1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * (C) 1997 Linus Torvalds
4 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5 */
6 #include <linux/export.h>
7 #include <linux/fs.h>
8 #include <linux/mm.h>
9 #include <linux/backing-dev.h>
10 #include <linux/hash.h>
11 #include <linux/swap.h>
12 #include <linux/security.h>
13 #include <linux/cdev.h>
14 #include <linux/memblock.h>
15 #include <linux/fscrypt.h>
16 #include <linux/fsnotify.h>
17 #include <linux/mount.h>
18 #include <linux/posix_acl.h>
19 #include <linux/prefetch.h>
20 #include <linux/buffer_head.h> /* for inode_has_buffers */
21 #include <linux/ratelimit.h>
22 #include <linux/list_lru.h>
23 #include <linux/iversion.h>
24 #include <trace/events/writeback.h>
25 #include "internal.h"
26
27 /*
28 * Inode locking rules:
29 *
30 * inode->i_lock protects:
31 * inode->i_state, inode->i_hash, __iget()
32 * Inode LRU list locks protect:
33 * inode->i_sb->s_inode_lru, inode->i_lru
34 * inode->i_sb->s_inode_list_lock protects:
35 * inode->i_sb->s_inodes, inode->i_sb_list
36 * bdi->wb.list_lock protects:
37 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
38 * inode_hash_lock protects:
39 * inode_hashtable, inode->i_hash
40 *
41 * Lock ordering:
42 *
43 * inode->i_sb->s_inode_list_lock
44 * inode->i_lock
45 * Inode LRU list locks
46 *
47 * bdi->wb.list_lock
48 * inode->i_lock
49 *
50 * inode_hash_lock
51 * inode->i_sb->s_inode_list_lock
52 * inode->i_lock
53 *
54 * iunique_lock
55 * inode_hash_lock
56 */
57
58 static unsigned int i_hash_mask __read_mostly;
59 static unsigned int i_hash_shift __read_mostly;
60 static struct hlist_head *inode_hashtable __read_mostly;
61 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
62
63 /*
64 * Empty aops. Can be used for the cases where the user does not
65 * define any of the address_space operations.
66 */
67 const struct address_space_operations empty_aops = {
68 };
69 EXPORT_SYMBOL(empty_aops);
70
71 /*
72 * Statistics gathering..
73 */
74 struct inodes_stat_t inodes_stat;
75
76 static DEFINE_PER_CPU(unsigned long, nr_inodes);
77 static DEFINE_PER_CPU(unsigned long, nr_unused);
78
79 static struct kmem_cache *inode_cachep __read_mostly;
80
get_nr_inodes(void)81 static long get_nr_inodes(void)
82 {
83 int i;
84 long sum = 0;
85 for_each_possible_cpu(i)
86 sum += per_cpu(nr_inodes, i);
87 return sum < 0 ? 0 : sum;
88 }
89
get_nr_inodes_unused(void)90 static inline long get_nr_inodes_unused(void)
91 {
92 int i;
93 long sum = 0;
94 for_each_possible_cpu(i)
95 sum += per_cpu(nr_unused, i);
96 return sum < 0 ? 0 : sum;
97 }
98
get_nr_dirty_inodes(void)99 long get_nr_dirty_inodes(void)
100 {
101 /* not actually dirty inodes, but a wild approximation */
102 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
103 return nr_dirty > 0 ? nr_dirty : 0;
104 }
105
106 /*
107 * Handle nr_inode sysctl
108 */
109 #ifdef CONFIG_SYSCTL
proc_nr_inodes(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)110 int proc_nr_inodes(struct ctl_table *table, int write,
111 void *buffer, size_t *lenp, loff_t *ppos)
112 {
113 inodes_stat.nr_inodes = get_nr_inodes();
114 inodes_stat.nr_unused = get_nr_inodes_unused();
115 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
116 }
117 #endif
118
no_open(struct inode * inode,struct file * file)119 static int no_open(struct inode *inode, struct file *file)
120 {
121 return -ENXIO;
122 }
123
124 /**
125 * inode_init_always - perform inode structure initialisation
126 * @sb: superblock inode belongs to
127 * @inode: inode to initialise
128 *
129 * These are initializations that need to be done on every inode
130 * allocation as the fields are not initialised by slab allocation.
131 */
inode_init_always(struct super_block * sb,struct inode * inode)132 int inode_init_always(struct super_block *sb, struct inode *inode)
133 {
134 static const struct inode_operations empty_iops;
135 static const struct file_operations no_open_fops = {.open = no_open};
136 struct address_space *const mapping = &inode->i_data;
137
138 inode->i_sb = sb;
139 inode->i_blkbits = sb->s_blocksize_bits;
140 inode->i_flags = 0;
141 atomic64_set(&inode->i_sequence, 0);
142 atomic_set(&inode->i_count, 1);
143 inode->i_op = &empty_iops;
144 inode->i_fop = &no_open_fops;
145 inode->__i_nlink = 1;
146 inode->i_opflags = 0;
147 if (sb->s_xattr)
148 inode->i_opflags |= IOP_XATTR;
149 i_uid_write(inode, 0);
150 i_gid_write(inode, 0);
151 atomic_set(&inode->i_writecount, 0);
152 inode->i_size = 0;
153 inode->i_write_hint = WRITE_LIFE_NOT_SET;
154 inode->i_blocks = 0;
155 inode->i_bytes = 0;
156 inode->i_generation = 0;
157 inode->i_pipe = NULL;
158 inode->i_bdev = NULL;
159 inode->i_cdev = NULL;
160 inode->i_link = NULL;
161 inode->i_dir_seq = 0;
162 inode->i_rdev = 0;
163 inode->dirtied_when = 0;
164
165 #ifdef CONFIG_CGROUP_WRITEBACK
166 inode->i_wb_frn_winner = 0;
167 inode->i_wb_frn_avg_time = 0;
168 inode->i_wb_frn_history = 0;
169 #endif
170
171 spin_lock_init(&inode->i_lock);
172 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
173
174 init_rwsem(&inode->i_rwsem);
175 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
176
177 atomic_set(&inode->i_dio_count, 0);
178
179 mapping->a_ops = &empty_aops;
180 mapping->host = inode;
181 mapping->flags = 0;
182 if (sb->s_type->fs_flags & FS_THP_SUPPORT)
183 __set_bit(AS_THP_SUPPORT, &mapping->flags);
184 mapping->wb_err = 0;
185 atomic_set(&mapping->i_mmap_writable, 0);
186 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
187 atomic_set(&mapping->nr_thps, 0);
188 #endif
189 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
190 mapping->private_data = NULL;
191 mapping->writeback_index = 0;
192 inode->i_private = NULL;
193 inode->i_mapping = mapping;
194 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
195 #ifdef CONFIG_FS_POSIX_ACL
196 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
197 #endif
198
199 #ifdef CONFIG_FSNOTIFY
200 inode->i_fsnotify_mask = 0;
201 #endif
202 inode->i_flctx = NULL;
203
204 if (unlikely(security_inode_alloc(inode)))
205 return -ENOMEM;
206 this_cpu_inc(nr_inodes);
207
208 return 0;
209 }
210 EXPORT_SYMBOL(inode_init_always);
211
free_inode_nonrcu(struct inode * inode)212 void free_inode_nonrcu(struct inode *inode)
213 {
214 kmem_cache_free(inode_cachep, inode);
215 }
216 EXPORT_SYMBOL(free_inode_nonrcu);
217
i_callback(struct rcu_head * head)218 static void i_callback(struct rcu_head *head)
219 {
220 struct inode *inode = container_of(head, struct inode, i_rcu);
221 if (inode->free_inode)
222 inode->free_inode(inode);
223 else
224 free_inode_nonrcu(inode);
225 }
226
alloc_inode(struct super_block * sb)227 static struct inode *alloc_inode(struct super_block *sb)
228 {
229 const struct super_operations *ops = sb->s_op;
230 struct inode *inode;
231
232 if (ops->alloc_inode)
233 inode = ops->alloc_inode(sb);
234 else
235 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
236
237 if (!inode)
238 return NULL;
239
240 if (unlikely(inode_init_always(sb, inode))) {
241 if (ops->destroy_inode) {
242 ops->destroy_inode(inode);
243 if (!ops->free_inode)
244 return NULL;
245 }
246 inode->free_inode = ops->free_inode;
247 i_callback(&inode->i_rcu);
248 return NULL;
249 }
250
251 return inode;
252 }
253
__destroy_inode(struct inode * inode)254 void __destroy_inode(struct inode *inode)
255 {
256 BUG_ON(inode_has_buffers(inode));
257 inode_detach_wb(inode);
258 security_inode_free(inode);
259 fsnotify_inode_delete(inode);
260 locks_free_lock_context(inode);
261 if (!inode->i_nlink) {
262 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
263 atomic_long_dec(&inode->i_sb->s_remove_count);
264 }
265
266 #ifdef CONFIG_FS_POSIX_ACL
267 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
268 posix_acl_release(inode->i_acl);
269 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
270 posix_acl_release(inode->i_default_acl);
271 #endif
272 this_cpu_dec(nr_inodes);
273 }
274 EXPORT_SYMBOL(__destroy_inode);
275
destroy_inode(struct inode * inode)276 static void destroy_inode(struct inode *inode)
277 {
278 const struct super_operations *ops = inode->i_sb->s_op;
279
280 BUG_ON(!list_empty(&inode->i_lru));
281 __destroy_inode(inode);
282 if (ops->destroy_inode) {
283 ops->destroy_inode(inode);
284 if (!ops->free_inode)
285 return;
286 }
287 inode->free_inode = ops->free_inode;
288 call_rcu(&inode->i_rcu, i_callback);
289 }
290
291 /**
292 * drop_nlink - directly drop an inode's link count
293 * @inode: inode
294 *
295 * This is a low-level filesystem helper to replace any
296 * direct filesystem manipulation of i_nlink. In cases
297 * where we are attempting to track writes to the
298 * filesystem, a decrement to zero means an imminent
299 * write when the file is truncated and actually unlinked
300 * on the filesystem.
