linux - fs/inode.c

3115 lines · 85.3 KiB

1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	* (C) 1997 Linus Torvalds
4	* (C) 1999 Andrea Arcangeli <[email protected]> (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/fsverity.h>
18	#include <linux/mount.h>
19	#include <linux/posix_acl.h>
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_lock
57	*/
58
59	static unsigned int i_hash_mask __ro_after_init;
60	static unsigned int i_hash_shift __ro_after_init;
61	static struct hlist_head *inode_hashtable __ro_after_init;
62	static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
63
64	/*
65	* Empty aops. Can be used for the cases where the user does not
66	* define any of the address_space operations.
67	*/
68	const struct address_space_operations empty_aops = {
69	};
70	EXPORT_SYMBOL(empty_aops);
71
72	static DEFINE_PER_CPU(unsigned long, nr_inodes);
73	static DEFINE_PER_CPU(unsigned long, nr_unused);
74
75	static struct kmem_cache *inode_cachep __ro_after_init;
76
77	static long get_nr_inodes(void)
78	{
79	int i;
80	long sum = 0;
81	for_each_possible_cpu(i)
82	sum += per_cpu(nr_inodes, i);
83	return sum < 0 ? 0 : sum;
84	}
85
86	static inline long get_nr_inodes_unused(void)
87	{
88	int i;
89	long sum = 0;
90	for_each_possible_cpu(i)
91	sum += per_cpu(nr_unused, i);
92	return sum < 0 ? 0 : sum;
93	}
94
95	long get_nr_dirty_inodes(void)
96	{
97	/* not actually dirty inodes, but a wild approximation */
98	long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
99	return nr_dirty > 0 ? nr_dirty : 0;
100	}
101
102	#ifdef CONFIG_DEBUG_FS
103	static DEFINE_PER_CPU(long, mg_ctime_updates);
104	static DEFINE_PER_CPU(long, mg_fine_stamps);
105	static DEFINE_PER_CPU(long, mg_ctime_swaps);
106
107	static unsigned long get_mg_ctime_updates(void)
108	{
109	unsigned long sum = 0;
110	int i;
111
112	for_each_possible_cpu(i)
113	sum += data_race(per_cpu(mg_ctime_updates, i));
114	return sum;
115	}
116
117	static unsigned long get_mg_fine_stamps(void)
118	{
119	unsigned long sum = 0;
120	int i;
121
122	for_each_possible_cpu(i)
123	sum += data_race(per_cpu(mg_fine_stamps, i));
124	return sum;
125	}
126
127	static unsigned long get_mg_ctime_swaps(void)
128	{
129	unsigned long sum = 0;
130	int i;
131
132	for_each_possible_cpu(i)
133	sum += data_race(per_cpu(mg_ctime_swaps, i));
134	return sum;
135	}
136
137	#define mgtime_counter_inc(__var) this_cpu_inc(__var)
138
139	static int mgts_show(struct seq_file s, void p)
140	{
141	unsigned long ctime_updates = get_mg_ctime_updates();
142	unsigned long ctime_swaps = get_mg_ctime_swaps();
143	unsigned long fine_stamps = get_mg_fine_stamps();
144	unsigned long floor_swaps = timekeeping_get_mg_floor_swaps();
145
146	seq_printf(s, "%lu %lu %lu %lu\n",
147	ctime_updates, ctime_swaps, fine_stamps, floor_swaps);
148	return 0;
149	}
150
151	DEFINE_SHOW_ATTRIBUTE(mgts);
152
153	static int __init mg_debugfs_init(void)
154	{
155	debugfs_create_file("multigrain_timestamps", S_IFREG \| S_IRUGO, NULL, NULL, &mgts_fops);
156	return 0;
157	}
158	late_initcall(mg_debugfs_init);
159
160	#else /* ! CONFIG_DEBUG_FS */
161
162	#define mgtime_counter_inc(__var) do { } while (0)
163
164	#endif /* CONFIG_DEBUG_FS */
165
166	/*
167	* Handle nr_inode sysctl
168	*/
169	#ifdef CONFIG_SYSCTL
170	/*
171	* Statistics gathering..
