xref: /linux/fs/inode.c (revision fcb3ad4366b9c810cbb9da34c076a9a52d8aa1e0)
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 <linux/rw_hint.h>
24 #include <linux/seq_file.h>
25 #include <linux/debugfs.h>
26 #include <trace/events/writeback.h>
27 #define CREATE_TRACE_POINTS
28 #include <trace/events/timestamp.h>
29 
30 #include "internal.h"
31 
32 /*
33  * Inode locking rules:
34  *
35  * inode->i_lock protects:
36  *   inode->i_state, inode->i_hash, __iget(), inode->i_io_list
37  * Inode LRU list locks protect:
38  *   inode->i_sb->s_inode_lru, inode->i_lru
39  * inode->i_sb->s_inode_list_lock protects:
40  *   inode->i_sb->s_inodes, inode->i_sb_list
41  * bdi->wb.list_lock protects:
42  *   bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
43  * inode_hash_lock protects:
44  *   inode_hashtable, inode->i_hash
45  *
46  * Lock ordering:
47  *
48  * inode->i_sb->s_inode_list_lock
49  *   inode->i_lock
50  *     Inode LRU list locks
51  *
52  * bdi->wb.list_lock
53  *   inode->i_lock
54  *
55  * inode_hash_lock
56  *   inode->i_sb->s_inode_list_lock
57  *   inode->i_lock
58  *
59  * iunique_lock
60  *   inode_hash_lock
61  */
62 
63 static unsigned int i_hash_mask __ro_after_init;
64 static unsigned int i_hash_shift __ro_after_init;
65 static struct hlist_head *inode_hashtable __ro_after_init;
66 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
67 
68 /*
69  * Empty aops. Can be used for the cases where the user does not
70  * define any of the address_space operations.
71  */
72 const struct address_space_operations empty_aops = {
73 };
74 EXPORT_SYMBOL(empty_aops);
75 
76 static DEFINE_PER_CPU(unsigned long, nr_inodes);
77 static DEFINE_PER_CPU(unsigned long, nr_unused);
78 
79 static struct kmem_cache *inode_cachep __ro_after_init;
80 
81 static long get_nr_inodes(void)
82 {
83 	int i;
84 	long sum = 0;
85 	for_each_possible_cpu(i)
86 		sum += per_cpu(nr_inodes, i);
87 	return sum < 0 ? 0 : sum;
88 }
89 
90 static inline long get_nr_inodes_unused(void)
91 {
92 	int i;
93 	long sum = 0;
94 	for_each_possible_cpu(i)
95 		sum += per_cpu(nr_unused, i);
96 	return sum < 0 ? 0 : sum;
97 }
98 
99 long get_nr_dirty_inodes(void)
100 {
101 	/* not actually dirty inodes, but a wild approximation */
102 	long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
103 	return nr_dirty > 0 ? nr_dirty : 0;
104 }
105 
106 #ifdef CONFIG_DEBUG_FS
107 static DEFINE_PER_CPU(long, mg_ctime_updates);
108 static DEFINE_PER_CPU(long, mg_fine_stamps);
109 static DEFINE_PER_CPU(long, mg_ctime_swaps);
110 
111 static unsigned long get_mg_ctime_updates(void)
112 {
113 	unsigned long sum = 0;
114 	int i;
115 
116 	for_each_possible_cpu(i)
117 		sum += data_race(per_cpu(mg_ctime_updates, i));
118 	return sum;
119 }
120 
121 static unsigned long get_mg_fine_stamps(void)
122 {
123 	unsigned long sum = 0;
124 	int i;
125 
126 	for_each_possible_cpu(i)
127 		sum += data_race(per_cpu(mg_fine_stamps, i));
128 	return sum;
129 }
130 
131 static unsigned long get_mg_ctime_swaps(void)
132 {
133 	unsigned long sum = 0;
134 	int i;
135 
136 	for_each_possible_cpu(i)
137 		sum += data_race(per_cpu(mg_ctime_swaps, i));
138 	return sum;
139 }
140 
141 #define mgtime_counter_inc(__var)	this_cpu_inc(__var)
142 
143 static int mgts_show(struct seq_file *s, void *p)
144 {
145 	unsigned long ctime_updates = get_mg_ctime_updates();
146 	unsigned long ctime_swaps = get_mg_ctime_swaps();
147 	unsigned long fine_stamps = get_mg_fine_stamps();
148 	unsigned long floor_swaps = timekeeping_get_mg_floor_swaps();
149 
150 	seq_printf(s, "%lu %lu %lu %lu\n",
151 		   ctime_updates, ctime_swaps, fine_stamps, floor_swaps);
152 	return 0;
153 }
154 
155 DEFINE_SHOW_ATTRIBUTE(mgts);
156 
157 static int __init mg_debugfs_init(void)
158 {
159 	debugfs_create_file("multigrain_timestamps", S_IFREG | S_IRUGO, NULL, NULL, &mgts_fops);
160 	return 0;
161 }
162 late_initcall(mg_debugfs_init);
163 
164 #else /* ! CONFIG_DEBUG_FS */
165 
166 #define mgtime_counter_inc(__var)	do { } while (0)
167 
168 #endif /* CONFIG_DEBUG_FS */
169 
170 /*
171  * Handle nr_inode sysctl
172  */
173 #ifdef CONFIG_SYSCTL
174 /*
175  * Statistics gathering..
176  */
177 static struct inodes_stat_t inodes_stat;
178 
179 static int proc_nr_inodes(const struct ctl_table *table, int write, void *buffer,
180 			  size_t *lenp, loff_t *ppos)
181 {
182 	inodes_stat.nr_inodes = get_nr_inodes();
183 	inodes_stat.nr_unused = get_nr_inodes_unused();
184 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
185 }
186 
187 static struct ctl_table inodes_sysctls[] = {
188 	{
189 		.procname	= "inode-nr",
190 		.data		= &inodes_stat,
191 		.maxlen		= 2*sizeof(long),
192 		.mode		= 0444,
193 		.proc_handler	= proc_nr_inodes,
194 	},
195 	{
196 		.procname	= "inode-state",
197 		.data		= &inodes_stat,
198 		.maxlen		= 7*sizeof(long),
199 		.mode		= 0444,
200 		.proc_handler	= proc_nr_inodes,
201 	},
202 };
203 
204 static int __init init_fs_inode_sysctls(void)
205 {
206 	register_sysctl_init("fs", inodes_sysctls);
207 	return 0;
208 }
209 early_initcall(init_fs_inode_sysctls);
210 #endif
211 
212 static int no_open(struct inode *inode, struct file *file)
213 {
214 	return -ENXIO;
215 }
216 
217 /**
218  * inode_init_always_gfp - perform inode structure initialisation
219  * @sb: superblock inode belongs to
220  * @inode: inode to initialise
221  * @gfp: allocation flags
222  *
223  * These are initializations that need to be done on every inode
224  * allocation as the fields are not initialised by slab allocation.
225  * If there are additional allocations required @gfp is used.
226  */
227 int inode_init_always_gfp(struct super_block *sb, struct inode *inode, gfp_t gfp)
228 {
229 	static const struct inode_operations empty_iops;
230 	static const struct file_operations no_open_fops = {.open = no_open};
231 	struct address_space *const mapping = &inode->i_data;
232 
233 	inode->i_sb = sb;
234 	inode->i_blkbits = sb->s_blocksize_bits;
235 	inode->i_flags = 0;
236 	inode->i_state = 0;
237 	atomic64_set(&inode->i_sequence, 0);
238 	atomic_set(&inode->i_count, 1);
239 	inode->i_op = &empty_iops;
240 	inode->i_fop = &no_open_fops;
241 	inode->i_ino = 0;
242 	inode->__i_nlink = 1;
243 	inode->i_opflags = 0;
244 	if (sb->s_xattr)
245 		inode->i_opflags |= IOP_XATTR;
246 	if (sb->s_type->fs_flags & FS_MGTIME)
247 		inode->i_opflags |= IOP_MGTIME;
248 	i_uid_write(inode, 0);
249 	i_gid_write(inode, 0);
250 	atomic_set(&inode->i_writecount, 0);
251 	inode->i_size = 0;
252 	inode->i_write_hint = WRITE_LIFE_NOT_SET;
253 	inode->i_blocks = 0;
254 	inode->i_bytes = 0;
255 	inode->i_generation = 0;
256 	inode->i_pipe = NULL;
257 	inode->i_cdev = NULL;
258 	inode->i_link = NULL;
259 	inode->i_dir_seq = 0;
260 	inode->i_rdev = 0;
261 	inode->dirtied_when = 0;
262 
263 #ifdef CONFIG_CGROUP_WRITEBACK
264 	inode->i_wb_frn_winner = 0;
265 	inode->i_wb_frn_avg_time = 0;
266 	inode->i_wb_frn_history = 0;
267 #endif
268 
269 	spin_lock_init(&inode->i_lock);
270 	lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
271 
272 	init_rwsem(&inode->i_rwsem);
273 	lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
274 
275 	atomic_set(&inode->i_dio_count, 0);
276 
277 	mapping->a_ops = &empty_aops;
278 	mapping->host = inode;
279 	mapping->flags = 0;
280 	mapping->wb_err = 0;
281 	atomic_set(&mapping->i_mmap_writable, 0);
282 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
283 	atomic_set(&mapping->nr_thps, 0);
284 #endif
285 	mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
286 	mapping->i_private_data = NULL;
287 	mapping->writeback_index = 0;
288 	init_rwsem(&mapping->invalidate_lock);
289 	lockdep_set_class_and_name(&mapping->invalidate_lock,
290 				   &sb->s_type->invalidate_lock_key,
291 				   "mapping.invalidate_lock");
292 	if (sb->s_iflags & SB_I_STABLE_WRITES)
293 		mapping_set_stable_writes(mapping);
294 	inode->i_private = NULL;
295 	inode->i_mapping = mapping;
296 	INIT_HLIST_HEAD(&inode->i_dentry);	/* buggered by rcu freeing */
297 #ifdef CONFIG_FS_POSIX_ACL
298 	inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
299 #endif
300 
301 #ifdef CONFIG_FSNOTIFY
302 	inode->i_fsnotify_mask = 0;
303 #endif
304 	inode->i_flctx = NULL;
305 
306 	if (unlikely(security_inode_alloc(inode, gfp)))
307 		return -ENOMEM;
308 
309 	this_cpu_inc(nr_inodes);
310 
311 	return 0;
312 }
313 EXPORT_SYMBOL(inode_init_always_gfp);
314 
315 void free_inode_nonrcu(struct inode *inode)
316 {
317 	kmem_cache_free(inode_cachep, inode);
318 }
319 EXPORT_SYMBOL(free_inode_nonrcu);
320 
321 static void i_callback(struct rcu_head *head)
322 {
323 	struct inode *inode = container_of(head, struct inode, i_rcu);
324 	if (inode->free_inode)
325 		inode->free_inode(inode);
326 	else
327 		free_inode_nonrcu(inode);
328 }
329 
330 static struct inode *alloc_inode(struct super_block *sb)
331 {
332 	const struct super_operations *ops = sb->s_op;
333 	struct inode *inode;
334 
335 	if (ops->alloc_inode)
336 		inode = ops->alloc_inode(sb);
337 	else
338 		inode = alloc_inode_sb(sb, inode_cachep, GFP_KERNEL);
339 
340 	if (!inode)
341 		return NULL;
342 
343 	if (unlikely(inode_init_always(sb, inode))) {
344 		if (ops->destroy_inode) {
345 			ops->destroy_inode(inode);
346 			if (!ops->free_inode)
347 				return NULL;
348 		}
349 		inode->free_inode = ops->free_inode;
350 		i_callback(&inode->i_rcu);
351 		return NULL;
352 	}
353 
354 	return inode;
355 }
356 
357 void __destroy_inode(struct inode *inode)
358 {
359 	BUG_ON(inode_has_buffers(inode));
360 	inode_detach_wb(inode);
361 	security_inode_free(inode);
362 	fsnotify_inode_delete(inode);
363 	locks_free_lock_context(inode);
364 	if (!inode->i_nlink) {
365 		WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
366 		atomic_long_dec(&inode->i_sb->s_remove_count);
367 	}
368 
369 #ifdef CONFIG_FS_POSIX_ACL
370 	if (inode->i_acl && !is_uncached_acl(inode->i_acl))
371 		posix_acl_release(inode->i_acl);
372 	if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
373 		posix_acl_release(inode->i_default_acl);
374 #endif
375 	this_cpu_dec(nr_inodes);
376 }
377 EXPORT_SYMBOL(__destroy_inode);
378 
379 static void destroy_inode(struct inode *inode)
380 {
381 	const struct super_operations *ops = inode->i_sb->s_op;
382 
383 	BUG_ON(!list_empty(&inode->i_lru));
384 	__destroy_inode(inode);
385 	if (ops->destroy_inode) {
386 		ops->destroy_inode(inode);
387 		if (!ops->free_inode)
388 			return;
389 	}
390 	inode->free_inode = ops->free_inode;
391 	call_rcu(&inode->i_rcu, i_callback);
392 }
393 
394 /**
395  * drop_nlink - directly drop an inode's link count
396  * @inode: inode
397  *
398  * This is a low-level filesystem helper to replace any
399  * direct filesystem manipulation of i_nlink.  In cases
400  * where we are attempting to track writes to the
401  * filesystem, a decrement to zero means an imminent
402  * write when the file is truncated and actually unlinked
403  * on the filesystem.