301 */
drop_nlink(struct inode * inode)302 void drop_nlink(struct inode *inode)
303 {
304 WARN_ON(inode->i_nlink == 0);
305 inode->__i_nlink--;
306 if (!inode->i_nlink)
307 atomic_long_inc(&inode->i_sb->s_remove_count);
308 }
309 EXPORT_SYMBOL(drop_nlink);
310
311 /**
312 * clear_nlink - directly zero an inode's link count
313 * @inode: inode
314 *
315 * This is a low-level filesystem helper to replace any
316 * direct filesystem manipulation of i_nlink. See
317 * drop_nlink() for why we care about i_nlink hitting zero.
318 */
clear_nlink(struct inode * inode)319 void clear_nlink(struct inode *inode)
320 {
321 if (inode->i_nlink) {
322 inode->__i_nlink = 0;
323 atomic_long_inc(&inode->i_sb->s_remove_count);
324 }
325 }
326 EXPORT_SYMBOL(clear_nlink);
327
328 /**
329 * set_nlink - directly set an inode's link count
330 * @inode: inode
331 * @nlink: new nlink (should be non-zero)
332 *
333 * This is a low-level filesystem helper to replace any
334 * direct filesystem manipulation of i_nlink.
335 */
set_nlink(struct inode * inode,unsigned int nlink)336 void set_nlink(struct inode *inode, unsigned int nlink)
337 {
338 if (!nlink) {
339 clear_nlink(inode);
340 } else {
341 /* Yes, some filesystems do change nlink from zero to one */
342 if (inode->i_nlink == 0)
343 atomic_long_dec(&inode->i_sb->s_remove_count);
344
345 inode->__i_nlink = nlink;
346 }
347 }
348 EXPORT_SYMBOL(set_nlink);
349
350 /**
351 * inc_nlink - directly increment an inode's link count
352 * @inode: inode
353 *
354 * This is a low-level filesystem helper to replace any
355 * direct filesystem manipulation of i_nlink. Currently,
356 * it is only here for parity with dec_nlink().
357 */
inc_nlink(struct inode * inode)358 void inc_nlink(struct inode *inode)
359 {
360 if (unlikely(inode->i_nlink == 0)) {
361 WARN_ON(!(inode->i_state & I_LINKABLE));
362 atomic_long_dec(&inode->i_sb->s_remove_count);
363 }
364
365 inode->__i_nlink++;
366 }
367 EXPORT_SYMBOL(inc_nlink);
368
__address_space_init_once(struct address_space * mapping)369 static void __address_space_init_once(struct address_space *mapping)
370 {
371 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
372 init_rwsem(&mapping->i_mmap_rwsem);
373 INIT_LIST_HEAD(&mapping->private_list);
374 spin_lock_init(&mapping->private_lock);
375 mapping->i_mmap = RB_ROOT_CACHED;
376 }
377
address_space_init_once(struct address_space * mapping)378 void address_space_init_once(struct address_space *mapping)
379 {
380 memset(mapping, 0, sizeof(*mapping));
381 __address_space_init_once(mapping);
382 }
383 EXPORT_SYMBOL(address_space_init_once);
384
385 /*
386 * These are initializations that only need to be done
387 * once, because the fields are idempotent across use
388 * of the inode, so let the slab aware of that.
389 */
inode_init_once(struct inode * inode)390 void inode_init_once(struct inode *inode)
391 {
392 memset(inode, 0, sizeof(*inode));
393 INIT_HLIST_NODE(&inode->i_hash);
394 INIT_LIST_HEAD(&inode->i_devices);
395 INIT_LIST_HEAD(&inode->i_io_list);
396 INIT_LIST_HEAD(&inode->i_wb_list);
397 INIT_LIST_HEAD(&inode->i_lru);
398 __address_space_init_once(&inode->i_data);
399 i_size_ordered_init(inode);
400 }
401 EXPORT_SYMBOL(inode_init_once);
402
init_once(void * foo)403 static void init_once(void *foo)
404 {
405 struct inode *inode = (struct inode *) foo;
406
407 inode_init_once(inode);
408 }
409
410 /*
411 * inode->i_lock must be held
412 */
__iget(struct inode * inode)413 void __iget(struct inode *inode)
414 {
415 atomic_inc(&inode->i_count);
416 }
417
418 /*
419 * get additional reference to inode; caller must already hold one.
420 */
ihold(struct inode * inode)421 void ihold(struct inode *inode)
422 {
423 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
424 }
425 EXPORT_SYMBOL(ihold);
426
inode_lru_list_add(struct inode * inode)427 static void inode_lru_list_add(struct inode *inode)
428 {
429 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
430 this_cpu_inc(nr_unused);
431 else
432 inode->i_state |= I_REFERENCED;
433 }
434
435 /*
436 * Add inode to LRU if needed (inode is unused and clean).
437 *
438 * Needs inode->i_lock held.
439 */
inode_add_lru(struct inode * inode)440 void inode_add_lru(struct inode *inode)
441 {
442 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
443 I_FREEING | I_WILL_FREE)) &&
444 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
445 inode_lru_list_add(inode);
446 }
447
448
inode_lru_list_del(struct inode * inode)449 static void inode_lru_list_del(struct inode *inode)
450 {
451
452 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
453 this_cpu_dec(nr_unused);
454 }
455
456 /**
457 * inode_sb_list_add - add inode to the superblock list of inodes
458 * @inode: inode to add
459 */
inode_sb_list_add(struct inode * inode)460 void inode_sb_list_add(struct inode *inode)
461 {
462 spin_lock(&inode->i_sb->s_inode_list_lock);
463 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
464 spin_unlock(&inode->i_sb->s_inode_list_lock);
465 }
466 EXPORT_SYMBOL_GPL(inode_sb_list_add);
467
inode_sb_list_del(struct inode * inode)468 static inline void inode_sb_list_del(struct inode *inode)
469 {
470 if (!list_empty(&inode->i_sb_list)) {
471 spin_lock(&inode->i_sb->s_inode_list_lock);
472 list_del_init(&inode->i_sb_list);
473 spin_unlock(&inode->i_sb->s_inode_list_lock);
474 }
475 }
476
hash(struct super_block * sb,unsigned long hashval)477 static unsigned long hash(struct super_block *sb, unsigned long hashval)
478 {
479 unsigned long tmp;
480
481 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
482 L1_CACHE_BYTES;
483 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
484 return tmp & i_hash_mask;
485 }
486
487 /**
488 * __insert_inode_hash - hash an inode
489 * @inode: unhashed inode
490 * @hashval: unsigned long value used to locate this object in the
491 * inode_hashtable.
492 *
493 * Add an inode to the inode hash for this superblock.
494 */
__insert_inode_hash(struct inode * inode,unsigned long hashval)495 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
496 {
497 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
498
499 spin_lock(&inode_hash_lock);
500 spin_lock(&inode->i_lock);
501 hlist_add_head_rcu(&inode->i_hash, b);
502 spin_unlock(&inode->i_lock);
503 spin_unlock(&inode_hash_lock);
504 }
505 EXPORT_SYMBOL(__insert_inode_hash);
506
507 /**
508 * __remove_inode_hash - remove an inode from the hash
509 * @inode: inode to unhash
510 *
511 * Remove an inode from the superblock.
512 */
__remove_inode_hash(struct inode * inode)513 void __remove_inode_hash(struct inode *inode)
514 {
515 spin_lock(&inode_hash_lock);
516 spin_lock(&inode->i_lock);
517 hlist_del_init_rcu(&inode->i_hash);
518 spin_unlock(&inode->i_lock);
519 spin_unlock(&inode_hash_lock);
520 }
521 EXPORT_SYMBOL(__remove_inode_hash);
522
clear_inode(struct inode * inode)523 void clear_inode(struct inode *inode)
524 {
525 /*
526 * We have to cycle the i_pages lock here because reclaim can be in the
527 * process of removing the last page (in __delete_from_page_cache())
528 * and we must not free the mapping under it.
529 */
530 xa_lock_irq(&inode->i_data.i_pages);
531 BUG_ON(inode->i_data.nrpages);
532 BUG_ON(inode->i_data.nrexceptional);
533 xa_unlock_irq(&inode->i_data.i_pages);
534 BUG_ON(!list_empty(&inode->i_data.private_list));
535 BUG_ON(!(inode->i_state & I_FREEING));
536 BUG_ON(inode->i_state & I_CLEAR);
537 BUG_ON(!list_empty(&inode->i_wb_list));
538 /* don't need i_lock here, no concurrent mods to i_state */
539 inode->i_state = I_FREEING | I_CLEAR;
540 }
541 EXPORT_SYMBOL(clear_inode);
542
543 /*
544 * Free the inode passed in, removing it from the lists it is still connected
545 * to. We remove any pages still attached to the inode and wait for any IO that
546 * is still in progress before finally destroying the inode.
547 *
548 * An inode must already be marked I_FREEING so that we avoid the inode being
549 * moved back onto lists if we race with other code that manipulates the lists
550 * (e.g. writeback_single_inode). The caller is responsible for setting this.
551 *
552 * An inode must already be removed from the LRU list before being evicted from
553 * the cache. This should occur atomically with setting the I_FREEING state
554 * flag, so no inodes here should ever be on the LRU when being evicted.
555 */
evict(struct inode * inode)556 static void evict(struct inode *inode)
557 {
558 const struct super_operations *op = inode->i_sb->s_op;
559
560 BUG_ON(!(inode->i_state & I_FREEING));
561 BUG_ON(!list_empty(&inode->i_lru));
562
563 if (!list_empty(&inode->i_io_list))
564 inode_io_list_del(inode);
565
566 inode_sb_list_del(inode);
567
568 /*
569 * Wait for flusher thread to be done with the inode so that filesystem
570 * does not start destroying it while writeback is still running. Since
571 * the inode has I_FREEING set, flusher thread won't start new work on
572 * the inode. We just have to wait for running writeback to finish.