172	*/
173	static struct inodes_stat_t inodes_stat;
174
175	static int proc_nr_inodes(const struct ctl_table table, int write, void buffer,
176	size_t lenp, loff_t ppos)
177	{
178	inodes_stat.nr_inodes = get_nr_inodes();
179	inodes_stat.nr_unused = get_nr_inodes_unused();
180	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
181	}
182
183	static const struct ctl_table inodes_sysctls[] = {
184	{
185	.procname = "inode-nr",
186	.data = &inodes_stat,
187	.maxlen = 2*sizeof(long),
188	.mode = 0444,
189	.proc_handler = proc_nr_inodes,
190	},
191	{
192	.procname = "inode-state",
193	.data = &inodes_stat,
194	.maxlen = 7*sizeof(long),
195	.mode = 0444,
196	.proc_handler = proc_nr_inodes,
197	},
198	};
199
200	static int __init init_fs_inode_sysctls(void)
201	{
202	register_sysctl_init("fs", inodes_sysctls);
203	return 0;
204	}
205	early_initcall(init_fs_inode_sysctls);
206	#endif
207
208	static int no_open(struct inode inode, struct file file)
209	{
210	return -ENXIO;
211	}
212
213	/**
214	* inode_init_always_gfp - perform inode structure initialisation
215	* @sb: superblock inode belongs to
216	* @inode: inode to initialise
217	* @gfp: allocation flags
218	*
219	* These are initializations that need to be done on every inode
220	* allocation as the fields are not initialised by slab allocation.
221	* If there are additional allocations required @gfp is used.
222	*/
223	int inode_init_always_gfp(struct super_block sb, struct inode inode, gfp_t gfp)
224	{
225	static const struct inode_operations empty_iops;
226	static const struct file_operations no_open_fops = {.open = no_open};
227	struct address_space *const mapping = &inode->i_data;
228
229	inode->i_sb = sb;
230	inode->i_blkbits = sb->s_blocksize_bits;
231	inode->i_flags = 0;
232	inode_state_assign_raw(inode, 0);
233	atomic64_set(&inode->i_sequence, 0);
234	atomic_set(&inode->i_count, 1);
235	inode->i_op = &empty_iops;
236	inode->i_fop = &no_open_fops;
237	inode->i_ino = 0;
238	inode->__i_nlink = 1;
239	inode->i_opflags = 0;
240	if (sb->s_xattr)
241	inode->i_opflags \|= IOP_XATTR;
242	if (sb->s_type->fs_flags & FS_MGTIME)
243	inode->i_opflags \|= IOP_MGTIME;
244	i_uid_write(inode, 0);
245	i_gid_write(inode, 0);
246	atomic_set(&inode->i_writecount, 0);
247	inode->i_size = 0;
248	inode->i_write_hint = WRITE_LIFE_NOT_SET;
249	inode->i_blocks = 0;
250	inode->i_bytes = 0;
251	inode->i_generation = 0;
252	inode->i_pipe = NULL;
253	inode->i_cdev = NULL;
254	inode->i_link = NULL;
255	inode->i_dir_seq = 0;
256	inode->i_rdev = 0;
257	inode->dirtied_when = 0;
258
259	#ifdef CONFIG_CGROUP_WRITEBACK
260	inode->i_wb_frn_winner = 0;
261	inode->i_wb_frn_avg_time = 0;
262	inode->i_wb_frn_history = 0;
263	#endif
264
265	spin_lock_init(&inode->i_lock);
266	lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
267
268	init_rwsem(&inode->i_rwsem);
269	lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
270
271	atomic_set(&inode->i_dio_count, 0);
272
273	mapping->a_ops = &empty_aops;
274	mapping->host = inode;
275	mapping->flags = 0;
276	mapping->wb_err = 0;
277	atomic_set(&mapping->i_mmap_writable, 0);
278	mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
279	mapping->writeback_index = 0;
280	init_rwsem(&mapping->invalidate_lock);
281	lockdep_set_class_and_name(&mapping->invalidate_lock,
282	&sb->s_type->invalidate_lock_key,
283	"mapping.invalidate_lock");
284	if (sb->s_iflags & SB_I_STABLE_WRITES)
285	mapping_set_stable_writes(mapping);
286	inode->i_private = NULL;
287	inode->i_mapping = mapping;
288	INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
289	#ifdef CONFIG_FS_POSIX_ACL
290	inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
291	#endif
292
293	#ifdef CONFIG_FSNOTIFY
294	inode->i_fsnotify_mask = 0;
295	#endif
296	inode->i_flctx = NULL;
297
298	if (unlikely(security_inode_alloc(inode, gfp)))
299	return -ENOMEM;
300
301	this_cpu_inc(nr_inodes);
302
303	return 0;
304	}
305	EXPORT_SYMBOL(inode_init_always_gfp);
306
307	void free_inode_nonrcu(struct inode *inode)
308	{
309	kmem_cache_free(inode_cachep, inode);
310	}
311	EXPORT_SYMBOL(free_inode_nonrcu);
312
313	static void i_callback(struct rcu_head *head)
314	{
315	struct inode *inode = container_of(head, struct inode, i_rcu);
316	if (inode->free_inode)
317	inode->free_inode(inode);
318	else
319	free_inode_nonrcu(inode);
320	}
321
322	/**
323	* alloc_inode - obtain an inode
324	* @sb: superblock
325	*
326	* Allocates a new inode for given superblock.