404  */
405 void drop_nlink(struct inode *inode)
406 {
407 	WARN_ON(inode->i_nlink == 0);
408 	inode->__i_nlink--;
409 	if (!inode->i_nlink)
410 		atomic_long_inc(&inode->i_sb->s_remove_count);
411 }
412 EXPORT_SYMBOL(drop_nlink);
413 
414 /**
415  * clear_nlink - directly zero an inode's link count
416  * @inode: inode
417  *
418  * This is a low-level filesystem helper to replace any
419  * direct filesystem manipulation of i_nlink.  See
420  * drop_nlink() for why we care about i_nlink hitting zero.
421  */
422 void clear_nlink(struct inode *inode)
423 {
424 	if (inode->i_nlink) {
425 		inode->__i_nlink = 0;
426 		atomic_long_inc(&inode->i_sb->s_remove_count);
427 	}
428 }
429 EXPORT_SYMBOL(clear_nlink);
430 
431 /**
432  * set_nlink - directly set an inode's link count
433  * @inode: inode
434  * @nlink: new nlink (should be non-zero)
435  *
436  * This is a low-level filesystem helper to replace any
437  * direct filesystem manipulation of i_nlink.
438  */
439 void set_nlink(struct inode *inode, unsigned int nlink)
440 {
441 	if (!nlink) {
442 		clear_nlink(inode);
443 	} else {
444 		/* Yes, some filesystems do change nlink from zero to one */
445 		if (inode->i_nlink == 0)
446 			atomic_long_dec(&inode->i_sb->s_remove_count);
447 
448 		inode->__i_nlink = nlink;
449 	}
450 }
451 EXPORT_SYMBOL(set_nlink);
452 
453 /**
454  * inc_nlink - directly increment an inode's link count
455  * @inode: inode
456  *
457  * This is a low-level filesystem helper to replace any
458  * direct filesystem manipulation of i_nlink.  Currently,
459  * it is only here for parity with dec_nlink().
460  */
461 void inc_nlink(struct inode *inode)
462 {
463 	if (unlikely(inode->i_nlink == 0)) {
464 		WARN_ON(!(inode->i_state & I_LINKABLE));
465 		atomic_long_dec(&inode->i_sb->s_remove_count);
466 	}
467 
468 	inode->__i_nlink++;
469 }
470 EXPORT_SYMBOL(inc_nlink);
471 
472 static void __address_space_init_once(struct address_space *mapping)
473 {
474 	xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
475 	init_rwsem(&mapping->i_mmap_rwsem);
476 	INIT_LIST_HEAD(&mapping->i_private_list);
477 	spin_lock_init(&mapping->i_private_lock);
478 	mapping->i_mmap = RB_ROOT_CACHED;
479 }
480 
481 void address_space_init_once(struct address_space *mapping)
482 {
483 	memset(mapping, 0, sizeof(*mapping));
484 	__address_space_init_once(mapping);
485 }
486 EXPORT_SYMBOL(address_space_init_once);
487 
488 /*
489  * These are initializations that only need to be done
490  * once, because the fields are idempotent across use
491  * of the inode, so let the slab aware of that.
492  */
493 void inode_init_once(struct inode *inode)
494 {
495 	memset(inode, 0, sizeof(*inode));
496 	INIT_HLIST_NODE(&inode->i_hash);
497 	INIT_LIST_HEAD(&inode->i_devices);
498 	INIT_LIST_HEAD(&inode->i_io_list);
499 	INIT_LIST_HEAD(&inode->i_wb_list);
500 	INIT_LIST_HEAD(&inode->i_lru);
501 	INIT_LIST_HEAD(&inode->i_sb_list);
502 	__address_space_init_once(&inode->i_data);
503 	i_size_ordered_init(inode);
504 }
505 EXPORT_SYMBOL(inode_init_once);
506 
507 static void init_once(void *foo)
508 {
509 	struct inode *inode = (struct inode *) foo;
510 
511 	inode_init_once(inode);
512 }
513 
514 /*
515  * get additional reference to inode; caller must already hold one.
516  */
517 void ihold(struct inode *inode)
518 {
519 	WARN_ON(atomic_inc_return(&inode->i_count) < 2);
520 }
521 EXPORT_SYMBOL(ihold);
522 
523 static void __inode_add_lru(struct inode *inode, bool rotate)
524 {
525 	if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
526 		return;
527 	if (atomic_read(&inode->i_count))
528 		return;
529 	if (!(inode->i_sb->s_flags & SB_ACTIVE))
530 		return;
531 	if (!mapping_shrinkable(&inode->i_data))
532 		return;
533 
534 	if (list_lru_add_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
535 		this_cpu_inc(nr_unused);
536 	else if (rotate)
537 		inode->i_state |= I_REFERENCED;
538 }
539 
540 struct wait_queue_head *inode_bit_waitqueue(struct wait_bit_queue_entry *wqe,
541 					    struct inode *inode, u32 bit)
542 {
543         void *bit_address;
544 
545         bit_address = inode_state_wait_address(inode, bit);
546         init_wait_var_entry(wqe, bit_address, 0);
547         return __var_waitqueue(bit_address);
548 }
549 EXPORT_SYMBOL(inode_bit_waitqueue);
550 
551 /*
552  * Add inode to LRU if needed (inode is unused and clean).
553  *
554  * Needs inode->i_lock held.
555  */
556 void inode_add_lru(struct inode *inode)
557 {
558 	__inode_add_lru(inode, false);
559 }
560 
561 static void inode_lru_list_del(struct inode *inode)
562 {
563 	if (list_lru_del_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
564 		this_cpu_dec(nr_unused);
565 }
566 
567 static void inode_pin_lru_isolating(struct inode *inode)
568 {
569 	lockdep_assert_held(&inode->i_lock);
570 	WARN_ON(inode->i_state & (I_LRU_ISOLATING | I_FREEING | I_WILL_FREE));
571 	inode->i_state |= I_LRU_ISOLATING;
572 }
573 
574 static void inode_unpin_lru_isolating(struct inode *inode)
575 {
576 	spin_lock(&inode->i_lock);
577 	WARN_ON(!(inode->i_state & I_LRU_ISOLATING));
578 	inode->i_state &= ~I_LRU_ISOLATING;
579 	/* Called with inode->i_lock which ensures memory ordering. */
580 	inode_wake_up_bit(inode, __I_LRU_ISOLATING);
581 	spin_unlock(&inode->i_lock);
582 }
583 
584 static void inode_wait_for_lru_isolating(struct inode *inode)
585 {
586 	struct wait_bit_queue_entry wqe;
587 	struct wait_queue_head *wq_head;
588 
589 	lockdep_assert_held(&inode->i_lock);
590 	if (!(inode->i_state & I_LRU_ISOLATING))
591 		return;
592 
593 	wq_head = inode_bit_waitqueue(&wqe, inode, __I_LRU_ISOLATING);
594 	for (;;) {
595 		prepare_to_wait_event(wq_head, &wqe.wq_entry, TASK_UNINTERRUPTIBLE);
596 		/*
597 		 * Checking I_LRU_ISOLATING with inode->i_lock guarantees
598 		 * memory ordering.
599 		 */
600 		if (!(inode->i_state & I_LRU_ISOLATING))
601 			break;
602 		spin_unlock(&inode->i_lock);
603 		schedule();
604 		spin_lock(&inode->i_lock);
605 	}
606 	finish_wait(wq_head, &wqe.wq_entry);
607 	WARN_ON(inode->i_state & I_LRU_ISOLATING);
608 }
609 
610 /**
611  * inode_sb_list_add - add inode to the superblock list of inodes
612  * @inode: inode to add
613  */
614 void inode_sb_list_add(struct inode *inode)
615 {
616 	spin_lock(&inode->i_sb->s_inode_list_lock);
617 	list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
618 	spin_unlock(&inode->i_sb->s_inode_list_lock);
619 }
620 EXPORT_SYMBOL_GPL(inode_sb_list_add);
621 
622 static inline void inode_sb_list_del(struct inode *inode)
623 {
624 	if (!list_empty(&inode->i_sb_list)) {
625 		spin_lock(&inode->i_sb->s_inode_list_lock);
626 		list_del_init(&inode->i_sb_list);
627 		spin_unlock(&inode->i_sb->s_inode_list_lock);
628 	}
629 }
630 
631 static unsigned long hash(struct super_block *sb, unsigned long hashval)
632 {
633 	unsigned long tmp;
634 
635 	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
636 			L1_CACHE_BYTES;
637 	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
638 	return tmp & i_hash_mask;
639 }
640 
641 /**
642  *	__insert_inode_hash - hash an inode
643  *	@inode: unhashed inode
644  *	@hashval: unsigned long value used to locate this object in the
645  *		inode_hashtable.
646  *
647  *	Add an inode to the inode hash for this superblock.
648  */
649 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
650 {
651 	struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
652 
653 	spin_lock(&inode_hash_lock);
654 	spin_lock(&inode->i_lock);
655 	hlist_add_head_rcu(&inode->i_hash, b);
656 	spin_unlock(&inode->i_lock);
657 	spin_unlock(&inode_hash_lock);
658 }
659 EXPORT_SYMBOL(__insert_inode_hash);
660 
661 /**
662  *	__remove_inode_hash - remove an inode from the hash
663  *	@inode: inode to unhash
664  *
665  *	Remove an inode from the superblock.
666  */
667 void __remove_inode_hash(struct inode *inode)
668 {
669 	spin_lock(&inode_hash_lock);
670 	spin_lock(&inode->i_lock);
671 	hlist_del_init_rcu(&inode->i_hash);
672 	spin_unlock(&inode->i_lock);
673 	spin_unlock(&inode_hash_lock);
674 }
675 EXPORT_SYMBOL(__remove_inode_hash);
676 
677 void dump_mapping(const struct address_space *mapping)
678 {
679 	struct inode *host;
680 	const struct address_space_operations *a_ops;
681 	struct hlist_node *dentry_first;
682 	struct dentry *dentry_ptr;
683 	struct dentry dentry;
684 	char fname[64] = {};
685 	unsigned long ino;
686 
687 	/*
688 	 * If mapping is an invalid pointer, we don't want to crash
689 	 * accessing it, so probe everything depending on it carefully.
690 	 */
691 	if (get_kernel_nofault(host, &mapping->host) ||
692 	    get_kernel_nofault(a_ops, &mapping->a_ops)) {
693 		pr_warn("invalid mapping:%px\n", mapping);
694 		return;
695 	}
696 
697 	if (!host) {
698 		pr_warn("aops:%ps\n", a_ops);
699 		return;
700 	}
701 
702 	if (get_kernel_nofault(dentry_first, &host->i_dentry.first) ||
703 	    get_kernel_nofault(ino, &host->i_ino)) {
704 		pr_warn("aops:%ps invalid inode:%px\n", a_ops, host);
705 		return;
706 	}
707 
708 	if (!dentry_first) {
709 		pr_warn("aops:%ps ino:%lx\n", a_ops, ino);
710 		return;
711 	}
712 
713 	dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
714 	if (get_kernel_nofault(dentry, dentry_ptr) ||
715 	    !dentry.d_parent || !dentry.d_name.name) {
716 		pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
717 				a_ops, ino, dentry_ptr);
718 		return;
719 	}
720 
721 	if (strncpy_from_kernel_nofault(fname, dentry.d_name.name, 63) < 0)
722 		strscpy(fname, "<invalid>");
723 	/*
724 	 * Even if strncpy_from_kernel_nofault() succeeded,
725 	 * the fname could be unreliable
726 	 */
727 	pr_warn("aops:%ps ino:%lx dentry name(?):\"%s\"\n",
728 		a_ops, ino, fname);
729 }
730 
731 void clear_inode(struct inode *inode)
732 {
733 	/*
734 	 * We have to cycle the i_pages lock here because reclaim can be in the
735 	 * process of removing the last page (in __filemap_remove_folio())
736 	 * and we must not free the mapping under it.
737 	 */
738 	xa_lock_irq(&inode->i_data.i_pages);
739 	BUG_ON(inode->i_data.nrpages);
740 	/*
741 	 * Almost always, mapping_empty(&inode->i_data) here; but there are
742 	 * two known and long-standing ways in which nodes may get left behind
743 	 * (when deep radix-tree node allocation failed partway; or when THP
744 	 * collapse_file() failed). Until those two known cases are cleaned up,
745 	 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
746 	 * nor even WARN_ON(!mapping_empty).
747 	 */
748 	xa_unlock_irq(&inode->i_data.i_pages);
749 	BUG_ON(!list_empty(&inode->i_data.i_private_list));
750 	BUG_ON(!(inode->i_state & I_FREEING));
751 	BUG_ON(inode->i_state & I_CLEAR);
752 	BUG_ON(!list_empty(&inode->i_wb_list));
753 	/* don't need i_lock here, no concurrent mods to i_state */
754 	inode->i_state = I_FREEING | I_CLEAR;
755 }
756 EXPORT_SYMBOL(clear_inode);
757 
758 /*
759  * Free the inode passed in, removing it from the lists it is still connected
760  * to. We remove any pages still attached to the inode and wait for any IO that
761  * is still in progress before finally destroying the inode.
