blob: 3844c3122f1a586b106c4bcf899f8a2b8867348a
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 | EXPORT_SYMBOL_GPL(evict_inodes); |
641 | |
642 | /** |
643 | * invalidate_inodes - attempt to free all inodes on a superblock |
644 | * @sb: superblock to operate on |
645 | * @kill_dirty: flag to guide handling of dirty inodes |
646 | * |
647 | * Attempts to free all inodes for a given superblock. If there were any |
648 | * busy inodes return a non-zero value, else zero. |
649 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat |
650 | * them as busy. |
651 | */ |
652 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) |
653 | { |
654 | int busy = 0; |
655 | struct inode *inode, *next; |
656 | LIST_HEAD(dispose); |
657 | |
658 | spin_lock(&sb->s_inode_list_lock); |
659 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
660 | spin_lock(&inode->i_lock); |
661 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
662 | spin_unlock(&inode->i_lock); |
663 | continue; |
664 | } |
665 | if (inode->i_state & I_DIRTY_ALL && !kill_dirty) { |
666 | spin_unlock(&inode->i_lock); |
667 | busy = 1; |
668 | continue; |
669 | } |
670 | if (atomic_read(&inode->i_count)) { |
671 | spin_unlock(&inode->i_lock); |
672 | busy = 1; |
673 | continue; |
674 | } |
675 | |
676 | inode->i_state |= I_FREEING; |
677 | inode_lru_list_del(inode); |
678 | spin_unlock(&inode->i_lock); |
679 | list_add(&inode->i_lru, &dispose); |
680 | } |
681 | spin_unlock(&sb->s_inode_list_lock); |
682 | |
683 | dispose_list(&dispose); |
684 | |
685 | return busy; |
686 | } |
687 | |
688 | /* |
689 | * Isolate the inode from the LRU in preparation for freeing it. |
690 | * |
691 | * Any inodes which are pinned purely because of attached pagecache have their |
692 | * pagecache removed. If the inode has metadata buffers attached to |
693 | * mapping->private_list then try to remove them. |
694 | * |
695 | * If the inode has the I_REFERENCED flag set, then it means that it has been |
696 | * used recently - the flag is set in iput_final(). When we encounter such an |
697 | * inode, clear the flag and move it to the back of the LRU so it gets another |
698 | * pass through the LRU before it gets reclaimed. This is necessary because of |
699 | * the fact we are doing lazy LRU updates to minimise lock contention so the |
700 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes |
701 | * with this flag set because they are the inodes that are out of order. |
702 | */ |
703 | static enum lru_status inode_lru_isolate(struct list_head *item, |
704 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
705 | { |
706 | struct list_head *freeable = arg; |
707 | struct inode *inode = container_of(item, struct inode, i_lru); |
708 | |
709 | /* |
710 | * we are inverting the lru lock/inode->i_lock here, so use a trylock. |
711 | * If we fail to get the lock, just skip it. |
712 | */ |
713 | if (!spin_trylock(&inode->i_lock)) |
714 | return LRU_SKIP; |
715 | |
716 | /* |
717 | * Referenced or dirty inodes are still in use. Give them another pass |
718 | * through the LRU as we canot reclaim them now. |
719 | */ |
720 | if (atomic_read(&inode->i_count) || |
721 | (inode->i_state & ~I_REFERENCED)) { |
722 | list_lru_isolate(lru, &inode->i_lru); |
723 | spin_unlock(&inode->i_lock); |
724 | this_cpu_dec(nr_unused); |
725 | return LRU_REMOVED; |
726 | } |
727 | |
728 | /* recently referenced inodes get one more pass */ |
729 | if (inode->i_state & I_REFERENCED) { |
730 | inode->i_state &= ~I_REFERENCED; |
731 | spin_unlock(&inode->i_lock); |
732 | return LRU_ROTATE; |
733 | } |
734 | |
735 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { |
736 | __iget(inode); |
737 | spin_unlock(&inode->i_lock); |
738 | spin_unlock(lru_lock); |
739 | if (remove_inode_buffers(inode)) { |
740 | unsigned long reap; |
741 | reap = invalidate_mapping_pages(&inode->i_data, 0, -1); |
742 | if (current_is_kswapd()) |
743 | __count_vm_events(KSWAPD_INODESTEAL, reap); |
744 | else |
745 | __count_vm_events(PGINODESTEAL, reap); |
746 | if (current->reclaim_state) |
747 | current->reclaim_state->reclaimed_slab += reap; |
748 | } |
749 | iput(inode); |
750 | spin_lock(lru_lock); |
751 | return LRU_RETRY; |
752 | } |
753 | |
754 | WARN_ON(inode->i_state & I_NEW); |
755 | inode->i_state |= I_FREEING; |
756 | list_lru_isolate_move(lru, &inode->i_lru, freeable); |
757 | spin_unlock(&inode->i_lock); |
758 | |
759 | this_cpu_dec(nr_unused); |
760 | return LRU_REMOVED; |
761 | } |
762 | |
763 | /* |
764 | * Walk the superblock inode LRU for freeable inodes and attempt to free them. |
765 | * This is called from the superblock shrinker function with a number of inodes |
766 | * to trim from the LRU. Inodes to be freed are moved to a temporary list and |
767 | * then are freed outside inode_lock by dispose_list(). |
768 | */ |
769 | long prune_icache_sb(struct super_block *sb, struct shrink_control *sc) |
770 | { |
771 | LIST_HEAD(freeable); |
772 | long freed; |
773 | |
774 | freed = list_lru_shrink_walk(&sb->s_inode_lru, sc, |
775 | inode_lru_isolate, &freeable); |
776 | dispose_list(&freeable); |
777 | return freed; |
778 | } |
779 | |
780 | static void __wait_on_freeing_inode(struct inode *inode); |
781 | /* |
782 | * Called with the inode lock held. |
783 | */ |
784 | static struct inode *find_inode(struct super_block *sb, |
785 | struct hlist_head *head, |
786 | int (*test)(struct inode *, void *), |
787 | void *data) |
788 | { |
789 | struct inode *inode = NULL; |
790 | |
791 | repeat: |
792 | hlist_for_each_entry(inode, head, i_hash) { |
793 | if (inode->i_sb != sb) |
794 | continue; |
795 | if (!test(inode, data)) |
796 | continue; |
797 | spin_lock(&inode->i_lock); |
798 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
799 | __wait_on_freeing_inode(inode); |
800 | goto repeat; |
801 | } |
802 | __iget(inode); |
803 | spin_unlock(&inode->i_lock); |
804 | return inode; |
805 | } |
806 | return NULL; |
807 | } |
808 | |
