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/filelock.h>
9 #include <linux/mm.h>
10 #include <linux/backing-dev.h>
11 #include <linux/hash.h>
12 #include <linux/swap.h>
13 #include <linux/security.h>
14 #include <linux/cdev.h>
15 #include <linux/memblock.h>
16 #include <linux/fsnotify.h>
17 #include <linux/mount.h>
18 #include <linux/posix_acl.h>
19 #include <linux/buffer_head.h> /* for inode_has_buffers */
20 #include <linux/ratelimit.h>
21 #include <linux/list_lru.h>
22 #include <linux/iversion.h>
23 #include <trace/events/writeback.h>
24 #include "internal.h"
25
26 /*
27 * Inode locking rules:
28 *
29 * inode->i_lock protects:
30 * inode->i_state, inode->i_hash, __iget(), inode->i_io_list
31 * Inode LRU list locks protect:
32 * inode->i_sb->s_inode_lru, inode->i_lru
33 * inode->i_sb->s_inode_list_lock protects:
34 * inode->i_sb->s_inodes, inode->i_sb_list
35 * bdi->wb.list_lock protects:
36 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
37 * inode_hash_lock protects:
38 * inode_hashtable, inode->i_hash
39 *
40 * Lock ordering:
41 *
42 * inode->i_sb->s_inode_list_lock
43 * inode->i_lock
44 * Inode LRU list locks
45 *
46 * bdi->wb.list_lock
47 * inode->i_lock
48 *
49 * inode_hash_lock
50 * inode->i_sb->s_inode_list_lock
51 * inode->i_lock
52 *
53 * iunique_lock
54 * inode_hash_lock
55 */
56
57 static unsigned int i_hash_mask __read_mostly;
58 static unsigned int i_hash_shift __read_mostly;
59 static struct hlist_head *inode_hashtable __read_mostly;
60 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
61
62 /*
63 * Empty aops. Can be used for the cases where the user does not
64 * define any of the address_space operations.
65 */
66 const struct address_space_operations empty_aops = {
67 };
68 EXPORT_SYMBOL(empty_aops);
69
70 static DEFINE_PER_CPU(unsigned long, nr_inodes);
71 static DEFINE_PER_CPU(unsigned long, nr_unused);
72
73 static struct kmem_cache *inode_cachep __read_mostly;
74
get_nr_inodes(void)75 static long get_nr_inodes(void)
76 {
77 int i;
78 long sum = 0;
79 for_each_possible_cpu(i)
80 sum += per_cpu(nr_inodes, i);
81 return sum < 0 ? 0 : sum;
82 }
83
get_nr_inodes_unused(void)84 static inline long get_nr_inodes_unused(void)
85 {
86 int i;
87 long sum = 0;
88 for_each_possible_cpu(i)
89 sum += per_cpu(nr_unused, i);
90 return sum < 0 ? 0 : sum;
91 }
92
get_nr_dirty_inodes(void)93 long get_nr_dirty_inodes(void)
94 {
95 /* not actually dirty inodes, but a wild approximation */
96 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
97 return nr_dirty > 0 ? nr_dirty : 0;
98 }
99
100 /*
101 * Handle nr_inode sysctl
102 */
103 #ifdef CONFIG_SYSCTL
104 /*
105 * Statistics gathering..
106 */
107 static struct inodes_stat_t inodes_stat;
108
proc_nr_inodes(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)109 static int proc_nr_inodes(struct ctl_table *table, int write, void *buffer,
110 size_t *lenp, loff_t *ppos)
111 {
112 inodes_stat.nr_inodes = get_nr_inodes();
113 inodes_stat.nr_unused = get_nr_inodes_unused();
114 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
115 }
116
117 static struct ctl_table inodes_sysctls[] = {
118 {
119 .procname = "inode-nr",
120 .data = &inodes_stat,
121 .maxlen = 2*sizeof(long),
122 .mode = 0444,
123 .proc_handler = proc_nr_inodes,
124 },
125 {
126 .procname = "inode-state",
127 .data = &inodes_stat,
128 .maxlen = 7*sizeof(long),
129 .mode = 0444,
130 .proc_handler = proc_nr_inodes,
131 },
132 { }
133 };
134
init_fs_inode_sysctls(void)135 static int __init init_fs_inode_sysctls(void)
136 {
137 register_sysctl_init("fs", inodes_sysctls);
138 return 0;
139 }
140 early_initcall(init_fs_inode_sysctls);
141 #endif
142
no_open(struct inode * inode,struct file * file)143 static int no_open(struct inode *inode, struct file *file)
144 {
145 return -ENXIO;
146 }
147
148 /**
149 * inode_init_always - perform inode structure initialisation
150 * @sb: superblock inode belongs to
151 * @inode: inode to initialise
152 *
153 * These are initializations that need to be done on every inode
154 * allocation as the fields are not initialised by slab allocation.
155 */
inode_init_always(struct super_block * sb,struct inode * inode)156 int inode_init_always(struct super_block *sb, struct inode *inode)
157 {
158 static const struct inode_operations empty_iops;
159 static const struct file_operations no_open_fops = {.open = no_open};
160 struct address_space *const mapping = &inode->i_data;
161
162 inode->i_sb = sb;
163 inode->i_blkbits = sb->s_blocksize_bits;
164 inode->i_flags = 0;
165 atomic64_set(&inode->i_sequence, 0);
166 atomic_set(&inode->i_count, 1);
167 inode->i_op = &empty_iops;
168 inode->i_fop = &no_open_fops;
169 inode->i_ino = 0;
170 inode->__i_nlink = 1;
171 inode->i_opflags = 0;
172 if (sb->s_xattr)
173 inode->i_opflags |= IOP_XATTR;
174 i_uid_write(inode, 0);
175 i_gid_write(inode, 0);
176 atomic_set(&inode->i_writecount, 0);
177 inode->i_size = 0;
178 inode->i_write_hint = WRITE_LIFE_NOT_SET;
179 inode->i_blocks = 0;
180 inode->i_bytes = 0;
181 inode->i_generation = 0;
182 inode->i_pipe = NULL;
183 inode->i_cdev = NULL;
184 inode->i_link = NULL;
185 inode->i_dir_seq = 0;
186 inode->i_rdev = 0;
187 inode->dirtied_when = 0;
188
189 #ifdef CONFIG_CGROUP_WRITEBACK
190 inode->i_wb_frn_winner = 0;
191 inode->i_wb_frn_avg_time = 0;
192 inode->i_wb_frn_history = 0;
193 #endif
194
195 spin_lock_init(&inode->i_lock);
196 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
197
198 init_rwsem(&inode->i_rwsem);
199 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
200
201 atomic_set(&inode->i_dio_count, 0);
202
203 mapping->a_ops = &empty_aops;
204 mapping->host = inode;
205 mapping->flags = 0;
206 mapping->wb_err = 0;
207 atomic_set(&mapping->i_mmap_writable, 0);
208 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
209 atomic_set(&mapping->nr_thps, 0);
210 #endif
211 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
212 mapping->private_data = NULL;
213 mapping->writeback_index = 0;
214 init_rwsem(&mapping->invalidate_lock);
215 lockdep_set_class_and_name(&mapping->invalidate_lock,
216 &sb->s_type->invalidate_lock_key,
217 "mapping.invalidate_lock");
218 if (sb->s_iflags & SB_I_STABLE_WRITES)
219 mapping_set_stable_writes(mapping);
220 inode->i_private = NULL;
221 inode->i_mapping = mapping;
222 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
223 #ifdef CONFIG_FS_POSIX_ACL
224 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
225 #endif
226
227 #ifdef CONFIG_FSNOTIFY
228 inode->i_fsnotify_mask = 0;
229 #endif
230 inode->i_flctx = NULL;
231
232 if (unlikely(security_inode_alloc(inode)))
233 return -ENOMEM;
234 this_cpu_inc(nr_inodes);
235
236 return 0;
237 }
238 EXPORT_SYMBOL(inode_init_always);
239
free_inode_nonrcu(struct inode * inode)240 void free_inode_nonrcu(struct inode *inode)
241 {
242 kmem_cache_free(inode_cachep, inode);
243 }
244 EXPORT_SYMBOL(free_inode_nonrcu);
245
i_callback(struct rcu_head * head)246 static void i_callback(struct rcu_head *head)
247 {
248 struct inode *inode = container_of(head, struct inode, i_rcu);
249 if (inode->free_inode)
250 inode->free_inode(inode);
251 else
252 free_inode_nonrcu(inode);
253 }
254
alloc_inode(struct super_block * sb)255 static struct inode *alloc_inode(struct super_block *sb)
256 {
257 const struct super_operations *ops = sb->s_op;
258 struct inode *inode;
259
260 if (ops->alloc_inode)
261 inode = ops->alloc_inode(sb);
262 else
263 inode = alloc_inode_sb(sb, inode_cachep, GFP_KERNEL);
264
265 if (!inode)
266 return NULL;
267
268 if (unlikely(inode_init_always(sb, inode))) {
269 if (ops->destroy_inode) {
270 ops->destroy_inode(inode);
271 if (!ops->free_inode)
272 return NULL;
273 }
274 inode->free_inode = ops->free_inode;
275 i_callback(&inode->i_rcu);
276 return NULL;
277 }
278
279 return inode;
280 }
281
__destroy_inode(struct inode * inode)282 void __destroy_inode(struct inode *inode)
283 {
284 BUG_ON(inode_has_buffers(inode));
285 inode_detach_wb(inode);
286 security_inode_free(inode);
287 fsnotify_inode_delete(inode);
288 locks_free_lock_context(inode);
289 if (!inode->i_nlink) {
290 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
291 atomic_long_dec(&inode->i_sb->s_remove_count);
292 }
293
294 #ifdef CONFIG_FS_POSIX_ACL
295 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
296 posix_acl_release(inode->i_acl);
297 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
298 posix_acl_release(inode->i_default_acl);
299 #endif
300 this_cpu_dec(nr_inodes);
301 }
302 EXPORT_SYMBOL(__destroy_inode);
303
destroy_inode(struct inode * inode)304 static void destroy_inode(struct inode *inode)
305 {
306 const struct super_operations *ops = inode->i_sb->s_op;
307
308 BUG_ON(!list_empty(&inode->i_lru));
309 __destroy_inode(inode);
310 if (ops->destroy_inode) {
311 ops->destroy_inode(inode);
312 if (!ops->free_inode)
313 return;
314 }
315 inode->free_inode = ops->free_inode;
316 call_rcu(&inode->i_rcu, i_callback);
317 }
318
319 /**
320 * drop_nlink - directly drop an inode's link count
321 * @inode: inode
322 *
323 * This is a low-level filesystem helper to replace any
324 * direct filesystem manipulation of i_nlink. In cases
325 * where we are attempting to track writes to the
326 * filesystem, a decrement to zero means an imminent
327 * write when the file is truncated and actually unlinked
328 * on the filesystem.
329 */
drop_nlink(struct inode * inode)330 void drop_nlink(struct inode *inode)
331 {
332 WARN_ON(inode->i_nlink == 0);
333 inode->__i_nlink--;
334 if (!inode->i_nlink)
335 atomic_long_inc(&inode->i_sb->s_remove_count);
336 }
337 EXPORT_SYMBOL(drop_nlink);
338
339 /**
340 * clear_nlink - directly zero an inode's link count
341 * @inode: inode
342 *
343 * This is a low-level filesystem helper to replace any
344 * direct filesystem manipulation of i_nlink. See
345 * drop_nlink() for why we care about i_nlink hitting zero.