762  *
763  * An inode must already be marked I_FREEING so that we avoid the inode being
764  * moved back onto lists if we race with other code that manipulates the lists
765  * (e.g. writeback_single_inode). The caller is responsible for setting this.
766  *
767  * An inode must already be removed from the LRU list before being evicted from
768  * the cache. This should occur atomically with setting the I_FREEING state
769  * flag, so no inodes here should ever be on the LRU when being evicted.
770  */
771 static void evict(struct inode *inode)
772 {
773 	const struct super_operations *op = inode->i_sb->s_op;
774 
775 	BUG_ON(!(inode->i_state & I_FREEING));
776 	BUG_ON(!list_empty(&inode->i_lru));
777 
778 	if (!list_empty(&inode->i_io_list))
779 		inode_io_list_del(inode);
780 
781 	inode_sb_list_del(inode);
782 
783 	spin_lock(&inode->i_lock);
784 	inode_wait_for_lru_isolating(inode);
785 
786 	/*
787 	 * Wait for flusher thread to be done with the inode so that filesystem
788 	 * does not start destroying it while writeback is still running. Since
789 	 * the inode has I_FREEING set, flusher thread won't start new work on
790 	 * the inode.  We just have to wait for running writeback to finish.
791 	 */
792 	inode_wait_for_writeback(inode);
793 	spin_unlock(&inode->i_lock);
794 
795 	if (op->evict_inode) {
796 		op->evict_inode(inode);
797 	} else {
798 		truncate_inode_pages_final(&inode->i_data);
799 		clear_inode(inode);
800 	}
801 	if (S_ISCHR(inode->i_mode) && inode->i_cdev)
802 		cd_forget(inode);
803 
804 	remove_inode_hash(inode);
805 
806 	/*
807 	 * Wake up waiters in __wait_on_freeing_inode().
808 	 *
809 	 * Lockless hash lookup may end up finding the inode before we removed
810 	 * it above, but only lock it *after* we are done with the wakeup below.
811 	 * In this case the potential waiter cannot safely block.
812 	 *
813 	 * The inode being unhashed after the call to remove_inode_hash() is
814 	 * used as an indicator whether blocking on it is safe.
815 	 */
816 	spin_lock(&inode->i_lock);
817 	/*
818 	 * Pairs with the barrier in prepare_to_wait_event() to make sure
819 	 * ___wait_var_event() either sees the bit cleared or
820 	 * waitqueue_active() check in wake_up_var() sees the waiter.
821 	 */
822 	smp_mb__after_spinlock();
823 	inode_wake_up_bit(inode, __I_NEW);
824 	BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
825 	spin_unlock(&inode->i_lock);
826 
827 	destroy_inode(inode);
828 }
829 
830 /*
831  * dispose_list - dispose of the contents of a local list
832  * @head: the head of the list to free
833  *
834  * Dispose-list gets a local list with local inodes in it, so it doesn't
835  * need to worry about list corruption and SMP locks.
836  */
837 static void dispose_list(struct list_head *head)
838 {
839 	while (!list_empty(head)) {
840 		struct inode *inode;
841 
842 		inode = list_first_entry(head, struct inode, i_lru);
843 		list_del_init(&inode->i_lru);
844 
845 		evict(inode);
846 		cond_resched();
847 	}
848 }
849 
850 /**
851  * evict_inodes	- evict all evictable inodes for a superblock
852  * @sb:		superblock to operate on
853  *
854  * Make sure that no inodes with zero refcount are retained.  This is
855  * called by superblock shutdown after having SB_ACTIVE flag removed,
856  * so any inode reaching zero refcount during or after that call will
857  * be immediately evicted.
858  */
859 void evict_inodes(struct super_block *sb)
860 {
861 	struct inode *inode, *next;
862 	LIST_HEAD(dispose);
863 
864 again:
865 	spin_lock(&sb->s_inode_list_lock);
866 	list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
867 		if (atomic_read(&inode->i_count))
868 			continue;
869 
870 		spin_lock(&inode->i_lock);
871 		if (atomic_read(&inode->i_count)) {
872 			spin_unlock(&inode->i_lock);
873 			continue;
874 		}
875 		if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
876 			spin_unlock(&inode->i_lock);
877 			continue;
878 		}
879 
880 		inode->i_state |= I_FREEING;
881 		inode_lru_list_del(inode);
882 		spin_unlock(&inode->i_lock);
883 		list_add(&inode->i_lru, &dispose);
884 
885 		/*
886 		 * We can have a ton of inodes to evict at unmount time given
887 		 * enough memory, check to see if we need to go to sleep for a
888 		 * bit so we don't livelock.
889 		 */
890 		if (need_resched()) {
891 			spin_unlock(&sb->s_inode_list_lock);
892 			cond_resched();
893 			dispose_list(&dispose);
894 			goto again;
895 		}
896 	}
897 	spin_unlock(&sb->s_inode_list_lock);
898 
899 	dispose_list(&dispose);
900 }
901 EXPORT_SYMBOL_GPL(evict_inodes);
902 
903 /**
904  * invalidate_inodes	- attempt to free all inodes on a superblock
905  * @sb:		superblock to operate on
906  *
907  * Attempts to free all inodes (including dirty inodes) for a given superblock.
908  */
909 void invalidate_inodes(struct super_block *sb)
910 {
911 	struct inode *inode, *next;
912 	LIST_HEAD(dispose);
913 
914 again:
915 	spin_lock(&sb->s_inode_list_lock);
916 	list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
917 		spin_lock(&inode->i_lock);
918 		if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
919 			spin_unlock(&inode->i_lock);
920 			continue;
921 		}
922 		if (atomic_read(&inode->i_count)) {
923 			spin_unlock(&inode->i_lock);
924 			continue;
925 		}
926 
927 		inode->i_state |= I_FREEING;
928 		inode_lru_list_del(inode);
929 		spin_unlock(&inode->i_lock);
930 		list_add(&inode->i_lru, &dispose);
931 		if (need_resched()) {
932 			spin_unlock(&sb->s_inode_list_lock);
933 			cond_resched();
934 			dispose_list(&dispose);
935 			goto again;
936 		}
937 	}
938 	spin_unlock(&sb->s_inode_list_lock);
939 
940 	dispose_list(&dispose);
941 }
942 
943 /*
944  * Isolate the inode from the LRU in preparation for freeing it.
945  *
946  * If the inode has the I_REFERENCED flag set, then it means that it has been
947  * used recently - the flag is set in iput_final(). When we encounter such an
948  * inode, clear the flag and move it to the back of the LRU so it gets another
949  * pass through the LRU before it gets reclaimed. This is necessary because of
950  * the fact we are doing lazy LRU updates to minimise lock contention so the
951  * LRU does not have strict ordering. Hence we don't want to reclaim inodes
952  * with this flag set because they are the inodes that are out of order.
953  */
954 static enum lru_status inode_lru_isolate(struct list_head *item,
955 		struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
956 {
957 	struct list_head *freeable = arg;
958 	struct inode	*inode = container_of(item, struct inode, i_lru);
959 
960 	/*
961 	 * We are inverting the lru lock/inode->i_lock here, so use a
962 	 * trylock. If we fail to get the lock, just skip it.
963 	 */
964 	if (!spin_trylock(&inode->i_lock))
965 		return LRU_SKIP;
966 
967 	/*
968 	 * Inodes can get referenced, redirtied, or repopulated while
969 	 * they're already on the LRU, and this can make them
970 	 * unreclaimable for a while. Remove them lazily here; iput,
971 	 * sync, or the last page cache deletion will requeue them.
972 	 */
973 	if (atomic_read(&inode->i_count) ||
974 	    (inode->i_state & ~I_REFERENCED) ||
975 	    !mapping_shrinkable(&inode->i_data)) {
976 		list_lru_isolate(lru, &inode->i_lru);
977 		spin_unlock(&inode->i_lock);
978 		this_cpu_dec(nr_unused);
979 		return LRU_REMOVED;
980 	}
981 
982 	/* Recently referenced inodes get one more pass */
983 	if (inode->i_state & I_REFERENCED) {
984 		inode->i_state &= ~I_REFERENCED;
985 		spin_unlock(&inode->i_lock);
986 		return LRU_ROTATE;
987 	}
988 
989 	/*
990 	 * On highmem systems, mapping_shrinkable() permits dropping
991 	 * page cache in order to free up struct inodes: lowmem might
992 	 * be under pressure before the cache inside the highmem zone.
993 	 */
994 	if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
995 		inode_pin_lru_isolating(inode);
996 		spin_unlock(&inode->i_lock);
997 		spin_unlock(lru_lock);
998 		if (remove_inode_buffers(inode)) {
999 			unsigned long reap;
1000 			reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
1001 			if (current_is_kswapd())
1002 				__count_vm_events(KSWAPD_INODESTEAL, reap);
1003 			else
1004 				__count_vm_events(PGINODESTEAL, reap);
1005 			mm_account_reclaimed_pages(reap);
1006 		}
1007 		inode_unpin_lru_isolating(inode);
1008 		spin_lock(lru_lock);
1009 		return LRU_RETRY;
1010 	}
1011 
1012 	WARN_ON(inode->i_state & I_NEW);
1013 	inode->i_state |= I_FREEING;
1014 	list_lru_isolate_move(lru, &inode->i_lru, freeable);
1015 	spin_unlock(&inode->i_lock);
1016 
1017 	this_cpu_dec(nr_unused);
1018 	return LRU_REMOVED;
1019 }
1020 
1021 /*
1022  * Walk the superblock inode LRU for freeable inodes and attempt to free them.
1023  * This is called from the superblock shrinker function with a number of inodes
1024  * to trim from the LRU. Inodes to be freed are moved to a temporary list and
1025  * then are freed outside inode_lock by dispose_list().
1026  */
1027 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
1028 {
1029 	LIST_HEAD(freeable);
1030 	long freed;
1031 
1032 	freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
1033 				     inode_lru_isolate, &freeable);
1034 	dispose_list(&freeable);
1035 	return freed;
1036 }
1037 
1038 static void __wait_on_freeing_inode(struct inode *inode, bool is_inode_hash_locked);
1039 /*
1040  * Called with the inode lock held.
1041  */
1042 static struct inode *find_inode(struct super_block *sb,
1043 				struct hlist_head *head,
1044 				int (*test)(struct inode *, void *),
1045 				void *data, bool is_inode_hash_locked)
1046 {
1047 	struct inode *inode = NULL;
1048 
1049 	if (is_inode_hash_locked)
1050 		lockdep_assert_held(&inode_hash_lock);
1051 	else
1052 		lockdep_assert_not_held(&inode_hash_lock);
1053 
1054 	rcu_read_lock();
1055 repeat:
1056 	hlist_for_each_entry_rcu(inode, head, i_hash) {
1057 		if (inode->i_sb != sb)
1058 			continue;
1059 		if (!test(inode, data))
1060 			continue;
1061 		spin_lock(&inode->i_lock);
1062 		if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1063 			__wait_on_freeing_inode(inode, is_inode_hash_locked);
1064 			goto repeat;
1065 		}
1066 		if (unlikely(inode->i_state & I_CREATING)) {
1067 			spin_unlock(&inode->i_lock);
1068 			rcu_read_unlock();
1069 			return ERR_PTR(-ESTALE);
1070 		}
1071 		__iget(inode);
1072 		spin_unlock(&inode->i_lock);
1073 		rcu_read_unlock();
1074 		return inode;
1075 	}
1076 	rcu_read_unlock();
1077 	return NULL;
1078 }
1079 
1080 /*
1081  * find_inode_fast is the fast path version of find_inode, see the comment at
1082  * iget_locked for details.
1083  */
1084 static struct inode *find_inode_fast(struct super_block *sb,
1085 				struct hlist_head *head, unsigned long ino,
1086 				bool is_inode_hash_locked)
1087 {
1088 	struct inode *inode = NULL;
1089 
1090 	if (is_inode_hash_locked)
1091 		lockdep_assert_held(&inode_hash_lock);
1092 	else
1093 		lockdep_assert_not_held(&inode_hash_lock);
1094 
1095 	rcu_read_lock();
1096 repeat:
1097 	hlist_for_each_entry_rcu(inode, head, i_hash) {
1098 		if (inode->i_ino != ino)
1099 			continue;
1100 		if (inode->i_sb != sb)
1101 			continue;
1102 		spin_lock(&inode->i_lock);
1103 		if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1104 			__wait_on_freeing_inode(inode, is_inode_hash_locked);
1105 			goto repeat;
1106 		}
1107 		if (unlikely(inode->i_state & I_CREATING)) {
1108 			spin_unlock(&inode->i_lock);
1109 			rcu_read_unlock();
1110 			return ERR_PTR(-ESTALE);
1111 		}
1112 		__iget(inode);
1113 		spin_unlock(&inode->i_lock);
1114 		rcu_read_unlock();
1115 		return inode;
1116 	}
1117 	rcu_read_unlock();
1118 	return NULL;
1119 }
1120 
1121 /*
1122  * Each cpu owns a range of LAST_INO_BATCH numbers.
1123  * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
1124  * to renew the exhausted range.
1125  *
1126  * This does not significantly increase overflow rate because every CPU can
1127  * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
1128  * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
1129  * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
1130  * overflow rate by 2x, which does not seem too significant.
1131  *
1132  * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1133  * error if st_ino won't fit in target struct field. Use 32bit counter
1134  * here to attempt to avoid that.