327	* Inode wont be chained in superblock s_inodes list
328	* This means :
329	* - fs can't be unmount
330	* - quotas, fsnotify, writeback can't work
331	*/
332	struct inode alloc_inode(struct super_block sb)
333	{
334	const struct super_operations *ops = sb->s_op;
335	struct inode *inode;
336
337	if (ops->alloc_inode)
338	inode = ops->alloc_inode(sb);
339	else
340	inode = alloc_inode_sb(sb, inode_cachep, GFP_KERNEL);
341
342	if (!inode)
343	return NULL;
344
345	if (unlikely(inode_init_always(sb, inode))) {
346	if (ops->destroy_inode) {
347	ops->destroy_inode(inode);
348	if (!ops->free_inode)
349	return NULL;
350	}
351	inode->free_inode = ops->free_inode;
352	i_callback(&inode->i_rcu);
353	return NULL;
354	}
355
356	return inode;
357	}
358
359	void __destroy_inode(struct inode *inode)
360	{
361	inode_detach_wb(inode);
362	security_inode_free(inode);
363	fsnotify_inode_delete(inode);
364	locks_free_lock_context(inode);
365	if (!inode->i_nlink) {
366	WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
367	atomic_long_dec(&inode->i_sb->s_remove_count);
368	}
369
370	#ifdef CONFIG_FS_POSIX_ACL
371	if (inode->i_acl && !is_uncached_acl(inode->i_acl))
372	posix_acl_release(inode->i_acl);
373	if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
374	posix_acl_release(inode->i_default_acl);
375	#endif
376	this_cpu_dec(nr_inodes);
377	}
378	EXPORT_SYMBOL(__destroy_inode);
379
380	static void destroy_inode(struct inode *inode)
381	{
382	const struct super_operations *ops = inode->i_sb->s_op;
383
384	BUG_ON(!list_empty(&inode->i_lru));
385	__destroy_inode(inode);
386	if (ops->destroy_inode) {
387	ops->destroy_inode(inode);
388	if (!ops->free_inode)
389	return;
390	}
391	inode->free_inode = ops->free_inode;
392	call_rcu(&inode->i_rcu, i_callback);
393	}
394
395	/**
396	* drop_nlink - directly drop an inode's link count
397	* @inode: inode
398	*
399	* This is a low-level filesystem helper to replace any
400	* direct filesystem manipulation of i_nlink. In cases
401	* where we are attempting to track writes to the
402	* filesystem, a decrement to zero means an imminent
403	* write when the file is truncated and actually unlinked
404	* on the filesystem.
405	*/
406	void drop_nlink(struct inode *inode)
407	{
408	WARN_ON(inode->i_nlink == 0);
409	inode->__i_nlink--;
410	if (!inode->i_nlink)
411	atomic_long_inc(&inode->i_sb->s_remove_count);
412	}
413	EXPORT_SYMBOL(drop_nlink);
414
415	/**
416	* clear_nlink - directly zero an inode's link count
417	* @inode: inode
418	*
419	* This is a low-level filesystem helper to replace any
420	* direct filesystem manipulation of i_nlink. See
421	* drop_nlink() for why we care about i_nlink hitting zero.
