carefully implemented. | |
1351 | */ | |
1352 | struct inode *find_inode_nowait(struct super_block *sb, | |
1353 | unsigned long hashval, | |
1354 | int (*match)(struct inode *, unsigned long, | |
1355 | void *), | |
1356 | void *data) | |
1357 | { | |
1358 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1359 | struct inode *inode, *ret_inode = NULL; | |
1360 | int mval; | |
1361 | ||
1362 | spin_lock(&inode_hash_lock); | |
1363 | hlist_for_each_entry(inode, head, i_hash) { | |
1364 | if (inode->i_sb != sb) | |
1365 | continue; | |
1366 | mval = match(inode, hashval, data); | |
1367 | if (mval == 0) | |
1368 | continue; | |
1369 | if (mval == 1) | |
1370 | ret_inode = inode; | |
1371 | goto out; | |
1372 | } | |
1373 | out: | |
1374 | spin_unlock(&inode_hash_lock); | |
1375 | return ret_inode; | |
1376 | } | |
1377 | EXPORT_SYMBOL(find_inode_nowait); | |
1378 | ||
1379 | int insert_inode_locked(struct inode *inode) | |
1380 | { | |
1381 | struct super_block *sb = inode->i_sb; | |
1382 | ino_t ino = inode->i_ino; | |
1383 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1384 | ||
1385 | while (1) { | |
1386 | struct inode *old = NULL; | |
1387 | spin_lock(&inode_hash_lock); | |
1388 | hlist_for_each_entry(old, head, i_hash) { | |
1389 | if (old->i_ino != ino) | |
1390 | continue; | |
1391 | if (old->i_sb != sb) | |
1392 | continue; | |
1393 | spin_lock(&old->i_lock); | |
1394 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1395 | spin_unlock(&old->i_lock); | |
1396 | continue; | |
1397 | } | |
1398 | break; | |
1399 | } | |
1400 | if (likely(!old)) { | |
1401 | spin_lock(&inode->i_lock); | |
1402 | inode->i_state |= I_NEW; | |
1403 | hlist_add_head(&inode->i_hash, head); | |
1404 | spin_unlock(&inode->i_lock); | |
1405 | spin_unlock(&inode_hash_lock); | |
1406 | return 0; | |
1407 | } | |
1408 | __iget(old); | |
1409 | spin_unlock(&old->i_lock); | |
1410 | spin_unlock(&inode_hash_lock); | |
1411 | wait_on_inode(old); | |
1412 | if (unlikely(!inode_unhashed(old))) { | |
1413 | iput(old); | |
1414 | return -EBUSY; | |
1415 | } | |
1416 | iput(old); | |
1417 | } | |
1418 | } | |
1419 | EXPORT_SYMBOL(insert_inode_locked); | |
1420 | ||
1421 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1422 | int (*test)(struct inode *, void *), void *data) | |
1423 | { | |
1424 | struct super_block *sb = inode->i_sb; | |
1425 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1426 | ||
1427 | while (1) { | |
1428 | struct inode *old = NULL; | |
1429 | ||
1430 | spin_lock(&inode_hash_lock); | |
1431 | hlist_for_each_entry(old, head, i_hash) { | |
1432 | if (old->i_sb != sb) | |
1433 | continue; | |
1434 | if (!test(old, data)) | |
1435 | continue; | |
1436 | spin_lock(&old->i_lock); | |
1437 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1438 | spin_unlock(&old->i_lock); | |
1439 | continue; | |
1440 | } | |
1441 | break; | |
1442 | } | |
1443 | if (likely(!old)) { | |
1444 | spin_lock(&inode->i_lock); | |
1445 | inode->i_state |= I_NEW; | |
1446 | hlist_add_head(&inode->i_hash, head); | |
1447 | spin_unlock(&inode->i_lock); | |
1448 | spin_unlock(&inode_hash_lock); | |
1449 | return 0; | |
1450 | } | |
1451 | __iget(old); | |
1452 | spin_unlock(&old->i_lock); | |
1453 | spin_unlock(&inode_hash_lock); | |
1454 | wait_on_inode(old); | |