573 */
574 inode_wait_for_writeback(inode);
575
576 if (op->evict_inode) {
577 op->evict_inode(inode);
578 } else {
579 truncate_inode_pages_final(&inode->i_data);
580 clear_inode(inode);
581 }
582 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
583 bd_forget(inode);
584 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
585 cd_forget(inode);
586
587 remove_inode_hash(inode);
588
589 spin_lock(&inode->i_lock);
590 wake_up_bit(&inode->i_state, __I_NEW);
591 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
592 spin_unlock(&inode->i_lock);
593
594 destroy_inode(inode);
595 }
596
597 /*
598 * dispose_list - dispose of the contents of a local list
599 * @head: the head of the list to free
600 *
601 * Dispose-list gets a local list with local inodes in it, so it doesn't
602 * need to worry about list corruption and SMP locks.
603 */
dispose_list(struct list_head * head)604 static void dispose_list(struct list_head *head)
605 {
606 while (!list_empty(head)) {
607 struct inode *inode;
608
609 inode = list_first_entry(head, struct inode, i_lru);
610 list_del_init(&inode->i_lru);
611
612 evict(inode);
613 cond_resched();
614 }
615 }
616
617 /**
618 * evict_inodes - evict all evictable inodes for a superblock
619 * @sb: superblock to operate on
620 *
621 * Make sure that no inodes with zero refcount are retained. This is
622 * called by superblock shutdown after having SB_ACTIVE flag removed,
623 * so any inode reaching zero refcount during or after that call will
624 * be immediately evicted.
625 */
evict_inodes(struct super_block * sb)626 void evict_inodes(struct super_block *sb)
627 {
628 struct inode *inode, *next;
629 LIST_HEAD(dispose);
630
631 again:
632 spin_lock(&sb->s_inode_list_lock);
633 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
634 if (atomic_read(&inode->i_count))
635 continue;
636
637 spin_lock(&inode->i_lock);
638 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
639 spin_unlock(&inode->i_lock);
640 continue;
641 }
642
643 inode->i_state |= I_FREEING;
644 inode_lru_list_del(inode);
645 spin_unlock(&inode->i_lock);
646 list_add(&inode->i_lru, &dispose);
647
648 /*
649 * We can have a ton of inodes to evict at unmount time given
650 * enough memory, check to see if we need to go to sleep for a
651 * bit so we don't livelock.
652 */
653 if (need_resched()) {
654 spin_unlock(&sb->s_inode_list_lock);
655 cond_resched();
656 dispose_list(&dispose);
657 goto again;
658 }
659 }
660 spin_unlock(&sb->s_inode_list_lock);
661
662 dispose_list(&dispose);
663 }
664 EXPORT_SYMBOL_GPL(evict_inodes);
665
666 /**
667 * invalidate_inodes - attempt to free all inodes on a superblock
668 * @sb: superblock to operate on
669 * @kill_dirty: flag to guide handling of dirty inodes
670 *
671 * Attempts to free all inodes for a given superblock. If there were any
672 * busy inodes return a non-zero value, else zero.
673 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
674 * them as busy.
675 */
invalidate_inodes(struct super_block * sb,bool kill_dirty)676 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
677 {
678 int busy = 0;
679 struct inode *inode, *next;
680 LIST_HEAD(dispose);
681
682 again:
683 spin_lock(&sb->s_inode_list_lock);
684 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
685 spin_lock(&inode->i_lock);
686 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
687 spin_unlock(&inode->i_lock);
688 continue;
689 }
690 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
691 spin_unlock(&inode->i_lock);
692 busy = 1;
693 continue;
694 }
695 if (atomic_read(&inode->i_count)) {
696 spin_unlock(&inode->i_lock);
697 busy = 1;
698 continue;
699 }
700
701 inode->i_state |= I_FREEING;
702 inode_lru_list_del(inode);
703 spin_unlock(&inode->i_lock);
704 list_add(&inode->i_lru, &dispose);
705 if (need_resched()) {
706 spin_unlock(&sb->s_inode_list_lock);
707 cond_resched();
708 dispose_list(&dispose);
709 goto again;
710 }
711 }
712 spin_unlock(&sb->s_inode_list_lock);
713
714 dispose_list(&dispose);
715
716 return busy;
717 }
718
719 /*
720 * Isolate the inode from the LRU in preparation for freeing it.
721 *
722 * Any inodes which are pinned purely because of attached pagecache have their
723 * pagecache removed. If the inode has metadata buffers attached to
724 * mapping->private_list then try to remove them.
725 *
726 * If the inode has the I_REFERENCED flag set, then it means that it has been
727 * used recently - the flag is set in iput_final(). When we encounter such an
728 * inode, clear the flag and move it to the back of the LRU so it gets another
729 * pass through the LRU before it gets reclaimed. This is necessary because of
730 * the fact we are doing lazy LRU updates to minimise lock contention so the
731 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
732 * with this flag set because they are the inodes that are out of order.
733 */
inode_lru_isolate(struct list_head * item,struct list_lru_one * lru,spinlock_t * lru_lock,void * arg)734 static enum lru_status inode_lru_isolate(struct list_head *item,
735 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
736 {
737 struct list_head *freeable = arg;
738 struct inode *inode = container_of(item, struct inode, i_lru);
739
740 /*
741 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
742 * If we fail to get the lock, just skip it.
743 */
744 if (!spin_trylock(&inode->i_lock))
745 return LRU_SKIP;
746
747 /*
748 * Referenced or dirty inodes are still in use. Give them another pass
749 * through the LRU as we canot reclaim them now.
750 */
751 if (atomic_read(&inode->i_count) ||
752 (inode->i_state & ~I_REFERENCED)) {
753 list_lru_isolate(lru, &inode->i_lru);
754 spin_unlock(&inode->i_lock);
755 this_cpu_dec(nr_unused);
756 return LRU_REMOVED;
757 }
758
759 /* recently referenced inodes get one more pass */
760 if (inode->i_state & I_REFERENCED) {
761 inode->i_state &= ~I_REFERENCED;
762 spin_unlock(&inode->i_lock);
763 return LRU_ROTATE;
764 }
765
766 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
767 __iget(inode);
768 spin_unlock(&inode->i_lock);
769 spin_unlock(lru_lock);
770 if (remove_inode_buffers(inode)) {
771 unsigned long reap;
772 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
773 if (current_is_kswapd())
774 __count_vm_events(KSWAPD_INODESTEAL, reap);
775 else
776 __count_vm_events(PGINODESTEAL, reap);
777 if (current->reclaim_state)
778 current->reclaim_state->reclaimed_slab += reap;
779 }
780 iput(inode);
781 spin_lock(lru_lock);
782 return LRU_RETRY;
783 }
784
785 WARN_ON(inode->i_state & I_NEW);
786 inode->i_state |= I_FREEING;
787 list_lru_isolate_move(lru, &inode->i_lru, freeable);
788 spin_unlock(&inode->i_lock);
789
790 this_cpu_dec(nr_unused);
791 return LRU_REMOVED;
792 }
793
794 /*
795 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
796 * This is called from the superblock shrinker function with a number of inodes
797 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
798 * then are freed outside inode_lock by dispose_list().
799 */
prune_icache_sb(struct super_block * sb,struct shrink_control * sc)800 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
801 {
802 LIST_HEAD(freeable);
803 long freed;
804
805 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
806 inode_lru_isolate, &freeable);
807 dispose_list(&freeable);
808 return freed;
809 }
810
811 static void __wait_on_freeing_inode(struct inode *inode);
812 /*
813 * Called with the inode lock held.
814 */
find_inode(struct super_block * sb,struct hlist_head * head,int (* test)(struct inode *,void *),void * data)815 static struct inode *find_inode(struct super_block *sb,
816 struct hlist_head *head,
817 int (*test)(struct inode *, void *),
818 void *data)
819 {
820 struct inode *inode = NULL;
821
822 repeat:
823 hlist_for_each_entry(inode, head, i_hash) {
824 if (inode->i_sb != sb)
825 continue;
826 if (!test(inode, data))
827 continue;
828 spin_lock(&inode->i_lock);
829 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
830 __wait_on_freeing_inode(inode);
831 goto repeat;
832 }
833 if (unlikely(inode->i_state & I_CREATING)) {
834 spin_unlock(&inode->i_lock);
835 return ERR_PTR(-ESTALE);
836 }
837 __iget(inode);
838 spin_unlock(&inode->i_lock);
839 return inode;
840 }
841 return NULL;
842 }
843
844 /*
845 * find_inode_fast is the fast path version of find_inode, see the comment at
846 * iget_locked for details.
847 */
find_inode_fast(struct super_block * sb,struct hlist_head * head,unsigned long ino)848 static struct inode *find_inode_fast(struct super_block *sb,
849 struct hlist_head *head, unsigned long ino)
850 {
851 struct inode *inode = NULL;
852
853 repeat:
854 hlist_for_each_entry(inode, head, i_hash) {
855 if (inode->i_ino != ino)
856 continue;
857 if (inode->i_sb != sb)
858 continue;
859 spin_lock(&inode->i_lock);
860 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
861 __wait_on_freeing_inode(inode);
862 goto repeat;
863 }
864 if (unlikely(inode->i_state & I_CREATING)) {
865 spin_unlock(&inode->i_lock);
866 return ERR_PTR(-ESTALE);
867 }
868 __iget(inode);
869 spin_unlock(&inode->i_lock);
870 return inode;
871 }
872 return NULL;
873 }
874
875 /*
876 * Each cpu owns a range of LAST_INO_BATCH numbers.
877 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
878 * to renew the exhausted range.
879 *
880 * This does not significantly increase overflow rate because every CPU can
881 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
882 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
883 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
884 * overflow rate by 2x, which does not seem too significant.
885 *
886 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
887 * error if st_ino won't fit in target struct field. Use 32bit counter
888 * here to attempt to avoid that.
889 */
890 #define LAST_INO_BATCH 1024
891 static DEFINE_PER_CPU(unsigned int, last_ino);
892
get_next_ino(void)893 unsigned int get_next_ino(void)
894 {
895 unsigned int *p = &get_cpu_var(last_ino);
896 unsigned int res = *p;
897
898 #ifdef CONFIG_SMP
899 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
900 static atomic_t shared_last_ino;
901 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
902
903 res = next - LAST_INO_BATCH;
904 }
905 #endif
906
907 res++;
908 /* get_next_ino should not provide a 0 inode number */
909 if (unlikely(!res))
910 res++;
911 *p = res;
912 put_cpu_var(last_ino);
913 return res;
914 }
915 EXPORT_SYMBOL(get_next_ino);
916
917 /**
918 * new_inode_pseudo - obtain an inode
919 * @sb: superblock
920 *
921 * Allocates a new inode for given superblock.