346 */
clear_nlink(struct inode * inode)347 void clear_nlink(struct inode *inode)
348 {
349 if (inode->i_nlink) {
350 inode->__i_nlink = 0;
351 atomic_long_inc(&inode->i_sb->s_remove_count);
352 }
353 }
354 EXPORT_SYMBOL(clear_nlink);
355
356 /**
357 * set_nlink - directly set an inode's link count
358 * @inode: inode
359 * @nlink: new nlink (should be non-zero)
360 *
361 * This is a low-level filesystem helper to replace any
362 * direct filesystem manipulation of i_nlink.
363 */
set_nlink(struct inode * inode,unsigned int nlink)364 void set_nlink(struct inode *inode, unsigned int nlink)
365 {
366 if (!nlink) {
367 clear_nlink(inode);
368 } else {
369 /* Yes, some filesystems do change nlink from zero to one */
370 if (inode->i_nlink == 0)
371 atomic_long_dec(&inode->i_sb->s_remove_count);
372
373 inode->__i_nlink = nlink;
374 }
375 }
376 EXPORT_SYMBOL(set_nlink);
377
378 /**
379 * inc_nlink - directly increment an inode's link count
380 * @inode: inode
381 *
382 * This is a low-level filesystem helper to replace any
383 * direct filesystem manipulation of i_nlink. Currently,
384 * it is only here for parity with dec_nlink().
385 */
inc_nlink(struct inode * inode)386 void inc_nlink(struct inode *inode)
387 {
388 if (unlikely(inode->i_nlink == 0)) {
389 WARN_ON(!(inode->i_state & I_LINKABLE));
390 atomic_long_dec(&inode->i_sb->s_remove_count);
391 }
392
393 inode->__i_nlink++;
394 }
395 EXPORT_SYMBOL(inc_nlink);
396
__address_space_init_once(struct address_space * mapping)397 static void __address_space_init_once(struct address_space *mapping)
398 {
399 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
400 init_rwsem(&mapping->i_mmap_rwsem);
401 INIT_LIST_HEAD(&mapping->private_list);
402 spin_lock_init(&mapping->private_lock);
403 mapping->i_mmap = RB_ROOT_CACHED;
404 }
405
address_space_init_once(struct address_space * mapping)406 void address_space_init_once(struct address_space *mapping)
407 {
408 memset(mapping, 0, sizeof(*mapping));
409 __address_space_init_once(mapping);
410 }
411 EXPORT_SYMBOL(address_space_init_once);
412
413 /*
414 * These are initializations that only need to be done
415 * once, because the fields are idempotent across use
416 * of the inode, so let the slab aware of that.
417 */
inode_init_once(struct inode * inode)418 void inode_init_once(struct inode *inode)
419 {
420 memset(inode, 0, sizeof(*inode));
421 INIT_HLIST_NODE(&inode->i_hash);
422 INIT_LIST_HEAD(&inode->i_devices);
423 INIT_LIST_HEAD(&inode->i_io_list);
424 INIT_LIST_HEAD(&inode->i_wb_list);
425 INIT_LIST_HEAD(&inode->i_lru);
426 INIT_LIST_HEAD(&inode->i_sb_list);
427 __address_space_init_once(&inode->i_data);
428 i_size_ordered_init(inode);
429 }
430 EXPORT_SYMBOL(inode_init_once);
431
init_once(void * foo)432 static void init_once(void *foo)
433 {
434 struct inode *inode = (struct inode *) foo;
435
436 inode_init_once(inode);
437 }
438
439 /*
440 * inode->i_lock must be held
441 */
__iget(struct inode * inode)442 void __iget(struct inode *inode)
443 {
444 atomic_inc(&inode->i_count);
445 }
446
447 /*
448 * get additional reference to inode; caller must already hold one.
449 */
ihold(struct inode * inode)450 void ihold(struct inode *inode)
451 {
452 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
453 }
454 EXPORT_SYMBOL(ihold);
455
__inode_add_lru(struct inode * inode,bool rotate)456 static void __inode_add_lru(struct inode *inode, bool rotate)
457 {
458 if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
459 return;
460 if (atomic_read(&inode->i_count))
461 return;
462 if (!(inode->i_sb->s_flags & SB_ACTIVE))
463 return;
464 if (!mapping_shrinkable(&inode->i_data))
465 return;
466
467 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
468 this_cpu_inc(nr_unused);
469 else if (rotate)
470 inode->i_state |= I_REFERENCED;
471 }
472
473 /*
474 * Add inode to LRU if needed (inode is unused and clean).
475 *
476 * Needs inode->i_lock held.
477 */
inode_add_lru(struct inode * inode)478 void inode_add_lru(struct inode *inode)
479 {
480 __inode_add_lru(inode, false);
481 }
482
inode_lru_list_del(struct inode * inode)483 static void inode_lru_list_del(struct inode *inode)
484 {
485 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
486 this_cpu_dec(nr_unused);
487 }
488
inode_pin_lru_isolating(struct inode * inode)489 static void inode_pin_lru_isolating(struct inode *inode)
490 {
491 lockdep_assert_held(&inode->i_lock);
492 WARN_ON(inode->i_state & (I_LRU_ISOLATING | I_FREEING | I_WILL_FREE));
493 inode->i_state |= I_LRU_ISOLATING;
494 }
495
inode_unpin_lru_isolating(struct inode * inode)496 static void inode_unpin_lru_isolating(struct inode *inode)
497 {
498 spin_lock(&inode->i_lock);
499 WARN_ON(!(inode->i_state & I_LRU_ISOLATING));
500 inode->i_state &= ~I_LRU_ISOLATING;
501 smp_mb();
502 wake_up_bit(&inode->i_state, __I_LRU_ISOLATING);
503 spin_unlock(&inode->i_lock);
504 }
505
inode_wait_for_lru_isolating(struct inode * inode)506 static void inode_wait_for_lru_isolating(struct inode *inode)
507 {
508 spin_lock(&inode->i_lock);
509 if (inode->i_state & I_LRU_ISOLATING) {
510 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_LRU_ISOLATING);
511 wait_queue_head_t *wqh;
512
513 wqh = bit_waitqueue(&inode->i_state, __I_LRU_ISOLATING);
514 spin_unlock(&inode->i_lock);
515 __wait_on_bit(wqh, &wq, bit_wait, TASK_UNINTERRUPTIBLE);
516 spin_lock(&inode->i_lock);
517 WARN_ON(inode->i_state & I_LRU_ISOLATING);
518 }
519 spin_unlock(&inode->i_lock);
520 }
521
522 /**
523 * inode_sb_list_add - add inode to the superblock list of inodes
524 * @inode: inode to add
525 */
inode_sb_list_add(struct inode * inode)526 void inode_sb_list_add(struct inode *inode)
527 {
528 spin_lock(&inode->i_sb->s_inode_list_lock);
529 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
530 spin_unlock(&inode->i_sb->s_inode_list_lock);
531 }
532 EXPORT_SYMBOL_GPL(inode_sb_list_add);
533
inode_sb_list_del(struct inode * inode)534 static inline void inode_sb_list_del(struct inode *inode)
535 {
536 if (!list_empty(&inode->i_sb_list)) {
537 spin_lock(&inode->i_sb->s_inode_list_lock);
538 list_del_init(&inode->i_sb_list);
539 spin_unlock(&inode->i_sb->s_inode_list_lock);
540 }
541 }
542
hash(struct super_block * sb,unsigned long hashval)543 static unsigned long hash(struct super_block *sb, unsigned long hashval)
544 {
545 unsigned long tmp;
546
547 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
548 L1_CACHE_BYTES;
549 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
550 return tmp & i_hash_mask;
551 }
552
553 /**
554 * __insert_inode_hash - hash an inode
555 * @inode: unhashed inode
556 * @hashval: unsigned long value used to locate this object in the
557 * inode_hashtable.
558 *
559 * Add an inode to the inode hash for this superblock.
560 */
__insert_inode_hash(struct inode * inode,unsigned long hashval)561 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
562 {
563 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
564
565 spin_lock(&inode_hash_lock);
566 spin_lock(&inode->i_lock);
567 hlist_add_head_rcu(&inode->i_hash, b);
568 spin_unlock(&inode->i_lock);
569 spin_unlock(&inode_hash_lock);
570 }
571 EXPORT_SYMBOL(__insert_inode_hash);
572
573 /**
574 * __remove_inode_hash - remove an inode from the hash
575 * @inode: inode to unhash
576 *
577 * Remove an inode from the superblock.
578 */
__remove_inode_hash(struct inode * inode)579 void __remove_inode_hash(struct inode *inode)
580 {
581 spin_lock(&inode_hash_lock);
582 spin_lock(&inode->i_lock);
583 hlist_del_init_rcu(&inode->i_hash);
584 spin_unlock(&inode->i_lock);
585 spin_unlock(&inode_hash_lock);
586 }
587 EXPORT_SYMBOL(__remove_inode_hash);
588
dump_mapping(const struct address_space * mapping)589 void dump_mapping(const struct address_space *mapping)
590 {
591 struct inode *host;
592 const struct address_space_operations *a_ops;
593 struct hlist_node *dentry_first;
594 struct dentry *dentry_ptr;
595 struct dentry dentry;
596 unsigned long ino;
597
598 /*
599 * If mapping is an invalid pointer, we don't want to crash
600 * accessing it, so probe everything depending on it carefully.
601 */
602 if (get_kernel_nofault(host, &mapping->host) ||
603 get_kernel_nofault(a_ops, &mapping->a_ops)) {
604 pr_warn("invalid mapping:%px\n", mapping);
605 return;
606 }
607
608 if (!host) {
609 pr_warn("aops:%ps\n", a_ops);
610 return;
611 }
612
613 if (get_kernel_nofault(dentry_first, &host->i_dentry.first) ||
614 get_kernel_nofault(ino, &host->i_ino)) {
615 pr_warn("aops:%ps invalid inode:%px\n", a_ops, host);
616 return;
617 }
618
619 if (!dentry_first) {
620 pr_warn("aops:%ps ino:%lx\n", a_ops, ino);
621 return;
622 }
623
624 dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
625 if (get_kernel_nofault(dentry, dentry_ptr)) {
626 pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
627 a_ops, ino, dentry_ptr);
628 return;
629 }
630
631 /*
632 * if dentry is corrupted, the %pd handler may still crash,
633 * but it's unlikely that we reach here with a corrupt mapping
634 */
635 pr_warn("aops:%ps ino:%lx dentry name:\"%pd\"\n", a_ops, ino, &dentry);
636 }
637
clear_inode(struct inode * inode)638 void clear_inode(struct inode *inode)
639 {
640 /*
641 * We have to cycle the i_pages lock here because reclaim can be in the
642 * process of removing the last page (in __filemap_remove_folio())
643 * and we must not free the mapping under it.
644 */
645 xa_lock_irq(&inode->i_data.i_pages);
646 BUG_ON(inode->i_data.nrpages);
647 /*
648 * Almost always, mapping_empty(&inode->i_data) here; but there are
649 * two known and long-standing ways in which nodes may get left behind
650 * (when deep radix-tree node allocation failed partway; or when THP
651 * collapse_file() failed). Until those two known cases are cleaned up,
652 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
653 * nor even WARN_ON(!mapping_empty).