422	*/
423	void clear_nlink(struct inode *inode)
424	{
425	if (inode->i_nlink) {
426	inode->__i_nlink = 0;
427	atomic_long_inc(&inode->i_sb->s_remove_count);
428	}
429	}
430	EXPORT_SYMBOL(clear_nlink);
431
432	/**
433	* set_nlink - directly set an inode's link count
434	* @inode: inode
435	* @nlink: new nlink (should be non-zero)
436	*
437	* This is a low-level filesystem helper to replace any
438	* direct filesystem manipulation of i_nlink.
439	*/
440	void set_nlink(struct inode *inode, unsigned int nlink)
441	{
442	if (!nlink) {
443	clear_nlink(inode);
444	} else {
445	/* Yes, some filesystems do change nlink from zero to one */
446	if (inode->i_nlink == 0)
447	atomic_long_dec(&inode->i_sb->s_remove_count);
448
449	inode->__i_nlink = nlink;
450	}
451	}
452	EXPORT_SYMBOL(set_nlink);
453
454	/**
455	* inc_nlink - directly increment an inode's link count
456	* @inode: inode
457	*
458	* This is a low-level filesystem helper to replace any
459	* direct filesystem manipulation of i_nlink. Currently,
460	* it is only here for parity with dec_nlink().
461	*/
462	void inc_nlink(struct inode *inode)
463	{
464	if (unlikely(inode->i_nlink == 0)) {
465	WARN_ON(!(inode_state_read_once(inode) & I_LINKABLE));
466	atomic_long_dec(&inode->i_sb->s_remove_count);
467	}
468
469	inode->__i_nlink++;
470	}
471	EXPORT_SYMBOL(inc_nlink);
472
473	static void __address_space_init_once(struct address_space *mapping)
474	{
475	xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ \| XA_FLAGS_ACCOUNT);
476	init_rwsem(&mapping->i_mmap_rwsem);
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	struct wait_queue_head inode_bit_waitqueue(struct wait_bit_queue_entry wqe,
515	struct inode *inode, u32 bit)
516	{
517	void *bit_address;
518
519	bit_address = inode_state_wait_address(inode, bit);
520	init_wait_var_entry(wqe, bit_address, 0);
521	return __var_waitqueue(bit_address);
522	}
523	EXPORT_SYMBOL(inode_bit_waitqueue);
524
525	void wait_on_new_inode(struct inode *inode)
526	{
527	struct wait_bit_queue_entry wqe;
528	struct wait_queue_head *wq_head;
529
530	spin_lock(&inode->i_lock);
531	if (!(inode_state_read(inode) & I_NEW)) {
532	spin_unlock(&inode->i_lock);
533	return;
534	}
535
536	wq_head = inode_bit_waitqueue(&wqe, inode, __I_NEW);
537	for (;;) {
538	prepare_to_wait_event(wq_head, &wqe.wq_entry, TASK_UNINTERRUPTIBLE);
539	if (!(inode_state_read(inode) & I_NEW))
540	break;
541	spin_unlock(&inode->i_lock);
542	schedule();
543	spin_lock(&inode->i_lock);
544	}
545	finish_wait(wq_head, &wqe.wq_entry);
546	WARN_ON(inode_state_read(inode) & I_NEW);
547	spin_unlock(&inode->i_lock);
548	}
549	EXPORT_SYMBOL(wait_on_new_inode);
550
551	static void __inode_lru_list_add(struct inode *inode, bool rotate)
552	{
553	lockdep_assert_held(&inode->i_lock);
554
555	if (inode_state_read(inode) & (I_DIRTY_ALL \| I_SYNC \| I_FREEING \| I_WILL_FREE))
556	return;
557	if (icount_read(inode))
558	return;
559	if (!(inode->i_sb->s_flags & SB_ACTIVE))
560	return;
561	if (!mapping_shrinkable(&inode->i_data))
562	return;
563
564	if (list_lru_add_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
565	this_cpu_inc(nr_unused);
566	else if (rotate)
567	inode_state_set(inode, I_REFERENCED);
568	}
569
570	/*
571	* Add inode to LRU if needed (inode is unused and clean).