922 * Inode wont be chained in superblock s_inodes list
923 * This means :
924 * - fs can't be unmount
925 * - quotas, fsnotify, writeback can't work
926 */
new_inode_pseudo(struct super_block * sb)927 struct inode *new_inode_pseudo(struct super_block *sb)
928 {
929 struct inode *inode = alloc_inode(sb);
930
931 if (inode) {
932 spin_lock(&inode->i_lock);
933 inode->i_state = 0;
934 spin_unlock(&inode->i_lock);
935 INIT_LIST_HEAD(&inode->i_sb_list);
936 }
937 return inode;
938 }
939
940 /**
941 * new_inode - obtain an inode
942 * @sb: superblock
943 *
944 * Allocates a new inode for given superblock. The default gfp_mask
945 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
946 * If HIGHMEM pages are unsuitable or it is known that pages allocated
947 * for the page cache are not reclaimable or migratable,
948 * mapping_set_gfp_mask() must be called with suitable flags on the
949 * newly created inode's mapping
950 *
951 */
new_inode(struct super_block * sb)952 struct inode *new_inode(struct super_block *sb)
953 {
954 struct inode *inode;
955
956 spin_lock_prefetch(&sb->s_inode_list_lock);
957
958 inode = new_inode_pseudo(sb);
959 if (inode)
960 inode_sb_list_add(inode);
961 return inode;
962 }
963 EXPORT_SYMBOL(new_inode);
964
965 #ifdef CONFIG_DEBUG_LOCK_ALLOC
lockdep_annotate_inode_mutex_key(struct inode * inode)966 void lockdep_annotate_inode_mutex_key(struct inode *inode)
967 {
968 if (S_ISDIR(inode->i_mode)) {
969 struct file_system_type *type = inode->i_sb->s_type;
970
971 /* Set new key only if filesystem hasn't already changed it */
972 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
973 /*
974 * ensure nobody is actually holding i_mutex
975 */
976 // mutex_destroy(&inode->i_mutex);
977 init_rwsem(&inode->i_rwsem);
978 lockdep_set_class(&inode->i_rwsem,
979 &type->i_mutex_dir_key);
980 }
981 }
982 }
983 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
984 #endif
985
986 /**
987 * unlock_new_inode - clear the I_NEW state and wake up any waiters
988 * @inode: new inode to unlock
989 *
990 * Called when the inode is fully initialised to clear the new state of the
991 * inode and wake up anyone waiting for the inode to finish initialisation.
992 */
unlock_new_inode(struct inode * inode)993 void unlock_new_inode(struct inode *inode)
994 {
995 lockdep_annotate_inode_mutex_key(inode);
996 spin_lock(&inode->i_lock);
997 WARN_ON(!(inode->i_state & I_NEW));
998 inode->i_state &= ~I_NEW & ~I_CREATING;
999 smp_mb();
1000 wake_up_bit(&inode->i_state, __I_NEW);
1001 spin_unlock(&inode->i_lock);
1002 }
1003 EXPORT_SYMBOL(unlock_new_inode);
1004
discard_new_inode(struct inode * inode)1005 void discard_new_inode(struct inode *inode)
1006 {
1007 lockdep_annotate_inode_mutex_key(inode);
1008 spin_lock(&inode->i_lock);
1009 WARN_ON(!(inode->i_state & I_NEW));
1010 inode->i_state &= ~I_NEW;
1011 smp_mb();
1012 wake_up_bit(&inode->i_state, __I_NEW);
1013 spin_unlock(&inode->i_lock);
1014 iput(inode);
1015 }
1016 EXPORT_SYMBOL(discard_new_inode);
1017
1018 /**
1019 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1020 *
1021 * Lock any non-NULL argument that is not a directory.
1022 * Zero, one or two objects may be locked by this function.
1023 *
1024 * @inode1: first inode to lock
1025 * @inode2: second inode to lock
1026 */
lock_two_nondirectories(struct inode * inode1,struct inode * inode2)1027 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1028 {
1029 if (inode1 > inode2)
1030 swap(inode1, inode2);
1031
1032 if (inode1 && !S_ISDIR(inode1->i_mode))
1033 inode_lock(inode1);
1034 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1035 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1036 }
1037 EXPORT_SYMBOL(lock_two_nondirectories);
1038
1039 /**
1040 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1041 * @inode1: first inode to unlock
1042 * @inode2: second inode to unlock
1043 */
unlock_two_nondirectories(struct inode * inode1,struct inode * inode2)1044 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1045 {
1046 if (inode1 && !S_ISDIR(inode1->i_mode))
1047 inode_unlock(inode1);
1048 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1049 inode_unlock(inode2);
1050 }
1051 EXPORT_SYMBOL(unlock_two_nondirectories);
1052
1053 /**
1054 * inode_insert5 - obtain an inode from a mounted file system
1055 * @inode: pre-allocated inode to use for insert to cache
1056 * @hashval: hash value (usually inode number) to get
1057 * @test: callback used for comparisons between inodes
1058 * @set: callback used to initialize a new struct inode
1059 * @data: opaque data pointer to pass to @test and @set
1060 *
1061 * Search for the inode specified by @hashval and @data in the inode cache,
1062 * and if present it is return it with an increased reference count. This is
1063 * a variant of iget5_locked() for callers that don't want to fail on memory
1064 * allocation of inode.
1065 *
1066 * If the inode is not in cache, insert the pre-allocated inode to cache and
1067 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1068 * to fill it in before unlocking it via unlock_new_inode().
1069 *
1070 * Note both @test and @set are called with the inode_hash_lock held, so can't
1071 * sleep.
1072 */
inode_insert5(struct inode * inode,unsigned long hashval,int (* test)(struct inode *,void *),int (* set)(struct inode *,void *),void * data)1073 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1074 int (*test)(struct inode *, void *),
1075 int (*set)(struct inode *, void *), void *data)
1076 {
1077 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1078 struct inode *old;
1079 bool creating = inode->i_state & I_CREATING;
1080
1081 again:
1082 spin_lock(&inode_hash_lock);
1083 old = find_inode(inode->i_sb, head, test, data);
1084 if (unlikely(old)) {
1085 /*
1086 * Uhhuh, somebody else created the same inode under us.
1087 * Use the old inode instead of the preallocated one.
1088 */
1089 spin_unlock(&inode_hash_lock);
1090 if (IS_ERR(old))
1091 return NULL;
1092 wait_on_inode(old);
1093 if (unlikely(inode_unhashed(old))) {
1094 iput(old);
1095 goto again;
1096 }
1097 return old;
1098 }
1099
1100 if (set && unlikely(set(inode, data))) {
1101 inode = NULL;
1102 goto unlock;
1103 }
1104
1105 /*
1106 * Return the locked inode with I_NEW set, the
1107 * caller is responsible for filling in the contents
1108 */
1109 spin_lock(&inode->i_lock);
1110 inode->i_state |= I_NEW;
1111 hlist_add_head_rcu(&inode->i_hash, head);
1112 spin_unlock(&inode->i_lock);
1113 if (!creating)
1114 inode_sb_list_add(inode);
1115 unlock:
1116 spin_unlock(&inode_hash_lock);
1117
1118 return inode;
1119 }
1120 EXPORT_SYMBOL(inode_insert5);
1121
1122 /**
1123 * iget5_locked - obtain an inode from a mounted file system
1124 * @sb: super block of file system
1125 * @hashval: hash value (usually inode number) to get
1126 * @test: callback used for comparisons between inodes
1127 * @set: callback used to initialize a new struct inode
1128 * @data: opaque data pointer to pass to @test and @set
1129 *
1130 * Search for the inode specified by @hashval and @data in the inode cache,
1131 * and if present it is return it with an increased reference count. This is
1132 * a generalized version of iget_locked() for file systems where the inode
1133 * number is not sufficient for unique identification of an inode.
1134 *
1135 * If the inode is not in cache, allocate a new inode and return it locked,
1136 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1137 * before unlocking it via unlock_new_inode().
1138 *
1139 * Note both @test and @set are called with the inode_hash_lock held, so can't
1140 * sleep.
1141 */
iget5_locked(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),int (* set)(struct inode *,void *),void * data)1142 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1143 int (*test)(struct inode *, void *),
1144 int (*set)(struct inode *, void *), void *data)
1145 {
1146 struct inode *inode = ilookup5(sb, hashval, test, data);
1147
1148 if (!inode) {
1149 struct inode *new = alloc_inode(sb);
1150
1151 if (new) {
1152 new->i_state = 0;
1153 inode = inode_insert5(new, hashval, test, set, data);
1154 if (unlikely(inode != new))
1155 destroy_inode(new);
1156 }
1157 }
1158 return inode;
1159 }
1160 EXPORT_SYMBOL(iget5_locked);
1161
1162 /**
1163 * iget_locked - obtain an inode from a mounted file system
1164 * @sb: super block of file system
1165 * @ino: inode number to get
1166 *
1167 * Search for the inode specified by @ino in the inode cache and if present
1168 * return it with an increased reference count. This is for file systems
1169 * where the inode number is sufficient for unique identification of an inode.
1170 *
1171 * If the inode is not in cache, allocate a new inode and return it locked,
1172 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1173 * before unlocking it via unlock_new_inode().
1174 */
iget_locked(struct super_block * sb,unsigned long ino)1175 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1176 {
1177 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1178 struct inode *inode;
1179 again:
1180 spin_lock(&inode_hash_lock);
1181 inode = find_inode_fast(sb, head, ino);
1182 spin_unlock(&inode_hash_lock);
1183 if (inode) {
1184 if (IS_ERR(inode))
1185 return NULL;
1186 wait_on_inode(inode);
1187 if (unlikely(inode_unhashed(inode))) {
1188 iput(inode);
1189 goto again;
1190 }
1191 return inode;
1192 }
1193
1194 inode = alloc_inode(sb);
1195 if (inode) {
1196 struct inode *old;
1197
1198 spin_lock(&inode_hash_lock);
1199 /* We released the lock, so.. */
1200 old = find_inode_fast(sb, head, ino);
1201 if (!old) {
1202 inode->i_ino = ino;
1203 spin_lock(&inode->i_lock);
1204 inode->i_state = I_NEW;
1205 hlist_add_head_rcu(&inode->i_hash, head);
1206 spin_unlock(&inode->i_lock);
1207 inode_sb_list_add(inode);
1208 spin_unlock(&inode_hash_lock);
1209
1210 /* Return the locked inode with I_NEW set, the
1211 * caller is responsible for filling in the contents
1212 */
1213 return inode;
1214 }
1215
1216 /*
1217 * Uhhuh, somebody else created the same inode under
1218 * us. Use the old inode instead of the one we just
1219 * allocated.