654 */
655 xa_unlock_irq(&inode->i_data.i_pages);
656 BUG_ON(!list_empty(&inode->i_data.private_list));
657 BUG_ON(!(inode->i_state & I_FREEING));
658 BUG_ON(inode->i_state & I_CLEAR);
659 BUG_ON(!list_empty(&inode->i_wb_list));
660 /* don't need i_lock here, no concurrent mods to i_state */
661 inode->i_state = I_FREEING | I_CLEAR;
662 }
663 EXPORT_SYMBOL(clear_inode);
664
665 /*
666 * Free the inode passed in, removing it from the lists it is still connected
667 * to. We remove any pages still attached to the inode and wait for any IO that
668 * is still in progress before finally destroying the inode.
669 *
670 * An inode must already be marked I_FREEING so that we avoid the inode being
671 * moved back onto lists if we race with other code that manipulates the lists
672 * (e.g. writeback_single_inode). The caller is responsible for setting this.
673 *
674 * An inode must already be removed from the LRU list before being evicted from
675 * the cache. This should occur atomically with setting the I_FREEING state
676 * flag, so no inodes here should ever be on the LRU when being evicted.
677 */
evict(struct inode * inode)678 static void evict(struct inode *inode)
679 {
680 const struct super_operations *op = inode->i_sb->s_op;
681
682 BUG_ON(!(inode->i_state & I_FREEING));
683 BUG_ON(!list_empty(&inode->i_lru));
684
685 if (!list_empty(&inode->i_io_list))
686 inode_io_list_del(inode);
687
688 inode_sb_list_del(inode);
689
690 inode_wait_for_lru_isolating(inode);
691
692 /*
693 * Wait for flusher thread to be done with the inode so that filesystem
694 * does not start destroying it while writeback is still running. Since
695 * the inode has I_FREEING set, flusher thread won't start new work on
696 * the inode. We just have to wait for running writeback to finish.
697 */
698 inode_wait_for_writeback(inode);
699
700 if (op->evict_inode) {
701 op->evict_inode(inode);
702 } else {
703 truncate_inode_pages_final(&inode->i_data);
704 clear_inode(inode);
705 }
706 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
707 cd_forget(inode);
708
709 remove_inode_hash(inode);
710
711 spin_lock(&inode->i_lock);
712 wake_up_bit(&inode->i_state, __I_NEW);
713 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
714 spin_unlock(&inode->i_lock);
715
716 destroy_inode(inode);
717 }
718
719 /*
720 * dispose_list - dispose of the contents of a local list
721 * @head: the head of the list to free
722 *
723 * Dispose-list gets a local list with local inodes in it, so it doesn't
724 * need to worry about list corruption and SMP locks.
725 */
dispose_list(struct list_head * head)726 static void dispose_list(struct list_head *head)
727 {
728 while (!list_empty(head)) {
729 struct inode *inode;
730
731 inode = list_first_entry(head, struct inode, i_lru);
732 list_del_init(&inode->i_lru);
733
734 evict(inode);
735 cond_resched();
736 }
737 }
738
739 /**
740 * evict_inodes - evict all evictable inodes for a superblock
741 * @sb: superblock to operate on
742 *
743 * Make sure that no inodes with zero refcount are retained. This is
744 * called by superblock shutdown after having SB_ACTIVE flag removed,
745 * so any inode reaching zero refcount during or after that call will
746 * be immediately evicted.
747 */
evict_inodes(struct super_block * sb)748 void evict_inodes(struct super_block *sb)
749 {
750 struct inode *inode, *next;
751 LIST_HEAD(dispose);
752
753 again:
754 spin_lock(&sb->s_inode_list_lock);
755 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
756 if (atomic_read(&inode->i_count))
757 continue;
758
759 spin_lock(&inode->i_lock);
760 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
761 spin_unlock(&inode->i_lock);
762 continue;
763 }
764
765 inode->i_state |= I_FREEING;
766 inode_lru_list_del(inode);
767 spin_unlock(&inode->i_lock);
768 list_add(&inode->i_lru, &dispose);
769
770 /*
771 * We can have a ton of inodes to evict at unmount time given
772 * enough memory, check to see if we need to go to sleep for a
773 * bit so we don't livelock.
774 */
775 if (need_resched()) {
776 spin_unlock(&sb->s_inode_list_lock);
777 cond_resched();
778 dispose_list(&dispose);
779 goto again;
780 }
781 }
782 spin_unlock(&sb->s_inode_list_lock);
783
784 dispose_list(&dispose);
785 }
786 EXPORT_SYMBOL_GPL(evict_inodes);
787
788 /**
789 * invalidate_inodes - attempt to free all inodes on a superblock
790 * @sb: superblock to operate on
791 *
792 * Attempts to free all inodes (including dirty inodes) for a given superblock.
793 */
invalidate_inodes(struct super_block * sb)794 void invalidate_inodes(struct super_block *sb)
795 {
796 struct inode *inode, *next;
797 LIST_HEAD(dispose);
798
799 again:
800 spin_lock(&sb->s_inode_list_lock);
801 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
802 spin_lock(&inode->i_lock);
803 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
804 spin_unlock(&inode->i_lock);
805 continue;
806 }
807 if (atomic_read(&inode->i_count)) {
808 spin_unlock(&inode->i_lock);
809 continue;
810 }
811
812 inode->i_state |= I_FREEING;
813 inode_lru_list_del(inode);
814 spin_unlock(&inode->i_lock);
815 list_add(&inode->i_lru, &dispose);
816 if (need_resched()) {
817 spin_unlock(&sb->s_inode_list_lock);
818 cond_resched();
819 dispose_list(&dispose);
820 goto again;
821 }
822 }
823 spin_unlock(&sb->s_inode_list_lock);
824
825 dispose_list(&dispose);
826 }
827
828 /*
829 * Isolate the inode from the LRU in preparation for freeing it.
830 *
831 * If the inode has the I_REFERENCED flag set, then it means that it has been
832 * used recently - the flag is set in iput_final(). When we encounter such an
833 * inode, clear the flag and move it to the back of the LRU so it gets another
834 * pass through the LRU before it gets reclaimed. This is necessary because of
835 * the fact we are doing lazy LRU updates to minimise lock contention so the
836 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
837 * with this flag set because they are the inodes that are out of order.
838 */
inode_lru_isolate(struct list_head * item,struct list_lru_one * lru,spinlock_t * lru_lock,void * arg)839 static enum lru_status inode_lru_isolate(struct list_head *item,
840 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
841 {
842 struct list_head *freeable = arg;
843 struct inode *inode = container_of(item, struct inode, i_lru);
844
845 /*
846 * We are inverting the lru lock/inode->i_lock here, so use a
847 * trylock. If we fail to get the lock, just skip it.
848 */
849 if (!spin_trylock(&inode->i_lock))
850 return LRU_SKIP;
851
852 /*
853 * Inodes can get referenced, redirtied, or repopulated while
854 * they're already on the LRU, and this can make them
855 * unreclaimable for a while. Remove them lazily here; iput,
856 * sync, or the last page cache deletion will requeue them.
857 */
858 if (atomic_read(&inode->i_count) ||
859 (inode->i_state & ~I_REFERENCED) ||
860 !mapping_shrinkable(&inode->i_data)) {
861 list_lru_isolate(lru, &inode->i_lru);
862 spin_unlock(&inode->i_lock);
863 this_cpu_dec(nr_unused);
864 return LRU_REMOVED;
865 }
866
867 /* Recently referenced inodes get one more pass */
868 if (inode->i_state & I_REFERENCED) {
869 inode->i_state &= ~I_REFERENCED;
870 spin_unlock(&inode->i_lock);
871 return LRU_ROTATE;
872 }
873
874 /*
875 * On highmem systems, mapping_shrinkable() permits dropping
876 * page cache in order to free up struct inodes: lowmem might
877 * be under pressure before the cache inside the highmem zone.
878 */
879 if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
880 inode_pin_lru_isolating(inode);
881 spin_unlock(&inode->i_lock);
882 spin_unlock(lru_lock);
883 if (remove_inode_buffers(inode)) {
884 unsigned long reap;
885 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
886 if (current_is_kswapd())
887 __count_vm_events(KSWAPD_INODESTEAL, reap);
888 else
889 __count_vm_events(PGINODESTEAL, reap);
890 mm_account_reclaimed_pages(reap);
891 }
892 inode_unpin_lru_isolating(inode);
893 spin_lock(lru_lock);
894 return LRU_RETRY;
895 }
896
897 WARN_ON(inode->i_state & I_NEW);
898 inode->i_state |= I_FREEING;
899 list_lru_isolate_move(lru, &inode->i_lru, freeable);
900 spin_unlock(&inode->i_lock);
901
902 this_cpu_dec(nr_unused);
903 return LRU_REMOVED;
904 }
905
906 /*
907 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
908 * This is called from the superblock shrinker function with a number of inodes
909 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
910 * then are freed outside inode_lock by dispose_list().
911 */
prune_icache_sb(struct super_block * sb,struct shrink_control * sc)912 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
913 {
914 LIST_HEAD(freeable);
915 long freed;
916
917 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
918 inode_lru_isolate, &freeable);
919 dispose_list(&freeable);
920 return freed;
921 }
922
923 static void __wait_on_freeing_inode(struct inode *inode);
924 /*
925 * Called with the inode lock held.
926 */
find_inode(struct super_block * sb,struct hlist_head * head,int (* test)(struct inode *,void *),void * data)927 static struct inode *find_inode(struct super_block *sb,
928 struct hlist_head *head,
929 int (*test)(struct inode *, void *),
930 void *data)
931 {
932 struct inode *inode = NULL;
933
934 repeat:
935 hlist_for_each_entry(inode, head, i_hash) {
936 if (inode->i_sb != sb)
937 continue;
938 if (!test(inode, data))
939 continue;
940 spin_lock(&inode->i_lock);
941 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
942 __wait_on_freeing_inode(inode);
943 goto repeat;
944 }
945 if (unlikely(inode->i_state & I_CREATING)) {
946 spin_unlock(&inode->i_lock);
947 return ERR_PTR(-ESTALE);
948 }
949 __iget(inode);
950 spin_unlock(&inode->i_lock);
951 return inode;
952 }
953 return NULL;
954 }
955
956 /*
957 * find_inode_fast is the fast path version of find_inode, see the comment at
958 * iget_locked for details.
959 */
find_inode_fast(struct super_block * sb,struct hlist_head * head,unsigned long ino)960 static struct inode *find_inode_fast(struct super_block *sb,
961 struct hlist_head *head, unsigned long ino)
962 {
963 struct inode *inode = NULL;
964
965 repeat:
966 hlist_for_each_entry(inode, head, i_hash) {
967 if (inode->i_ino != ino)
968 continue;
969 if (inode->i_sb != sb)
970 continue;
971 spin_lock(&inode->i_lock);
972 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
973 __wait_on_freeing_inode(inode);
974 goto repeat;
975 }
976 if (unlikely(inode->i_state & I_CREATING)) {
977 spin_unlock(&inode->i_lock);
978 return ERR_PTR(-ESTALE);
979 }
980 __iget(inode);
981 spin_unlock(&inode->i_lock);
982 return inode;
983 }
984 return NULL;
985 }
986
987 /*
988 * Each cpu owns a range of LAST_INO_BATCH numbers.
989 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
990 * to renew the exhausted range.
991 *
992 * This does not significantly increase overflow rate because every CPU can
993 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
994 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
995 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
996 * overflow rate by 2x, which does not seem too significant.
997 *
998 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
999 * error if st_ino won't fit in target struct field. Use 32bit counter
1000 * here to attempt to avoid that.