809 | /* |
810 | * find_inode_fast is the fast path version of find_inode, see the comment at |
811 | * iget_locked for details. |
812 | */ |
813 | static struct inode *find_inode_fast(struct super_block *sb, |
814 | struct hlist_head *head, unsigned long ino) |
815 | { |
816 | struct inode *inode = NULL; |
817 | |
818 | repeat: |
819 | hlist_for_each_entry(inode, head, i_hash) { |
820 | if (inode->i_ino != ino) |
821 | continue; |
822 | if (inode->i_sb != sb) |
823 | continue; |
824 | spin_lock(&inode->i_lock); |
825 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
826 | __wait_on_freeing_inode(inode); |
827 | goto repeat; |
828 | } |
829 | __iget(inode); |
830 | spin_unlock(&inode->i_lock); |
831 | return inode; |
832 | } |
833 | return NULL; |
834 | } |
835 | |
836 | /* |
837 | * Each cpu owns a range of LAST_INO_BATCH numbers. |
838 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, |
839 | * to renew the exhausted range. |
840 | * |
841 | * This does not significantly increase overflow rate because every CPU can |
842 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is |
843 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the |
844 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase |
845 | * overflow rate by 2x, which does not seem too significant. |
846 | * |
847 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW |
848 | * error if st_ino won't fit in target struct field. Use 32bit counter |
849 | * here to attempt to avoid that. |
850 | */ |
851 | #define LAST_INO_BATCH 1024 |
852 | static DEFINE_PER_CPU(unsigned int, last_ino); |
853 | |
854 | unsigned int get_next_ino(void) |
855 | { |
856 | unsigned int *p = &get_cpu_var(last_ino); |
857 | unsigned int res = *p; |
858 | |
859 | #ifdef CONFIG_SMP |
860 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { |
861 | static atomic_t shared_last_ino; |
862 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); |
863 | |
864 | res = next - LAST_INO_BATCH; |
865 | } |
866 | #endif |
867 | |
868 | res++; |
869 | /* get_next_ino should not provide a 0 inode number */ |
870 | if (unlikely(!res)) |
871 | res++; |
872 | *p = res; |
873 | put_cpu_var(last_ino); |
874 | return res; |
875 | } |
876 | EXPORT_SYMBOL(get_next_ino); |
877 | |
878 | /** |
879 | * new_inode_pseudo - obtain an inode |
880 | * @sb: superblock |
881 | * |
882 | * Allocates a new inode for given superblock. |
883 | * Inode wont be chained in superblock s_inodes list |
884 | * This means : |
885 | * - fs can't be unmount |
886 | * - quotas, fsnotify, writeback can't work |
887 | */ |
888 | struct inode *new_inode_pseudo(struct super_block *sb) |
889 | { |
890 | struct inode *inode = alloc_inode(sb); |
891 | |
892 | if (inode) { |
893 | spin_lock(&inode->i_lock); |
894 | inode->i_state = 0; |
895 | spin_unlock(&inode->i_lock); |
896 | INIT_LIST_HEAD(&inode->i_sb_list); |
897 | } |
898 | return inode; |
899 | } |
900 | |
901 | /** |
902 | * new_inode - obtain an inode |
903 | * @sb: superblock |
904 | * |
905 | * Allocates a new inode for given superblock. The default gfp_mask |
906 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
907 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
908 | * for the page cache are not reclaimable or migratable, |
909 | * mapping_set_gfp_mask() must be called with suitable flags on the |
910 | * newly created inode's mapping |
911 | * |
912 | */ |
913 | struct inode *new_inode(struct super_block *sb) |
914 | { |
915 | struct inode *inode; |
916 | |
917 | spin_lock_prefetch(&sb->s_inode_list_lock); |
918 | |
919 | inode = new_inode_pseudo(sb); |
920 | if (inode) |
921 | inode_sb_list_add(inode); |
922 | return inode; |
923 | } |
924 | EXPORT_SYMBOL(new_inode); |
925 | |
926 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
927 | void lockdep_annotate_inode_mutex_key(struct inode *inode) |
928 | { |
929 | if (S_ISDIR(inode->i_mode)) { |
930 | struct file_system_type *type = inode->i_sb->s_type; |
931 | |
932 | /* Set new key only if filesystem hasn't already changed it */ |
933 | if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) { |
934 | /* |
935 | * ensure nobody is actually holding i_mutex |
936 | */ |
937 | // mutex_destroy(&inode->i_mutex); |
938 | init_rwsem(&inode->i_rwsem); |
939 | lockdep_set_class(&inode->i_rwsem, |
940 | &type->i_mutex_dir_key); |
941 | } |
942 | } |
943 | } |
944 | EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key); |
945 | #endif |
946 | |
947 | /** |
948 | * unlock_new_inode - clear the I_NEW state and wake up any waiters |
949 | * @inode: new inode to unlock |
950 | * |
951 | * Called when the inode is fully initialised to clear the new state of the |
952 | * inode and wake up anyone waiting for the inode to finish initialisation. |
953 | */ |
954 | void unlock_new_inode(struct inode *inode) |
955 | { |
956 | lockdep_annotate_inode_mutex_key(inode); |
957 | spin_lock(&inode->i_lock); |
958 | WARN_ON(!(inode->i_state & I_NEW)); |
959 | inode->i_state &= ~I_NEW; |
960 | smp_mb(); |
961 | wake_up_bit(&inode->i_state, __I_NEW); |
962 | spin_unlock(&inode->i_lock); |
963 | } |
964 | EXPORT_SYMBOL(unlock_new_inode); |
965 | |
966 | /** |
967 | * lock_two_nondirectories - take two i_mutexes on non-directory objects |
968 | * |
969 | * Lock any non-NULL argument that is not a directory. |
970 | * Zero, one or two objects may be locked by this function. |
971 | * |
972 | * @inode1: first inode to lock |
973 | * @inode2: second inode to lock |
974 | */ |
975 | void lock_two_nondirectories(struct inode *inode1, struct inode *inode2) |
976 | { |
977 | if (inode1 > inode2) |
978 | swap(inode1, inode2); |
979 | |
980 | if (inode1 && !S_ISDIR(inode1->i_mode)) |
981 | inode_lock(inode1); |
982 | if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1) |
983 | inode_lock_nested(inode2, I_MUTEX_NONDIR2); |
984 | } |
985 | EXPORT_SYMBOL(lock_two_nondirectories); |
986 | |
987 | /** |
988 | * unlock_two_nondirectories - release locks from lock_two_nondirectories() |