1220 */
1221 spin_unlock(&inode_hash_lock);
1222 destroy_inode(inode);
1223 if (IS_ERR(old))
1224 return NULL;
1225 inode = old;
1226 wait_on_inode(inode);
1227 if (unlikely(inode_unhashed(inode))) {
1228 iput(inode);
1229 goto again;
1230 }
1231 }
1232 return inode;
1233 }
1234 EXPORT_SYMBOL(iget_locked);
1235
1236 /*
1237 * search the inode cache for a matching inode number.
1238 * If we find one, then the inode number we are trying to
1239 * allocate is not unique and so we should not use it.
1240 *
1241 * Returns 1 if the inode number is unique, 0 if it is not.
1242 */
test_inode_iunique(struct super_block * sb,unsigned long ino)1243 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1244 {
1245 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1246 struct inode *inode;
1247
1248 hlist_for_each_entry_rcu(inode, b, i_hash) {
1249 if (inode->i_ino == ino && inode->i_sb == sb)
1250 return 0;
1251 }
1252 return 1;
1253 }
1254
1255 /**
1256 * iunique - get a unique inode number
1257 * @sb: superblock
1258 * @max_reserved: highest reserved inode number
1259 *
1260 * Obtain an inode number that is unique on the system for a given
1261 * superblock. This is used by file systems that have no natural
1262 * permanent inode numbering system. An inode number is returned that
1263 * is higher than the reserved limit but unique.
1264 *
1265 * BUGS:
1266 * With a large number of inodes live on the file system this function
1267 * currently becomes quite slow.
1268 */
iunique(struct super_block * sb,ino_t max_reserved)1269 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1270 {
1271 /*
1272 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1273 * error if st_ino won't fit in target struct field. Use 32bit counter
1274 * here to attempt to avoid that.
1275 */
1276 static DEFINE_SPINLOCK(iunique_lock);
1277 static unsigned int counter;
1278 ino_t res;
1279
1280 rcu_read_lock();
1281 spin_lock(&iunique_lock);
1282 do {
1283 if (counter <= max_reserved)
1284 counter = max_reserved + 1;
1285 res = counter++;
1286 } while (!test_inode_iunique(sb, res));
1287 spin_unlock(&iunique_lock);
1288 rcu_read_unlock();
1289
1290 return res;
1291 }
1292 EXPORT_SYMBOL(iunique);
1293
igrab(struct inode * inode)1294 struct inode *igrab(struct inode *inode)
1295 {
1296 spin_lock(&inode->i_lock);
1297 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1298 __iget(inode);
1299 spin_unlock(&inode->i_lock);
1300 } else {
1301 spin_unlock(&inode->i_lock);
1302 /*
1303 * Handle the case where s_op->clear_inode is not been
1304 * called yet, and somebody is calling igrab
1305 * while the inode is getting freed.
1306 */
1307 inode = NULL;
1308 }
1309 return inode;
1310 }
1311 EXPORT_SYMBOL(igrab);
1312
1313 /**
1314 * ilookup5_nowait - search for an inode in the inode cache
1315 * @sb: super block of file system to search
1316 * @hashval: hash value (usually inode number) to search for
1317 * @test: callback used for comparisons between inodes
1318 * @data: opaque data pointer to pass to @test
1319 *
1320 * Search for the inode specified by @hashval and @data in the inode cache.
1321 * If the inode is in the cache, the inode is returned with an incremented
1322 * reference count.
1323 *
1324 * Note: I_NEW is not waited upon so you have to be very careful what you do
1325 * with the returned inode. You probably should be using ilookup5() instead.
1326 *
1327 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1328 */
ilookup5_nowait(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1329 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1330 int (*test)(struct inode *, void *), void *data)
1331 {
1332 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1333 struct inode *inode;
1334
1335 spin_lock(&inode_hash_lock);
1336 inode = find_inode(sb, head, test, data);
1337 spin_unlock(&inode_hash_lock);
1338
1339 return IS_ERR(inode) ? NULL : inode;
1340 }
1341 EXPORT_SYMBOL(ilookup5_nowait);
1342
1343 /**
1344 * ilookup5 - search for an inode in the inode cache
1345 * @sb: super block of file system to search
1346 * @hashval: hash value (usually inode number) to search for
1347 * @test: callback used for comparisons between inodes
1348 * @data: opaque data pointer to pass to @test
1349 *
1350 * Search for the inode specified by @hashval and @data in the inode cache,
1351 * and if the inode is in the cache, return the inode with an incremented
1352 * reference count. Waits on I_NEW before returning the inode.
1353 * returned with an incremented reference count.
1354 *
1355 * This is a generalized version of ilookup() for file systems where the
1356 * inode number is not sufficient for unique identification of an inode.
1357 *
1358 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1359 */
ilookup5(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1360 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1361 int (*test)(struct inode *, void *), void *data)
1362 {
1363 struct inode *inode;
1364 again:
1365 inode = ilookup5_nowait(sb, hashval, test, data);
1366 if (inode) {
1367 wait_on_inode(inode);
1368 if (unlikely(inode_unhashed(inode))) {
1369 iput(inode);
1370 goto again;
1371 }
1372 }
1373 return inode;
1374 }
1375 EXPORT_SYMBOL(ilookup5);
1376
1377 /**
1378 * ilookup - search for an inode in the inode cache
1379 * @sb: super block of file system to search
1380 * @ino: inode number to search for
1381 *
1382 * Search for the inode @ino in the inode cache, and if the inode is in the
1383 * cache, the inode is returned with an incremented reference count.
1384 */
ilookup(struct super_block * sb,unsigned long ino)1385 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1386 {
1387 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1388 struct inode *inode;
1389 again:
1390 spin_lock(&inode_hash_lock);
1391 inode = find_inode_fast(sb, head, ino);
1392 spin_unlock(&inode_hash_lock);
1393
1394 if (inode) {
1395 if (IS_ERR(inode))
1396 return NULL;
1397 wait_on_inode(inode);
1398 if (unlikely(inode_unhashed(inode))) {
1399 iput(inode);
1400 goto again;
1401 }
1402 }
1403 return inode;
1404 }
1405 EXPORT_SYMBOL(ilookup);
1406
1407 /**
1408 * find_inode_nowait - find an inode in the inode cache
1409 * @sb: super block of file system to search
1410 * @hashval: hash value (usually inode number) to search for
1411 * @match: callback used for comparisons between inodes
1412 * @data: opaque data pointer to pass to @match
1413 *
1414 * Search for the inode specified by @hashval and @data in the inode
1415 * cache, where the helper function @match will return 0 if the inode
1416 * does not match, 1 if the inode does match, and -1 if the search
1417 * should be stopped. The @match function must be responsible for
1418 * taking the i_lock spin_lock and checking i_state for an inode being
1419 * freed or being initialized, and incrementing the reference count
1420 * before returning 1. It also must not sleep, since it is called with
1421 * the inode_hash_lock spinlock held.
1422 *
1423 * This is a even more generalized version of ilookup5() when the
1424 * function must never block --- find_inode() can block in
1425 * __wait_on_freeing_inode() --- or when the caller can not increment
1426 * the reference count because the resulting iput() might cause an
1427 * inode eviction. The tradeoff is that the @match funtion must be
1428 * very carefully implemented.
1429 */
find_inode_nowait(struct super_block * sb,unsigned long hashval,int (* match)(struct inode *,unsigned long,void *),void * data)1430 struct inode *find_inode_nowait(struct super_block *sb,
1431 unsigned long hashval,
1432 int (*match)(struct inode *, unsigned long,
1433 void *),
1434 void *data)
1435 {
1436 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1437 struct inode *inode, *ret_inode = NULL;
1438 int mval;
1439
1440 spin_lock(&inode_hash_lock);
1441 hlist_for_each_entry(inode, head, i_hash) {
1442 if (inode->i_sb != sb)
1443 continue;
1444 mval = match(inode, hashval, data);
1445 if (mval == 0)
1446 continue;
1447 if (mval == 1)
1448 ret_inode = inode;
1449 goto out;
1450 }
1451 out:
1452 spin_unlock(&inode_hash_lock);
1453 return ret_inode;
1454 }
1455 EXPORT_SYMBOL(find_inode_nowait);
1456
1457 /**
1458 * find_inode_rcu - find an inode in the inode cache
1459 * @sb: Super block of file system to search
1460 * @hashval: Key to hash
1461 * @test: Function to test match on an inode
1462 * @data: Data for test function
1463 *
1464 * Search for the inode specified by @hashval and @data in the inode cache,
1465 * where the helper function @test will return 0 if the inode does not match
1466 * and 1 if it does. The @test function must be responsible for taking the
1467 * i_lock spin_lock and checking i_state for an inode being freed or being
1468 * initialized.
1469 *
1470 * If successful, this will return the inode for which the @test function
1471 * returned 1 and NULL otherwise.
1472 *
1473 * The @test function is not permitted to take a ref on any inode presented.
1474 * It is also not permitted to sleep.
1475 *
1476 * The caller must hold the RCU read lock.
1477 */
find_inode_rcu(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1478 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1479 int (*test)(struct inode *, void *), void *data)
1480 {
1481 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1482 struct inode *inode;
1483
1484 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1485 "suspicious find_inode_rcu() usage");
1486
1487 hlist_for_each_entry_rcu(inode, head, i_hash) {
1488 if (inode->i_sb == sb &&
1489 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1490 test(inode, data))
1491 return inode;
1492 }
1493 return NULL;
1494 }
1495 EXPORT_SYMBOL(find_inode_rcu);
1496
1497 /**
1498 * find_inode_by_rcu - Find an inode in the inode cache
1499 * @sb: Super block of file system to search
1500 * @ino: The inode number to match
1501 *
1502 * Search for the inode specified by @hashval and @data in the inode cache,
1503 * where the helper function @test will return 0 if the inode does not match
1504 * and 1 if it does. The @test function must be responsible for taking the
1505 * i_lock spin_lock and checking i_state for an inode being freed or being
1506 * initialized.