1001 */
1002 #define LAST_INO_BATCH 1024
1003 static DEFINE_PER_CPU(unsigned int, last_ino);
1004
get_next_ino(void)1005 unsigned int get_next_ino(void)
1006 {
1007 unsigned int *p = &get_cpu_var(last_ino);
1008 unsigned int res = *p;
1009
1010 #ifdef CONFIG_SMP
1011 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
1012 static atomic_t shared_last_ino;
1013 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
1014
1015 res = next - LAST_INO_BATCH;
1016 }
1017 #endif
1018
1019 res++;
1020 /* get_next_ino should not provide a 0 inode number */
1021 if (unlikely(!res))
1022 res++;
1023 *p = res;
1024 put_cpu_var(last_ino);
1025 return res;
1026 }
1027 EXPORT_SYMBOL(get_next_ino);
1028
1029 /**
1030 * new_inode_pseudo - obtain an inode
1031 * @sb: superblock
1032 *
1033 * Allocates a new inode for given superblock.
1034 * Inode wont be chained in superblock s_inodes list
1035 * This means :
1036 * - fs can't be unmount
1037 * - quotas, fsnotify, writeback can't work
1038 */
new_inode_pseudo(struct super_block * sb)1039 struct inode *new_inode_pseudo(struct super_block *sb)
1040 {
1041 struct inode *inode = alloc_inode(sb);
1042
1043 if (inode) {
1044 spin_lock(&inode->i_lock);
1045 inode->i_state = 0;
1046 spin_unlock(&inode->i_lock);
1047 }
1048 return inode;
1049 }
1050
1051 /**
1052 * new_inode - obtain an inode
1053 * @sb: superblock
1054 *
1055 * Allocates a new inode for given superblock. The default gfp_mask
1056 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1057 * If HIGHMEM pages are unsuitable or it is known that pages allocated
1058 * for the page cache are not reclaimable or migratable,
1059 * mapping_set_gfp_mask() must be called with suitable flags on the
1060 * newly created inode's mapping
1061 *
1062 */
new_inode(struct super_block * sb)1063 struct inode *new_inode(struct super_block *sb)
1064 {
1065 struct inode *inode;
1066
1067 inode = new_inode_pseudo(sb);
1068 if (inode)
1069 inode_sb_list_add(inode);
1070 return inode;
1071 }
1072 EXPORT_SYMBOL(new_inode);
1073
1074 #ifdef CONFIG_DEBUG_LOCK_ALLOC
lockdep_annotate_inode_mutex_key(struct inode * inode)1075 void lockdep_annotate_inode_mutex_key(struct inode *inode)
1076 {
1077 if (S_ISDIR(inode->i_mode)) {
1078 struct file_system_type *type = inode->i_sb->s_type;
1079
1080 /* Set new key only if filesystem hasn't already changed it */
1081 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1082 /*
1083 * ensure nobody is actually holding i_mutex
1084 */
1085 // mutex_destroy(&inode->i_mutex);
1086 init_rwsem(&inode->i_rwsem);
1087 lockdep_set_class(&inode->i_rwsem,
1088 &type->i_mutex_dir_key);
1089 }
1090 }
1091 }
1092 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1093 #endif
1094
1095 /**
1096 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1097 * @inode: new inode to unlock
1098 *
1099 * Called when the inode is fully initialised to clear the new state of the
1100 * inode and wake up anyone waiting for the inode to finish initialisation.
1101 */
unlock_new_inode(struct inode * inode)1102 void unlock_new_inode(struct inode *inode)
1103 {
1104 lockdep_annotate_inode_mutex_key(inode);
1105 spin_lock(&inode->i_lock);
1106 WARN_ON(!(inode->i_state & I_NEW));
1107 inode->i_state &= ~I_NEW & ~I_CREATING;
1108 smp_mb();
1109 wake_up_bit(&inode->i_state, __I_NEW);
1110 spin_unlock(&inode->i_lock);
1111 }
1112 EXPORT_SYMBOL(unlock_new_inode);
1113
discard_new_inode(struct inode * inode)1114 void discard_new_inode(struct inode *inode)
1115 {
1116 lockdep_annotate_inode_mutex_key(inode);
1117 spin_lock(&inode->i_lock);
1118 WARN_ON(!(inode->i_state & I_NEW));
1119 inode->i_state &= ~I_NEW;
1120 smp_mb();
1121 wake_up_bit(&inode->i_state, __I_NEW);
1122 spin_unlock(&inode->i_lock);
1123 iput(inode);
1124 }
1125 EXPORT_SYMBOL(discard_new_inode);
1126
1127 /**
1128 * lock_two_inodes - lock two inodes (may be regular files but also dirs)
1129 *
1130 * Lock any non-NULL argument. The caller must make sure that if he is passing
1131 * in two directories, one is not ancestor of the other. Zero, one or two
1132 * objects may be locked by this function.
1133 *
1134 * @inode1: first inode to lock
1135 * @inode2: second inode to lock
1136 * @subclass1: inode lock subclass for the first lock obtained
1137 * @subclass2: inode lock subclass for the second lock obtained
1138 */
lock_two_inodes(struct inode * inode1,struct inode * inode2,unsigned subclass1,unsigned subclass2)1139 void lock_two_inodes(struct inode *inode1, struct inode *inode2,
1140 unsigned subclass1, unsigned subclass2)
1141 {
1142 if (!inode1 || !inode2) {
1143 /*
1144 * Make sure @subclass1 will be used for the acquired lock.
1145 * This is not strictly necessary (no current caller cares) but
1146 * let's keep things consistent.
1147 */
1148 if (!inode1)
1149 swap(inode1, inode2);
1150 goto lock;
1151 }
1152
1153 /*
1154 * If one object is directory and the other is not, we must make sure
1155 * to lock directory first as the other object may be its child.
1156 */
1157 if (S_ISDIR(inode2->i_mode) == S_ISDIR(inode1->i_mode)) {
1158 if (inode1 > inode2)
1159 swap(inode1, inode2);
1160 } else if (!S_ISDIR(inode1->i_mode))
1161 swap(inode1, inode2);
1162 lock:
1163 if (inode1)
1164 inode_lock_nested(inode1, subclass1);
1165 if (inode2 && inode2 != inode1)
1166 inode_lock_nested(inode2, subclass2);
1167 }
1168
1169 /**
1170 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1171 *
1172 * Lock any non-NULL argument. Passed objects must not be directories.
1173 * Zero, one or two objects may be locked by this function.
1174 *
1175 * @inode1: first inode to lock
1176 * @inode2: second inode to lock
1177 */
lock_two_nondirectories(struct inode * inode1,struct inode * inode2)1178 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1179 {
1180 if (inode1)
1181 WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1182 if (inode2)
1183 WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1184 lock_two_inodes(inode1, inode2, I_MUTEX_NORMAL, I_MUTEX_NONDIR2);
1185 }
1186 EXPORT_SYMBOL(lock_two_nondirectories);
1187
1188 /**
1189 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1190 * @inode1: first inode to unlock
1191 * @inode2: second inode to unlock
1192 */
unlock_two_nondirectories(struct inode * inode1,struct inode * inode2)1193 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1194 {
1195 if (inode1) {
1196 WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1197 inode_unlock(inode1);
1198 }
1199 if (inode2 && inode2 != inode1) {
1200 WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1201 inode_unlock(inode2);
1202 }
1203 }
1204 EXPORT_SYMBOL(unlock_two_nondirectories);
1205
1206 /**
1207 * inode_insert5 - obtain an inode from a mounted file system
1208 * @inode: pre-allocated inode to use for insert to cache
1209 * @hashval: hash value (usually inode number) to get
1210 * @test: callback used for comparisons between inodes
1211 * @set: callback used to initialize a new struct inode
1212 * @data: opaque data pointer to pass to @test and @set
1213 *
1214 * Search for the inode specified by @hashval and @data in the inode cache,
1215 * and if present it is return it with an increased reference count. This is
1216 * a variant of iget5_locked() for callers that don't want to fail on memory
1217 * allocation of inode.
1218 *
1219 * If the inode is not in cache, insert the pre-allocated inode to cache and
1220 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1221 * to fill it in before unlocking it via unlock_new_inode().
1222 *
1223 * Note both @test and @set are called with the inode_hash_lock held, so can't
1224 * sleep.
1225 */
inode_insert5(struct inode * inode,unsigned long hashval,int (* test)(struct inode *,void *),int (* set)(struct inode *,void *),void * data)1226 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1227 int (*test)(struct inode *, void *),
1228 int (*set)(struct inode *, void *), void *data)
1229 {
1230 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1231 struct inode *old;
1232
1233 again:
1234 spin_lock(&inode_hash_lock);
1235 old = find_inode(inode->i_sb, head, test, data);
1236 if (unlikely(old)) {
1237 /*
1238 * Uhhuh, somebody else created the same inode under us.
1239 * Use the old inode instead of the preallocated one.
1240 */
1241 spin_unlock(&inode_hash_lock);
1242 if (IS_ERR(old))
1243 return NULL;
1244 wait_on_inode(old);
1245 if (unlikely(inode_unhashed(old))) {
1246 iput(old);
1247 goto again;
1248 }
1249 return old;
1250 }
1251
1252 if (set && unlikely(set(inode, data))) {
1253 inode = NULL;
1254 goto unlock;
1255 }
1256
1257 /*
1258 * Return the locked inode with I_NEW set, the
1259 * caller is responsible for filling in the contents
1260 */
1261 spin_lock(&inode->i_lock);
1262 inode->i_state |= I_NEW;
1263 hlist_add_head_rcu(&inode->i_hash, head);
1264 spin_unlock(&inode->i_lock);
1265
1266 /*
1267 * Add inode to the sb list if it's not already. It has I_NEW at this
1268 * point, so it should be safe to test i_sb_list locklessly.
1269 */
1270 if (list_empty(&inode->i_sb_list))
1271 inode_sb_list_add(inode);
1272 unlock:
1273 spin_unlock(&inode_hash_lock);
1274
1275 return inode;
1276 }
1277 EXPORT_SYMBOL(inode_insert5);
1278
1279 /**
1280 * iget5_locked - obtain an inode from a mounted file system
1281 * @sb: super block of file system
1282 * @hashval: hash value (usually inode number) to get
1283 * @test: callback used for comparisons between inodes
1284 * @set: callback used to initialize a new struct inode
1285 * @data: opaque data pointer to pass to @test and @set
1286 *
1287 * Search for the inode specified by @hashval and @data in the inode cache,
1288 * and if present it is return it with an increased reference count. This is
1289 * a generalized version of iget_locked() for file systems where the inode
1290 * number is not sufficient for unique identification of an inode.
1291 *
1292 * If the inode is not in cache, allocate a new inode and return it locked,
1293 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1294 * before unlocking it via unlock_new_inode().
1295 *
1296 * Note both @test and @set are called with the inode_hash_lock held, so can't
1297 * sleep.
1298 */
iget5_locked(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),int (* set)(struct inode *,void *),void * data)1299 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1300 int (*test)(struct inode *, void *),
1301 int (*set)(struct inode *, void *), void *data)
1302 {
1303 struct inode *inode = ilookup5(sb, hashval, test, data);
1304
1305 if (!inode) {
1306 struct inode *new = alloc_inode(sb);
1307
1308 if (new) {
1309 new->i_state = 0;
1310 inode = inode_insert5(new, hashval, test, set, data);
1311 if (unlikely(inode != new))
1312 destroy_inode(new);
1313 }
1314 }
1315 return inode;
1316 }
1317 EXPORT_SYMBOL(iget5_locked);
1318
1319 /**
1320 * iget_locked - obtain an inode from a mounted file system
1321 * @sb: super block of file system
1322 * @ino: inode number to get
1323 *
1324 * Search for the inode specified by @ino in the inode cache and if present
1325 * return it with an increased reference count. This is for file systems
1326 * where the inode number is sufficient for unique identification of an inode.