1455 | if (unlikely(!inode_unhashed(old))) { | |
1456 | iput(old); | |
1457 | return -EBUSY; | |
1458 | } | |
1459 | iput(old); | |
1460 | } | |
1461 | } | |
1462 | EXPORT_SYMBOL(insert_inode_locked4); | |
1463 | ||
1464 | ||
1465 | int generic_delete_inode(struct inode *inode) | |
1466 | { | |
1467 | return 1; | |
1468 | } | |
1469 | EXPORT_SYMBOL(generic_delete_inode); | |
1470 | ||
1471 | /* | |
1472 | * Called when we're dropping the last reference | |
1473 | * to an inode. | |
1474 | * | |
1475 | * Call the FS "drop_inode()" function, defaulting to | |
1476 | * the legacy UNIX filesystem behaviour. If it tells | |
1477 | * us to evict inode, do so. Otherwise, retain inode | |
1478 | * in cache if fs is alive, sync and evict if fs is | |
1479 | * shutting down. | |
1480 | */ | |
1481 | static void iput_final(struct inode *inode) | |
1482 | { | |
1483 | struct super_block *sb = inode->i_sb; | |
1484 | const struct super_operations *op = inode->i_sb->s_op; | |
1485 | int drop; | |
1486 | ||
1487 | WARN_ON(inode->i_state & I_NEW); | |
1488 | ||
1489 | if (op->drop_inode) | |
1490 | drop = op->drop_inode(inode); | |
1491 | else | |
1492 | drop = generic_drop_inode(inode); | |
1493 | ||
1494 | if (!drop && (sb->s_flags & MS_ACTIVE)) { | |
1495 | inode->i_state |= I_REFERENCED; | |
1496 | inode_add_lru(inode); | |
1497 | spin_unlock(&inode->i_lock); | |
1498 | return; | |
1499 | } | |
1500 | ||
1501 | if (!drop) { | |
1502 | inode->i_state |= I_WILL_FREE; | |
1503 | spin_unlock(&inode->i_lock); | |
1504 | write_inode_now(inode, 1); | |
1505 | spin_lock(&inode->i_lock); | |
1506 | WARN_ON(inode->i_state & I_NEW); | |
1507 | inode->i_state &= ~I_WILL_FREE; | |
1508 | } | |
1509 | ||
1510 | inode->i_state |= I_FREEING; | |
1511 | if (!list_empty(&inode->i_lru)) | |
1512 | inode_lru_list_del(inode); | |
1513 | spin_unlock(&inode->i_lock); | |
1514 | ||
1515 | evict(inode); | |
1516 | } | |
1517 | ||
1518 | /** | |
1519 | * iput - put an inode | |
1520 | * @inode: inode to put | |
1521 | * | |
1522 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1523 | * zero, the inode is then freed and may also be destroyed. | |
1524 | * | |
1525 | * Consequently, iput() can sleep. | |
1526 | */ | |
1527 | void iput(struct inode *inode) | |
1528 | { | |
1529 | if (!inode) | |
1530 | return; | |
1531 | BUG_ON(inode->i_state & I_CLEAR); | |
1532 | retry: | |
1533 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) { | |
1534 | if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) { | |
1535 | atomic_inc(&inode->i_count); | |
1536 | inode->i_state &= ~I_DIRTY_TIME; | |
1537 | spin_unlock(&inode->i_lock); | |
1538 | trace_writeback_lazytime_iput(inode); | |
1539 | mark_inode_dirty_sync(inode); | |
1540 | goto retry; | |
1541 | } | |
1542 | iput_final(inode); | |
1543 | } | |
1544 | } | |
1545 | EXPORT_SYMBOL(iput); | |
1546 | ||
1547 | /** | |
1548 | * bmap - find a block number in a file | |
1549 | * @inode: inode of file | |
1550 | * @block: block to find | |
1551 | * | |
1552 | * Returns the block number on the device holding the inode that | |
1553 | * is the disk block number for the block of the file requested. | |
1554 | * That is, asked for block 4 of inode 1 the function will return the | |
1555 | * disk block relative to the disk start that holds that block of the | |