989 | * @inode1: first inode to unlock |
990 | * @inode2: second inode to unlock |
991 | */ |
992 | void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2) |
993 | { |
994 | if (inode1 && !S_ISDIR(inode1->i_mode)) |
995 | inode_unlock(inode1); |
996 | if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1) |
997 | inode_unlock(inode2); |
998 | } |
999 | EXPORT_SYMBOL(unlock_two_nondirectories); |
1000 | |
1001 | /** |
1002 | * iget5_locked - obtain an inode from a mounted file system |
1003 | * @sb: super block of file system |
1004 | * @hashval: hash value (usually inode number) to get |
1005 | * @test: callback used for comparisons between inodes |
1006 | * @set: callback used to initialize a new struct inode |
1007 | * @data: opaque data pointer to pass to @test and @set |
1008 | * |
1009 | * Search for the inode specified by @hashval and @data in the inode cache, |
1010 | * and if present it is return it with an increased reference count. This is |
1011 | * a generalized version of iget_locked() for file systems where the inode |
1012 | * number is not sufficient for unique identification of an inode. |
1013 | * |
1014 | * If the inode is not in cache, allocate a new inode and return it locked, |
1015 | * hashed, and with the I_NEW flag set. The file system gets to fill it in |
1016 | * before unlocking it via unlock_new_inode(). |
1017 | * |
1018 | * Note both @test and @set are called with the inode_hash_lock held, so can't |
1019 | * sleep. |
1020 | */ |
1021 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, |
1022 | int (*test)(struct inode *, void *), |
1023 | int (*set)(struct inode *, void *), void *data) |
1024 | { |
1025 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1026 | struct inode *inode; |
1027 | again: |
1028 | spin_lock(&inode_hash_lock); |
1029 | inode = find_inode(sb, head, test, data); |
1030 | spin_unlock(&inode_hash_lock); |
1031 | |
1032 | if (inode) { |
1033 | wait_on_inode(inode); |
1034 | if (unlikely(inode_unhashed(inode))) { |
1035 | iput(inode); |
1036 | goto again; |
1037 | } |
1038 | return inode; |
1039 | } |
1040 | |
1041 | inode = alloc_inode(sb); |
1042 | if (inode) { |
1043 | struct inode *old; |
1044 | |
1045 | spin_lock(&inode_hash_lock); |
1046 | /* We released the lock, so.. */ |
1047 | old = find_inode(sb, head, test, data); |
1048 | if (!old) { |
1049 | if (set(inode, data)) |
1050 | goto set_failed; |
1051 | |
1052 | spin_lock(&inode->i_lock); |
1053 | inode->i_state = I_NEW; |
1054 | hlist_add_head(&inode->i_hash, head); |
1055 | spin_unlock(&inode->i_lock); |
1056 | inode_sb_list_add(inode); |
1057 | spin_unlock(&inode_hash_lock); |
1058 | |
1059 | /* Return the locked inode with I_NEW set, the |
1060 | * caller is responsible for filling in the contents |
1061 | */ |
1062 | return inode; |
1063 | } |
1064 | |
1065 | /* |
1066 | * Uhhuh, somebody else created the same inode under |
1067 | * us. Use the old inode instead of the one we just |
1068 | * allocated. |
1069 | */ |
1070 | spin_unlock(&inode_hash_lock); |
1071 | destroy_inode(inode); |
1072 | inode = old; |
1073 | wait_on_inode(inode); |
1074 | if (unlikely(inode_unhashed(inode))) { |
1075 | iput(inode); |
1076 | goto again; |
1077 | } |
1078 | } |
1079 | return inode; |
1080 | |
1081 | set_failed: |
1082 | spin_unlock(&inode_hash_lock); |
1083 | destroy_inode(inode); |
1084 | return NULL; |
1085 | } |
1086 | EXPORT_SYMBOL(iget5_locked); |
1087 | |
1088 | /** |
1089 | * iget_locked - obtain an inode from a mounted file system |
1090 | * @sb: super block of file system |
1091 | * @ino: inode number to get |
1092 | * |
1093 | * Search for the inode specified by @ino in the inode cache and if present |
1094 | * return it with an increased reference count. This is for file systems |
1095 | * where the inode number is sufficient for unique identification of an inode. |
1096 | * |
1097 | * If the inode is not in cache, allocate a new inode and return it locked, |
1098 | * hashed, and with the I_NEW flag set. The file system gets to fill it in |
1099 | * before unlocking it via unlock_new_inode(). |
1100 | */ |
1101 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) |
1102 | { |
1103 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
1104 | struct inode *inode; |
1105 | again: |
1106 | spin_lock(&inode_hash_lock); |
1107 | inode = find_inode_fast(sb, head, ino); |
1108 | spin_unlock(&inode_hash_lock); |
1109 | if (inode) { |
1110 | wait_on_inode(inode); |
1111 | if (unlikely(inode_unhashed(inode))) { |
1112 | iput(inode); |
1113 | goto again; |
1114 | } |
1115 | return inode; |
1116 | } |
1117 | |
1118 | inode = alloc_inode(sb); |
1119 | if (inode) { |
1120 | struct inode *old; |
1121 | |
1122 | spin_lock(&inode_hash_lock); |
1123 | /* We released the lock, so.. */ |
1124 | old = find_inode_fast(sb, head, ino); |
1125 | if (!old) { |
1126 | inode->i_ino = ino; |
1127 | spin_lock(&inode->i_lock); |
1128 | inode->i_state = I_NEW; |
1129 | hlist_add_head(&inode->i_hash, head); |
1130 | spin_unlock(&inode->i_lock); |
1131 | inode_sb_list_add(inode); |
1132 | spin_unlock(&inode_hash_lock); |
1133 | |
1134 | /* Return the locked inode with I_NEW set, the |
1135 | * caller is responsible for filling in the contents |
1136 | */ |
1137 | return inode; |
1138 | } |
1139 | |
1140 | /* |
1141 | * Uhhuh, somebody else created the same inode under |
1142 | * us. Use the old inode instead of the one we just |
1143 | * allocated. |
1144 | */ |
1145 | spin_unlock(&inode_hash_lock); |
1146 | destroy_inode(inode); |
1147 | inode = old; |
1148 | wait_on_inode(inode); |
1149 | if (unlikely(inode_unhashed(inode))) { |
1150 | iput(inode); |
1151 | goto again; |
1152 | } |
1153 | } |
1154 | return inode; |
1155 | } |
1156 | EXPORT_SYMBOL(iget_locked); |
1157 | |
1158 | /* |
1159 | * search the inode cache for a matching inode number. |
1160 | * If we find one, then the inode number we are trying to |