1507 *
1508 * If successful, this will return the inode for which the @test function
1509 * returned 1 and NULL otherwise.
1510 *
1511 * The @test function is not permitted to take a ref on any inode presented.
1512 * It is also not permitted to sleep.
1513 *
1514 * The caller must hold the RCU read lock.
1515 */
find_inode_by_ino_rcu(struct super_block * sb,unsigned long ino)1516 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1517 unsigned long ino)
1518 {
1519 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1520 struct inode *inode;
1521
1522 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1523 "suspicious find_inode_by_ino_rcu() usage");
1524
1525 hlist_for_each_entry_rcu(inode, head, i_hash) {
1526 if (inode->i_ino == ino &&
1527 inode->i_sb == sb &&
1528 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1529 return inode;
1530 }
1531 return NULL;
1532 }
1533 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1534
insert_inode_locked(struct inode * inode)1535 int insert_inode_locked(struct inode *inode)
1536 {
1537 struct super_block *sb = inode->i_sb;
1538 ino_t ino = inode->i_ino;
1539 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1540
1541 while (1) {
1542 struct inode *old = NULL;
1543 spin_lock(&inode_hash_lock);
1544 hlist_for_each_entry(old, head, i_hash) {
1545 if (old->i_ino != ino)
1546 continue;
1547 if (old->i_sb != sb)
1548 continue;
1549 spin_lock(&old->i_lock);
1550 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1551 spin_unlock(&old->i_lock);
1552 continue;
1553 }
1554 break;
1555 }
1556 if (likely(!old)) {
1557 spin_lock(&inode->i_lock);
1558 inode->i_state |= I_NEW | I_CREATING;
1559 hlist_add_head_rcu(&inode->i_hash, head);
1560 spin_unlock(&inode->i_lock);
1561 spin_unlock(&inode_hash_lock);
1562 return 0;
1563 }
1564 if (unlikely(old->i_state & I_CREATING)) {
1565 spin_unlock(&old->i_lock);
1566 spin_unlock(&inode_hash_lock);
1567 return -EBUSY;
1568 }
1569 __iget(old);
1570 spin_unlock(&old->i_lock);
1571 spin_unlock(&inode_hash_lock);
1572 wait_on_inode(old);
1573 if (unlikely(!inode_unhashed(old))) {
1574 iput(old);
1575 return -EBUSY;
1576 }
1577 iput(old);
1578 }
1579 }
1580 EXPORT_SYMBOL(insert_inode_locked);
1581
insert_inode_locked4(struct inode * inode,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1582 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1583 int (*test)(struct inode *, void *), void *data)
1584 {
1585 struct inode *old;
1586
1587 inode->i_state |= I_CREATING;
1588 old = inode_insert5(inode, hashval, test, NULL, data);
1589
1590 if (old != inode) {
1591 iput(old);
1592 return -EBUSY;
1593 }
1594 return 0;
1595 }
1596 EXPORT_SYMBOL(insert_inode_locked4);
1597
1598
generic_delete_inode(struct inode * inode)1599 int generic_delete_inode(struct inode *inode)
1600 {
1601 return 1;
1602 }
1603 EXPORT_SYMBOL(generic_delete_inode);
1604
1605 /*
1606 * Called when we're dropping the last reference
1607 * to an inode.
1608 *
1609 * Call the FS "drop_inode()" function, defaulting to
1610 * the legacy UNIX filesystem behaviour. If it tells
1611 * us to evict inode, do so. Otherwise, retain inode
1612 * in cache if fs is alive, sync and evict if fs is
1613 * shutting down.
1614 */
iput_final(struct inode * inode)1615 static void iput_final(struct inode *inode)
1616 {
1617 struct super_block *sb = inode->i_sb;
1618 const struct super_operations *op = inode->i_sb->s_op;
1619 unsigned long state;
1620 int drop;
1621
1622 WARN_ON(inode->i_state & I_NEW);
1623
1624 if (op->drop_inode)
1625 drop = op->drop_inode(inode);
1626 else
1627 drop = generic_drop_inode(inode);
1628
1629 if (!drop &&
1630 !(inode->i_state & I_DONTCACHE) &&
1631 (sb->s_flags & SB_ACTIVE)) {
1632 inode_add_lru(inode);
1633 spin_unlock(&inode->i_lock);
1634 return;
1635 }
1636
1637 state = inode->i_state;
1638 if (!drop) {
1639 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1640 spin_unlock(&inode->i_lock);
1641
1642 write_inode_now(inode, 1);
1643
1644 spin_lock(&inode->i_lock);
1645 state = inode->i_state;
1646 WARN_ON(state & I_NEW);
1647 state &= ~I_WILL_FREE;
1648 }
1649
1650 WRITE_ONCE(inode->i_state, state | I_FREEING);
1651 if (!list_empty(&inode->i_lru))
1652 inode_lru_list_del(inode);
1653 spin_unlock(&inode->i_lock);
1654
1655 evict(inode);
1656 }
1657
1658 /**
1659 * iput - put an inode
1660 * @inode: inode to put
1661 *
1662 * Puts an inode, dropping its usage count. If the inode use count hits
1663 * zero, the inode is then freed and may also be destroyed.
1664 *
1665 * Consequently, iput() can sleep.
1666 */
iput(struct inode * inode)1667 void iput(struct inode *inode)
1668 {
1669 if (!inode)
1670 return;
1671 BUG_ON(inode->i_state & I_CLEAR);
1672 retry:
1673 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1674 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1675 atomic_inc(&inode->i_count);
1676 spin_unlock(&inode->i_lock);
1677 trace_writeback_lazytime_iput(inode);
1678 mark_inode_dirty_sync(inode);
1679 goto retry;
1680 }
1681 iput_final(inode);
1682 }
1683 }
1684 EXPORT_SYMBOL(iput);
1685
1686 #ifdef CONFIG_BLOCK
1687 /**
1688 * bmap - find a block number in a file
1689 * @inode: inode owning the block number being requested
1690 * @block: pointer containing the block to find
1691 *
1692 * Replaces the value in ``*block`` with the block number on the device holding
1693 * corresponding to the requested block number in the file.
1694 * That is, asked for block 4 of inode 1 the function will replace the
1695 * 4 in ``*block``, with disk block relative to the disk start that holds that
1696 * block of the file.
1697 *
1698 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1699 * hole, returns 0 and ``*block`` is also set to 0.
1700 */
bmap(struct inode * inode,sector_t * block)1701 int bmap(struct inode *inode, sector_t *block)
1702 {
1703 if (!inode->i_mapping->a_ops->bmap)
1704 return -EINVAL;
1705
1706 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1707 return 0;
1708 }
1709 EXPORT_SYMBOL(bmap);
1710 #endif
1711
1712 /*
1713 * With relative atime, only update atime if the previous atime is
1714 * earlier than either the ctime or mtime or if at least a day has
1715 * passed since the last atime update.
1716 */
relatime_need_update(struct vfsmount * mnt,struct inode * inode,struct timespec64 now)1717 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1718 struct timespec64 now)
1719 {
1720
1721 if (!(mnt->mnt_flags & MNT_RELATIME))
1722 return 1;
1723 /*
1724 * Is mtime younger than atime? If yes, update atime:
1725 */
1726 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1727 return 1;
1728 /*
1729 * Is ctime younger than atime? If yes, update atime:
1730 */
1731 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1732 return 1;
1733
1734 /*
1735 * Is the previous atime value older than a day? If yes,
1736 * update atime:
1737 */
1738 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1739 return 1;
1740 /*
1741 * Good, we can skip the atime update:
1742 */
1743 return 0;
1744 }
1745
generic_update_time(struct inode * inode,struct timespec64 * time,int flags)1746 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1747 {
1748 int iflags = I_DIRTY_TIME;
1749 bool dirty = false;
1750
1751 if (flags & S_ATIME)
1752 inode->i_atime = *time;
1753 if (flags & S_VERSION)
1754 dirty = inode_maybe_inc_iversion(inode, false);
1755 if (flags & S_CTIME)
1756 inode->i_ctime = *time;
1757 if (flags & S_MTIME)
1758 inode->i_mtime = *time;
1759 if ((flags & (S_ATIME | S_CTIME | S_MTIME)) &&
1760 !(inode->i_sb->s_flags & SB_LAZYTIME))
1761 dirty = true;
1762
1763 if (dirty)
1764 iflags |= I_DIRTY_SYNC;
1765 __mark_inode_dirty(inode, iflags);
1766 return 0;
1767 }
1768 EXPORT_SYMBOL(generic_update_time);
1769
1770 /*
1771 * This does the actual work of updating an inodes time or version. Must have
1772 * had called mnt_want_write() before calling this.
1773 */
inode_update_time(struct inode * inode,struct timespec64 * time,int flags)1774 int inode_update_time(struct inode *inode, struct timespec64 *time, int flags)
1775 {
1776 if (inode->i_op->update_time)
1777 return inode->i_op->update_time(inode, time, flags);
1778 return generic_update_time(inode, time, flags);
1779 }
1780 EXPORT_SYMBOL(inode_update_time);
1781
1782 /**
1783 * touch_atime - update the access time
1784 * @path: the &struct path to update
1785 * @inode: inode to update
1786 *
1787 * Update the accessed time on an inode and mark it for writeback.
1788 * This function automatically handles read only file systems and media,
1789 * as well as the "noatime" flag and inode specific "noatime" markers.
1790 */
atime_needs_update(const struct path * path,struct inode * inode)1791 bool atime_needs_update(const struct path *path, struct inode *inode)
1792 {
1793 struct vfsmount *mnt = path->mnt;
1794 struct timespec64 now;
1795
1796 if (inode->i_flags & S_NOATIME)
1797 return false;
1798
1799 /* Atime updates will likely cause i_uid and i_gid to be written
1800 * back improprely if their true value is unknown to the vfs.