1135  */
1136 #define LAST_INO_BATCH 1024
1137 static DEFINE_PER_CPU(unsigned int, last_ino);
1138 
1139 unsigned int get_next_ino(void)
1140 {
1141 	unsigned int *p = &get_cpu_var(last_ino);
1142 	unsigned int res = *p;
1143 
1144 #ifdef CONFIG_SMP
1145 	if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
1146 		static atomic_t shared_last_ino;
1147 		int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
1148 
1149 		res = next - LAST_INO_BATCH;
1150 	}
1151 #endif
1152 
1153 	res++;
1154 	/* get_next_ino should not provide a 0 inode number */
1155 	if (unlikely(!res))
1156 		res++;
1157 	*p = res;
1158 	put_cpu_var(last_ino);
1159 	return res;
1160 }
1161 EXPORT_SYMBOL(get_next_ino);
1162 
1163 /**
1164  *	new_inode_pseudo 	- obtain an inode
1165  *	@sb: superblock
1166  *
1167  *	Allocates a new inode for given superblock.
1168  *	Inode wont be chained in superblock s_inodes list
1169  *	This means :
1170  *	- fs can't be unmount
1171  *	- quotas, fsnotify, writeback can't work
1172  */
1173 struct inode *new_inode_pseudo(struct super_block *sb)
1174 {
1175 	return alloc_inode(sb);
1176 }
1177 
1178 /**
1179  *	new_inode 	- obtain an inode
1180  *	@sb: superblock
1181  *
1182  *	Allocates a new inode for given superblock. The default gfp_mask
1183  *	for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1184  *	If HIGHMEM pages are unsuitable or it is known that pages allocated
1185  *	for the page cache are not reclaimable or migratable,
1186  *	mapping_set_gfp_mask() must be called with suitable flags on the
1187  *	newly created inode's mapping
1188  *
1189  */
1190 struct inode *new_inode(struct super_block *sb)
1191 {
1192 	struct inode *inode;
1193 
1194 	inode = new_inode_pseudo(sb);
1195 	if (inode)
1196 		inode_sb_list_add(inode);
1197 	return inode;
1198 }
1199 EXPORT_SYMBOL(new_inode);
1200 
1201 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1202 void lockdep_annotate_inode_mutex_key(struct inode *inode)
1203 {
1204 	if (S_ISDIR(inode->i_mode)) {
1205 		struct file_system_type *type = inode->i_sb->s_type;
1206 
1207 		/* Set new key only if filesystem hasn't already changed it */
1208 		if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1209 			/*
1210 			 * ensure nobody is actually holding i_mutex
1211 			 */
1212 			// mutex_destroy(&inode->i_mutex);
1213 			init_rwsem(&inode->i_rwsem);
1214 			lockdep_set_class(&inode->i_rwsem,
1215 					  &type->i_mutex_dir_key);
1216 		}
1217 	}
1218 }
1219 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1220 #endif
1221 
1222 /**
1223  * unlock_new_inode - clear the I_NEW state and wake up any waiters
1224  * @inode:	new inode to unlock
1225  *
1226  * Called when the inode is fully initialised to clear the new state of the
1227  * inode and wake up anyone waiting for the inode to finish initialisation.
1228  */
1229 void unlock_new_inode(struct inode *inode)
1230 {
1231 	lockdep_annotate_inode_mutex_key(inode);
1232 	spin_lock(&inode->i_lock);
1233 	WARN_ON(!(inode->i_state & I_NEW));
1234 	inode->i_state &= ~I_NEW & ~I_CREATING;
1235 	/*
1236 	 * Pairs with the barrier in prepare_to_wait_event() to make sure
1237 	 * ___wait_var_event() either sees the bit cleared or
1238 	 * waitqueue_active() check in wake_up_var() sees the waiter.
1239 	 */
1240 	smp_mb();
1241 	inode_wake_up_bit(inode, __I_NEW);
1242 	spin_unlock(&inode->i_lock);
1243 }
1244 EXPORT_SYMBOL(unlock_new_inode);
1245 
1246 void discard_new_inode(struct inode *inode)
1247 {
1248 	lockdep_annotate_inode_mutex_key(inode);
1249 	spin_lock(&inode->i_lock);
1250 	WARN_ON(!(inode->i_state & I_NEW));
1251 	inode->i_state &= ~I_NEW;
1252 	/*
1253 	 * Pairs with the barrier in prepare_to_wait_event() to make sure
1254 	 * ___wait_var_event() either sees the bit cleared or
1255 	 * waitqueue_active() check in wake_up_var() sees the waiter.
1256 	 */
1257 	smp_mb();
1258 	inode_wake_up_bit(inode, __I_NEW);
1259 	spin_unlock(&inode->i_lock);
1260 	iput(inode);
1261 }
1262 EXPORT_SYMBOL(discard_new_inode);
1263 
1264 /**
1265  * lock_two_nondirectories - take two i_mutexes on non-directory objects
1266  *
1267  * Lock any non-NULL argument. Passed objects must not be directories.
1268  * Zero, one or two objects may be locked by this function.
1269  *
1270  * @inode1: first inode to lock
1271  * @inode2: second inode to lock
1272  */
1273 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1274 {
1275 	if (inode1)
1276 		WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1277 	if (inode2)
1278 		WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1279 	if (inode1 > inode2)
1280 		swap(inode1, inode2);
1281 	if (inode1)
1282 		inode_lock(inode1);
1283 	if (inode2 && inode2 != inode1)
1284 		inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1285 }
1286 EXPORT_SYMBOL(lock_two_nondirectories);
1287 
1288 /**
1289  * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1290  * @inode1: first inode to unlock
1291  * @inode2: second inode to unlock
1292  */
1293 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1294 {
1295 	if (inode1) {
1296 		WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1297 		inode_unlock(inode1);
1298 	}
1299 	if (inode2 && inode2 != inode1) {
1300 		WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1301 		inode_unlock(inode2);
1302 	}
1303 }
1304 EXPORT_SYMBOL(unlock_two_nondirectories);
1305 
1306 /**
1307  * inode_insert5 - obtain an inode from a mounted file system
1308  * @inode:	pre-allocated inode to use for insert to cache
1309  * @hashval:	hash value (usually inode number) to get
1310  * @test:	callback used for comparisons between inodes
1311  * @set:	callback used to initialize a new struct inode
1312  * @data:	opaque data pointer to pass to @test and @set
1313  *
1314  * Search for the inode specified by @hashval and @data in the inode cache,
1315  * and if present return it with an increased reference count. This is a
1316  * variant of iget5_locked() that doesn't allocate an inode.
1317  *
1318  * If the inode is not present in the cache, insert the pre-allocated inode and
1319  * return it locked, hashed, and with the I_NEW flag set. The file system gets
1320  * to fill it in before unlocking it via unlock_new_inode().
1321  *
1322  * Note that both @test and @set are called with the inode_hash_lock held, so
1323  * they can't sleep.
1324  */
1325 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1326 			    int (*test)(struct inode *, void *),
1327 			    int (*set)(struct inode *, void *), void *data)
1328 {
1329 	struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1330 	struct inode *old;
1331 
1332 again:
1333 	spin_lock(&inode_hash_lock);
1334 	old = find_inode(inode->i_sb, head, test, data, true);
1335 	if (unlikely(old)) {
1336 		/*
1337 		 * Uhhuh, somebody else created the same inode under us.
1338 		 * Use the old inode instead of the preallocated one.
1339 		 */
1340 		spin_unlock(&inode_hash_lock);
1341 		if (IS_ERR(old))
1342 			return NULL;
1343 		wait_on_inode(old);
1344 		if (unlikely(inode_unhashed(old))) {
1345 			iput(old);
1346 			goto again;
1347 		}
1348 		return old;
1349 	}
1350 
1351 	if (set && unlikely(set(inode, data))) {
1352 		inode = NULL;
1353 		goto unlock;
1354 	}
1355 
1356 	/*
1357 	 * Return the locked inode with I_NEW set, the
1358 	 * caller is responsible for filling in the contents
1359 	 */
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 
1365 	/*
1366 	 * Add inode to the sb list if it's not already. It has I_NEW at this
1367 	 * point, so it should be safe to test i_sb_list locklessly.
1368 	 */
1369 	if (list_empty(&inode->i_sb_list))
1370 		inode_sb_list_add(inode);
1371 unlock:
1372 	spin_unlock(&inode_hash_lock);
1373 
1374 	return inode;
1375 }
1376 EXPORT_SYMBOL(inode_insert5);
1377 
1378 /**
1379  * iget5_locked - obtain an inode from a mounted file system
1380  * @sb:		super block of file system
1381  * @hashval:	hash value (usually inode number) to get
1382  * @test:	callback used for comparisons between inodes
1383  * @set:	callback used to initialize a new struct inode
1384  * @data:	opaque data pointer to pass to @test and @set
1385  *
1386  * Search for the inode specified by @hashval and @data in the inode cache,
1387  * and if present return it with an increased reference count. This is a
1388  * generalized version of iget_locked() for file systems where the inode
1389  * number is not sufficient for unique identification of an inode.
1390  *
1391  * If the inode is not present in the cache, allocate and insert a new inode
1392  * and return it locked, hashed, and with the I_NEW flag set. The file system
1393  * gets to fill it in before unlocking it via unlock_new_inode().
1394  *
1395  * Note that both @test and @set are called with the inode_hash_lock held, so
1396  * they can't sleep.
1397  */
1398 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1399 		int (*test)(struct inode *, void *),
1400 		int (*set)(struct inode *, void *), void *data)
1401 {
1402 	struct inode *inode = ilookup5(sb, hashval, test, data);
1403 
1404 	if (!inode) {
1405 		struct inode *new = alloc_inode(sb);
1406 
1407 		if (new) {
1408 			inode = inode_insert5(new, hashval, test, set, data);
1409 			if (unlikely(inode != new))
1410 				destroy_inode(new);
1411 		}
1412 	}
1413 	return inode;
1414 }
1415 EXPORT_SYMBOL(iget5_locked);
1416 
1417 /**
1418  * iget5_locked_rcu - obtain an inode from a mounted file system
1419  * @sb:		super block of file system
1420  * @hashval:	hash value (usually inode number) to get
1421  * @test:	callback used for comparisons between inodes
1422  * @set:	callback used to initialize a new struct inode
1423  * @data:	opaque data pointer to pass to @test and @set
1424  *
1425  * This is equivalent to iget5_locked, except the @test callback must
1426  * tolerate the inode not being stable, including being mid-teardown.
1427  */
1428 struct inode *iget5_locked_rcu(struct super_block *sb, unsigned long hashval,
1429 		int (*test)(struct inode *, void *),
1430 		int (*set)(struct inode *, void *), void *data)
1431 {
1432 	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1433 	struct inode *inode, *new;
1434 
1435 again:
1436 	inode = find_inode(sb, head, test, data, false);
1437 	if (inode) {
1438 		if (IS_ERR(inode))
1439 			return NULL;
1440 		wait_on_inode(inode);
1441 		if (unlikely(inode_unhashed(inode))) {
1442 			iput(inode);
1443 			goto again;
1444 		}
1445 		return inode;
1446 	}
1447 
1448 	new = alloc_inode(sb);
1449 	if (new) {
1450 		inode = inode_insert5(new, hashval, test, set, data);
1451 		if (unlikely(inode != new))
1452 			destroy_inode(new);
1453 	}
1454 	return inode;
1455 }
1456 EXPORT_SYMBOL_GPL(iget5_locked_rcu);
1457 
1458 /**
1459  * iget_locked - obtain an inode from a mounted file system
1460  * @sb:		super block of file system
1461  * @ino:	inode number to get
1462  *
1463  * Search for the inode specified by @ino in the inode cache and if present
1464  * return it with an increased reference count. This is for file systems
1465  * where the inode number is sufficient for unique identification of an inode.
1466  *
1467  * If the inode is not in cache, allocate a new inode and return it locked,
1468  * hashed, and with the I_NEW flag set.  The file system gets to fill it in
1469  * before unlocking it via unlock_new_inode().
1470  */
1471 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1472 {
1473 	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1474 	struct inode *inode;
1475 again:
1476 	inode = find_inode_fast(sb, head, ino, false);
1477 	if (inode) {
1478 		if (IS_ERR(inode))
1479 			return NULL;
1480 		wait_on_inode(inode);
1481 		if (unlikely(inode_unhashed(inode))) {
1482 			iput(inode);
1483 			goto again;
1484 		}
1485 		return inode;
1486 	}
1487 
1488 	inode = alloc_inode(sb);
1489 	if (inode) {
1490 		struct inode *old;
1491 
1492 		spin_lock(&inode_hash_lock);
1493 		/* We released the lock, so.. */
1494 		old = find_inode_fast(sb, head, ino, true);
1495 		if (!old) {
1496 			inode->i_ino = ino;
1497 			spin_lock(&inode->i_lock);
1498 			inode->i_state = I_NEW;
1499 			hlist_add_head_rcu(&inode->i_hash, head);
1500 			spin_unlock(&inode->i_lock);
1501 			inode_sb_list_add(inode);
1502 			spin_unlock(&inode_hash_lock);
1503 
1504 			/* Return the locked inode with I_NEW set, the
1505 			 * caller is responsible for filling in the contents
1506 			 */
1507 			return inode;
1508 		}
1509 
1510 		/*
1511 		 * Uhhuh, somebody else created the same inode under
1512 		 * us. Use the old inode instead of the one we just
1513 		 * allocated.