1556 | * file. | |
1557 | */ | |
1558 | sector_t bmap(struct inode *inode, sector_t block) | |
1559 | { | |
1560 | sector_t res = 0; | |
1561 | if (inode->i_mapping->a_ops->bmap) | |
1562 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1563 | return res; | |
1564 | } | |
1565 | EXPORT_SYMBOL(bmap); | |
1566 | ||
1567 | /* | |
1568 | * Update times in overlayed inode from underlying real inode | |
1569 | */ | |
1570 | static void update_ovl_inode_times(struct dentry *dentry, struct inode *inode, | |
1571 | bool rcu) | |
1572 | { | |
1573 | if (!rcu) { | |
1574 | struct inode *realinode = d_real_inode(dentry); | |
1575 | ||
1576 | if (unlikely(inode != realinode) && | |
1577 | (!timespec_equal(&inode->i_mtime, &realinode->i_mtime) || | |
1578 | !timespec_equal(&inode->i_ctime, &realinode->i_ctime))) { | |
1579 | inode->i_mtime = realinode->i_mtime; | |
1580 | inode->i_ctime = realinode->i_ctime; | |
1581 | } | |
1582 | } | |
1583 | } | |
1584 | ||
1585 | /* | |
1586 | * With relative atime, only update atime if the previous atime is | |
1587 | * earlier than either the ctime or mtime or if at least a day has | |
1588 | * passed since the last atime update. | |
1589 | */ | |
1590 | static int relatime_need_update(const struct path *path, struct inode *inode, | |
1591 | struct timespec now, bool rcu) | |
1592 | { | |
1593 | ||
1594 | if (!(path->mnt->mnt_flags & MNT_RELATIME)) | |
1595 | return 1; | |
1596 | ||
1597 | update_ovl_inode_times(path->dentry, inode, rcu); | |
1598 | /* | |
1599 | * Is mtime younger than atime? If yes, update atime: | |
1600 | */ | |
1601 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1602 | return 1; | |
1603 | /* | |
1604 | * Is ctime younger than atime? If yes, update atime: | |
1605 | */ | |
1606 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1607 | return 1; | |
1608 | ||
1609 | /* | |
1610 | * Is the previous atime value older than a day? If yes, | |
1611 | * update atime: | |
1612 | */ | |
1613 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1614 | return 1; | |
1615 | /* | |
1616 | * Good, we can skip the atime update: | |
1617 | */ | |
1618 | return 0; | |
1619 | } | |
1620 | ||
1621 | int generic_update_time(struct inode *inode, struct timespec *time, int flags) | |
1622 | { | |
1623 | int iflags = I_DIRTY_TIME; | |
1624 | ||
1625 | if (flags & S_ATIME) | |
1626 | inode->i_atime = *time; | |
1627 | if (flags & S_VERSION) | |
1628 | inode_inc_iversion(inode); | |
1629 | if (flags & S_CTIME) | |
1630 | inode->i_ctime = *time; | |
1631 | if (flags & S_MTIME) | |
1632 | inode->i_mtime = *time; | |
1633 | ||
1634 | if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION)) | |
1635 | iflags |= I_DIRTY_SYNC; | |
1636 | __mark_inode_dirty(inode, iflags); | |
1637 | return 0; | |
1638 | } | |
1639 | EXPORT_SYMBOL(generic_update_time); | |
1640 | ||
1641 | /* | |
1642 | * This does the actual work of updating an inodes time or version. Must have | |
1643 | * had called mnt_want_write() before calling this. | |
1644 | */ | |
1645 | int update_time(struct inode *inode, struct timespec *time, int flags) | |
1646 | { | |
1647 | int (*update_time)(struct inode *, struct timespec *, int); | |
1648 | ||
1649 | update_time = inode->i_op->update_time ? inode->i_op->update_time : | |
1650 | generic_update_time; | |