1161 | * allocate is not unique and so we should not use it. |
1162 | * |
1163 | * Returns 1 if the inode number is unique, 0 if it is not. |
1164 | */ |
1165 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) |
1166 | { |
1167 | struct hlist_head *b = inode_hashtable + hash(sb, ino); |
1168 | struct inode *inode; |
1169 | |
1170 | spin_lock(&inode_hash_lock); |
1171 | hlist_for_each_entry(inode, b, i_hash) { |
1172 | if (inode->i_ino == ino && inode->i_sb == sb) { |
1173 | spin_unlock(&inode_hash_lock); |
1174 | return 0; |
1175 | } |
1176 | } |
1177 | spin_unlock(&inode_hash_lock); |
1178 | |
1179 | return 1; |
1180 | } |
1181 | |
1182 | /** |
1183 | * iunique - get a unique inode number |
1184 | * @sb: superblock |
1185 | * @max_reserved: highest reserved inode number |
1186 | * |
1187 | * Obtain an inode number that is unique on the system for a given |
1188 | * superblock. This is used by file systems that have no natural |
1189 | * permanent inode numbering system. An inode number is returned that |
1190 | * is higher than the reserved limit but unique. |
1191 | * |
1192 | * BUGS: |
1193 | * With a large number of inodes live on the file system this function |
1194 | * currently becomes quite slow. |
1195 | */ |
1196 | ino_t iunique(struct super_block *sb, ino_t max_reserved) |
1197 | { |
1198 | /* |
1199 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW |
1200 | * error if st_ino won't fit in target struct field. Use 32bit counter |
1201 | * here to attempt to avoid that. |
1202 | */ |
1203 | static DEFINE_SPINLOCK(iunique_lock); |
1204 | static unsigned int counter; |
1205 | ino_t res; |
1206 | |
1207 | spin_lock(&iunique_lock); |
1208 | do { |
1209 | if (counter <= max_reserved) |
1210 | counter = max_reserved + 1; |
1211 | res = counter++; |
1212 | } while (!test_inode_iunique(sb, res)); |
1213 | spin_unlock(&iunique_lock); |
1214 | |
1215 | return res; |
1216 | } |
1217 | EXPORT_SYMBOL(iunique); |
1218 | |
1219 | struct inode *igrab(struct inode *inode) |
1220 | { |
1221 | spin_lock(&inode->i_lock); |
1222 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { |
1223 | __iget(inode); |
1224 | spin_unlock(&inode->i_lock); |
1225 | } else { |
1226 | spin_unlock(&inode->i_lock); |
1227 | /* |
1228 | * Handle the case where s_op->clear_inode is not been |
1229 | * called yet, and somebody is calling igrab |
1230 | * while the inode is getting freed. |
1231 | */ |
1232 | inode = NULL; |
1233 | } |
1234 | return inode; |
1235 | } |
1236 | EXPORT_SYMBOL(igrab); |
1237 | |
1238 | /** |
1239 | * ilookup5_nowait - search for an inode in the inode cache |
1240 | * @sb: super block of file system to search |
1241 | * @hashval: hash value (usually inode number) to search for |
1242 | * @test: callback used for comparisons between inodes |
1243 | * @data: opaque data pointer to pass to @test |
1244 | * |
1245 | * Search for the inode specified by @hashval and @data in the inode cache. |
1246 | * If the inode is in the cache, the inode is returned with an incremented |
1247 | * reference count. |
1248 | * |
1249 | * Note: I_NEW is not waited upon so you have to be very careful what you do |
1250 | * with the returned inode. You probably should be using ilookup5() instead. |
1251 | * |
1252 | * Note2: @test is called with the inode_hash_lock held, so can't sleep. |
1253 | */ |
1254 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, |
1255 | int (*test)(struct inode *, void *), void *data) |
1256 | { |
1257 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1258 | struct inode *inode; |
1259 | |
1260 | spin_lock(&inode_hash_lock); |
1261 | inode = find_inode(sb, head, test, data); |
1262 | spin_unlock(&inode_hash_lock); |
1263 | |
1264 | return inode; |
1265 | } |
1266 | EXPORT_SYMBOL(ilookup5_nowait); |
1267 | |
1268 | /** |
1269 | * ilookup5 - search for an inode in the inode cache |
1270 | * @sb: super block of file system to search |
1271 | * @hashval: hash value (usually inode number) to search for |
1272 | * @test: callback used for comparisons between inodes |
1273 | * @data: opaque data pointer to pass to @test |
1274 | * |
1275 | * Search for the inode specified by @hashval and @data in the inode cache, |
1276 | * and if the inode is in the cache, return the inode with an incremented |
1277 | * reference count. Waits on I_NEW before returning the inode. |
1278 | * returned with an incremented reference count. |
1279 | * |
1280 | * This is a generalized version of ilookup() for file systems where the |
1281 | * inode number is not sufficient for unique identification of an inode. |
1282 | * |
1283 | * Note: @test is called with the inode_hash_lock held, so can't sleep. |
1284 | */ |
1285 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, |
1286 | int (*test)(struct inode *, void *), void *data) |
1287 | { |
1288 | struct inode *inode; |
1289 | again: |
1290 | inode = ilookup5_nowait(sb, hashval, test, data); |
1291 | if (inode) { |
1292 | wait_on_inode(inode); |
1293 | if (unlikely(inode_unhashed(inode))) { |
1294 | iput(inode); |
1295 | goto again; |
1296 | } |
1297 | } |
1298 | return inode; |
1299 | } |
1300 | EXPORT_SYMBOL(ilookup5); |
1301 | |
1302 | /** |
1303 | * ilookup - search for an inode in the inode cache |
1304 | * @sb: super block of file system to search |
1305 | * @ino: inode number to search for |
1306 | * |
1307 | * Search for the inode @ino in the inode cache, and if the inode is in the |
1308 | * cache, the inode is returned with an incremented reference count. |
1309 | */ |
1310 | struct inode *ilookup(struct super_block *sb, unsigned long ino) |
1311 | { |
1312 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
1313 | struct inode *inode; |
1314 | again: |
1315 | spin_lock(&inode_hash_lock); |
1316 | inode = find_inode_fast(sb, head, ino); |
1317 | spin_unlock(&inode_hash_lock); |
1318 | |
1319 | if (inode) { |
1320 | wait_on_inode(inode); |
1321 | if (unlikely(inode_unhashed(inode))) { |
1322 | iput(inode); |