1514 		 */
1515 		spin_unlock(&inode_hash_lock);
1516 		destroy_inode(inode);
1517 		if (IS_ERR(old))
1518 			return NULL;
1519 		inode = old;
1520 		wait_on_inode(inode);
1521 		if (unlikely(inode_unhashed(inode))) {
1522 			iput(inode);
1523 			goto again;
1524 		}
1525 	}
1526 	return inode;
1527 }
1528 EXPORT_SYMBOL(iget_locked);
1529 
1530 /*
1531  * search the inode cache for a matching inode number.
1532  * If we find one, then the inode number we are trying to
1533  * allocate is not unique and so we should not use it.
1534  *
1535  * Returns 1 if the inode number is unique, 0 if it is not.
1536  */
1537 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1538 {
1539 	struct hlist_head *b = inode_hashtable + hash(sb, ino);
1540 	struct inode *inode;
1541 
1542 	hlist_for_each_entry_rcu(inode, b, i_hash) {
1543 		if (inode->i_ino == ino && inode->i_sb == sb)
1544 			return 0;
1545 	}
1546 	return 1;
1547 }
1548 
1549 /**
1550  *	iunique - get a unique inode number
1551  *	@sb: superblock
1552  *	@max_reserved: highest reserved inode number
1553  *
1554  *	Obtain an inode number that is unique on the system for a given
1555  *	superblock. This is used by file systems that have no natural
1556  *	permanent inode numbering system. An inode number is returned that
1557  *	is higher than the reserved limit but unique.
1558  *
1559  *	BUGS:
1560  *	With a large number of inodes live on the file system this function
1561  *	currently becomes quite slow.
1562  */
1563 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1564 {
1565 	/*
1566 	 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1567 	 * error if st_ino won't fit in target struct field. Use 32bit counter
1568 	 * here to attempt to avoid that.
1569 	 */
1570 	static DEFINE_SPINLOCK(iunique_lock);
1571 	static unsigned int counter;
1572 	ino_t res;
1573 
1574 	rcu_read_lock();
1575 	spin_lock(&iunique_lock);
1576 	do {
1577 		if (counter <= max_reserved)
1578 			counter = max_reserved + 1;
1579 		res = counter++;
1580 	} while (!test_inode_iunique(sb, res));
1581 	spin_unlock(&iunique_lock);
1582 	rcu_read_unlock();
1583 
1584 	return res;
1585 }
1586 EXPORT_SYMBOL(iunique);
1587 
1588 struct inode *igrab(struct inode *inode)
1589 {
1590 	spin_lock(&inode->i_lock);
1591 	if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1592 		__iget(inode);
1593 		spin_unlock(&inode->i_lock);
1594 	} else {
1595 		spin_unlock(&inode->i_lock);
1596 		/*
1597 		 * Handle the case where s_op->clear_inode is not been
1598 		 * called yet, and somebody is calling igrab
1599 		 * while the inode is getting freed.
1600 		 */
1601 		inode = NULL;
1602 	}
1603 	return inode;
1604 }
1605 EXPORT_SYMBOL(igrab);
1606 
1607 /**
1608  * ilookup5_nowait - search for an inode in the inode cache
1609  * @sb:		super block of file system to search
1610  * @hashval:	hash value (usually inode number) to search for
1611  * @test:	callback used for comparisons between inodes
1612  * @data:	opaque data pointer to pass to @test
1613  *
1614  * Search for the inode specified by @hashval and @data in the inode cache.
1615  * If the inode is in the cache, the inode is returned with an incremented
1616  * reference count.
1617  *
1618  * Note: I_NEW is not waited upon so you have to be very careful what you do
1619  * with the returned inode.  You probably should be using ilookup5() instead.
1620  *
1621  * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1622  */
1623 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1624 		int (*test)(struct inode *, void *), void *data)
1625 {
1626 	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1627 	struct inode *inode;
1628 
1629 	spin_lock(&inode_hash_lock);
1630 	inode = find_inode(sb, head, test, data, true);
1631 	spin_unlock(&inode_hash_lock);
1632 
1633 	return IS_ERR(inode) ? NULL : inode;
1634 }
1635 EXPORT_SYMBOL(ilookup5_nowait);
1636 
1637 /**
1638  * ilookup5 - search for an inode in the inode cache
1639  * @sb:		super block of file system to search
1640  * @hashval:	hash value (usually inode number) to search for
1641  * @test:	callback used for comparisons between inodes
1642  * @data:	opaque data pointer to pass to @test
1643  *
1644  * Search for the inode specified by @hashval and @data in the inode cache,
1645  * and if the inode is in the cache, return the inode with an incremented
1646  * reference count.  Waits on I_NEW before returning the inode.
1647  * returned with an incremented reference count.
1648  *
1649  * This is a generalized version of ilookup() for file systems where the
1650  * inode number is not sufficient for unique identification of an inode.
1651  *
1652  * Note: @test is called with the inode_hash_lock held, so can't sleep.
1653  */
1654 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1655 		int (*test)(struct inode *, void *), void *data)
1656 {
1657 	struct inode *inode;
1658 again:
1659 	inode = ilookup5_nowait(sb, hashval, test, data);
1660 	if (inode) {
1661 		wait_on_inode(inode);
1662 		if (unlikely(inode_unhashed(inode))) {
1663 			iput(inode);
1664 			goto again;
1665 		}
1666 	}
1667 	return inode;
1668 }
1669 EXPORT_SYMBOL(ilookup5);
1670 
1671 /**
1672  * ilookup - search for an inode in the inode cache
1673  * @sb:		super block of file system to search
1674  * @ino:	inode number to search for
1675  *
1676  * Search for the inode @ino in the inode cache, and if the inode is in the
1677  * cache, the inode is returned with an incremented reference count.
1678  */
1679 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1680 {
1681 	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1682 	struct inode *inode;
1683 again:
1684 	inode = find_inode_fast(sb, head, ino, false);
1685 
1686 	if (inode) {
1687 		if (IS_ERR(inode))
1688 			return NULL;
1689 		wait_on_inode(inode);
1690 		if (unlikely(inode_unhashed(inode))) {
1691 			iput(inode);
1692 			goto again;
1693 		}
1694 	}
1695 	return inode;
1696 }
1697 EXPORT_SYMBOL(ilookup);
1698 
1699 /**
1700  * find_inode_nowait - find an inode in the inode cache
1701  * @sb:		super block of file system to search
1702  * @hashval:	hash value (usually inode number) to search for
1703  * @match:	callback used for comparisons between inodes
1704  * @data:	opaque data pointer to pass to @match
1705  *
1706  * Search for the inode specified by @hashval and @data in the inode
1707  * cache, where the helper function @match will return 0 if the inode
1708  * does not match, 1 if the inode does match, and -1 if the search
1709  * should be stopped.  The @match function must be responsible for
1710  * taking the i_lock spin_lock and checking i_state for an inode being
1711  * freed or being initialized, and incrementing the reference count
1712  * before returning 1.  It also must not sleep, since it is called with
1713  * the inode_hash_lock spinlock held.
1714  *
1715  * This is a even more generalized version of ilookup5() when the
1716  * function must never block --- find_inode() can block in
1717  * __wait_on_freeing_inode() --- or when the caller can not increment
1718  * the reference count because the resulting iput() might cause an
1719  * inode eviction.  The tradeoff is that the @match funtion must be
1720  * very carefully implemented.
1721  */
1722 struct inode *find_inode_nowait(struct super_block *sb,
1723 				unsigned long hashval,
1724 				int (*match)(struct inode *, unsigned long,
1725 					     void *),
1726 				void *data)
1727 {
1728 	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1729 	struct inode *inode, *ret_inode = NULL;
1730 	int mval;
1731 
1732 	spin_lock(&inode_hash_lock);
1733 	hlist_for_each_entry(inode, head, i_hash) {
1734 		if (inode->i_sb != sb)
1735 			continue;
1736 		mval = match(inode, hashval, data);
1737 		if (mval == 0)
1738 			continue;
1739 		if (mval == 1)
1740 			ret_inode = inode;
1741 		goto out;
1742 	}
1743 out:
1744 	spin_unlock(&inode_hash_lock);
1745 	return ret_inode;
1746 }
1747 EXPORT_SYMBOL(find_inode_nowait);
1748 
1749 /**
1750  * find_inode_rcu - find an inode in the inode cache
1751  * @sb:		Super block of file system to search
1752  * @hashval:	Key to hash
1753  * @test:	Function to test match on an inode
1754  * @data:	Data for test function
1755  *
1756  * Search for the inode specified by @hashval and @data in the inode cache,
1757  * where the helper function @test will return 0 if the inode does not match
1758  * and 1 if it does.  The @test function must be responsible for taking the
1759  * i_lock spin_lock and checking i_state for an inode being freed or being
1760  * initialized.
1761  *
1762  * If successful, this will return the inode for which the @test function
1763  * returned 1 and NULL otherwise.
1764  *
1765  * The @test function is not permitted to take a ref on any inode presented.
1766  * It is also not permitted to sleep.
1767  *
1768  * The caller must hold the RCU read lock.
1769  */
1770 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1771 			     int (*test)(struct inode *, void *), void *data)
1772 {
1773 	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1774 	struct inode *inode;
1775 
1776 	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1777 			 "suspicious find_inode_rcu() usage");
1778 
1779 	hlist_for_each_entry_rcu(inode, head, i_hash) {
1780 		if (inode->i_sb == sb &&
1781 		    !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1782 		    test(inode, data))
1783 			return inode;
1784 	}
1785 	return NULL;
1786 }
1787 EXPORT_SYMBOL(find_inode_rcu);
1788 
1789 /**
1790  * find_inode_by_ino_rcu - Find an inode in the inode cache
1791  * @sb:		Super block of file system to search
1792  * @ino:	The inode number to match
1793  *
1794  * Search for the inode specified by @hashval and @data in the inode cache,
1795  * where the helper function @test will return 0 if the inode does not match
1796  * and 1 if it does.  The @test function must be responsible for taking the
1797  * i_lock spin_lock and checking i_state for an inode being freed or being
1798  * initialized.
1799  *
1800  * If successful, this will return the inode for which the @test function
1801  * returned 1 and NULL otherwise.
1802  *
1803  * The @test function is not permitted to take a ref on any inode presented.
1804  * It is also not permitted to sleep.
1805  *
1806  * The caller must hold the RCU read lock.
1807  */
1808 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1809 				    unsigned long ino)
1810 {
1811 	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1812 	struct inode *inode;
1813 
1814 	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1815 			 "suspicious find_inode_by_ino_rcu() usage");
1816 
1817 	hlist_for_each_entry_rcu(inode, head, i_hash) {
1818 		if (inode->i_ino == ino &&
1819 		    inode->i_sb == sb &&
1820 		    !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1821 		    return inode;
1822 	}
1823 	return NULL;
1824 }
1825 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1826 
1827 int insert_inode_locked(struct inode *inode)
1828 {
1829 	struct super_block *sb = inode->i_sb;
1830 	ino_t ino = inode->i_ino;
1831 	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1832 
1833 	while (1) {
1834 		struct inode *old = NULL;
1835 		spin_lock(&inode_hash_lock);
1836 		hlist_for_each_entry(old, head, i_hash) {
1837 			if (old->i_ino != ino)
1838 				continue;
1839 			if (old->i_sb != sb)
1840 				continue;
1841 			spin_lock(&old->i_lock);
1842 			if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1843 				spin_unlock(&old->i_lock);
1844 				continue;
1845 			}
1846 			break;
1847 		}
1848 		if (likely(!old)) {
1849 			spin_lock(&inode->i_lock);
1850 			inode->i_state |= I_NEW | I_CREATING;
1851 			hlist_add_head_rcu(&inode->i_hash, head);
1852 			spin_unlock(&inode->i_lock);
1853 			spin_unlock(&inode_hash_lock);
1854 			return 0;
1855 		}
1856 		if (unlikely(old->i_state & I_CREATING)) {
1857 			spin_unlock(&old->i_lock);
1858 			spin_unlock(&inode_hash_lock);
1859 			return -EBUSY;
1860 		}
1861 		__iget(old);
1862 		spin_unlock(&old->i_lock);
1863 		spin_unlock(&inode_hash_lock);
1864 		wait_on_inode(old);
1865 		if (unlikely(!inode_unhashed(old))) {
1866 			iput(old);
1867 			return -EBUSY;
1868 		}
1869 		iput(old);
1870 	}
1871 }
1872 EXPORT_SYMBOL(insert_inode_locked);
1873 
1874 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1875 		int (*test)(struct inode *, void *), void *data)
1876 {
1877 	struct inode *old;
1878 
1879 	inode->i_state |= I_CREATING;
1880 	old = inode_insert5(inode, hashval, test, NULL, data);
1881 
1882 	if (old != inode) {
1883 		iput(old);
1884 		return -EBUSY;
1885 	}
1886 	return 0;
1887 }
1888 EXPORT_SYMBOL(insert_inode_locked4);
1889 
1890 
1891 int generic_delete_inode(struct inode *inode)
1892 {
1893 	return 1;
1894 }
1895 EXPORT_SYMBOL(generic_delete_inode);
1896 
1897 /*
1898  * Called when we're dropping the last reference
1899  * to an inode.
