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