572	*/
573	void inode_lru_list_add(struct inode *inode)
574	{
575	__inode_lru_list_add(inode, false);
576	}
577
578	static void inode_lru_list_del(struct inode *inode)
579	{
580	if (list_empty(&inode->i_lru))
581	return;
582
583	if (list_lru_del_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
584	this_cpu_dec(nr_unused);
585	}
586
587	static void inode_pin_lru_isolating(struct inode *inode)
588	{
589	lockdep_assert_held(&inode->i_lock);
590	WARN_ON(inode_state_read(inode) & (I_LRU_ISOLATING \| I_FREEING \| I_WILL_FREE));
591	inode_state_set(inode, I_LRU_ISOLATING);
592	}
593
594	static void inode_unpin_lru_isolating(struct inode *inode)
595	{
596	spin_lock(&inode->i_lock);
597	WARN_ON(!(inode_state_read(inode) & I_LRU_ISOLATING));
598	inode_state_clear(inode, I_LRU_ISOLATING);
599	/* Called with inode->i_lock which ensures memory ordering. */
600	inode_wake_up_bit(inode, __I_LRU_ISOLATING);
601	spin_unlock(&inode->i_lock);
602	}
603
604	static void inode_wait_for_lru_isolating(struct inode *inode)
605	{
606	struct wait_bit_queue_entry wqe;
607	struct wait_queue_head *wq_head;
608
609	lockdep_assert_held(&inode->i_lock);
610	if (!(inode_state_read(inode) & I_LRU_ISOLATING))
611	return;
612
613	wq_head = inode_bit_waitqueue(&wqe, inode, __I_LRU_ISOLATING);
614	for (;;) {
615	prepare_to_wait_event(wq_head, &wqe.wq_entry, TASK_UNINTERRUPTIBLE);
616	/*
617	* Checking I_LRU_ISOLATING with inode->i_lock guarantees
618	* memory ordering.
619	*/
620	if (!(inode_state_read(inode) & I_LRU_ISOLATING))
621	break;
622	spin_unlock(&inode->i_lock);
623	schedule();
624	spin_lock(&inode->i_lock);
625	}
626	finish_wait(wq_head, &wqe.wq_entry);
627	WARN_ON(inode_state_read(inode) & I_LRU_ISOLATING);
628	}
629
630	/**
631	* inode_sb_list_add - add inode to the superblock list of inodes
632	* @inode: inode to add
633	*/
634	void inode_sb_list_add(struct inode *inode)
635	{
636	struct super_block *sb = inode->i_sb;
637
638	spin_lock(&sb->s_inode_list_lock);
639	list_add(&inode->i_sb_list, &sb->s_inodes);
640	spin_unlock(&sb->s_inode_list_lock);
641	}
642	EXPORT_SYMBOL_GPL(inode_sb_list_add);
643
644	static inline void inode_sb_list_del(struct inode *inode)
645	{
646	struct super_block *sb = inode->i_sb;
647
648	if (!list_empty(&inode->i_sb_list)) {
649	spin_lock(&sb->s_inode_list_lock);
650	list_del_init(&inode->i_sb_list);
651	spin_unlock(&sb->s_inode_list_lock);
652	}
653	}
654
655	static unsigned long hash(struct super_block *sb, u64 hashval)
656	{
657	unsigned long tmp;
658
659	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
660	L1_CACHE_BYTES;
661	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
662	return tmp & i_hash_mask;
663	}
664
665	/**
666	* __insert_inode_hash - hash an inode
667	* @inode: unhashed inode
668	* @hashval: u64 value used to locate this object in the
669	* inode_hashtable.
670	*
671	* Add an inode to the inode hash for this superblock.
672	*/
673	void __insert_inode_hash(struct inode *inode, u64 hashval)
674	{
675	struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
676
677	spin_lock(&inode_hash_lock);
678	spin_lock(&inode->i_lock);
679	hlist_add_head_rcu(&inode->i_hash, b);
680	spin_unlock(&inode->i_lock);
681	spin_unlock(&inode_hash_lock);
682	}
683	EXPORT_SYMBOL(__insert_inode_hash);
684
685	/**
686	* __remove_inode_hash - remove an inode from the hash
687	* @inode: inode to unhash
688	*
689	* Remove an inode from the superblock.