1651 | ||
1652 | return update_time(inode, time, flags); | |
1653 | } | |
1654 | EXPORT_SYMBOL_GPL(update_time); | |
1655 | ||
1656 | /** | |
1657 | * touch_atime - update the access time | |
1658 | * @path: the &struct path to update | |
1659 | * @inode: inode to update | |
1660 | * | |
1661 | * Update the accessed time on an inode and mark it for writeback. | |
1662 | * This function automatically handles read only file systems and media, | |
1663 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1664 | */ | |
1665 | bool __atime_needs_update(const struct path *path, struct inode *inode, | |
1666 | bool rcu) | |
1667 | { | |
1668 | struct vfsmount *mnt = path->mnt; | |
1669 | struct timespec now; | |
1670 | ||
1671 | if (inode->i_flags & S_NOATIME) | |
1672 | return false; | |
1673 | ||
1674 | /* Atime updates will likely cause i_uid and i_gid to be written | |
1675 | * back improprely if their true value is unknown to the vfs. | |
1676 | */ | |
1677 | if (HAS_UNMAPPED_ID(inode)) | |
1678 | return false; | |
1679 | ||
1680 | if (IS_NOATIME(inode)) | |
1681 | return false; | |
1682 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1683 | return false; | |
1684 | ||
1685 | if (mnt->mnt_flags & MNT_NOATIME) | |
1686 | return false; | |
1687 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1688 | return false; | |
1689 | ||
1690 | now = current_time(inode); | |
1691 | ||
1692 | if (!relatime_need_update(path, inode, now, rcu)) | |
1693 | return false; | |
1694 | ||
1695 | if (timespec_equal(&inode->i_atime, &now)) | |
1696 | return false; | |
1697 | ||
1698 | return true; | |
1699 | } | |
1700 | ||
1701 | void touch_atime(const struct path *path) | |
1702 | { | |
1703 | struct vfsmount *mnt = path->mnt; | |
1704 | struct inode *inode = d_inode(path->dentry); | |
1705 | struct timespec now; | |
1706 | ||
1707 | if (!__atime_needs_update(path, inode, false)) | |
1708 | return; | |
1709 | ||
1710 | if (!sb_start_write_trylock(inode->i_sb)) | |
1711 | return; | |
1712 | ||
1713 | if (__mnt_want_write(mnt) != 0) | |
1714 | goto skip_update; | |
1715 | /* | |
1716 | * File systems can error out when updating inodes if they need to | |
1717 | * allocate new space to modify an inode (such is the case for | |
1718 | * Btrfs), but since we touch atime while walking down the path we | |
1719 | * really don't care if we failed to update the atime of the file, | |
1720 | * so just ignore the return value. | |
1721 | * We may also fail on filesystems that have the ability to make parts | |
1722 | * of the fs read only, e.g. subvolumes in Btrfs. | |
1723 | */ | |
1724 | now = current_time(inode); | |
1725 | update_time(inode, &now, S_ATIME); | |
1726 | __mnt_drop_write(mnt); | |
1727 | skip_update: | |
1728 | sb_end_write(inode->i_sb); | |
1729 | } | |
1730 | EXPORT_SYMBOL(touch_atime); | |
1731 | ||
1732 | /* | |
1733 | * The logic we want is | |
1734 | * | |
1735 | * if suid or (sgid and xgrp) | |
1736 | * remove privs | |
1737 | */ | |
1738 | int should_remove_suid(struct dentry *dentry) | |
1739 | { | |
1740 | struct inode *inode = d_inode(dentry); | |
1741 | umode_t mode = inode->i_mode; | |
1742 | int kill = 0; | |
1743 | ||
1744 | /* suid always must be killed */ | |
1745 | if (unlikely(mode & S_ISUID)) | |
1746 | kill = ATTR_KILL_SUID; | |
1747 | ||