1323 | goto again; |
1324 | } |
1325 | } |
1326 | return inode; |
1327 | } |
1328 | EXPORT_SYMBOL(ilookup); |
1329 | |
1330 | /** |
1331 | * find_inode_nowait - find an inode in the inode cache |
1332 | * @sb: super block of file system to search |
1333 | * @hashval: hash value (usually inode number) to search for |
1334 | * @match: callback used for comparisons between inodes |
1335 | * @data: opaque data pointer to pass to @match |
1336 | * |
1337 | * Search for the inode specified by @hashval and @data in the inode |
1338 | * cache, where the helper function @match will return 0 if the inode |
1339 | * does not match, 1 if the inode does match, and -1 if the search |
1340 | * should be stopped. The @match function must be responsible for |
1341 | * taking the i_lock spin_lock and checking i_state for an inode being |
1342 | * freed or being initialized, and incrementing the reference count |
1343 | * before returning 1. It also must not sleep, since it is called with |
1344 | * the inode_hash_lock spinlock held. |
1345 | * |
1346 | * This is a even more generalized version of ilookup5() when the |
1347 | * function must never block --- find_inode() can block in |
1348 | * __wait_on_freeing_inode() --- or when the caller can not increment |
1349 | * the reference count because the resulting iput() might cause an |
1350 | * inode eviction. The tradeoff is that the @match funtion must be |
1351 | * very carefully implemented. |
1352 | */ |
1353 | struct inode *find_inode_nowait(struct super_block *sb, |
1354 | unsigned long hashval, |
1355 | int (*match)(struct inode *, unsigned long, |
1356 | void *), |
1357 | void *data) |
1358 | { |
1359 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1360 | struct inode *inode, *ret_inode = NULL; |
1361 | int mval; |
1362 | |
1363 | spin_lock(&inode_hash_lock); |
1364 | hlist_for_each_entry(inode, head, i_hash) { |
1365 | if (inode->i_sb != sb) |
1366 | continue; |
1367 | mval = match(inode, hashval, data); |
1368 | if (mval == 0) |
1369 | continue; |
1370 | if (mval == 1) |
1371 | ret_inode = inode; |
1372 | goto out; |
1373 | } |
1374 | out: |
1375 | spin_unlock(&inode_hash_lock); |
1376 | return ret_inode; |
1377 | } |
1378 | EXPORT_SYMBOL(find_inode_nowait); |
1379 | |
1380 | int insert_inode_locked(struct inode *inode) |
1381 | { |
1382 | struct super_block *sb = inode->i_sb; |
1383 | ino_t ino = inode->i_ino; |
1384 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
1385 | |
1386 | while (1) { |
1387 | struct inode *old = NULL; |
1388 | spin_lock(&inode_hash_lock); |
1389 | hlist_for_each_entry(old, head, i_hash) { |
1390 | if (old->i_ino != ino) |
1391 | continue; |
1392 | if (old->i_sb != sb) |
1393 | continue; |
1394 | spin_lock(&old->i_lock); |
1395 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { |
1396 | spin_unlock(&old->i_lock); |
1397 | continue; |
1398 | } |
1399 | break; |
1400 | } |
1401 | if (likely(!old)) { |
1402 | spin_lock(&inode->i_lock); |
1403 | inode->i_state |= I_NEW; |
1404 | hlist_add_head(&inode->i_hash, head); |
1405 | spin_unlock(&inode->i_lock); |
1406 | spin_unlock(&inode_hash_lock); |
1407 | return 0; |
1408 | } |
1409 | __iget(old); |
1410 | spin_unlock(&old->i_lock); |
1411 | spin_unlock(&inode_hash_lock); |
1412 | wait_on_inode(old); |
1413 | if (unlikely(!inode_unhashed(old))) { |
1414 | iput(old); |
1415 | return -EBUSY; |
1416 | } |
1417 | iput(old); |
1418 | } |
1419 | } |
1420 | EXPORT_SYMBOL(insert_inode_locked); |
1421 | |
1422 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, |
1423 | int (*test)(struct inode *, void *), void *data) |
1424 | { |
1425 | struct super_block *sb = inode->i_sb; |
1426 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1427 | |
1428 | while (1) { |
1429 | struct inode *old = NULL; |
1430 | |
1431 | spin_lock(&inode_hash_lock); |
1432 | hlist_for_each_entry(old, head, i_hash) { |
1433 | if (old->i_sb != sb) |
1434 | continue; |
1435 | if (!test(old, data)) |
1436 | continue; |
1437 | spin_lock(&old->i_lock); |
1438 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { |
1439 | spin_unlock(&old->i_lock); |
1440 | continue; |
1441 | } |
1442 | break; |
1443 | } |
1444 | if (likely(!old)) { |
1445 | spin_lock(&inode->i_lock); |
1446 | inode->i_state |= I_NEW; |
1447 | hlist_add_head(&inode->i_hash, head); |
1448 | spin_unlock(&inode->i_lock); |
1449 | spin_unlock(&inode_hash_lock); |
1450 | return 0; |
1451 | } |
1452 | __iget(old); |
1453 | spin_unlock(&old->i_lock); |
1454 | spin_unlock(&inode_hash_lock); |
1455 | wait_on_inode(old); |
1456 | if (unlikely(!inode_unhashed(old))) { |
1457 | iput(old); |
1458 | return -EBUSY; |
1459 | } |
1460 | iput(old); |
1461 | } |
1462 | } |
1463 | EXPORT_SYMBOL(insert_inode_locked4); |
1464 | |
1465 | |
1466 | int generic_delete_inode(struct inode *inode) |
1467 | { |
1468 | return 1; |
1469 | } |
1470 | EXPORT_SYMBOL(generic_delete_inode); |
1471 | |
1472 | /* |
1473 | * Called when we're dropping the last reference |
1474 | * to an inode. |
1475 | * |
1476 | * Call the FS "drop_inode()" function, defaulting to |
1477 | * the legacy UNIX filesystem behaviour. If it tells |
1478 | * us to evict inode, do so. Otherwise, retain inode |
1479 | * in cache if fs is alive, sync and evict if fs is |
1480 | * shutting down. |
1481 | */ |
1482 | static void iput_final(struct inode *inode) |
1483 | { |
1484 | struct super_block *sb = inode->i_sb; |
1485 | const struct super_operations *op = inode->i_sb->s_op; |
1486 | int drop; |
1487 | |
1488 | WARN_ON(inode->i_state & I_NEW); |
1489 | |
1490 | if (op->drop_inode) |
1491 | drop = op->drop_inode(inode); |
1492 | else |
1493 | drop = generic_drop_inode(inode); |
1494 | |
1495 | if (!drop && (sb->s_flags & MS_ACTIVE)) { |
1496 | inode->i_state |= I_REFERENCED; |
1497 | inode_add_lru(inode); |
1498 | spin_unlock(&inode->i_lock); |
1499 | return; |
1500 | } |
1501 | |
1502 | if (!drop) { |
1503 | inode->i_state |= I_WILL_FREE; |