690	*/
691	void __remove_inode_hash(struct inode *inode)
692	{
693	spin_lock(&inode_hash_lock);
694	spin_lock(&inode->i_lock);
695	hlist_del_init_rcu(&inode->i_hash);
696	spin_unlock(&inode->i_lock);
697	spin_unlock(&inode_hash_lock);
698	}
699	EXPORT_SYMBOL(__remove_inode_hash);
700
701	void dump_mapping(const struct address_space *mapping)
702	{
703	struct inode *host;
704	const struct address_space_operations *a_ops;
705	struct hlist_node *dentry_first;
706	struct dentry *dentry_ptr;
707	struct dentry dentry;
708	char fname[64] = {};
709	u64 ino;
710
711	/*
712	* If mapping is an invalid pointer, we don't want to crash
713	* accessing it, so probe everything depending on it carefully.
714	*/
715	if (get_kernel_nofault(host, &mapping->host) \|\|
716	get_kernel_nofault(a_ops, &mapping->a_ops)) {
717	pr_warn("invalid mapping:%px\n", mapping);
718	return;
719	}
720
721	if (!host) {
722	pr_warn("aops:%ps\n", a_ops);
723	return;
724	}
725
726	if (get_kernel_nofault(dentry_first, &host->i_dentry.first) \|\|
727	get_kernel_nofault(ino, &host->i_ino)) {
728	pr_warn("aops:%ps invalid inode:%px\n", a_ops, host);
729	return;
730	}
731
732	if (!dentry_first) {
733	pr_warn("aops:%ps ino:%llx\n", a_ops, ino);
734	return;
735	}
736
737	dentry_ptr = container_of(dentry_first, struct dentry, d_alias);
738	if (get_kernel_nofault(dentry, dentry_ptr) \|\|
739	!dentry.d_parent \|\| !dentry.d_name.name) {
740	pr_warn("aops:%ps ino:%llx invalid dentry:%px\n",
741	a_ops, ino, dentry_ptr);
742	return;
743	}
744
745	if (strncpy_from_kernel_nofault(fname, dentry.d_name.name, 63) < 0)
746	strscpy(fname, "<invalid>");
747	/*
748	* Even if strncpy_from_kernel_nofault() succeeded,
749	* the fname could be unreliable
750	*/
751	pr_warn("aops:%ps ino:%llx dentry name(?):\"%s\"\n",
752	a_ops, ino, fname);
753	}
754
755	void clear_inode(struct inode *inode)
756	{
757	/*
758	* Only IS_VERITY() inodes can have verity info, so start by checking
759	* for IS_VERITY() (which is faster than retrieving the pointer to the
760	* verity info). This minimizes overhead for non-verity inodes.
761	*/
762	if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
763	fsverity_cleanup_inode(inode);
764
765	/*
766	* We have to cycle the i_pages lock here because reclaim
767	* can be in the process of removing the last page (in
768	* __filemap_remove_folio()) and we must not free the mapping
769	* under it. We also remove nodes which are empty; these
770	* can occur in two different ways. The first is that radix
771	* tree expansion can fail partway and the second is that THP
772	* collapse_file() can allocate some temporary nodes and not
773	* clean them up.
774	*/
775	xa_destroy(&inode->i_data.i_pages);
776
777	BUG_ON(inode->i_data.nrpages);
778	BUG_ON(!(inode_state_read_once(inode) & I_FREEING));
779	BUG_ON(inode_state_read_once(inode) & I_CLEAR);
780	BUG_ON(!list_empty(&inode->i_wb_list));
781	/* don't need i_lock here, no concurrent mods to i_state */
782	inode_state_assign_raw(inode, I_FREEING \| I_CLEAR);
783	}
784	EXPORT_SYMBOL(clear_inode);
785
786	/*
787	* Free the inode passed in, removing it from the lists it is still connected
788	* to. We remove any pages still attached to the inode and wait for any IO that
789	* is still in progress before finally destroying the inode.
790	*
791	* An inode must already be marked I_FREEING so that we avoid the inode being
792	* moved back onto lists if we race with other code that manipulates the lists
793	* (e.g. writeback_single_inode). The caller is responsible for setting this.
794	*
795	* An inode must already be removed from the LRU list before being evicted from
796	* the cache. This should occur atomically with setting the I_FREEING state
797	* flag, so no inodes here should ever be on the LRU when being evicted.