1748 | /* | |
1749 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1750 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1751 | */ | |
1752 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1753 | kill |= ATTR_KILL_SGID; | |
1754 | ||
1755 | if (unlikely(kill && !capable_wrt_inode_uidgid(inode, CAP_FSETID) && | |
1756 | S_ISREG(mode))) | |
1757 | return kill; | |
1758 | ||
1759 | return 0; | |
1760 | } | |
1761 | EXPORT_SYMBOL(should_remove_suid); | |
1762 | ||
1763 | /* | |
1764 | * Return mask of changes for notify_change() that need to be done as a | |
1765 | * response to write or truncate. Return 0 if nothing has to be changed. | |
1766 | * Negative value on error (change should be denied). | |
1767 | */ | |
1768 | int dentry_needs_remove_privs(struct dentry *dentry) | |
1769 | { | |
1770 | struct inode *inode = d_inode(dentry); | |
1771 | int mask = 0; | |
1772 | int ret; | |
1773 | ||
1774 | if (IS_NOSEC(inode)) | |
1775 | return 0; | |
1776 | ||
1777 | mask = should_remove_suid(dentry); | |
1778 | ret = security_inode_need_killpriv(dentry); | |
1779 | if (ret < 0) | |
1780 | return ret; | |
1781 | if (ret) | |
1782 | mask |= ATTR_KILL_PRIV; | |
1783 | return mask; | |
1784 | } | |
1785 | ||
1786 | static int __remove_privs(struct dentry *dentry, int kill) | |
1787 | { | |
1788 | struct iattr newattrs; | |
1789 | ||
1790 | newattrs.ia_valid = ATTR_FORCE | kill; | |
1791 | /* | |
1792 | * Note we call this on write, so notify_change will not | |
1793 | * encounter any conflicting delegations: | |
1794 | */ | |
1795 | return notify_change(dentry, &newattrs, NULL); | |
1796 | } | |
1797 | ||
1798 | /* | |
1799 | * Remove special file priviledges (suid, capabilities) when file is written | |
1800 | * to or truncated. | |
1801 | */ | |
1802 | int file_remove_privs(struct file *file) | |
1803 | { | |
1804 | struct dentry *dentry = file_dentry(file); | |
1805 | struct inode *inode = file_inode(file); | |
1806 | int kill; | |
1807 | int error = 0; | |
1808 | ||
1809 | /* Fast path for nothing security related */ | |
1810 | if (IS_NOSEC(inode)) | |
1811 | return 0; | |
1812 | ||
1813 | kill = dentry_needs_remove_privs(dentry); | |
1814 | if (kill < 0) | |
1815 | return kill; | |
1816 | if (kill) | |
1817 | error = __remove_privs(dentry, kill); | |
1818 | if (!error) | |
1819 | inode_has_no_xattr(inode); | |
1820 | ||
1821 | return error; | |
1822 | } | |
1823 | EXPORT_SYMBOL(file_remove_privs); | |
1824 | ||
1825 | /** | |
1826 | * file_update_time - update mtime and ctime time | |
1827 | * @file: file accessed | |
1828 | * | |
1829 | * Update the mtime and ctime members of an inode and mark the inode | |
1830 | * for writeback. Note that this function is meant exclusively for | |
1831 | * usage in the file write path of filesystems, and filesystems may | |
1832 | * choose to explicitly ignore update via this function with the | |
1833 | * S_NOCMTIME inode flag, e.g. for network filesystem where these | |
1834 | * timestamps are handled by the server. This can return an error for | |
1835 | * file systems who need to allocate space in order to update an inode. | |
1836 | */ | |
1837 | ||
1838 | int file_update_time(struct file *file) | |
1839 | { | |
1840 | struct inode *inode = file_inode(file); | |
1841 | struct timespec now; | |