1504 | spin_unlock(&inode->i_lock); |
1505 | write_inode_now(inode, 1); |
1506 | spin_lock(&inode->i_lock); |
1507 | WARN_ON(inode->i_state & I_NEW); |
1508 | inode->i_state &= ~I_WILL_FREE; |
1509 | } |
1510 | |
1511 | inode->i_state |= I_FREEING; |
1512 | if (!list_empty(&inode->i_lru)) |
1513 | inode_lru_list_del(inode); |
1514 | spin_unlock(&inode->i_lock); |
1515 | |
1516 | evict(inode); |
1517 | } |
1518 | |
1519 | /** |
1520 | * iput - put an inode |
1521 | * @inode: inode to put |
1522 | * |
1523 | * Puts an inode, dropping its usage count. If the inode use count hits |
1524 | * zero, the inode is then freed and may also be destroyed. |
1525 | * |
1526 | * Consequently, iput() can sleep. |
1527 | */ |
1528 | void iput(struct inode *inode) |
1529 | { |
1530 | if (!inode) |
1531 | return; |
1532 | BUG_ON(inode->i_state & I_CLEAR); |
1533 | retry: |
1534 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) { |
1535 | if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) { |
1536 | atomic_inc(&inode->i_count); |
1537 | inode->i_state &= ~I_DIRTY_TIME; |
1538 | spin_unlock(&inode->i_lock); |
1539 | trace_writeback_lazytime_iput(inode); |
1540 | mark_inode_dirty_sync(inode); |
1541 | goto retry; |
1542 | } |
1543 | iput_final(inode); |
1544 | } |
1545 | } |
1546 | EXPORT_SYMBOL(iput); |
1547 | |
1548 | /** |
1549 | * bmap - find a block number in a file |
1550 | * @inode: inode of file |
1551 | * @block: block to find |
1552 | * |
1553 | * Returns the block number on the device holding the inode that |
1554 | * is the disk block number for the block of the file requested. |
1555 | * That is, asked for block 4 of inode 1 the function will return the |
1556 | * disk block relative to the disk start that holds that block of the |
1557 | * file. |
1558 | */ |
1559 | sector_t bmap(struct inode *inode, sector_t block) |
1560 | { |
1561 | sector_t res = 0; |
1562 | if (inode->i_mapping->a_ops->bmap) |
1563 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); |
1564 | return res; |
1565 | } |
1566 | EXPORT_SYMBOL(bmap); |
1567 | |
1568 | /* |
1569 | * Update times in overlayed inode from underlying real inode |
1570 | */ |
1571 | static void update_ovl_inode_times(struct dentry *dentry, struct inode *inode, |
1572 | bool rcu) |
1573 | { |
1574 | if (!rcu) { |
1575 | struct inode *realinode = d_real_inode(dentry); |
1576 | |
1577 | if (unlikely(inode != realinode) && |
1578 | (!timespec_equal(&inode->i_mtime, &realinode->i_mtime) || |
1579 | !timespec_equal(&inode->i_ctime, &realinode->i_ctime))) { |
1580 | inode->i_mtime = realinode->i_mtime; |
1581 | inode->i_ctime = realinode->i_ctime; |
1582 | } |
1583 | } |
1584 | } |
1585 | |
1586 | /* |
1587 | * With relative atime, only update atime if the previous atime is |
1588 | * earlier than either the ctime or mtime or if at least a day has |
1589 | * passed since the last atime update. |
1590 | */ |
1591 | static int relatime_need_update(const struct path *path, struct inode *inode, |
1592 | struct timespec now, bool rcu) |
1593 | { |
1594 | |
1595 | if (!(path->mnt->mnt_flags & MNT_RELATIME)) |
1596 | return 1; |
1597 | |
1598 | update_ovl_inode_times(path->dentry, inode, rcu); |
1599 | /* |
1600 | * Is mtime younger than atime? If yes, update atime: |
1601 | */ |
1602 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) |
1603 | return 1; |
1604 | /* |
1605 | * Is ctime younger than atime? If yes, update atime: |
1606 | */ |
1607 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) |
1608 | return 1; |
1609 | |
1610 | /* |
1611 | * Is the previous atime value older than a day? If yes, |
1612 | * update atime: |
1613 | */ |
1614 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) |
1615 | return 1; |
1616 | /* |
1617 | * Good, we can skip the atime update: |
1618 | */ |
1619 | return 0; |
1620 | } |
1621 | |
1622 | int generic_update_time(struct inode *inode, struct timespec *time, int flags) |
1623 | { |
1624 | int iflags = I_DIRTY_TIME; |
1625 | |
1626 | if (flags & S_ATIME) |
1627 | inode->i_atime = *time; |
1628 | if (flags & S_VERSION) |
1629 | inode_inc_iversion(inode); |
1630 | if (flags & S_CTIME) |
1631 | inode->i_ctime = *time; |
1632 | if (flags & S_MTIME) |
1633 | inode->i_mtime = *time; |
1634 | |
1635 | if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION)) |
1636 | iflags |= I_DIRTY_SYNC; |
1637 | __mark_inode_dirty(inode, iflags); |
1638 | return 0; |
1639 | } |
1640 | EXPORT_SYMBOL(generic_update_time); |
1641 | |
1642 | /* |
1643 | * This does the actual work of updating an inodes time or version. Must have |
1644 | * had called mnt_want_write() before calling this. |
1645 | */ |
1646 | static int update_time(struct inode *inode, struct timespec *time, int flags) |
1647 | { |
1648 | int (*update_time)(struct inode *, struct timespec *, int); |
1649 | |
1650 | update_time = inode->i_op->update_time ? inode->i_op->update_time : |
1651 | generic_update_time; |
1652 | |
1653 | return update_time(inode, time, flags); |
1654 | } |
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 | umode_t mode = d_inode(dentry)->i_mode; |
1741 | int kill = 0; |
1742 | |
1743 | /* suid always must be killed */ |
1744 | if (unlikely(mode & S_ISUID)) |
1745 | kill = ATTR_KILL_SUID; |
1746 | |
1747 | /* |
1748 | * sgid without any exec bits is just a mandatory locking mark; leave |
1749 | * it alone. If some exec bits are set, it's a real sgid; kill it. |
1750 | */ |
1751 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) |
1752 | kill |= ATTR_KILL_SGID; |
1753 | |
1754 | if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) |
1755 | return kill; |
1756 | |
1757 | return 0; |
1758 | } |
1759 | EXPORT_SYMBOL(should_remove_suid); |
1760 | |
1761 | /* |
1762 | * Return mask of changes for notify_change() that need to be done as a |
1763 | * response to write or truncate. Return 0 if nothing has to be changed. |
1764 | * Negative value on error (change should be denied). |
1765 | */ |