1801 */
1802 if (HAS_UNMAPPED_ID(inode))
1803 return false;
1804
1805 if (IS_NOATIME(inode))
1806 return false;
1807 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1808 return false;
1809
1810 if (mnt->mnt_flags & MNT_NOATIME)
1811 return false;
1812 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1813 return false;
1814
1815 now = current_time(inode);
1816
1817 if (!relatime_need_update(mnt, inode, now))
1818 return false;
1819
1820 if (timespec64_equal(&inode->i_atime, &now))
1821 return false;
1822
1823 return true;
1824 }
1825
touch_atime(const struct path * path)1826 void touch_atime(const struct path *path)
1827 {
1828 struct vfsmount *mnt = path->mnt;
1829 struct inode *inode = d_inode(path->dentry);
1830 struct timespec64 now;
1831
1832 if (!atime_needs_update(path, inode))
1833 return;
1834
1835 if (!sb_start_write_trylock(inode->i_sb))
1836 return;
1837
1838 if (__mnt_want_write(mnt) != 0)
1839 goto skip_update;
1840 /*
1841 * File systems can error out when updating inodes if they need to
1842 * allocate new space to modify an inode (such is the case for
1843 * Btrfs), but since we touch atime while walking down the path we
1844 * really don't care if we failed to update the atime of the file,
1845 * so just ignore the return value.
1846 * We may also fail on filesystems that have the ability to make parts
1847 * of the fs read only, e.g. subvolumes in Btrfs.
1848 */
1849 now = current_time(inode);
1850 inode_update_time(inode, &now, S_ATIME);
1851 __mnt_drop_write(mnt);
1852 skip_update:
1853 sb_end_write(inode->i_sb);
1854 }
1855 EXPORT_SYMBOL(touch_atime);
1856
1857 /*
1858 * The logic we want is
1859 *
1860 * if suid or (sgid and xgrp)
1861 * remove privs
1862 */
should_remove_suid(struct dentry * dentry)1863 int should_remove_suid(struct dentry *dentry)
1864 {
1865 umode_t mode = d_inode(dentry)->i_mode;
1866 int kill = 0;
1867
1868 /* suid always must be killed */
1869 if (unlikely(mode & S_ISUID))
1870 kill = ATTR_KILL_SUID;
1871
1872 /*
1873 * sgid without any exec bits is just a mandatory locking mark; leave
1874 * it alone. If some exec bits are set, it's a real sgid; kill it.
1875 */
1876 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1877 kill |= ATTR_KILL_SGID;
1878
1879 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1880 return kill;
1881
1882 return 0;
1883 }
1884 EXPORT_SYMBOL(should_remove_suid);
1885
1886 /*
1887 * Return mask of changes for notify_change() that need to be done as a
1888 * response to write or truncate. Return 0 if nothing has to be changed.
1889 * Negative value on error (change should be denied).
1890 */
dentry_needs_remove_privs(struct dentry * dentry)1891 int dentry_needs_remove_privs(struct dentry *dentry)
1892 {
1893 struct inode *inode = d_inode(dentry);
1894 int mask = 0;
1895 int ret;
1896
1897 if (IS_NOSEC(inode))
1898 return 0;
1899
1900 mask = should_remove_suid(dentry);
1901 ret = security_inode_need_killpriv(dentry);
1902 if (ret < 0)
1903 return ret;
1904 if (ret)
1905 mask |= ATTR_KILL_PRIV;
1906 return mask;
1907 }
1908
__remove_privs(struct dentry * dentry,int kill)1909 static int __remove_privs(struct dentry *dentry, int kill)
1910 {
1911 struct iattr newattrs;
1912
1913 newattrs.ia_valid = ATTR_FORCE | kill;
1914 /*
1915 * Note we call this on write, so notify_change will not
1916 * encounter any conflicting delegations:
1917 */
1918 return notify_change(dentry, &newattrs, NULL);
1919 }
1920
1921 /*
1922 * Remove special file priviledges (suid, capabilities) when file is written
1923 * to or truncated.
1924 */
file_remove_privs(struct file * file)1925 int file_remove_privs(struct file *file)
1926 {
1927 struct dentry *dentry = file_dentry(file);
1928 struct inode *inode = file_inode(file);
1929 int kill;
1930 int error = 0;
1931
1932 /*
1933 * Fast path for nothing security related.
1934 * As well for non-regular files, e.g. blkdev inodes.
1935 * For example, blkdev_write_iter() might get here
1936 * trying to remove privs which it is not allowed to.
1937 */
1938 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1939 return 0;
1940
1941 kill = dentry_needs_remove_privs(dentry);
1942 if (kill < 0)
1943 return kill;
1944 if (kill)
1945 error = __remove_privs(dentry, kill);
1946 if (!error)
1947 inode_has_no_xattr(inode);
1948
1949 return error;
1950 }
1951 EXPORT_SYMBOL(file_remove_privs);
1952
1953 /**
1954 * file_update_time - update mtime and ctime time
1955 * @file: file accessed
1956 *
1957 * Update the mtime and ctime members of an inode and mark the inode
1958 * for writeback. Note that this function is meant exclusively for
1959 * usage in the file write path of filesystems, and filesystems may
1960 * choose to explicitly ignore update via this function with the
1961 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1962 * timestamps are handled by the server. This can return an error for
1963 * file systems who need to allocate space in order to update an inode.
1964 */
1965
file_update_time(struct file * file)1966 int file_update_time(struct file *file)
1967 {
1968 struct inode *inode = file_inode(file);
1969 struct timespec64 now;
1970 int sync_it = 0;
1971 int ret;
1972
1973 /* First try to exhaust all avenues to not sync */
1974 if (IS_NOCMTIME(inode))
1975 return 0;
1976
1977 now = current_time(inode);
1978 if (!timespec64_equal(&inode->i_mtime, &now))
1979 sync_it = S_MTIME;
1980
1981 if (!timespec64_equal(&inode->i_ctime, &now))
1982 sync_it |= S_CTIME;
1983
1984 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1985 sync_it |= S_VERSION;
1986
1987 if (!sync_it)
1988 return 0;
1989
1990 /* Finally allowed to write? Takes lock. */
1991 if (__mnt_want_write_file(file))
1992 return 0;
1993
1994 ret = inode_update_time(inode, &now, sync_it);
1995 __mnt_drop_write_file(file);
1996
1997 return ret;
1998 }
1999 EXPORT_SYMBOL(file_update_time);
2000
2001 /* Caller must hold the file's inode lock */
file_modified(struct file * file)2002 int file_modified(struct file *file)
2003 {
2004 int err;
2005
2006 /*
2007 * Clear the security bits if the process is not being run by root.
2008 * This keeps people from modifying setuid and setgid binaries.
2009 */
2010 err = file_remove_privs(file);
2011 if (err)
2012 return err;
2013
2014 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2015 return 0;
2016
2017 return file_update_time(file);
2018 }
2019 EXPORT_SYMBOL(file_modified);
2020
inode_needs_sync(struct inode * inode)2021 int inode_needs_sync(struct inode *inode)
2022 {
2023 if (IS_SYNC(inode))
2024 return 1;
2025 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2026 return 1;
2027 return 0;
2028 }
2029 EXPORT_SYMBOL(inode_needs_sync);
2030
2031 /*
2032 * If we try to find an inode in the inode hash while it is being
2033 * deleted, we have to wait until the filesystem completes its
2034 * deletion before reporting that it isn't found. This function waits
2035 * until the deletion _might_ have completed. Callers are responsible
2036 * to recheck inode state.
2037 *
2038 * It doesn't matter if I_NEW is not set initially, a call to
2039 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2040 * will DTRT.
2041 */
__wait_on_freeing_inode(struct inode * inode)2042 static void __wait_on_freeing_inode(struct inode *inode)
2043 {
2044 wait_queue_head_t *wq;
2045 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2046 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2047 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2048 spin_unlock(&inode->i_lock);
2049 spin_unlock(&inode_hash_lock);
2050 schedule();
2051 finish_wait(wq, &wait.wq_entry);
2052 spin_lock(&inode_hash_lock);
2053 }
2054
2055 static __initdata unsigned long ihash_entries;
set_ihash_entries(char * str)2056 static int __init set_ihash_entries(char *str)
2057 {
2058 if (!str)
2059 return 0;
2060 ihash_entries = simple_strtoul(str, &str, 0);
2061 return 1;
2062 }
2063 __setup("ihash_entries=", set_ihash_entries);
2064
2065 /*
2066 * Initialize the waitqueues and inode hash table.
2067 */
inode_init_early(void)2068 void __init inode_init_early(void)
2069 {
2070 /* If hashes are distributed across NUMA nodes, defer
2071 * hash allocation until vmalloc space is available.