1842 | int sync_it = 0; | |
1843 | int ret; | |
1844 | ||
1845 | /* First try to exhaust all avenues to not sync */ | |
1846 | if (IS_NOCMTIME(inode)) | |
1847 | return 0; | |
1848 | ||
1849 | now = current_time(inode); | |
1850 | if (!timespec_equal(&inode->i_mtime, &now)) | |
1851 | sync_it = S_MTIME; | |
1852 | ||
1853 | if (!timespec_equal(&inode->i_ctime, &now)) | |
1854 | sync_it |= S_CTIME; | |
1855 | ||
1856 | if (IS_I_VERSION(inode)) | |
1857 | sync_it |= S_VERSION; | |
1858 | ||
1859 | if (!sync_it) | |
1860 | return 0; | |
1861 | ||
1862 | /* Finally allowed to write? Takes lock. */ | |
1863 | if (__mnt_want_write_file(file)) | |
1864 | return 0; | |
1865 | ||
1866 | ret = update_time(inode, &now, sync_it); | |
1867 | __mnt_drop_write_file(file); | |
1868 | ||
1869 | return ret; | |
1870 | } | |
1871 | EXPORT_SYMBOL(file_update_time); | |
1872 | ||
1873 | int inode_needs_sync(struct inode *inode) | |
1874 | { | |
1875 | if (IS_SYNC(inode)) | |
1876 | return 1; | |
1877 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1878 | return 1; | |
1879 | return 0; | |
1880 | } | |
1881 | EXPORT_SYMBOL(inode_needs_sync); | |
1882 | ||
1883 | /* | |
1884 | * If we try to find an inode in the inode hash while it is being | |
1885 | * deleted, we have to wait until the filesystem completes its | |
1886 | * deletion before reporting that it isn't found. This function waits | |
1887 | * until the deletion _might_ have completed. Callers are responsible | |
1888 | * to recheck inode state. | |
1889 | * | |
1890 | * It doesn't matter if I_NEW is not set initially, a call to | |
1891 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list | |
1892 | * will DTRT. | |
1893 | */ | |
1894 | static void __wait_on_freeing_inode(struct inode *inode) | |
1895 | { | |
1896 | wait_queue_head_t *wq; | |
1897 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); | |
1898 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1899 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
1900 | spin_unlock(&inode->i_lock); | |
1901 | spin_unlock(&inode_hash_lock); | |
1902 | schedule(); | |
1903 | finish_wait(wq, &wait.wait); | |
1904 | spin_lock(&inode_hash_lock); | |
1905 | } | |
1906 | ||
1907 | static __initdata unsigned long ihash_entries; | |
1908 | static int __init set_ihash_entries(char *str) | |
1909 | { | |
1910 | if (!str) | |
1911 | return 0; | |
1912 | ihash_entries = simple_strtoul(str, &str, 0); | |
1913 | return 1; | |
1914 | } | |
1915 | __setup("ihash_entries=", set_ihash_entries); | |
1916 | ||
1917 | /* | |
1918 | * Initialize the waitqueues and inode hash table. | |
1919 | */ | |
1920 | void __init inode_init_early(void) | |
1921 | { | |
1922 | unsigned int loop; | |
1923 | ||
1924 | /* If hashes are distributed across NUMA nodes, defer | |
1925 | * hash allocation until vmalloc space is available. | |
1926 | */ | |
1927 | if (hashdist) | |
1928 | return; | |
1929 | ||
1930 | inode_hashtable = | |
1931 | alloc_large_system_hash("Inode-cache", | |
1932 | sizeof(struct hlist_head), | |
1933 | ihash_entries, | |
1934 | 14, | |
1935 | HASH_EARLY, | |
1936 | &i_hash_shift, | |
1937 | &i_hash_mask, | |
1938 | 0, | |
1939 | 0); | |
1940 | ||
1941 | for (loop = 0; loop < (1U << i_hash_shift); loop++) | |
1942 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1943 | } | |