1327 *
1328 * If the inode is not in cache, allocate a new inode and return it locked,
1329 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1330 * before unlocking it via unlock_new_inode().
1331 */
iget_locked(struct super_block * sb,unsigned long ino)1332 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1333 {
1334 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1335 struct inode *inode;
1336 again:
1337 spin_lock(&inode_hash_lock);
1338 inode = find_inode_fast(sb, head, ino);
1339 spin_unlock(&inode_hash_lock);
1340 if (inode) {
1341 if (IS_ERR(inode))
1342 return NULL;
1343 wait_on_inode(inode);
1344 if (unlikely(inode_unhashed(inode))) {
1345 iput(inode);
1346 goto again;
1347 }
1348 return inode;
1349 }
1350
1351 inode = alloc_inode(sb);
1352 if (inode) {
1353 struct inode *old;
1354
1355 spin_lock(&inode_hash_lock);
1356 /* We released the lock, so.. */
1357 old = find_inode_fast(sb, head, ino);
1358 if (!old) {
1359 inode->i_ino = ino;
1360 spin_lock(&inode->i_lock);
1361 inode->i_state = I_NEW;
1362 hlist_add_head_rcu(&inode->i_hash, head);
1363 spin_unlock(&inode->i_lock);
1364 inode_sb_list_add(inode);
1365 spin_unlock(&inode_hash_lock);
1366
1367 /* Return the locked inode with I_NEW set, the
1368 * caller is responsible for filling in the contents
1369 */
1370 return inode;
1371 }
1372
1373 /*
1374 * Uhhuh, somebody else created the same inode under
1375 * us. Use the old inode instead of the one we just
1376 * allocated.
1377 */
1378 spin_unlock(&inode_hash_lock);
1379 destroy_inode(inode);
1380 if (IS_ERR(old))
1381 return NULL;
1382 inode = old;
1383 wait_on_inode(inode);
1384 if (unlikely(inode_unhashed(inode))) {
1385 iput(inode);
1386 goto again;
1387 }
1388 }
1389 return inode;
1390 }
1391 EXPORT_SYMBOL(iget_locked);
1392
1393 /*
1394 * search the inode cache for a matching inode number.
1395 * If we find one, then the inode number we are trying to
1396 * allocate is not unique and so we should not use it.
1397 *
1398 * Returns 1 if the inode number is unique, 0 if it is not.
1399 */
test_inode_iunique(struct super_block * sb,unsigned long ino)1400 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1401 {
1402 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1403 struct inode *inode;
1404
1405 hlist_for_each_entry_rcu(inode, b, i_hash) {
1406 if (inode->i_ino == ino && inode->i_sb == sb)
1407 return 0;
1408 }
1409 return 1;
1410 }
1411
1412 /**
1413 * iunique - get a unique inode number
1414 * @sb: superblock
1415 * @max_reserved: highest reserved inode number
1416 *
1417 * Obtain an inode number that is unique on the system for a given
1418 * superblock. This is used by file systems that have no natural
1419 * permanent inode numbering system. An inode number is returned that
1420 * is higher than the reserved limit but unique.
1421 *
1422 * BUGS:
1423 * With a large number of inodes live on the file system this function
1424 * currently becomes quite slow.
1425 */
iunique(struct super_block * sb,ino_t max_reserved)1426 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1427 {
1428 /*
1429 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1430 * error if st_ino won't fit in target struct field. Use 32bit counter
1431 * here to attempt to avoid that.
1432 */
1433 static DEFINE_SPINLOCK(iunique_lock);
1434 static unsigned int counter;
1435 ino_t res;
1436
1437 rcu_read_lock();
1438 spin_lock(&iunique_lock);
1439 do {
1440 if (counter <= max_reserved)
1441 counter = max_reserved + 1;
1442 res = counter++;
1443 } while (!test_inode_iunique(sb, res));
1444 spin_unlock(&iunique_lock);
1445 rcu_read_unlock();
1446
1447 return res;
1448 }
1449 EXPORT_SYMBOL(iunique);
1450
igrab(struct inode * inode)1451 struct inode *igrab(struct inode *inode)
1452 {
1453 spin_lock(&inode->i_lock);
1454 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1455 __iget(inode);
1456 spin_unlock(&inode->i_lock);
1457 } else {
1458 spin_unlock(&inode->i_lock);
1459 /*
1460 * Handle the case where s_op->clear_inode is not been
1461 * called yet, and somebody is calling igrab
1462 * while the inode is getting freed.
1463 */
1464 inode = NULL;
1465 }
1466 return inode;
1467 }
1468 EXPORT_SYMBOL(igrab);
1469
1470 /**
1471 * ilookup5_nowait - search for an inode in the inode cache
1472 * @sb: super block of file system to search
1473 * @hashval: hash value (usually inode number) to search for
1474 * @test: callback used for comparisons between inodes
1475 * @data: opaque data pointer to pass to @test
1476 *
1477 * Search for the inode specified by @hashval and @data in the inode cache.
1478 * If the inode is in the cache, the inode is returned with an incremented
1479 * reference count.
1480 *
1481 * Note: I_NEW is not waited upon so you have to be very careful what you do
1482 * with the returned inode. You probably should be using ilookup5() instead.
1483 *
1484 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1485 */
ilookup5_nowait(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1486 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1487 int (*test)(struct inode *, void *), void *data)
1488 {
1489 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1490 struct inode *inode;
1491
1492 spin_lock(&inode_hash_lock);
1493 inode = find_inode(sb, head, test, data);
1494 spin_unlock(&inode_hash_lock);
1495
1496 return IS_ERR(inode) ? NULL : inode;
1497 }
1498 EXPORT_SYMBOL(ilookup5_nowait);
1499
1500 /**
1501 * ilookup5 - search for an inode in the inode cache
1502 * @sb: super block of file system to search
1503 * @hashval: hash value (usually inode number) to search for
1504 * @test: callback used for comparisons between inodes
1505 * @data: opaque data pointer to pass to @test
1506 *
1507 * Search for the inode specified by @hashval and @data in the inode cache,
1508 * and if the inode is in the cache, return the inode with an incremented
1509 * reference count. Waits on I_NEW before returning the inode.
1510 * returned with an incremented reference count.
1511 *
1512 * This is a generalized version of ilookup() for file systems where the
1513 * inode number is not sufficient for unique identification of an inode.
1514 *
1515 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1516 */
ilookup5(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1517 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1518 int (*test)(struct inode *, void *), void *data)
1519 {
1520 struct inode *inode;
1521 again:
1522 inode = ilookup5_nowait(sb, hashval, test, data);
1523 if (inode) {
1524 wait_on_inode(inode);
1525 if (unlikely(inode_unhashed(inode))) {
1526 iput(inode);
1527 goto again;
1528 }
1529 }
1530 return inode;
1531 }
1532 EXPORT_SYMBOL(ilookup5);
1533
1534 /**
1535 * ilookup - search for an inode in the inode cache
1536 * @sb: super block of file system to search
1537 * @ino: inode number to search for
1538 *
1539 * Search for the inode @ino in the inode cache, and if the inode is in the
1540 * cache, the inode is returned with an incremented reference count.
1541 */
ilookup(struct super_block * sb,unsigned long ino)1542 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1543 {
1544 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1545 struct inode *inode;
1546 again:
1547 spin_lock(&inode_hash_lock);
1548 inode = find_inode_fast(sb, head, ino);
1549 spin_unlock(&inode_hash_lock);
1550
1551 if (inode) {
1552 if (IS_ERR(inode))
1553 return NULL;
1554 wait_on_inode(inode);
1555 if (unlikely(inode_unhashed(inode))) {
1556 iput(inode);
1557 goto again;
1558 }
1559 }
1560 return inode;
1561 }
1562 EXPORT_SYMBOL(ilookup);
1563
1564 /**
1565 * find_inode_nowait - find an inode in the inode cache
1566 * @sb: super block of file system to search
1567 * @hashval: hash value (usually inode number) to search for
1568 * @match: callback used for comparisons between inodes
1569 * @data: opaque data pointer to pass to @match
1570 *
1571 * Search for the inode specified by @hashval and @data in the inode
1572 * cache, where the helper function @match will return 0 if the inode
1573 * does not match, 1 if the inode does match, and -1 if the search
1574 * should be stopped. The @match function must be responsible for
1575 * taking the i_lock spin_lock and checking i_state for an inode being
1576 * freed or being initialized, and incrementing the reference count
1577 * before returning 1. It also must not sleep, since it is called with
1578 * the inode_hash_lock spinlock held.
1579 *
1580 * This is a even more generalized version of ilookup5() when the
1581 * function must never block --- find_inode() can block in
1582 * __wait_on_freeing_inode() --- or when the caller can not increment
1583 * the reference count because the resulting iput() might cause an
1584 * inode eviction. The tradeoff is that the @match funtion must be
1585 * very carefully implemented.
1586 */
find_inode_nowait(struct super_block * sb,unsigned long hashval,int (* match)(struct inode *,unsigned long,void *),void * data)1587 struct inode *find_inode_nowait(struct super_block *sb,
1588 unsigned long hashval,
1589 int (*match)(struct inode *, unsigned long,
1590 void *),
1591 void *data)
1592 {
1593 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1594 struct inode *inode, *ret_inode = NULL;
1595 int mval;
1596
1597 spin_lock(&inode_hash_lock);
1598 hlist_for_each_entry(inode, head, i_hash) {
1599 if (inode->i_sb != sb)
1600 continue;
1601 mval = match(inode, hashval, data);
1602 if (mval == 0)
1603 continue;
1604 if (mval == 1)
1605 ret_inode = inode;
1606 goto out;
1607 }
1608 out:
1609 spin_unlock(&inode_hash_lock);
1610 return ret_inode;
1611 }
1612 EXPORT_SYMBOL(find_inode_nowait);
1613
1614 /**
1615 * find_inode_rcu - find an inode in the inode cache
1616 * @sb: Super block of file system to search
1617 * @hashval: Key to hash
1618 * @test: Function to test match on an inode
1619 * @data: Data for test function
1620 *
1621 * Search for the inode specified by @hashval and @data in the inode cache,
1622 * where the helper function @test will return 0 if the inode does not match
1623 * and 1 if it does. The @test function must be responsible for taking the
1624 * i_lock spin_lock and checking i_state for an inode being freed or being
1625 * initialized.
1626 *
1627 * If successful, this will return the inode for which the @test function
1628 * returned 1 and NULL otherwise.
1629 *
1630 * The @test function is not permitted to take a ref on any inode presented.
1631 * It is also not permitted to sleep.
1632 *
1633 * The caller must hold the RCU read lock.
1634 */
find_inode_rcu(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1635 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1636 int (*test)(struct inode *, void *), void *data)
1637 {
1638 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1639 struct inode *inode;
1640
1641 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1642 "suspicious find_inode_rcu() usage");
1643
1644 hlist_for_each_entry_rcu(inode, head, i_hash) {
1645 if (inode->i_sb == sb &&
1646 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1647 test(inode, data))
1648 return inode;
1649 }
1650 return NULL;
1651 }
1652 EXPORT_SYMBOL(find_inode_rcu);
1653
1654 /**
1655 * find_inode_by_ino_rcu - Find an inode in the inode cache
1656 * @sb: Super block of file system to search
1657 * @ino: The inode number to match
1658 *
1659 * Search for the inode specified by @hashval and @data in the inode cache,
1660 * where the helper function @test will return 0 if the inode does not match
1661 * and 1 if it does. The @test function must be responsible for taking the
1662 * i_lock spin_lock and checking i_state for an inode being freed or being
1663 * initialized.