2072 */
2073 if (hashdist)
2074 return;
2075
2076 inode_hashtable =
2077 alloc_large_system_hash("Inode-cache",
2078 sizeof(struct hlist_head),
2079 ihash_entries,
2080 14,
2081 HASH_EARLY | HASH_ZERO,
2082 &i_hash_shift,
2083 &i_hash_mask,
2084 0,
2085 0);
2086 }
2087
inode_init(void)2088 void __init inode_init(void)
2089 {
2090 /* inode slab cache */
2091 inode_cachep = kmem_cache_create("inode_cache",
2092 sizeof(struct inode),
2093 0,
2094 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2095 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2096 init_once);
2097
2098 /* Hash may have been set up in inode_init_early */
2099 if (!hashdist)
2100 return;
2101
2102 inode_hashtable =
2103 alloc_large_system_hash("Inode-cache",
2104 sizeof(struct hlist_head),
2105 ihash_entries,
2106 14,
2107 HASH_ZERO,
2108 &i_hash_shift,
2109 &i_hash_mask,
2110 0,
2111 0);
2112 }
2113
init_special_inode(struct inode * inode,umode_t mode,dev_t rdev)2114 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2115 {
2116 inode->i_mode = mode;
2117 if (S_ISCHR(mode)) {
2118 inode->i_fop = &def_chr_fops;
2119 inode->i_rdev = rdev;
2120 } else if (S_ISBLK(mode)) {
2121 inode->i_fop = &def_blk_fops;
2122 inode->i_rdev = rdev;
2123 } else if (S_ISFIFO(mode))
2124 inode->i_fop = &pipefifo_fops;
2125 else if (S_ISSOCK(mode))
2126 ; /* leave it no_open_fops */
2127 else
2128 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2129 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2130 inode->i_ino);
2131 }
2132 EXPORT_SYMBOL(init_special_inode);
2133
2134 /**
2135 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2136 * @inode: New inode
2137 * @dir: Directory inode
2138 * @mode: mode of the new inode
2139 */
inode_init_owner(struct inode * inode,const struct inode * dir,umode_t mode)2140 void inode_init_owner(struct inode *inode, const struct inode *dir,
2141 umode_t mode)
2142 {
2143 inode->i_uid = current_fsuid();
2144 if (dir && dir->i_mode & S_ISGID) {
2145 inode->i_gid = dir->i_gid;
2146
2147 /* Directories are special, and always inherit S_ISGID */
2148 if (S_ISDIR(mode))
2149 mode |= S_ISGID;
2150 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2151 !in_group_p(inode->i_gid) &&
2152 !capable_wrt_inode_uidgid(dir, CAP_FSETID))
2153 mode &= ~S_ISGID;
2154 } else
2155 inode->i_gid = current_fsgid();
2156 inode->i_mode = mode;
2157 }
2158 EXPORT_SYMBOL(inode_init_owner);
2159
2160 /**
2161 * inode_owner_or_capable - check current task permissions to inode
2162 * @inode: inode being checked
2163 *
2164 * Return true if current either has CAP_FOWNER in a namespace with the
2165 * inode owner uid mapped, or owns the file.
2166 */
inode_owner_or_capable(const struct inode * inode)2167 bool inode_owner_or_capable(const struct inode *inode)
2168 {
2169 struct user_namespace *ns;
2170
2171 if (uid_eq(current_fsuid(), inode->i_uid))
2172 return true;
2173
2174 ns = current_user_ns();
2175 if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2176 return true;
2177 return false;
2178 }
2179 EXPORT_SYMBOL(inode_owner_or_capable);
2180
2181 /*
2182 * Direct i/o helper functions
2183 */
__inode_dio_wait(struct inode * inode)2184 static void __inode_dio_wait(struct inode *inode)
2185 {
2186 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2187 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2188
2189 do {
2190 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2191 if (atomic_read(&inode->i_dio_count))
2192 schedule();
2193 } while (atomic_read(&inode->i_dio_count));
2194 finish_wait(wq, &q.wq_entry);
2195 }
2196
2197 /**
2198 * inode_dio_wait - wait for outstanding DIO requests to finish
2199 * @inode: inode to wait for
2200 *
2201 * Waits for all pending direct I/O requests to finish so that we can
2202 * proceed with a truncate or equivalent operation.
2203 *
2204 * Must be called under a lock that serializes taking new references
2205 * to i_dio_count, usually by inode->i_mutex.
2206 */
inode_dio_wait(struct inode * inode)2207 void inode_dio_wait(struct inode *inode)
2208 {
2209 if (atomic_read(&inode->i_dio_count))
2210 __inode_dio_wait(inode);
2211 }
2212 EXPORT_SYMBOL(inode_dio_wait);
2213
2214 /*
2215 * inode_set_flags - atomically set some inode flags
2216 *
2217 * Note: the caller should be holding i_mutex, or else be sure that
2218 * they have exclusive access to the inode structure (i.e., while the
2219 * inode is being instantiated). The reason for the cmpxchg() loop
2220 * --- which wouldn't be necessary if all code paths which modify
2221 * i_flags actually followed this rule, is that there is at least one
2222 * code path which doesn't today so we use cmpxchg() out of an abundance
2223 * of caution.
2224 *
2225 * In the long run, i_mutex is overkill, and we should probably look
2226 * at using the i_lock spinlock to protect i_flags, and then make sure
2227 * it is so documented in include/linux/fs.h and that all code follows
2228 * the locking convention!!
2229 */
inode_set_flags(struct inode * inode,unsigned int flags,unsigned int mask)2230 void inode_set_flags(struct inode *inode, unsigned int flags,
2231 unsigned int mask)
2232 {
2233 WARN_ON_ONCE(flags & ~mask);
2234 set_mask_bits(&inode->i_flags, mask, flags);
2235 }
2236 EXPORT_SYMBOL(inode_set_flags);
2237
inode_nohighmem(struct inode * inode)2238 void inode_nohighmem(struct inode *inode)
2239 {
2240 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2241 }
2242 EXPORT_SYMBOL(inode_nohighmem);
2243
2244 /**
2245 * timestamp_truncate - Truncate timespec to a granularity
2246 * @t: Timespec
2247 * @inode: inode being updated
2248 *
2249 * Truncate a timespec to the granularity supported by the fs
2250 * containing the inode. Always rounds down. gran must
2251 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2252 */
timestamp_truncate(struct timespec64 t,struct inode * inode)2253 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2254 {
2255 struct super_block *sb = inode->i_sb;
2256 unsigned int gran = sb->s_time_gran;
2257
2258 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2259 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2260 t.tv_nsec = 0;
2261
2262 /* Avoid division in the common cases 1 ns and 1 s. */
2263 if (gran == 1)
2264 ; /* nothing */
2265 else if (gran == NSEC_PER_SEC)
2266 t.tv_nsec = 0;
2267 else if (gran > 1 && gran < NSEC_PER_SEC)
2268 t.tv_nsec -= t.tv_nsec % gran;
2269 else
2270 WARN(1, "invalid file time granularity: %u", gran);
2271 return t;
2272 }
2273 EXPORT_SYMBOL(timestamp_truncate);
2274
2275 /**
2276 * current_time - Return FS time
2277 * @inode: inode.
2278 *
2279 * Return the current time truncated to the time granularity supported by
2280 * the fs.
2281 *
2282 * Note that inode and inode->sb cannot be NULL.
2283 * Otherwise, the function warns and returns time without truncation.
2284 */
current_time(struct inode * inode)2285 struct timespec64 current_time(struct inode *inode)
2286 {
2287 struct timespec64 now;
2288
2289 ktime_get_coarse_real_ts64(&now);
2290
2291 if (unlikely(!inode->i_sb)) {
2292 WARN(1, "current_time() called with uninitialized super_block in the inode");
2293 return now;
2294 }
2295
2296 return timestamp_truncate(now, inode);
2297 }
2298 EXPORT_SYMBOL(current_time);
2299
2300 /*
2301 * Generic function to check FS_IOC_SETFLAGS values and reject any invalid
2302 * configurations.
2303 *
2304 * Note: the caller should be holding i_mutex, or else be sure that they have
2305 * exclusive access to the inode structure.
2306 */
vfs_ioc_setflags_prepare(struct inode * inode,unsigned int oldflags,unsigned int flags)2307 int vfs_ioc_setflags_prepare(struct inode *inode, unsigned int oldflags,
2308 unsigned int flags)
2309 {
2310 /*
2311 * The IMMUTABLE and APPEND_ONLY flags can only be changed by
2312 * the relevant capability.
2313 *
2314 * This test looks nicer. Thanks to Pauline Middelink
2315 */
2316 if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL) &&
2317 !capable(CAP_LINUX_IMMUTABLE))
2318 return -EPERM;
2319
2320 return fscrypt_prepare_setflags(inode, oldflags, flags);
2321 }
2322 EXPORT_SYMBOL(vfs_ioc_setflags_prepare);
2323
2324 /*
2325 * Generic function to check FS_IOC_FSSETXATTR values and reject any invalid
2326 * configurations.
2327 *
2328 * Note: the caller should be holding i_mutex, or else be sure that they have
2329 * exclusive access to the inode structure.
2330 */
vfs_ioc_fssetxattr_check(struct inode * inode,const struct fsxattr * old_fa,struct fsxattr * fa)2331 int vfs_ioc_fssetxattr_check(struct inode *inode, const struct fsxattr *old_fa,
2332 struct fsxattr *fa)
2333 {
2334 /*
2335 * Can't modify an immutable/append-only file unless we have
2336 * appropriate permission.
2337 */
2338 if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2339 (FS_XFLAG_IMMUTABLE | FS_XFLAG_APPEND) &&
2340 !capable(CAP_LINUX_IMMUTABLE))
2341 return -EPERM;
2342
2343 /*
2344 * Project Quota ID state is only allowed to change from within the init
2345 * namespace. Enforce that restriction only if we are trying to change
2346 * the quota ID state. Everything else is allowed in user namespaces.
2347 */
2348 if (current_user_ns() != &init_user_ns) {
2349 if (old_fa->fsx_projid != fa->fsx_projid)
2350 return -EINVAL;
2351 if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2352 FS_XFLAG_PROJINHERIT)
2353 return -EINVAL;
2354 }
2355
2356 /* Check extent size hints. */
2357 if ((fa->fsx_xflags & FS_XFLAG_EXTSIZE) && !S_ISREG(inode->i_mode))
2358 return -EINVAL;
2359
2360 if ((fa->fsx_xflags & FS_XFLAG_EXTSZINHERIT) &&
2361 !S_ISDIR(inode->i_mode))
2362 return -EINVAL;
2363
2364 if ((fa->fsx_xflags & FS_XFLAG_COWEXTSIZE) &&
2365 !S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2366 return -EINVAL;
2367
2368 /*
2369 * It is only valid to set the DAX flag on regular files and
2370 * directories on filesystems.
2371 */
2372 if ((fa->fsx_xflags & FS_XFLAG_DAX) &&
2373 !(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
2374 return -EINVAL;
2375
2376 /* Extent size hints of zero turn off the flags. */
2377 if (fa->fsx_extsize == 0)
2378 fa->fsx_xflags &= ~(FS_XFLAG_EXTSIZE | FS_XFLAG_EXTSZINHERIT);
2379 if (fa->fsx_cowextsize == 0)
2380 fa->fsx_xflags &= ~FS_XFLAG_COWEXTSIZE;
2381
2382 return 0;
2383 }
2384 EXPORT_SYMBOL(vfs_ioc_fssetxattr_check);
2385