1944 | ||
1945 | void __init inode_init(void) | |
1946 | { | |
1947 | unsigned int loop; | |
1948 | ||
1949 | /* inode slab cache */ | |
1950 | inode_cachep = kmem_cache_create("inode_cache", | |
1951 | sizeof(struct inode), | |
1952 | 0, | |
1953 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1954 | SLAB_MEM_SPREAD|SLAB_ACCOUNT), | |
1955 | init_once); | |
1956 | ||
1957 | /* Hash may have been set up in inode_init_early */ | |
1958 | if (!hashdist) | |
1959 | return; | |
1960 | ||
1961 | inode_hashtable = | |
1962 | alloc_large_system_hash("Inode-cache", | |
1963 | sizeof(struct hlist_head), | |
1964 | ihash_entries, | |
1965 | 14, | |
1966 | 0, | |
1967 | &i_hash_shift, | |
1968 | &i_hash_mask, | |
1969 | 0, | |
1970 | 0); | |
1971 | ||
1972 | for (loop = 0; loop < (1U << i_hash_shift); loop++) | |
1973 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1974 | } | |
1975 | ||
1976 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1977 | { | |
1978 | inode->i_mode = mode; | |
1979 | if (S_ISCHR(mode)) { | |
1980 | inode->i_fop = &def_chr_fops; | |
1981 | inode->i_rdev = rdev; | |
1982 | } else if (S_ISBLK(mode)) { | |
1983 | inode->i_fop = &def_blk_fops; | |
1984 | inode->i_rdev = rdev; | |
1985 | } else if (S_ISFIFO(mode)) | |
1986 | inode->i_fop = &pipefifo_fops; | |
1987 | else if (S_ISSOCK(mode)) | |
1988 | ; /* leave it no_open_fops */ | |
1989 | else | |
1990 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" | |
1991 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1992 | inode->i_ino); | |
1993 | } | |
1994 | EXPORT_SYMBOL(init_special_inode); | |
1995 | ||
1996 | /** | |
1997 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards | |
1998 | * @inode: New inode | |
1999 | * @dir: Directory inode | |
2000 | * @mode: mode of the new inode | |
2001 | */ | |
2002 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
2003 | umode_t mode) | |
2004 | { | |
2005 | inode->i_uid = current_fsuid(); | |
2006 | if (dir && dir->i_mode & S_ISGID) { | |
2007 | inode->i_gid = dir->i_gid; | |
2008 | if (S_ISDIR(mode)) | |
2009 | mode |= S_ISGID; | |
2010 | } else | |
2011 | inode->i_gid = current_fsgid(); | |
2012 | inode->i_mode = mode; | |
2013 | } | |
2014 | EXPORT_SYMBOL(inode_init_owner); | |
2015 | ||
2016 | /** | |
2017 | * inode_owner_or_capable - check current task permissions to inode | |
2018 | * @inode: inode being checked | |
2019 | * | |
2020 | * Return true if current either has CAP_FOWNER in a namespace with the | |
2021 | * inode owner uid mapped, or owns the file. | |
2022 | */ | |
2023 | bool inode_owner_or_capable(const struct inode *inode) | |
2024 | { | |
2025 | struct user_namespace *ns; | |
2026 | ||
2027 | if (uid_eq(current_fsuid(), inode->i_uid)) | |
2028 | return true; | |
2029 | ||
2030 | ns = current_user_ns(); | |
2031 | if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid)) | |
2032 | return true; | |
2033 | return false; | |
2034 | } | |
2035 | EXPORT_SYMBOL(inode_owner_or_capable); | |
2036 | ||
2037 | /* | |
2038 | * Direct i/o helper functions | |
2039 | */ | |
2040 | static void __inode_dio_wait(struct inode *inode) | |
2041 | { | |
2042 | wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP); | |
2043 | DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP); | |
2044 | ||
2045 | do { | |