1900  *
1901  * Call the FS "drop_inode()" function, defaulting to
1902  * the legacy UNIX filesystem behaviour.  If it tells
1903  * us to evict inode, do so.  Otherwise, retain inode
1904  * in cache if fs is alive, sync and evict if fs is
1905  * shutting down.
1906  */
1907 static void iput_final(struct inode *inode)
1908 {
1909 	struct super_block *sb = inode->i_sb;
1910 	const struct super_operations *op = inode->i_sb->s_op;
1911 	unsigned long state;
1912 	int drop;
1913 
1914 	WARN_ON(inode->i_state & I_NEW);
1915 
1916 	if (op->drop_inode)
1917 		drop = op->drop_inode(inode);
1918 	else
1919 		drop = generic_drop_inode(inode);
1920 
1921 	if (!drop &&
1922 	    !(inode->i_state & I_DONTCACHE) &&
1923 	    (sb->s_flags & SB_ACTIVE)) {
1924 		__inode_add_lru(inode, true);
1925 		spin_unlock(&inode->i_lock);
1926 		return;
1927 	}
1928 
1929 	state = inode->i_state;
1930 	if (!drop) {
1931 		WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1932 		spin_unlock(&inode->i_lock);
1933 
1934 		write_inode_now(inode, 1);
1935 
1936 		spin_lock(&inode->i_lock);
1937 		state = inode->i_state;
1938 		WARN_ON(state & I_NEW);
1939 		state &= ~I_WILL_FREE;
1940 	}
1941 
1942 	WRITE_ONCE(inode->i_state, state | I_FREEING);
1943 	if (!list_empty(&inode->i_lru))
1944 		inode_lru_list_del(inode);
1945 	spin_unlock(&inode->i_lock);
1946 
1947 	evict(inode);
1948 }
1949 
1950 /**
1951  *	iput	- put an inode
1952  *	@inode: inode to put
1953  *
1954  *	Puts an inode, dropping its usage count. If the inode use count hits
1955  *	zero, the inode is then freed and may also be destroyed.
1956  *
1957  *	Consequently, iput() can sleep.
1958  */
1959 void iput(struct inode *inode)
1960 {
1961 	if (!inode)
1962 		return;
1963 	BUG_ON(inode->i_state & I_CLEAR);
1964 retry:
1965 	if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1966 		if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1967 			atomic_inc(&inode->i_count);
1968 			spin_unlock(&inode->i_lock);
1969 			trace_writeback_lazytime_iput(inode);
1970 			mark_inode_dirty_sync(inode);
1971 			goto retry;
1972 		}
1973 		iput_final(inode);
1974 	}
1975 }
1976 EXPORT_SYMBOL(iput);
1977 
1978 #ifdef CONFIG_BLOCK
1979 /**
1980  *	bmap	- find a block number in a file
1981  *	@inode:  inode owning the block number being requested
1982  *	@block: pointer containing the block to find
1983  *
1984  *	Replaces the value in ``*block`` with the block number on the device holding
1985  *	corresponding to the requested block number in the file.
1986  *	That is, asked for block 4 of inode 1 the function will replace the
1987  *	4 in ``*block``, with disk block relative to the disk start that holds that
1988  *	block of the file.
1989  *
1990  *	Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1991  *	hole, returns 0 and ``*block`` is also set to 0.
1992  */
1993 int bmap(struct inode *inode, sector_t *block)
1994 {
1995 	if (!inode->i_mapping->a_ops->bmap)
1996 		return -EINVAL;
1997 
1998 	*block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1999 	return 0;
2000 }
2001 EXPORT_SYMBOL(bmap);
2002 #endif
2003 
2004 /*
2005  * With relative atime, only update atime if the previous atime is
2006  * earlier than or equal to either the ctime or mtime,
2007  * or if at least a day has passed since the last atime update.
2008  */
2009 static bool relatime_need_update(struct vfsmount *mnt, struct inode *inode,
2010 			     struct timespec64 now)
2011 {
2012 	struct timespec64 atime, mtime, ctime;
2013 
2014 	if (!(mnt->mnt_flags & MNT_RELATIME))
2015 		return true;
2016 	/*
2017 	 * Is mtime younger than or equal to atime? If yes, update atime:
2018 	 */
2019 	atime = inode_get_atime(inode);
2020 	mtime = inode_get_mtime(inode);
2021 	if (timespec64_compare(&mtime, &atime) >= 0)
2022 		return true;
2023 	/*
2024 	 * Is ctime younger than or equal to atime? If yes, update atime:
2025 	 */
2026 	ctime = inode_get_ctime(inode);
2027 	if (timespec64_compare(&ctime, &atime) >= 0)
2028 		return true;
2029 
2030 	/*
2031 	 * Is the previous atime value older than a day? If yes,
2032 	 * update atime:
2033 	 */
2034 	if ((long)(now.tv_sec - atime.tv_sec) >= 24*60*60)
2035 		return true;
2036 	/*
2037 	 * Good, we can skip the atime update:
2038 	 */
2039 	return false;
2040 }
2041 
2042 /**
2043  * inode_update_timestamps - update the timestamps on the inode
2044  * @inode: inode to be updated
2045  * @flags: S_* flags that needed to be updated
2046  *
2047  * The update_time function is called when an inode's timestamps need to be
2048  * updated for a read or write operation. This function handles updating the
2049  * actual timestamps. It's up to the caller to ensure that the inode is marked
2050  * dirty appropriately.
2051  *
2052  * In the case where any of S_MTIME, S_CTIME, or S_VERSION need to be updated,
2053  * attempt to update all three of them. S_ATIME updates can be handled
2054  * independently of the rest.
2055  *
2056  * Returns a set of S_* flags indicating which values changed.
2057  */
2058 int inode_update_timestamps(struct inode *inode, int flags)
2059 {
2060 	int updated = 0;
2061 	struct timespec64 now;
2062 
2063 	if (flags & (S_MTIME|S_CTIME|S_VERSION)) {
2064 		struct timespec64 ctime = inode_get_ctime(inode);
2065 		struct timespec64 mtime = inode_get_mtime(inode);
2066 
2067 		now = inode_set_ctime_current(inode);
2068 		if (!timespec64_equal(&now, &ctime))
2069 			updated |= S_CTIME;
2070 		if (!timespec64_equal(&now, &mtime)) {
2071 			inode_set_mtime_to_ts(inode, now);
2072 			updated |= S_MTIME;
2073 		}
2074 		if (IS_I_VERSION(inode) && inode_maybe_inc_iversion(inode, updated))
2075 			updated |= S_VERSION;
2076 	} else {
2077 		now = current_time(inode);
2078 	}
2079 
2080 	if (flags & S_ATIME) {
2081 		struct timespec64 atime = inode_get_atime(inode);
2082 
2083 		if (!timespec64_equal(&now, &atime)) {
2084 			inode_set_atime_to_ts(inode, now);
2085 			updated |= S_ATIME;
2086 		}
2087 	}
2088 	return updated;
2089 }
2090 EXPORT_SYMBOL(inode_update_timestamps);
2091 
2092 /**
2093  * generic_update_time - update the timestamps on the inode
2094  * @inode: inode to be updated
2095  * @flags: S_* flags that needed to be updated
2096  *
2097  * The update_time function is called when an inode's timestamps need to be
2098  * updated for a read or write operation. In the case where any of S_MTIME, S_CTIME,
2099  * or S_VERSION need to be updated we attempt to update all three of them. S_ATIME
2100  * updates can be handled done independently of the rest.
2101  *
2102  * Returns a S_* mask indicating which fields were updated.
2103  */
2104 int generic_update_time(struct inode *inode, int flags)
2105 {
2106 	int updated = inode_update_timestamps(inode, flags);
2107 	int dirty_flags = 0;
2108 
2109 	if (updated & (S_ATIME|S_MTIME|S_CTIME))
2110 		dirty_flags = inode->i_sb->s_flags & SB_LAZYTIME ? I_DIRTY_TIME : I_DIRTY_SYNC;
2111 	if (updated & S_VERSION)
2112 		dirty_flags |= I_DIRTY_SYNC;
2113 	__mark_inode_dirty(inode, dirty_flags);
2114 	return updated;
2115 }
2116 EXPORT_SYMBOL(generic_update_time);
2117 
2118 /*
2119  * This does the actual work of updating an inodes time or version.  Must have
2120  * had called mnt_want_write() before calling this.
2121  */
2122 int inode_update_time(struct inode *inode, int flags)
2123 {
2124 	if (inode->i_op->update_time)
2125 		return inode->i_op->update_time(inode, flags);
2126 	generic_update_time(inode, flags);
2127 	return 0;
2128 }
2129 EXPORT_SYMBOL(inode_update_time);
2130 
2131 /**
2132  *	atime_needs_update	-	update the access time
2133  *	@path: the &struct path to update
2134  *	@inode: inode to update
2135  *
2136  *	Update the accessed time on an inode and mark it for writeback.
2137  *	This function automatically handles read only file systems and media,
2138  *	as well as the "noatime" flag and inode specific "noatime" markers.
2139  */
2140 bool atime_needs_update(const struct path *path, struct inode *inode)
2141 {
2142 	struct vfsmount *mnt = path->mnt;
2143 	struct timespec64 now, atime;
2144 
2145 	if (inode->i_flags & S_NOATIME)
2146 		return false;
2147 
2148 	/* Atime updates will likely cause i_uid and i_gid to be written
2149 	 * back improprely if their true value is unknown to the vfs.
2150 	 */
2151 	if (HAS_UNMAPPED_ID(mnt_idmap(mnt), inode))
2152 		return false;
2153 
2154 	if (IS_NOATIME(inode))
2155 		return false;
2156 	if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
2157 		return false;
2158 
2159 	if (mnt->mnt_flags & MNT_NOATIME)
2160 		return false;
2161 	if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
2162 		return false;
2163 
2164 	now = current_time(inode);
2165 
2166 	if (!relatime_need_update(mnt, inode, now))
2167 		return false;
2168 
2169 	atime = inode_get_atime(inode);
2170 	if (timespec64_equal(&atime, &now))
2171 		return false;
2172 
2173 	return true;
2174 }
2175 
2176 void touch_atime(const struct path *path)
2177 {
2178 	struct vfsmount *mnt = path->mnt;
2179 	struct inode *inode = d_inode(path->dentry);
2180 
2181 	if (!atime_needs_update(path, inode))
2182 		return;
2183 
2184 	if (!sb_start_write_trylock(inode->i_sb))
2185 		return;
2186 
2187 	if (mnt_get_write_access(mnt) != 0)
2188 		goto skip_update;
2189 	/*
2190 	 * File systems can error out when updating inodes if they need to
2191 	 * allocate new space to modify an inode (such is the case for
2192 	 * Btrfs), but since we touch atime while walking down the path we
2193 	 * really don't care if we failed to update the atime of the file,
2194 	 * so just ignore the return value.
2195 	 * We may also fail on filesystems that have the ability to make parts
2196 	 * of the fs read only, e.g. subvolumes in Btrfs.
2197 	 */
2198 	inode_update_time(inode, S_ATIME);
2199 	mnt_put_write_access(mnt);
2200 skip_update:
2201 	sb_end_write(inode->i_sb);
2202 }
2203 EXPORT_SYMBOL(touch_atime);
2204 
2205 /*
2206  * Return mask of changes for notify_change() that need to be done as a
2207  * response to write or truncate. Return 0 if nothing has to be changed.
2208  * Negative value on error (change should be denied).
2209  */
2210 int dentry_needs_remove_privs(struct mnt_idmap *idmap,
2211 			      struct dentry *dentry)
2212 {
2213 	struct inode *inode = d_inode(dentry);
2214 	int mask = 0;
2215 	int ret;
2216 
2217 	if (IS_NOSEC(inode))
2218 		return 0;
2219 
2220 	mask = setattr_should_drop_suidgid(idmap, inode);
2221 	ret = security_inode_need_killpriv(dentry);
2222 	if (ret < 0)
2223 		return ret;
2224 	if (ret)
2225 		mask |= ATTR_KILL_PRIV;
2226 	return mask;
2227 }
2228 
2229 static int __remove_privs(struct mnt_idmap *idmap,
2230 			  struct dentry *dentry, int kill)
2231 {
2232 	struct iattr newattrs;
2233 
2234 	newattrs.ia_valid = ATTR_FORCE | kill;
2235 	/*
2236 	 * Note we call this on write, so notify_change will not
2237 	 * encounter any conflicting delegations:
2238 	 */
2239 	return notify_change(idmap, dentry, &newattrs, NULL);
2240 }
2241 
2242 int file_remove_privs_flags(struct file *file, unsigned int flags)
2243 {
2244 	struct dentry *dentry = file_dentry(file);
2245 	struct inode *inode = file_inode(file);
2246 	int error = 0;
2247 	int kill;
2248 
2249 	if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
2250 		return 0;
2251 
2252 	kill = dentry_needs_remove_privs(file_mnt_idmap(file), dentry);
2253 	if (kill < 0)
2254 		return kill;
2255 
2256 	if (kill) {
2257 		if (flags & IOCB_NOWAIT)
2258 			return -EAGAIN;
2259 
2260 		error = __remove_privs(file_mnt_idmap(file), dentry, kill);
2261 	}
2262 
2263 	if (!error)
2264 		inode_has_no_xattr(inode);
2265 	return error;
2266 }
2267 EXPORT_SYMBOL_GPL(file_remove_privs_flags);
2268 
2269 /**
2270  * file_remove_privs - remove special file privileges (suid, capabilities)
2271  * @file: file to remove privileges from
2272  *
2273  * When file is modified by a write or truncation ensure that special
2274  * file privileges are removed.