1664 *
1665 * If successful, this will return the inode for which the @test function
1666 * returned 1 and NULL otherwise.
1667 *
1668 * The @test function is not permitted to take a ref on any inode presented.
1669 * It is also not permitted to sleep.
1670 *
1671 * The caller must hold the RCU read lock.
1672 */
find_inode_by_ino_rcu(struct super_block * sb,unsigned long ino)1673 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1674 unsigned long ino)
1675 {
1676 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1677 struct inode *inode;
1678
1679 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1680 "suspicious find_inode_by_ino_rcu() usage");
1681
1682 hlist_for_each_entry_rcu(inode, head, i_hash) {
1683 if (inode->i_ino == ino &&
1684 inode->i_sb == sb &&
1685 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1686 return inode;
1687 }
1688 return NULL;
1689 }
1690 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1691
insert_inode_locked(struct inode * inode)1692 int insert_inode_locked(struct inode *inode)
1693 {
1694 struct super_block *sb = inode->i_sb;
1695 ino_t ino = inode->i_ino;
1696 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1697
1698 while (1) {
1699 struct inode *old = NULL;
1700 spin_lock(&inode_hash_lock);
1701 hlist_for_each_entry(old, head, i_hash) {
1702 if (old->i_ino != ino)
1703 continue;
1704 if (old->i_sb != sb)
1705 continue;
1706 spin_lock(&old->i_lock);
1707 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1708 spin_unlock(&old->i_lock);
1709 continue;
1710 }
1711 break;
1712 }
1713 if (likely(!old)) {
1714 spin_lock(&inode->i_lock);
1715 inode->i_state |= I_NEW | I_CREATING;
1716 hlist_add_head_rcu(&inode->i_hash, head);
1717 spin_unlock(&inode->i_lock);
1718 spin_unlock(&inode_hash_lock);
1719 return 0;
1720 }
1721 if (unlikely(old->i_state & I_CREATING)) {
1722 spin_unlock(&old->i_lock);
1723 spin_unlock(&inode_hash_lock);
1724 return -EBUSY;
1725 }
1726 __iget(old);
1727 spin_unlock(&old->i_lock);
1728 spin_unlock(&inode_hash_lock);
1729 wait_on_inode(old);
1730 if (unlikely(!inode_unhashed(old))) {
1731 iput(old);
1732 return -EBUSY;
1733 }
1734 iput(old);
1735 }
1736 }
1737 EXPORT_SYMBOL(insert_inode_locked);
1738
insert_inode_locked4(struct inode * inode,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1739 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1740 int (*test)(struct inode *, void *), void *data)
1741 {
1742 struct inode *old;
1743
1744 inode->i_state |= I_CREATING;
1745 old = inode_insert5(inode, hashval, test, NULL, data);
1746
1747 if (old != inode) {
1748 iput(old);
1749 return -EBUSY;
1750 }
1751 return 0;
1752 }
1753 EXPORT_SYMBOL(insert_inode_locked4);
1754
1755
generic_delete_inode(struct inode * inode)1756 int generic_delete_inode(struct inode *inode)
1757 {
1758 return 1;
1759 }
1760 EXPORT_SYMBOL(generic_delete_inode);
1761
1762 /*
1763 * Called when we're dropping the last reference
1764 * to an inode.
1765 *
1766 * Call the FS "drop_inode()" function, defaulting to
1767 * the legacy UNIX filesystem behaviour. If it tells
1768 * us to evict inode, do so. Otherwise, retain inode
1769 * in cache if fs is alive, sync and evict if fs is
1770 * shutting down.
1771 */
iput_final(struct inode * inode)1772 static void iput_final(struct inode *inode)
1773 {
1774 struct super_block *sb = inode->i_sb;
1775 const struct super_operations *op = inode->i_sb->s_op;
1776 unsigned long state;
1777 int drop;
1778
1779 WARN_ON(inode->i_state & I_NEW);
1780
1781 if (op->drop_inode)
1782 drop = op->drop_inode(inode);
1783 else
1784 drop = generic_drop_inode(inode);
1785
1786 if (!drop &&
1787 !(inode->i_state & I_DONTCACHE) &&
1788 (sb->s_flags & SB_ACTIVE)) {
1789 __inode_add_lru(inode, true);
1790 spin_unlock(&inode->i_lock);
1791 return;
1792 }
1793
1794 state = inode->i_state;
1795 if (!drop) {
1796 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1797 spin_unlock(&inode->i_lock);
1798
1799 write_inode_now(inode, 1);
1800
1801 spin_lock(&inode->i_lock);
1802 state = inode->i_state;
1803 WARN_ON(state & I_NEW);
1804 state &= ~I_WILL_FREE;
1805 }
1806
1807 WRITE_ONCE(inode->i_state, state | I_FREEING);
1808 if (!list_empty(&inode->i_lru))
1809 inode_lru_list_del(inode);
1810 spin_unlock(&inode->i_lock);
1811
1812 evict(inode);
1813 }
1814
1815 /**
1816 * iput - put an inode
1817 * @inode: inode to put
1818 *
1819 * Puts an inode, dropping its usage count. If the inode use count hits
1820 * zero, the inode is then freed and may also be destroyed.
1821 *
1822 * Consequently, iput() can sleep.
1823 */
iput(struct inode * inode)1824 void iput(struct inode *inode)
1825 {
1826 if (!inode)
1827 return;
1828 BUG_ON(inode->i_state & I_CLEAR);
1829 retry:
1830 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1831 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1832 atomic_inc(&inode->i_count);
1833 spin_unlock(&inode->i_lock);
1834 trace_writeback_lazytime_iput(inode);
1835 mark_inode_dirty_sync(inode);
1836 goto retry;
1837 }
1838 iput_final(inode);
1839 }
1840 }
1841 EXPORT_SYMBOL(iput);
1842
1843 #ifdef CONFIG_BLOCK
1844 /**
1845 * bmap - find a block number in a file
1846 * @inode: inode owning the block number being requested
1847 * @block: pointer containing the block to find
1848 *
1849 * Replaces the value in ``*block`` with the block number on the device holding
1850 * corresponding to the requested block number in the file.
1851 * That is, asked for block 4 of inode 1 the function will replace the
1852 * 4 in ``*block``, with disk block relative to the disk start that holds that
1853 * block of the file.
1854 *
1855 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1856 * hole, returns 0 and ``*block`` is also set to 0.
1857 */
bmap(struct inode * inode,sector_t * block)1858 int bmap(struct inode *inode, sector_t *block)
1859 {
1860 if (!inode->i_mapping->a_ops->bmap)
1861 return -EINVAL;
1862
1863 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1864 return 0;
1865 }
1866 EXPORT_SYMBOL(bmap);
1867 #endif
1868
1869 /*
1870 * With relative atime, only update atime if the previous atime is
1871 * earlier than or equal to either the ctime or mtime,
1872 * or if at least a day has passed since the last atime update.
1873 */
relatime_need_update(struct vfsmount * mnt,struct inode * inode,struct timespec64 now)1874 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1875 struct timespec64 now)
1876 {
1877 struct timespec64 ctime;
1878
1879 if (!(mnt->mnt_flags & MNT_RELATIME))
1880 return 1;
1881 /*
1882 * Is mtime younger than or equal to atime? If yes, update atime:
1883 */
1884 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1885 return 1;
1886 /*
1887 * Is ctime younger than or equal to atime? If yes, update atime:
1888 */
1889 ctime = inode_get_ctime(inode);
1890 if (timespec64_compare(&ctime, &inode->i_atime) >= 0)
1891 return 1;
1892
1893 /*
1894 * Is the previous atime value older than a day? If yes,
1895 * update atime:
1896 */
1897 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1898 return 1;
1899 /*
1900 * Good, we can skip the atime update:
1901 */
1902 return 0;
1903 }
1904
1905 /**
1906 * inode_update_timestamps - update the timestamps on the inode
1907 * @inode: inode to be updated
1908 * @flags: S_* flags that needed to be updated
1909 *
1910 * The update_time function is called when an inode's timestamps need to be
1911 * updated for a read or write operation. This function handles updating the
1912 * actual timestamps. It's up to the caller to ensure that the inode is marked
1913 * dirty appropriately.
1914 *
1915 * In the case where any of S_MTIME, S_CTIME, or S_VERSION need to be updated,
1916 * attempt to update all three of them. S_ATIME updates can be handled
1917 * independently of the rest.
1918 *
1919 * Returns a set of S_* flags indicating which values changed.
1920 */
inode_update_timestamps(struct inode * inode,int flags)1921 int inode_update_timestamps(struct inode *inode, int flags)
1922 {
1923 int updated = 0;
1924 struct timespec64 now;
1925
1926 if (flags & (S_MTIME|S_CTIME|S_VERSION)) {
1927 struct timespec64 ctime = inode_get_ctime(inode);
1928
1929 now = inode_set_ctime_current(inode);
1930 if (!timespec64_equal(&now, &ctime))
1931 updated |= S_CTIME;
1932 if (!timespec64_equal(&now, &inode->i_mtime)) {
1933 inode->i_mtime = now;
1934 updated |= S_MTIME;
1935 }
1936 if (IS_I_VERSION(inode) && inode_maybe_inc_iversion(inode, updated))
1937 updated |= S_VERSION;
1938 } else {
1939 now = current_time(inode);
1940 }
1941
1942 if (flags & S_ATIME) {
1943 if (!timespec64_equal(&now, &inode->i_atime)) {
1944 inode->i_atime = now;
1945 updated |= S_ATIME;
1946 }
1947 }
1948 return updated;
1949 }
1950 EXPORT_SYMBOL(inode_update_timestamps);
1951
1952 /**
1953 * generic_update_time - update the timestamps on the inode
1954 * @inode: inode to be updated
1955 * @flags: S_* flags that needed to be updated
1956 *
1957 * The update_time function is called when an inode's timestamps need to be
1958 * updated for a read or write operation. In the case where any of S_MTIME, S_CTIME,
1959 * or S_VERSION need to be updated we attempt to update all three of them. S_ATIME
1960 * updates can be handled done independently of the rest.
1961 *
1962 * Returns a S_* mask indicating which fields were updated.
1963 */
generic_update_time(struct inode * inode,int flags)1964 int generic_update_time(struct inode *inode, int flags)
1965 {
1966 int updated = inode_update_timestamps(inode, flags);
1967 int dirty_flags = 0;
1968
1969 if (updated & (S_ATIME|S_MTIME|S_CTIME))
1970 dirty_flags = inode->i_sb->s_flags & SB_LAZYTIME ? I_DIRTY_TIME : I_DIRTY_SYNC;
1971 if (updated & S_VERSION)
1972 dirty_flags |= I_DIRTY_SYNC;
1973 __mark_inode_dirty(inode, dirty_flags);
1974 return updated;
1975 }
1976 EXPORT_SYMBOL(generic_update_time);
1977
1978 /*
1979 * This does the actual work of updating an inodes time or version. Must have
1980 * had called mnt_want_write() before calling this.
1981 */
inode_update_time(struct inode * inode,int flags)1982 int inode_update_time(struct inode *inode, int flags)
1983 {
1984 if (inode->i_op->update_time)
1985 return inode->i_op->update_time(inode, flags);
1986 generic_update_time(inode, flags);
1987 return 0;
1988 }
1989 EXPORT_SYMBOL(inode_update_time);
1990
1991 /**
1992 * atime_needs_update - update the access time
1993 * @path: the &struct path to update
1994 * @inode: inode to update
1995 *
1996 * Update the accessed time on an inode and mark it for writeback.