2046 | prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE); | |
2047 | if (atomic_read(&inode->i_dio_count)) | |
2048 | schedule(); | |
2049 | } while (atomic_read(&inode->i_dio_count)); | |
2050 | finish_wait(wq, &q.wait); | |
2051 | } | |
2052 | ||
2053 | /** | |
2054 | * inode_dio_wait - wait for outstanding DIO requests to finish | |
2055 | * @inode: inode to wait for | |
2056 | * | |
2057 | * Waits for all pending direct I/O requests to finish so that we can | |
2058 | * proceed with a truncate or equivalent operation. | |
2059 | * | |
2060 | * Must be called under a lock that serializes taking new references | |
2061 | * to i_dio_count, usually by inode->i_mutex. | |
2062 | */ | |
2063 | void inode_dio_wait(struct inode *inode) | |
2064 | { | |
2065 | if (atomic_read(&inode->i_dio_count)) | |
2066 | __inode_dio_wait(inode); | |
2067 | } | |
2068 | EXPORT_SYMBOL(inode_dio_wait); | |
2069 | ||
2070 | /* | |
2071 | * inode_set_flags - atomically set some inode flags | |
2072 | * | |
2073 | * Note: the caller should be holding i_mutex, or else be sure that | |
2074 | * they have exclusive access to the inode structure (i.e., while the | |
2075 | * inode is being instantiated). The reason for the cmpxchg() loop | |
2076 | * --- which wouldn't be necessary if all code paths which modify | |
2077 | * i_flags actually followed this rule, is that there is at least one | |
2078 | * code path which doesn't today so we use cmpxchg() out of an abundance | |
2079 | * of caution. | |
2080 | * | |
2081 | * In the long run, i_mutex is overkill, and we should probably look | |
2082 | * at using the i_lock spinlock to protect i_flags, and then make sure | |
2083 | * it is so documented in include/linux/fs.h and that all code follows | |
2084 | * the locking convention!! | |
2085 | */ | |
2086 | void inode_set_flags(struct inode *inode, unsigned int flags, | |
2087 | unsigned int mask) | |
2088 | { | |
2089 | unsigned int old_flags, new_flags; | |
2090 | ||
2091 | WARN_ON_ONCE(flags & ~mask); | |
2092 | do { | |
2093 | old_flags = ACCESS_ONCE(inode->i_flags); | |
2094 | new_flags = (old_flags & ~mask) | flags; | |
2095 | } while (unlikely(cmpxchg(&inode->i_flags, old_flags, | |
2096 | new_flags) != old_flags)); | |
2097 | } | |
2098 | EXPORT_SYMBOL(inode_set_flags); | |
2099 | ||
2100 | void inode_nohighmem(struct inode *inode) | |
2101 | { | |
2102 | mapping_set_gfp_mask(inode->i_mapping, GFP_USER); | |
2103 | } | |
2104 | EXPORT_SYMBOL(inode_nohighmem); | |
2105 | ||
2106 | /** | |
2107 | * current_time - Return FS time | |
2108 | * @inode: inode. | |
2109 | * | |
2110 | * Return the current time truncated to the time granularity supported by | |
2111 | * the fs. | |
2112 | * | |
2113 | * Note that inode and inode->sb cannot be NULL. | |
2114 | * Otherwise, the function warns and returns time without truncation. | |
2115 | */ | |
2116 | struct timespec current_time(struct inode *inode) | |
2117 | { | |
2118 | struct timespec now = current_kernel_time(); | |
2119 | ||
2120 | if (unlikely(!inode->i_sb)) { | |
2121 | WARN(1, "current_time() called with uninitialized super_block in the inode"); | |
2122 | return now; | |
2123 | } | |
2124 | ||
2125 | return timespec_trunc(now, inode->i_sb->s_time_gran); | |
2126 | } | |
2127 | EXPORT_SYMBOL(current_time); |