2275  *
2276  * Return: 0 on success, negative errno on failure.
2277  */
2278 int file_remove_privs(struct file *file)
2279 {
2280 	return file_remove_privs_flags(file, 0);
2281 }
2282 EXPORT_SYMBOL(file_remove_privs);
2283 
2284 /**
2285  * current_time - Return FS time (possibly fine-grained)
2286  * @inode: inode.
2287  *
2288  * Return the current time truncated to the time granularity supported by
2289  * the fs, as suitable for a ctime/mtime change. If the ctime is flagged
2290  * as having been QUERIED, get a fine-grained timestamp, but don't update
2291  * the floor.
2292  *
2293  * For a multigrain inode, this is effectively an estimate of the timestamp
2294  * that a file would receive. An actual update must go through
2295  * inode_set_ctime_current().
2296  */
2297 struct timespec64 current_time(struct inode *inode)
2298 {
2299 	struct timespec64 now;
2300 	u32 cns;
2301 
2302 	ktime_get_coarse_real_ts64_mg(&now);
2303 
2304 	if (!is_mgtime(inode))
2305 		goto out;
2306 
2307 	/* If nothing has queried it, then coarse time is fine */
2308 	cns = smp_load_acquire(&inode->i_ctime_nsec);
2309 	if (cns & I_CTIME_QUERIED) {
2310 		/*
2311 		 * If there is no apparent change, then get a fine-grained
2312 		 * timestamp.
2313 		 */
2314 		if (now.tv_nsec == (cns & ~I_CTIME_QUERIED))
2315 			ktime_get_real_ts64(&now);
2316 	}
2317 out:
2318 	return timestamp_truncate(now, inode);
2319 }
2320 EXPORT_SYMBOL(current_time);
2321 
2322 static int inode_needs_update_time(struct inode *inode)
2323 {
2324 	struct timespec64 now, ts;
2325 	int sync_it = 0;
2326 
2327 	/* First try to exhaust all avenues to not sync */
2328 	if (IS_NOCMTIME(inode))
2329 		return 0;
2330 
2331 	now = current_time(inode);
2332 
2333 	ts = inode_get_mtime(inode);
2334 	if (!timespec64_equal(&ts, &now))
2335 		sync_it |= S_MTIME;
2336 
2337 	ts = inode_get_ctime(inode);
2338 	if (!timespec64_equal(&ts, &now))
2339 		sync_it |= S_CTIME;
2340 
2341 	if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2342 		sync_it |= S_VERSION;
2343 
2344 	return sync_it;
2345 }
2346 
2347 static int __file_update_time(struct file *file, int sync_mode)
2348 {
2349 	int ret = 0;
2350 	struct inode *inode = file_inode(file);
2351 
2352 	/* try to update time settings */
2353 	if (!mnt_get_write_access_file(file)) {
2354 		ret = inode_update_time(inode, sync_mode);
2355 		mnt_put_write_access_file(file);
2356 	}
2357 
2358 	return ret;
2359 }
2360 
2361 /**
2362  * file_update_time - update mtime and ctime time
2363  * @file: file accessed
2364  *
2365  * Update the mtime and ctime members of an inode and mark the inode for
2366  * writeback. Note that this function is meant exclusively for usage in
2367  * the file write path of filesystems, and filesystems may choose to
2368  * explicitly ignore updates via this function with the _NOCMTIME inode
2369  * flag, e.g. for network filesystem where these imestamps are handled
2370  * by the server. This can return an error for file systems who need to
2371  * allocate space in order to update an inode.
2372  *
2373  * Return: 0 on success, negative errno on failure.
2374  */
2375 int file_update_time(struct file *file)
2376 {
2377 	int ret;
2378 	struct inode *inode = file_inode(file);
2379 
2380 	ret = inode_needs_update_time(inode);
2381 	if (ret <= 0)
2382 		return ret;
2383 
2384 	return __file_update_time(file, ret);
2385 }
2386 EXPORT_SYMBOL(file_update_time);
2387 
2388 /**
2389  * file_modified_flags - handle mandated vfs changes when modifying a file
2390  * @file: file that was modified
2391  * @flags: kiocb flags
2392  *
2393  * When file has been modified ensure that special
2394  * file privileges are removed and time settings are updated.
2395  *
2396  * If IOCB_NOWAIT is set, special file privileges will not be removed and
2397  * time settings will not be updated. It will return -EAGAIN.
2398  *
2399  * Context: Caller must hold the file's inode lock.
2400  *
2401  * Return: 0 on success, negative errno on failure.
2402  */
2403 static int file_modified_flags(struct file *file, int flags)
2404 {
2405 	int ret;
2406 	struct inode *inode = file_inode(file);
2407 
2408 	/*
2409 	 * Clear the security bits if the process is not being run by root.
2410 	 * This keeps people from modifying setuid and setgid binaries.
2411 	 */
2412 	ret = file_remove_privs_flags(file, flags);
2413 	if (ret)
2414 		return ret;
2415 
2416 	if (unlikely(file->f_mode & FMODE_NOCMTIME))
2417 		return 0;
2418 
2419 	ret = inode_needs_update_time(inode);
2420 	if (ret <= 0)
2421 		return ret;
2422 	if (flags & IOCB_NOWAIT)
2423 		return -EAGAIN;
2424 
2425 	return __file_update_time(file, ret);
2426 }
2427 
2428 /**
2429  * file_modified - handle mandated vfs changes when modifying a file
2430  * @file: file that was modified
2431  *
2432  * When file has been modified ensure that special
2433  * file privileges are removed and time settings are updated.
2434  *
2435  * Context: Caller must hold the file's inode lock.
2436  *
2437  * Return: 0 on success, negative errno on failure.
2438  */
2439 int file_modified(struct file *file)
2440 {
2441 	return file_modified_flags(file, 0);
2442 }
2443 EXPORT_SYMBOL(file_modified);
2444 
2445 /**
2446  * kiocb_modified - handle mandated vfs changes when modifying a file
2447  * @iocb: iocb that was modified
2448  *
2449  * When file has been modified ensure that special
2450  * file privileges are removed and time settings are updated.
2451  *
2452  * Context: Caller must hold the file's inode lock.
2453  *
2454  * Return: 0 on success, negative errno on failure.
2455  */
2456 int kiocb_modified(struct kiocb *iocb)
2457 {
2458 	return file_modified_flags(iocb->ki_filp, iocb->ki_flags);
2459 }
2460 EXPORT_SYMBOL_GPL(kiocb_modified);
2461 
2462 int inode_needs_sync(struct inode *inode)
2463 {
2464 	if (IS_SYNC(inode))
2465 		return 1;
2466 	if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2467 		return 1;
2468 	return 0;
2469 }
2470 EXPORT_SYMBOL(inode_needs_sync);
2471 
2472 /*
2473  * If we try to find an inode in the inode hash while it is being
2474  * deleted, we have to wait until the filesystem completes its
2475  * deletion before reporting that it isn't found.  This function waits
2476  * until the deletion _might_ have completed.  Callers are responsible
2477  * to recheck inode state.
2478  *
2479  * It doesn't matter if I_NEW is not set initially, a call to
2480  * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2481  * will DTRT.
2482  */
2483 static void __wait_on_freeing_inode(struct inode *inode, bool is_inode_hash_locked)
2484 {
2485 	struct wait_bit_queue_entry wqe;
2486 	struct wait_queue_head *wq_head;
2487 
2488 	/*
2489 	 * Handle racing against evict(), see that routine for more details.
2490 	 */
2491 	if (unlikely(inode_unhashed(inode))) {
2492 		WARN_ON(is_inode_hash_locked);
2493 		spin_unlock(&inode->i_lock);
2494 		return;
2495 	}
2496 
2497 	wq_head = inode_bit_waitqueue(&wqe, inode, __I_NEW);
2498 	prepare_to_wait_event(wq_head, &wqe.wq_entry, TASK_UNINTERRUPTIBLE);
2499 	spin_unlock(&inode->i_lock);
2500 	rcu_read_unlock();
2501 	if (is_inode_hash_locked)
2502 		spin_unlock(&inode_hash_lock);
2503 	schedule();
2504 	finish_wait(wq_head, &wqe.wq_entry);
2505 	if (is_inode_hash_locked)
2506 		spin_lock(&inode_hash_lock);
2507 	rcu_read_lock();
2508 }
2509 
2510 static __initdata unsigned long ihash_entries;
2511 static int __init set_ihash_entries(char *str)
2512 {
2513 	if (!str)
2514 		return 0;
2515 	ihash_entries = simple_strtoul(str, &str, 0);
2516 	return 1;
2517 }
2518 __setup("ihash_entries=", set_ihash_entries);
2519 
2520 /*
2521  * Initialize the waitqueues and inode hash table.
2522  */
2523 void __init inode_init_early(void)
2524 {
2525 	/* If hashes are distributed across NUMA nodes, defer
2526 	 * hash allocation until vmalloc space is available.
2527 	 */
2528 	if (hashdist)
2529 		return;
2530 
2531 	inode_hashtable =
2532 		alloc_large_system_hash("Inode-cache",
2533 					sizeof(struct hlist_head),
2534 					ihash_entries,
2535 					14,
2536 					HASH_EARLY | HASH_ZERO,
2537 					&i_hash_shift,
2538 					&i_hash_mask,
2539 					0,
2540 					0);
2541 }
2542 
2543 void __init inode_init(void)
2544 {
2545 	/* inode slab cache */
2546 	inode_cachep = kmem_cache_create("inode_cache",
2547 					 sizeof(struct inode),
2548 					 0,
2549 					 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2550 					 SLAB_ACCOUNT),
2551 					 init_once);
2552 
2553 	/* Hash may have been set up in inode_init_early */
2554 	if (!hashdist)
2555 		return;
2556 
2557 	inode_hashtable =
2558 		alloc_large_system_hash("Inode-cache",
2559 					sizeof(struct hlist_head),
2560 					ihash_entries,
2561 					14,
2562 					HASH_ZERO,
2563 					&i_hash_shift,
2564 					&i_hash_mask,
2565 					0,
2566 					0);
2567 }
2568 
2569 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2570 {
2571 	inode->i_mode = mode;
2572 	if (S_ISCHR(mode)) {
2573 		inode->i_fop = &def_chr_fops;
2574 		inode->i_rdev = rdev;
2575 	} else if (S_ISBLK(mode)) {
2576 		if (IS_ENABLED(CONFIG_BLOCK))
2577 			inode->i_fop = &def_blk_fops;
2578 		inode->i_rdev = rdev;
2579 	} else if (S_ISFIFO(mode))
2580 		inode->i_fop = &pipefifo_fops;
2581 	else if (S_ISSOCK(mode))
2582 		;	/* leave it no_open_fops */
2583 	else
2584 		printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2585 				  " inode %s:%lu\n", mode, inode->i_sb->s_id,
2586 				  inode->i_ino);
2587 }
2588 EXPORT_SYMBOL(init_special_inode);
2589 
2590 /**
2591  * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2592  * @idmap: idmap of the mount the inode was created from
2593  * @inode: New inode
2594  * @dir: Directory inode
2595  * @mode: mode of the new inode
2596  *
2597  * If the inode has been created through an idmapped mount the idmap of
2598  * the vfsmount must be passed through @idmap. This function will then take
2599  * care to map the inode according to @idmap before checking permissions
2600  * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2601  * checking is to be performed on the raw inode simply pass @nop_mnt_idmap.
2602  */
2603 void inode_init_owner(struct mnt_idmap *idmap, struct inode *inode,
2604 		      const struct inode *dir, umode_t mode)
2605 {
2606 	inode_fsuid_set(inode, idmap);
2607 	if (dir && dir->i_mode & S_ISGID) {
2608 		inode->i_gid = dir->i_gid;
2609 
2610 		/* Directories are special, and always inherit S_ISGID */
2611 		if (S_ISDIR(mode))
2612 			mode |= S_ISGID;
2613 	} else
2614 		inode_fsgid_set(inode, idmap);
2615 	inode->i_mode = mode;
2616 }
2617 EXPORT_SYMBOL(inode_init_owner);
2618 
2619 /**
2620  * inode_owner_or_capable - check current task permissions to inode
2621  * @idmap: idmap of the mount the inode was found from
2622  * @inode: inode being checked
2623  *
2624  * Return true if current either has CAP_FOWNER in a namespace with the
2625  * inode owner uid mapped, or owns the file.
2626  *
2627  * If the inode has been found through an idmapped mount the idmap of
2628  * the vfsmount must be passed through @idmap. This function will then take
2629  * care to map the inode according to @idmap before checking permissions.
2630  * On non-idmapped mounts or if permission checking is to be performed on the
2631  * raw inode simply pass @nop_mnt_idmap.