1766 | int dentry_needs_remove_privs(struct dentry *dentry) |
1767 | { |
1768 | struct inode *inode = d_inode(dentry); |
1769 | int mask = 0; |
1770 | int ret; |
1771 | |
1772 | if (IS_NOSEC(inode)) |
1773 | return 0; |
1774 | |
1775 | mask = should_remove_suid(dentry); |
1776 | ret = security_inode_need_killpriv(dentry); |
1777 | if (ret < 0) |
1778 | return ret; |
1779 | if (ret) |
1780 | mask |= ATTR_KILL_PRIV; |
1781 | return mask; |
1782 | } |
1783 | |
1784 | static int __remove_privs(struct vfsmount *mnt, struct dentry *dentry, int kill) |
1785 | { |
1786 | struct iattr newattrs; |
1787 | |
1788 | newattrs.ia_valid = ATTR_FORCE | kill; |
1789 | /* |
1790 | * Note we call this on write, so notify_change will not |
1791 | * encounter any conflicting delegations: |
1792 | */ |
1793 | return notify_change2(mnt, dentry, &newattrs, NULL); |
1794 | } |
1795 | |
1796 | /* |
1797 | * Remove special file priviledges (suid, capabilities) when file is written |
1798 | * to or truncated. |
1799 | */ |
1800 | int file_remove_privs(struct file *file) |
1801 | { |
1802 | struct dentry *dentry = file_dentry(file); |
1803 | struct inode *inode = file_inode(file); |
1804 | int kill; |
1805 | int error = 0; |
1806 | |
1807 | /* Fast path for nothing security related */ |
1808 | if (IS_NOSEC(inode)) |
1809 | return 0; |
1810 | |
1811 | kill = dentry_needs_remove_privs(dentry); |
1812 | if (kill < 0) |
1813 | return kill; |
1814 | if (kill) |
1815 | error = __remove_privs(file->f_path.mnt, dentry, kill); |
1816 | if (!error) |
1817 | inode_has_no_xattr(inode); |
1818 | |
1819 | return error; |
1820 | } |
1821 | EXPORT_SYMBOL(file_remove_privs); |
1822 | |
1823 | /** |
1824 | * file_update_time - update mtime and ctime time |
1825 | * @file: file accessed |
1826 | * |
1827 | * Update the mtime and ctime members of an inode and mark the inode |
1828 | * for writeback. Note that this function is meant exclusively for |
1829 | * usage in the file write path of filesystems, and filesystems may |
1830 | * choose to explicitly ignore update via this function with the |
1831 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
1832 | * timestamps are handled by the server. This can return an error for |
1833 | * file systems who need to allocate space in order to update an inode. |
1834 | */ |
1835 | |
1836 | int file_update_time(struct file *file) |
1837 | { |
1838 | struct inode *inode = file_inode(file); |
1839 | struct timespec now; |
1840 | int sync_it = 0; |
1841 | int ret; |
1842 | |
1843 | /* First try to exhaust all avenues to not sync */ |
1844 | if (IS_NOCMTIME(inode)) |
1845 | return 0; |
1846 | |
1847 | now = current_time(inode); |
1848 | if (!timespec_equal(&inode->i_mtime, &now)) |
1849 | sync_it = S_MTIME; |
1850 | |
1851 | if (!timespec_equal(&inode->i_ctime, &now)) |
1852 | sync_it |= S_CTIME; |
1853 | |
1854 | if (IS_I_VERSION(inode)) |
1855 | sync_it |= S_VERSION; |
1856 | |
1857 | if (!sync_it) |
1858 | return 0; |
1859 | |
1860 | /* Finally allowed to write? Takes lock. */ |
1861 | if (__mnt_want_write_file(file)) |
1862 | return 0; |
1863 | |
1864 | ret = update_time(inode, &now, sync_it); |
1865 | __mnt_drop_write_file(file); |
1866 | |
1867 | return ret; |
1868 | } |
1869 | EXPORT_SYMBOL(file_update_time); |
1870 | |
1871 | int inode_needs_sync(struct inode *inode) |
1872 | { |
1873 | if (IS_SYNC(inode)) |
1874 | return 1; |
1875 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) |
1876 | return 1; |
1877 | return 0; |
1878 | } |
1879 | EXPORT_SYMBOL(inode_needs_sync); |
1880 | |
1881 | /* |
1882 | * If we try to find an inode in the inode hash while it is being |
1883 | * deleted, we have to wait until the filesystem completes its |
1884 | * deletion before reporting that it isn't found. This function waits |
1885 | * until the deletion _might_ have completed. Callers are responsible |
1886 | * to recheck inode state. |
1887 | * |
1888 | * It doesn't matter if I_NEW is not set initially, a call to |
1889 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list |
1890 | * will DTRT. |
1891 | */ |
1892 | static void __wait_on_freeing_inode(struct inode *inode) |
1893 | { |
1894 | wait_queue_head_t *wq; |
1895 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1896 | wq = bit_waitqueue(&inode->i_state, __I_NEW); |
1897 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
1898 | spin_unlock(&inode->i_lock); |
1899 | spin_unlock(&inode_hash_lock); |
1900 | schedule(); |
1901 | finish_wait(wq, &wait.wait); |
1902 | spin_lock(&inode_hash_lock); |
1903 | } |
1904 | |
1905 | static __initdata unsigned long ihash_entries; |
1906 | static int __init set_ihash_entries(char *str) |
1907 | { |
1908 | if (!str) |
1909 | return 0; |
1910 | ihash_entries = simple_strtoul(str, &str, 0); |
1911 | return 1; |
1912 | } |
1913 | __setup("ihash_entries=", set_ihash_entries); |
1914 | |
1915 | /* |
1916 | * Initialize the waitqueues and inode hash table. |
1917 | */ |
1918 | void __init inode_init_early(void) |
1919 | { |
1920 | unsigned int loop; |
1921 | |
1922 | /* If hashes are distributed across NUMA nodes, defer |
1923 | * hash allocation until vmalloc space is available. |
1924 | */ |
1925 | if (hashdist) |
1926 | return; |
1927 | |
1928 | inode_hashtable = |
1929 | alloc_large_system_hash("Inode-cache", |
1930 | sizeof(struct hlist_head), |
1931 | ihash_entries, |
1932 | 14, |
1933 | HASH_EARLY, |
1934 | &i_hash_shift, |
1935 | &i_hash_mask, |
1936 | 0, |
1937 | 0); |
1938 | |
1939 | for (loop = 0; loop < (1U << i_hash_shift); loop++) |
1940 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1941 | } |
1942 | |
1943 | void __init inode_init(void) |
1944 | { |
1945 | unsigned int loop; |
1946 | |
1947 | /* inode slab cache */ |
1948 | inode_cachep = kmem_cache_create("inode_cache", |
1949 | sizeof(struct inode), |
1950 | 0, |
1951 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| |
1952 | SLAB_MEM_SPREAD|SLAB_ACCOUNT), |
1953 | init_once); |