1997 * This function automatically handles read only file systems and media,
1998 * as well as the "noatime" flag and inode specific "noatime" markers.
1999 */
atime_needs_update(const struct path * path,struct inode * inode)2000 bool atime_needs_update(const struct path *path, struct inode *inode)
2001 {
2002 struct vfsmount *mnt = path->mnt;
2003 struct timespec64 now;
2004
2005 if (inode->i_flags & S_NOATIME)
2006 return false;
2007
2008 /* Atime updates will likely cause i_uid and i_gid to be written
2009 * back improprely if their true value is unknown to the vfs.
2010 */
2011 if (HAS_UNMAPPED_ID(mnt_idmap(mnt), inode))
2012 return false;
2013
2014 if (IS_NOATIME(inode))
2015 return false;
2016 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
2017 return false;
2018
2019 if (mnt->mnt_flags & MNT_NOATIME)
2020 return false;
2021 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
2022 return false;
2023
2024 now = current_time(inode);
2025
2026 if (!relatime_need_update(mnt, inode, now))
2027 return false;
2028
2029 if (timespec64_equal(&inode->i_atime, &now))
2030 return false;
2031
2032 return true;
2033 }
2034
touch_atime(const struct path * path)2035 void touch_atime(const struct path *path)
2036 {
2037 struct vfsmount *mnt = path->mnt;
2038 struct inode *inode = d_inode(path->dentry);
2039
2040 if (!atime_needs_update(path, inode))
2041 return;
2042
2043 if (!sb_start_write_trylock(inode->i_sb))
2044 return;
2045
2046 if (__mnt_want_write(mnt) != 0)
2047 goto skip_update;
2048 /*
2049 * File systems can error out when updating inodes if they need to
2050 * allocate new space to modify an inode (such is the case for
2051 * Btrfs), but since we touch atime while walking down the path we
2052 * really don't care if we failed to update the atime of the file,
2053 * so just ignore the return value.
2054 * We may also fail on filesystems that have the ability to make parts
2055 * of the fs read only, e.g. subvolumes in Btrfs.
2056 */
2057 inode_update_time(inode, S_ATIME);
2058 __mnt_drop_write(mnt);
2059 skip_update:
2060 sb_end_write(inode->i_sb);
2061 }
2062 EXPORT_SYMBOL(touch_atime);
2063
2064 /*
2065 * Return mask of changes for notify_change() that need to be done as a
2066 * response to write or truncate. Return 0 if nothing has to be changed.
2067 * Negative value on error (change should be denied).
2068 */
dentry_needs_remove_privs(struct mnt_idmap * idmap,struct dentry * dentry)2069 int dentry_needs_remove_privs(struct mnt_idmap *idmap,
2070 struct dentry *dentry)
2071 {
2072 struct inode *inode = d_inode(dentry);
2073 int mask = 0;
2074 int ret;
2075
2076 if (IS_NOSEC(inode))
2077 return 0;
2078
2079 mask = setattr_should_drop_suidgid(idmap, inode);
2080 ret = security_inode_need_killpriv(dentry);
2081 if (ret < 0)
2082 return ret;
2083 if (ret)
2084 mask |= ATTR_KILL_PRIV;
2085 return mask;
2086 }
2087
__remove_privs(struct mnt_idmap * idmap,struct dentry * dentry,int kill)2088 static int __remove_privs(struct mnt_idmap *idmap,
2089 struct dentry *dentry, int kill)
2090 {
2091 struct iattr newattrs;
2092
2093 newattrs.ia_valid = ATTR_FORCE | kill;
2094 /*
2095 * Note we call this on write, so notify_change will not
2096 * encounter any conflicting delegations:
2097 */
2098 return notify_change(idmap, dentry, &newattrs, NULL);
2099 }
2100
__file_remove_privs(struct file * file,unsigned int flags)2101 static int __file_remove_privs(struct file *file, unsigned int flags)
2102 {
2103 struct dentry *dentry = file_dentry(file);
2104 struct inode *inode = file_inode(file);
2105 int error = 0;
2106 int kill;
2107
2108 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
2109 return 0;
2110
2111 kill = dentry_needs_remove_privs(file_mnt_idmap(file), dentry);
2112 if (kill < 0)
2113 return kill;
2114
2115 if (kill) {
2116 if (flags & IOCB_NOWAIT)
2117 return -EAGAIN;
2118
2119 error = __remove_privs(file_mnt_idmap(file), dentry, kill);
2120 }
2121
2122 if (!error)
2123 inode_has_no_xattr(inode);
2124 return error;
2125 }
2126
2127 /**
2128 * file_remove_privs - remove special file privileges (suid, capabilities)
2129 * @file: file to remove privileges from
2130 *
2131 * When file is modified by a write or truncation ensure that special
2132 * file privileges are removed.
2133 *
2134 * Return: 0 on success, negative errno on failure.
2135 */
file_remove_privs(struct file * file)2136 int file_remove_privs(struct file *file)
2137 {
2138 return __file_remove_privs(file, 0);
2139 }
2140 EXPORT_SYMBOL(file_remove_privs);
2141
inode_needs_update_time(struct inode * inode)2142 static int inode_needs_update_time(struct inode *inode)
2143 {
2144 int sync_it = 0;
2145 struct timespec64 now = current_time(inode);
2146 struct timespec64 ctime;
2147
2148 /* First try to exhaust all avenues to not sync */
2149 if (IS_NOCMTIME(inode))
2150 return 0;
2151
2152 if (!timespec64_equal(&inode->i_mtime, &now))
2153 sync_it = S_MTIME;
2154
2155 ctime = inode_get_ctime(inode);
2156 if (!timespec64_equal(&ctime, &now))
2157 sync_it |= S_CTIME;
2158
2159 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2160 sync_it |= S_VERSION;
2161
2162 return sync_it;
2163 }
2164
__file_update_time(struct file * file,int sync_mode)2165 static int __file_update_time(struct file *file, int sync_mode)
2166 {
2167 int ret = 0;
2168 struct inode *inode = file_inode(file);
2169
2170 /* try to update time settings */
2171 if (!__mnt_want_write_file(file)) {
2172 ret = inode_update_time(inode, sync_mode);
2173 __mnt_drop_write_file(file);
2174 }
2175
2176 return ret;
2177 }
2178
2179 /**
2180 * file_update_time - update mtime and ctime time
2181 * @file: file accessed
2182 *
2183 * Update the mtime and ctime members of an inode and mark the inode for
2184 * writeback. Note that this function is meant exclusively for usage in
2185 * the file write path of filesystems, and filesystems may choose to
2186 * explicitly ignore updates via this function with the _NOCMTIME inode
2187 * flag, e.g. for network filesystem where these imestamps are handled
2188 * by the server. This can return an error for file systems who need to
2189 * allocate space in order to update an inode.
2190 *
2191 * Return: 0 on success, negative errno on failure.
2192 */
file_update_time(struct file * file)2193 int file_update_time(struct file *file)
2194 {
2195 int ret;
2196 struct inode *inode = file_inode(file);
2197
2198 ret = inode_needs_update_time(inode);
2199 if (ret <= 0)
2200 return ret;
2201
2202 return __file_update_time(file, ret);
2203 }
2204 EXPORT_SYMBOL(file_update_time);
2205
2206 /**
2207 * file_modified_flags - handle mandated vfs changes when modifying a file
2208 * @file: file that was modified
2209 * @flags: kiocb flags
2210 *
2211 * When file has been modified ensure that special
2212 * file privileges are removed and time settings are updated.
2213 *
2214 * If IOCB_NOWAIT is set, special file privileges will not be removed and
2215 * time settings will not be updated. It will return -EAGAIN.
2216 *
2217 * Context: Caller must hold the file's inode lock.
2218 *
2219 * Return: 0 on success, negative errno on failure.
2220 */
file_modified_flags(struct file * file,int flags)2221 static int file_modified_flags(struct file *file, int flags)
2222 {
2223 int ret;
2224 struct inode *inode = file_inode(file);
2225
2226 /*
2227 * Clear the security bits if the process is not being run by root.
2228 * This keeps people from modifying setuid and setgid binaries.
2229 */
2230 ret = __file_remove_privs(file, flags);
2231 if (ret)
2232 return ret;
2233
2234 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2235 return 0;
2236
2237 ret = inode_needs_update_time(inode);
2238 if (ret <= 0)
2239 return ret;
2240 if (flags & IOCB_NOWAIT)
2241 return -EAGAIN;
2242
2243 return __file_update_time(file, ret);
2244 }
2245
2246 /**
2247 * file_modified - handle mandated vfs changes when modifying a file
2248 * @file: file that was modified
2249 *
2250 * When file has been modified ensure that special
2251 * file privileges are removed and time settings are updated.
2252 *
2253 * Context: Caller must hold the file's inode lock.
2254 *
2255 * Return: 0 on success, negative errno on failure.
2256 */
file_modified(struct file * file)2257 int file_modified(struct file *file)
2258 {
2259 return file_modified_flags(file, 0);
2260 }
2261 EXPORT_SYMBOL(file_modified);
2262
2263 /**
2264 * kiocb_modified - handle mandated vfs changes when modifying a file
2265 * @iocb: iocb that was modified
2266 *
2267 * When file has been modified ensure that special
2268 * file privileges are removed and time settings are updated.
2269 *
2270 * Context: Caller must hold the file's inode lock.
2271 *
2272 * Return: 0 on success, negative errno on failure.
2273 */
kiocb_modified(struct kiocb * iocb)2274 int kiocb_modified(struct kiocb *iocb)
2275 {
2276 return file_modified_flags(iocb->ki_filp, iocb->ki_flags);
2277 }
2278 EXPORT_SYMBOL_GPL(kiocb_modified);
2279
inode_needs_sync(struct inode * inode)2280 int inode_needs_sync(struct inode *inode)
2281 {
2282 if (IS_SYNC(inode))
2283 return 1;
2284 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2285 return 1;
2286 return 0;
2287 }
2288 EXPORT_SYMBOL(inode_needs_sync);
2289
2290 /*
2291 * If we try to find an inode in the inode hash while it is being
2292 * deleted, we have to wait until the filesystem completes its
2293 * deletion before reporting that it isn't found. This function waits
2294 * until the deletion _might_ have completed. Callers are responsible
2295 * to recheck inode state.
2296 *
2297 * It doesn't matter if I_NEW is not set initially, a call to
2298 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2299 * will DTRT.
2300 */
__wait_on_freeing_inode(struct inode * inode)2301 static void __wait_on_freeing_inode(struct inode *inode)
2302 {
2303 wait_queue_head_t *wq;
2304 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2305 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2306 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2307 spin_unlock(&inode->i_lock);
2308 spin_unlock(&inode_hash_lock);
2309 schedule();
2310 finish_wait(wq, &wait.wq_entry);
2311 spin_lock(&inode_hash_lock);
2312 }
2313
2314 static __initdata unsigned long ihash_entries;
set_ihash_entries(char * str)2315 static int __init set_ihash_entries(char *str)
2316 {
2317 if (!str)
2318 return 0;
2319 ihash_entries = simple_strtoul(str, &str, 0);
2320 return 1;
2321 }
2322 __setup("ihash_entries=", set_ihash_entries);
2323
2324 /*
2325 * Initialize the waitqueues and inode hash table.