2632  */
2633 bool inode_owner_or_capable(struct mnt_idmap *idmap,
2634 			    const struct inode *inode)
2635 {
2636 	vfsuid_t vfsuid;
2637 	struct user_namespace *ns;
2638 
2639 	vfsuid = i_uid_into_vfsuid(idmap, inode);
2640 	if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
2641 		return true;
2642 
2643 	ns = current_user_ns();
2644 	if (vfsuid_has_mapping(ns, vfsuid) && ns_capable(ns, CAP_FOWNER))
2645 		return true;
2646 	return false;
2647 }
2648 EXPORT_SYMBOL(inode_owner_or_capable);
2649 
2650 /*
2651  * Direct i/o helper functions
2652  */
2653 bool inode_dio_finished(const struct inode *inode)
2654 {
2655 	return atomic_read(&inode->i_dio_count) == 0;
2656 }
2657 EXPORT_SYMBOL(inode_dio_finished);
2658 
2659 /**
2660  * inode_dio_wait - wait for outstanding DIO requests to finish
2661  * @inode: inode to wait for
2662  *
2663  * Waits for all pending direct I/O requests to finish so that we can
2664  * proceed with a truncate or equivalent operation.
2665  *
2666  * Must be called under a lock that serializes taking new references
2667  * to i_dio_count, usually by inode->i_mutex.
2668  */
2669 void inode_dio_wait(struct inode *inode)
2670 {
2671 	wait_var_event(&inode->i_dio_count, inode_dio_finished(inode));
2672 }
2673 EXPORT_SYMBOL(inode_dio_wait);
2674 
2675 void inode_dio_wait_interruptible(struct inode *inode)
2676 {
2677 	wait_var_event_interruptible(&inode->i_dio_count,
2678 				     inode_dio_finished(inode));
2679 }
2680 EXPORT_SYMBOL(inode_dio_wait_interruptible);
2681 
2682 /*
2683  * inode_set_flags - atomically set some inode flags
2684  *
2685  * Note: the caller should be holding i_mutex, or else be sure that
2686  * they have exclusive access to the inode structure (i.e., while the
2687  * inode is being instantiated).  The reason for the cmpxchg() loop
2688  * --- which wouldn't be necessary if all code paths which modify
2689  * i_flags actually followed this rule, is that there is at least one
2690  * code path which doesn't today so we use cmpxchg() out of an abundance
2691  * of caution.
2692  *
2693  * In the long run, i_mutex is overkill, and we should probably look
2694  * at using the i_lock spinlock to protect i_flags, and then make sure
2695  * it is so documented in include/linux/fs.h and that all code follows
2696  * the locking convention!!
2697  */
2698 void inode_set_flags(struct inode *inode, unsigned int flags,
2699 		     unsigned int mask)
2700 {
2701 	WARN_ON_ONCE(flags & ~mask);
2702 	set_mask_bits(&inode->i_flags, mask, flags);
2703 }
2704 EXPORT_SYMBOL(inode_set_flags);
2705 
2706 void inode_nohighmem(struct inode *inode)
2707 {
2708 	mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2709 }
2710 EXPORT_SYMBOL(inode_nohighmem);
2711 
2712 struct timespec64 inode_set_ctime_to_ts(struct inode *inode, struct timespec64 ts)
2713 {
2714 	trace_inode_set_ctime_to_ts(inode, &ts);
2715 	set_normalized_timespec64(&ts, ts.tv_sec, ts.tv_nsec);
2716 	inode->i_ctime_sec = ts.tv_sec;
2717 	inode->i_ctime_nsec = ts.tv_nsec;
2718 	return ts;
2719 }
2720 EXPORT_SYMBOL(inode_set_ctime_to_ts);
2721 
2722 /**
2723  * timestamp_truncate - Truncate timespec to a granularity
2724  * @t: Timespec
2725  * @inode: inode being updated
2726  *
2727  * Truncate a timespec to the granularity supported by the fs
2728  * containing the inode. Always rounds down. gran must
2729  * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2730  */
2731 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2732 {
2733 	struct super_block *sb = inode->i_sb;
2734 	unsigned int gran = sb->s_time_gran;
2735 
2736 	t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2737 	if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2738 		t.tv_nsec = 0;
2739 
2740 	/* Avoid division in the common cases 1 ns and 1 s. */
2741 	if (gran == 1)
2742 		; /* nothing */
2743 	else if (gran == NSEC_PER_SEC)
2744 		t.tv_nsec = 0;
2745 	else if (gran > 1 && gran < NSEC_PER_SEC)
2746 		t.tv_nsec -= t.tv_nsec % gran;
2747 	else
2748 		WARN(1, "invalid file time granularity: %u", gran);
2749 	return t;
2750 }
2751 EXPORT_SYMBOL(timestamp_truncate);
2752 
2753 /**
2754  * inode_set_ctime_current - set the ctime to current_time
2755  * @inode: inode
2756  *
2757  * Set the inode's ctime to the current value for the inode. Returns the
2758  * current value that was assigned. If this is not a multigrain inode, then we
2759  * set it to the later of the coarse time and floor value.
2760  *
2761  * If it is multigrain, then we first see if the coarse-grained timestamp is
2762  * distinct from what is already there. If so, then use that. Otherwise, get a
2763  * fine-grained timestamp.
2764  *
2765  * After that, try to swap the new value into i_ctime_nsec. Accept the
2766  * resulting ctime, regardless of the outcome of the swap. If it has
2767  * already been replaced, then that timestamp is later than the earlier
2768  * unacceptable one, and is thus acceptable.
2769  */
2770 struct timespec64 inode_set_ctime_current(struct inode *inode)
2771 {
2772 	struct timespec64 now;
2773 	u32 cns, cur;
2774 
2775 	ktime_get_coarse_real_ts64_mg(&now);
2776 	now = timestamp_truncate(now, inode);
2777 
2778 	/* Just return that if this is not a multigrain fs */
2779 	if (!is_mgtime(inode)) {
2780 		inode_set_ctime_to_ts(inode, now);
2781 		goto out;
2782 	}
2783 
2784 	/*
2785 	 * A fine-grained time is only needed if someone has queried
2786 	 * for timestamps, and the current coarse grained time isn't
2787 	 * later than what's already there.
2788 	 */
2789 	cns = smp_load_acquire(&inode->i_ctime_nsec);
2790 	if (cns & I_CTIME_QUERIED) {
2791 		struct timespec64 ctime = { .tv_sec = inode->i_ctime_sec,
2792 					    .tv_nsec = cns & ~I_CTIME_QUERIED };
2793 
2794 		if (timespec64_compare(&now, &ctime) <= 0) {
2795 			ktime_get_real_ts64_mg(&now);
2796 			now = timestamp_truncate(now, inode);
2797 			mgtime_counter_inc(mg_fine_stamps);
2798 		}
2799 	}
2800 	mgtime_counter_inc(mg_ctime_updates);
2801 
2802 	/* No need to cmpxchg if it's exactly the same */
2803 	if (cns == now.tv_nsec && inode->i_ctime_sec == now.tv_sec) {
2804 		trace_ctime_xchg_skip(inode, &now);
2805 		goto out;
2806 	}
2807 	cur = cns;
2808 retry:
2809 	/* Try to swap the nsec value into place. */
2810 	if (try_cmpxchg(&inode->i_ctime_nsec, &cur, now.tv_nsec)) {
2811 		/* If swap occurred, then we're (mostly) done */
2812 		inode->i_ctime_sec = now.tv_sec;
2813 		trace_ctime_ns_xchg(inode, cns, now.tv_nsec, cur);
2814 		mgtime_counter_inc(mg_ctime_swaps);
2815 	} else {
2816 		/*
2817 		 * Was the change due to someone marking the old ctime QUERIED?
2818 		 * If so then retry the swap. This can only happen once since
2819 		 * the only way to clear I_CTIME_QUERIED is to stamp the inode
2820 		 * with a new ctime.
2821 		 */
2822 		if (!(cns & I_CTIME_QUERIED) && (cns | I_CTIME_QUERIED) == cur) {
2823 			cns = cur;
2824 			goto retry;
2825 		}
2826 		/* Otherwise, keep the existing ctime */
2827 		now.tv_sec = inode->i_ctime_sec;
2828 		now.tv_nsec = cur & ~I_CTIME_QUERIED;
2829 	}
2830 out:
2831 	return now;
2832 }
2833 EXPORT_SYMBOL(inode_set_ctime_current);
2834 
2835 /**
2836  * inode_set_ctime_deleg - try to update the ctime on a delegated inode
2837  * @inode: inode to update
2838  * @update: timespec64 to set the ctime
2839  *
2840  * Attempt to atomically update the ctime on behalf of a delegation holder.
2841  *
2842  * The nfs server can call back the holder of a delegation to get updated
2843  * inode attributes, including the mtime. When updating the mtime, update
2844  * the ctime to a value at least equal to that.
2845  *
2846  * This can race with concurrent updates to the inode, in which
2847  * case the update is skipped.
2848  *
2849  * Note that this works even when multigrain timestamps are not enabled,
2850  * so it is used in either case.
2851  */
2852 struct timespec64 inode_set_ctime_deleg(struct inode *inode, struct timespec64 update)
2853 {
2854 	struct timespec64 now, cur_ts;
2855 	u32 cur, old;
2856 
2857 	/* pairs with try_cmpxchg below */
2858 	cur = smp_load_acquire(&inode->i_ctime_nsec);
2859 	cur_ts.tv_nsec = cur & ~I_CTIME_QUERIED;
2860 	cur_ts.tv_sec = inode->i_ctime_sec;
2861 
2862 	/* If the update is older than the existing value, skip it. */
2863 	if (timespec64_compare(&update, &cur_ts) <= 0)
2864 		return cur_ts;
2865 
2866 	ktime_get_coarse_real_ts64_mg(&now);
2867 
2868 	/* Clamp the update to "now" if it's in the future */
2869 	if (timespec64_compare(&update, &now) > 0)
2870 		update = now;
2871 
2872 	update = timestamp_truncate(update, inode);
2873 
2874 	/* No need to update if the values are already the same */
2875 	if (timespec64_equal(&update, &cur_ts))
2876 		return cur_ts;
2877 
2878 	/*
2879 	 * Try to swap the nsec value into place. If it fails, that means
2880 	 * it raced with an update due to a write or similar activity. That
2881 	 * stamp takes precedence, so just skip the update.
2882 	 */
2883 retry:
2884 	old = cur;
2885 	if (try_cmpxchg(&inode->i_ctime_nsec, &cur, update.tv_nsec)) {
2886 		inode->i_ctime_sec = update.tv_sec;
2887 		mgtime_counter_inc(mg_ctime_swaps);
2888 		return update;
2889 	}
2890 
2891 	/*
2892 	 * Was the change due to another task marking the old ctime QUERIED?
2893 	 *
2894 	 * If so, then retry the swap. This can only happen once since
2895 	 * the only way to clear I_CTIME_QUERIED is to stamp the inode
2896 	 * with a new ctime.
2897 	 */
2898 	if (!(old & I_CTIME_QUERIED) && (cur == (old | I_CTIME_QUERIED)))
2899 		goto retry;
2900 
2901 	/* Otherwise, it was a new timestamp. */
2902 	cur_ts.tv_sec = inode->i_ctime_sec;
2903 	cur_ts.tv_nsec = cur & ~I_CTIME_QUERIED;
2904 	return cur_ts;
2905 }
2906 EXPORT_SYMBOL(inode_set_ctime_deleg);
2907 
2908 /**
2909  * in_group_or_capable - check whether caller is CAP_FSETID privileged
2910  * @idmap:	idmap of the mount @inode was found from
2911  * @inode:	inode to check
2912  * @vfsgid:	the new/current vfsgid of @inode
2913  *
2914  * Check whether @vfsgid is in the caller's group list or if the caller is
2915  * privileged with CAP_FSETID over @inode. This can be used to determine
2916  * whether the setgid bit can be kept or must be dropped.
2917  *
2918  * Return: true if the caller is sufficiently privileged, false if not.
2919  */
2920 bool in_group_or_capable(struct mnt_idmap *idmap,
2921 			 const struct inode *inode, vfsgid_t vfsgid)
2922 {
2923 	if (vfsgid_in_group_p(vfsgid))
2924 		return true;
2925 	if (capable_wrt_inode_uidgid(idmap, inode, CAP_FSETID))
2926 		return true;
2927 	return false;
2928 }
2929 EXPORT_SYMBOL(in_group_or_capable);
2930 
2931 /**
2932  * mode_strip_sgid - handle the sgid bit for non-directories
2933  * @idmap: idmap of the mount the inode was created from
2934  * @dir: parent directory inode
2935  * @mode: mode of the file to be created in @dir
2936  *
2937  * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2938  * raised and @dir has the S_ISGID bit raised ensure that the caller is
2939  * either in the group of the parent directory or they have CAP_FSETID
2940  * in their user namespace and are privileged over the parent directory.
2941  * In all other cases, strip the S_ISGID bit from @mode.
2942  *
2943  * Return: the new mode to use for the file
2944  */
2945 umode_t mode_strip_sgid(struct mnt_idmap *idmap,
2946 			const struct inode *dir, umode_t mode)
2947 {
2948 	if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2949 		return mode;
2950 	if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2951 		return mode;
2952 	if (in_group_or_capable(idmap, dir, i_gid_into_vfsgid(idmap, dir)))
2953 		return mode;
2954 	return mode & ~S_ISGID;
2955 }
2956 EXPORT_SYMBOL(mode_strip_sgid);
2957