1954 | |
1955 | /* Hash may have been set up in inode_init_early */ |
1956 | if (!hashdist) |
1957 | return; |
1958 | |
1959 | inode_hashtable = |
1960 | alloc_large_system_hash("Inode-cache", |
1961 | sizeof(struct hlist_head), |
1962 | ihash_entries, |
1963 | 14, |
1964 | 0, |
1965 | &i_hash_shift, |
1966 | &i_hash_mask, |
1967 | 0, |
1968 | 0); |
1969 | |
1970 | for (loop = 0; loop < (1U << i_hash_shift); loop++) |
1971 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1972 | } |
1973 | |
1974 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) |
1975 | { |
1976 | inode->i_mode = mode; |
1977 | if (S_ISCHR(mode)) { |
1978 | inode->i_fop = &def_chr_fops; |
1979 | inode->i_rdev = rdev; |
1980 | } else if (S_ISBLK(mode)) { |
1981 | inode->i_fop = &def_blk_fops; |
1982 | inode->i_rdev = rdev; |
1983 | } else if (S_ISFIFO(mode)) |
1984 | inode->i_fop = &pipefifo_fops; |
1985 | else if (S_ISSOCK(mode)) |
1986 | ; /* leave it no_open_fops */ |
1987 | else |
1988 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1989 | " inode %s:%lu\n", mode, inode->i_sb->s_id, |
1990 | inode->i_ino); |
1991 | } |
1992 | EXPORT_SYMBOL(init_special_inode); |
1993 | |
1994 | /** |
1995 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards |
1996 | * @inode: New inode |
1997 | * @dir: Directory inode |
1998 | * @mode: mode of the new inode |
1999 | */ |
2000 | void inode_init_owner(struct inode *inode, const struct inode *dir, |
2001 | umode_t mode) |
2002 | { |
2003 | inode->i_uid = current_fsuid(); |
2004 | if (dir && dir->i_mode & S_ISGID) { |
2005 | inode->i_gid = dir->i_gid; |
2006 | if (S_ISDIR(mode)) |
2007 | mode |= S_ISGID; |
2008 | } else |
2009 | inode->i_gid = current_fsgid(); |
2010 | inode->i_mode = mode; |
2011 | } |
2012 | EXPORT_SYMBOL(inode_init_owner); |
2013 | |
2014 | /** |
2015 | * inode_owner_or_capable - check current task permissions to inode |
2016 | * @inode: inode being checked |
2017 | * |
2018 | * Return true if current either has CAP_FOWNER in a namespace with the |
2019 | * inode owner uid mapped, or owns the file. |
2020 | */ |
2021 | bool inode_owner_or_capable(const struct inode *inode) |
2022 | { |
2023 | struct user_namespace *ns; |
2024 | |
2025 | if (uid_eq(current_fsuid(), inode->i_uid)) |
2026 | return true; |
2027 | |
2028 | ns = current_user_ns(); |
2029 | if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid)) |
2030 | return true; |
2031 | return false; |
2032 | } |
2033 | EXPORT_SYMBOL(inode_owner_or_capable); |
2034 | |
2035 | /* |
2036 | * Direct i/o helper functions |
2037 | */ |
2038 | static void __inode_dio_wait(struct inode *inode) |
2039 | { |
2040 | wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP); |
2041 | DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP); |
2042 | |
2043 | do { |
2044 | prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE); |
2045 | if (atomic_read(&inode->i_dio_count)) |
2046 | schedule(); |
2047 | } while (atomic_read(&inode->i_dio_count)); |
2048 | finish_wait(wq, &q.wait); |
2049 | } |
2050 | |
2051 | /** |
2052 | * inode_dio_wait - wait for outstanding DIO requests to finish |
2053 | * @inode: inode to wait for |
2054 | * |
2055 | * Waits for all pending direct I/O requests to finish so that we can |
2056 | * proceed with a truncate or equivalent operation. |
2057 | * |
2058 | * Must be called under a lock that serializes taking new references |
2059 | * to i_dio_count, usually by inode->i_mutex. |
2060 | */ |
2061 | void inode_dio_wait(struct inode *inode) |
2062 | { |
2063 | if (atomic_read(&inode->i_dio_count)) |
2064 | __inode_dio_wait(inode); |
2065 | } |
2066 | EXPORT_SYMBOL(inode_dio_wait); |
2067 | |
2068 | /* |
2069 | * inode_set_flags - atomically set some inode flags |
2070 | * |
2071 | * Note: the caller should be holding i_mutex, or else be sure that |
2072 | * they have exclusive access to the inode structure (i.e., while the |
2073 | * inode is being instantiated). The reason for the cmpxchg() loop |
2074 | * --- which wouldn't be necessary if all code paths which modify |
2075 | * i_flags actually followed this rule, is that there is at least one |
2076 | * code path which doesn't today so we use cmpxchg() out of an abundance |
2077 | * of caution. |
2078 | * |
2079 | * In the long run, i_mutex is overkill, and we should probably look |
2080 | * at using the i_lock spinlock to protect i_flags, and then make sure |
2081 | * it is so documented in include/linux/fs.h and that all code follows |
2082 | * the locking convention!! |
2083 | */ |
2084 | void inode_set_flags(struct inode *inode, unsigned int flags, |
2085 | unsigned int mask) |
2086 | { |
2087 | unsigned int old_flags, new_flags; |
2088 | |
2089 | WARN_ON_ONCE(flags & ~mask); |
2090 | do { |
2091 | old_flags = ACCESS_ONCE(inode->i_flags); |
2092 | new_flags = (old_flags & ~mask) | flags; |
2093 | } while (unlikely(cmpxchg(&inode->i_flags, old_flags, |
2094 | new_flags) != old_flags)); |
2095 | } |
2096 | EXPORT_SYMBOL(inode_set_flags); |
2097 | |
2098 | void inode_nohighmem(struct inode *inode) |
2099 | { |
2100 | mapping_set_gfp_mask(inode->i_mapping, GFP_USER); |
2101 | } |
2102 | EXPORT_SYMBOL(inode_nohighmem); |
2103 | |
2104 | /** |
2105 | * current_time - Return FS time |
2106 | * @inode: inode. |
2107 | * |
2108 | * Return the current time truncated to the time granularity supported by |
2109 | * the fs. |
2110 | * |
2111 | * Note that inode and inode->sb cannot be NULL. |
2112 | * Otherwise, the function warns and returns time without truncation. |
2113 | */ |
2114 | struct timespec current_time(struct inode *inode) |
2115 | { |
2116 | struct timespec now = current_kernel_time(); |
2117 | |
2118 | if (unlikely(!inode->i_sb)) { |
2119 | WARN(1, "current_time() called with uninitialized super_block in the inode"); |
2120 | return now; |
2121 | } |
2122 | |
2123 | return timespec_trunc(now, inode->i_sb->s_time_gran); |
2124 | } |
2125 | EXPORT_SYMBOL(current_time); |
2126 |