2326 */
inode_init_early(void)2327 void __init inode_init_early(void)
2328 {
2329 /* If hashes are distributed across NUMA nodes, defer
2330 * hash allocation until vmalloc space is available.
2331 */
2332 if (hashdist)
2333 return;
2334
2335 inode_hashtable =
2336 alloc_large_system_hash("Inode-cache",
2337 sizeof(struct hlist_head),
2338 ihash_entries,
2339 14,
2340 HASH_EARLY | HASH_ZERO,
2341 &i_hash_shift,
2342 &i_hash_mask,
2343 0,
2344 0);
2345 }
2346
inode_init(void)2347 void __init inode_init(void)
2348 {
2349 /* inode slab cache */
2350 inode_cachep = kmem_cache_create("inode_cache",
2351 sizeof(struct inode),
2352 0,
2353 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2354 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2355 init_once);
2356
2357 /* Hash may have been set up in inode_init_early */
2358 if (!hashdist)
2359 return;
2360
2361 inode_hashtable =
2362 alloc_large_system_hash("Inode-cache",
2363 sizeof(struct hlist_head),
2364 ihash_entries,
2365 14,
2366 HASH_ZERO,
2367 &i_hash_shift,
2368 &i_hash_mask,
2369 0,
2370 0);
2371 }
2372
init_special_inode(struct inode * inode,umode_t mode,dev_t rdev)2373 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2374 {
2375 inode->i_mode = mode;
2376 if (S_ISCHR(mode)) {
2377 inode->i_fop = &def_chr_fops;
2378 inode->i_rdev = rdev;
2379 } else if (S_ISBLK(mode)) {
2380 if (IS_ENABLED(CONFIG_BLOCK))
2381 inode->i_fop = &def_blk_fops;
2382 inode->i_rdev = rdev;
2383 } else if (S_ISFIFO(mode))
2384 inode->i_fop = &pipefifo_fops;
2385 else if (S_ISSOCK(mode))
2386 ; /* leave it no_open_fops */
2387 else
2388 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2389 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2390 inode->i_ino);
2391 }
2392 EXPORT_SYMBOL(init_special_inode);
2393
2394 /**
2395 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2396 * @idmap: idmap of the mount the inode was created from
2397 * @inode: New inode
2398 * @dir: Directory inode
2399 * @mode: mode of the new inode
2400 *
2401 * If the inode has been created through an idmapped mount the idmap of
2402 * the vfsmount must be passed through @idmap. This function will then take
2403 * care to map the inode according to @idmap before checking permissions
2404 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2405 * checking is to be performed on the raw inode simply pass @nop_mnt_idmap.
2406 */
inode_init_owner(struct mnt_idmap * idmap,struct inode * inode,const struct inode * dir,umode_t mode)2407 void inode_init_owner(struct mnt_idmap *idmap, struct inode *inode,
2408 const struct inode *dir, umode_t mode)
2409 {
2410 inode_fsuid_set(inode, idmap);
2411 if (dir && dir->i_mode & S_ISGID) {
2412 inode->i_gid = dir->i_gid;
2413
2414 /* Directories are special, and always inherit S_ISGID */
2415 if (S_ISDIR(mode))
2416 mode |= S_ISGID;
2417 } else
2418 inode_fsgid_set(inode, idmap);
2419 inode->i_mode = mode;
2420 }
2421 EXPORT_SYMBOL(inode_init_owner);
2422
2423 /**
2424 * inode_owner_or_capable - check current task permissions to inode
2425 * @idmap: idmap of the mount the inode was found from
2426 * @inode: inode being checked
2427 *
2428 * Return true if current either has CAP_FOWNER in a namespace with the
2429 * inode owner uid mapped, or owns the file.
2430 *
2431 * If the inode has been found through an idmapped mount the idmap of
2432 * the vfsmount must be passed through @idmap. This function will then take
2433 * care to map the inode according to @idmap before checking permissions.
2434 * On non-idmapped mounts or if permission checking is to be performed on the
2435 * raw inode simply passs @nop_mnt_idmap.
2436 */
inode_owner_or_capable(struct mnt_idmap * idmap,const struct inode * inode)2437 bool inode_owner_or_capable(struct mnt_idmap *idmap,
2438 const struct inode *inode)
2439 {
2440 vfsuid_t vfsuid;
2441 struct user_namespace *ns;
2442
2443 vfsuid = i_uid_into_vfsuid(idmap, inode);
2444 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
2445 return true;
2446
2447 ns = current_user_ns();
2448 if (vfsuid_has_mapping(ns, vfsuid) && ns_capable(ns, CAP_FOWNER))
2449 return true;
2450 return false;
2451 }
2452 EXPORT_SYMBOL(inode_owner_or_capable);
2453
2454 /*
2455 * Direct i/o helper functions
2456 */
__inode_dio_wait(struct inode * inode)2457 static void __inode_dio_wait(struct inode *inode)
2458 {
2459 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2460 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2461
2462 do {
2463 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2464 if (atomic_read(&inode->i_dio_count))
2465 schedule();
2466 } while (atomic_read(&inode->i_dio_count));
2467 finish_wait(wq, &q.wq_entry);
2468 }
2469
2470 /**
2471 * inode_dio_wait - wait for outstanding DIO requests to finish
2472 * @inode: inode to wait for
2473 *
2474 * Waits for all pending direct I/O requests to finish so that we can
2475 * proceed with a truncate or equivalent operation.
2476 *
2477 * Must be called under a lock that serializes taking new references
2478 * to i_dio_count, usually by inode->i_mutex.
2479 */
inode_dio_wait(struct inode * inode)2480 void inode_dio_wait(struct inode *inode)
2481 {
2482 if (atomic_read(&inode->i_dio_count))
2483 __inode_dio_wait(inode);
2484 }
2485 EXPORT_SYMBOL(inode_dio_wait);
2486
2487 /*
2488 * inode_set_flags - atomically set some inode flags
2489 *
2490 * Note: the caller should be holding i_mutex, or else be sure that
2491 * they have exclusive access to the inode structure (i.e., while the
2492 * inode is being instantiated). The reason for the cmpxchg() loop
2493 * --- which wouldn't be necessary if all code paths which modify
2494 * i_flags actually followed this rule, is that there is at least one
2495 * code path which doesn't today so we use cmpxchg() out of an abundance
2496 * of caution.
2497 *
2498 * In the long run, i_mutex is overkill, and we should probably look
2499 * at using the i_lock spinlock to protect i_flags, and then make sure
2500 * it is so documented in include/linux/fs.h and that all code follows
2501 * the locking convention!!
2502 */
inode_set_flags(struct inode * inode,unsigned int flags,unsigned int mask)2503 void inode_set_flags(struct inode *inode, unsigned int flags,
2504 unsigned int mask)
2505 {
2506 WARN_ON_ONCE(flags & ~mask);
2507 set_mask_bits(&inode->i_flags, mask, flags);
2508 }
2509 EXPORT_SYMBOL(inode_set_flags);
2510
inode_nohighmem(struct inode * inode)2511 void inode_nohighmem(struct inode *inode)
2512 {
2513 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2514 }
2515 EXPORT_SYMBOL(inode_nohighmem);
2516
2517 /**
2518 * timestamp_truncate - Truncate timespec to a granularity
2519 * @t: Timespec
2520 * @inode: inode being updated
2521 *
2522 * Truncate a timespec to the granularity supported by the fs
2523 * containing the inode. Always rounds down. gran must
2524 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2525 */
timestamp_truncate(struct timespec64 t,struct inode * inode)2526 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2527 {
2528 struct super_block *sb = inode->i_sb;
2529 unsigned int gran = sb->s_time_gran;
2530
2531 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2532 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2533 t.tv_nsec = 0;
2534
2535 /* Avoid division in the common cases 1 ns and 1 s. */
2536 if (gran == 1)
2537 ; /* nothing */
2538 else if (gran == NSEC_PER_SEC)
2539 t.tv_nsec = 0;
2540 else if (gran > 1 && gran < NSEC_PER_SEC)
2541 t.tv_nsec -= t.tv_nsec % gran;
2542 else
2543 WARN(1, "invalid file time granularity: %u", gran);
2544 return t;
2545 }
2546 EXPORT_SYMBOL(timestamp_truncate);
2547
2548 /**
2549 * current_time - Return FS time
2550 * @inode: inode.
2551 *
2552 * Return the current time truncated to the time granularity supported by
2553 * the fs.
2554 *
2555 * Note that inode and inode->sb cannot be NULL.
2556 * Otherwise, the function warns and returns time without truncation.
2557 */
current_time(struct inode * inode)2558 struct timespec64 current_time(struct inode *inode)
2559 {
2560 struct timespec64 now;
2561
2562 ktime_get_coarse_real_ts64(&now);
2563 return timestamp_truncate(now, inode);
2564 }
2565 EXPORT_SYMBOL(current_time);
2566
2567 /**
2568 * inode_set_ctime_current - set the ctime to current_time
2569 * @inode: inode
2570 *
2571 * Set the inode->i_ctime to the current value for the inode. Returns
2572 * the current value that was assigned to i_ctime.
2573 */
inode_set_ctime_current(struct inode * inode)2574 struct timespec64 inode_set_ctime_current(struct inode *inode)
2575 {
2576 struct timespec64 now = current_time(inode);
2577
2578 inode_set_ctime(inode, now.tv_sec, now.tv_nsec);
2579 return now;
2580 }
2581 EXPORT_SYMBOL(inode_set_ctime_current);
2582
2583 /**
2584 * in_group_or_capable - check whether caller is CAP_FSETID privileged
2585 * @idmap: idmap of the mount @inode was found from
2586 * @inode: inode to check
2587 * @vfsgid: the new/current vfsgid of @inode
2588 *
2589 * Check wether @vfsgid is in the caller's group list or if the caller is
2590 * privileged with CAP_FSETID over @inode. This can be used to determine
2591 * whether the setgid bit can be kept or must be dropped.
2592 *
2593 * Return: true if the caller is sufficiently privileged, false if not.
2594 */
in_group_or_capable(struct mnt_idmap * idmap,const struct inode * inode,vfsgid_t vfsgid)2595 bool in_group_or_capable(struct mnt_idmap *idmap,
2596 const struct inode *inode, vfsgid_t vfsgid)
2597 {
2598 if (vfsgid_in_group_p(vfsgid))
2599 return true;
2600 if (capable_wrt_inode_uidgid(idmap, inode, CAP_FSETID))
2601 return true;
2602 return false;
2603 }
2604
2605 /**
2606 * mode_strip_sgid - handle the sgid bit for non-directories
2607 * @idmap: idmap of the mount the inode was created from
2608 * @dir: parent directory inode
2609 * @mode: mode of the file to be created in @dir
2610 *
2611 * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2612 * raised and @dir has the S_ISGID bit raised ensure that the caller is
2613 * either in the group of the parent directory or they have CAP_FSETID
2614 * in their user namespace and are privileged over the parent directory.
2615 * In all other cases, strip the S_ISGID bit from @mode.
2616 *
2617 * Return: the new mode to use for the file
2618 */
mode_strip_sgid(struct mnt_idmap * idmap,const struct inode * dir,umode_t mode)2619 umode_t mode_strip_sgid(struct mnt_idmap *idmap,
2620 const struct inode *dir, umode_t mode)
2621 {
2622 if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2623 return mode;
2624 if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2625 return mode;
2626 if (in_group_or_capable(idmap, dir, i_gid_into_vfsgid(idmap, dir)))
2627 return mode;
2628 return mode & ~S_ISGID;
2629 }
2630 EXPORT_SYMBOL(mode_strip_sgid);
2631