798	*/
799	static void evict(struct inode *inode)
800	{
801	const struct super_operations *op = inode->i_sb->s_op;
802
803	BUG_ON(!(inode_state_read_once(inode) & I_FREEING));
804	BUG_ON(!list_empty(&inode->i_lru));
805
806	inode_io_list_del(inode);
807	inode_sb_list_del(inode);
808
809	spin_lock(&inode->i_lock);
810	inode_wait_for_lru_isolating(inode);
811
812	/*
813	* Wait for flusher thread to be done with the inode so that filesystem
814	* does not start destroying it while writeback is still running. Since
815	* the inode has I_FREEING set, flusher thread won't start new work on
816	* the inode. We just have to wait for running writeback to finish.
817	*/
818	inode_wait_for_writeback(inode);
819	spin_unlock(&inode->i_lock);
820
821	if (op->evict_inode) {
822	op->evict_inode(inode);
823	} else {
824	truncate_inode_pages_final(&inode->i_data);
825	clear_inode(inode);
826	}
827	if (S_ISCHR(inode->i_mode) && inode->i_cdev)
828	cd_forget(inode);
829
830	remove_inode_hash(inode);
831
832	/*
833	* Wake up waiters in __wait_on_freeing_inode().
834	*
835	* It is an invariant that any thread we need to wake up is already
836	* accounted for before remove_inode_hash() acquires ->i_lock -- both
837	* sides take the lock and sleep is aborted if the inode is found
838	* unhashed. Thus either the sleeper wins and goes off CPU, or removal
839	* wins and the sleeper aborts after testing with the lock.
840	*
841	* This also means we don't need any fences for the call below.
842	*/
843	inode_wake_up_bit(inode, __I_NEW);
844	BUG_ON(inode_state_read_once(inode) != (I_FREEING \| I_CLEAR));
845
846	destroy_inode(inode);
847	}
848
849	/*
850	* dispose_list - dispose of the contents of a local list
851	* @head: the head of the list to free
852	*
853	* Dispose-list gets a local list with local inodes in it, so it doesn't
854	* need to worry about list corruption and SMP locks.
855	*/
856	static void dispose_list(struct list_head *head)
857	{
858	while (!list_empty(head)) {
859	struct inode *inode;
860
861	inode = list_first_entry(head, struct inode, i_lru);
862	list_del_init(&inode->i_lru);
863
864	evict(inode);
865	cond_resched();
866	}
867	}
868
869	/**
870	* evict_inodes - evict all evictable inodes for a superblock
871	* @sb: superblock to operate on
872	*
873	* Make sure that no inodes with zero refcount are retained. This is
874	* called by superblock shutdown after having SB_ACTIVE flag removed,
875	* so any inode reaching zero refcount during or after that call will
876	* be immediately evicted.
877	*/
878	void evict_inodes(struct super_block *sb)
879	{
880	struct inode *inode;
881	LIST_HEAD(dispose);
882
883	spin_lock(&sb->s_inode_list_lock);
884	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
885	if (icount_read_once(inode))
886	continue;
887
888	spin_lock(&inode->i_lock);
889	if (icount_read(inode)) {
890	spin_unlock(&inode->i_lock);
891	continue;
892	}
893	if (inode_state_read(inode) & (I_NEW \| I_FREEING \| I_WILL_FREE)) {
894	spin_unlock(&inode->i_lock);
895	continue;
896	}
897
898	inode_state_set(inode, I_FREEING);
899	inode_lru_list_del(inode);
900	spin_unlock(&inode->i_lock);
901
902	/*
903	* Keep this inode out of dispose so it stays on s_inodes while
904	* the list lock is dropped. I_FREEING prevents new references
905	* and leaves eviction to us, so we can resume the walk from it.
906	*/
907	if (need_resched()) {
908	spin_unlock(&sb->s_inode_list_lock);
909	cond_resched();
910	dispose_list(&dispose);
911	spin_lock(&sb->s_inode_list_lock);
912	}
913	list_add(&inode->i_lru, &dispose);
914	}
915	spin_unlock(&sb->s_inode_list_lock);
916
917	dispose_list(&dispose);
918	}
919	EXPORT_SYMBOL_GPL(evict_inodes);
920
921	/*
922	* Isolate the inode from the LRU in preparation for freeing it.
923	*
924	* If the inode has the I_REFERENCED flag set, then it means that it has been
925	* used recently - the flag is set in iput_final(). When we encounter such an
926	* inode, clear the flag and move it to the back of the LRU so it gets another
927	* pass through the LRU before it gets reclaimed. This is necessary because of
928	* the fact we are doing lazy LRU updates to minimise lock contention so the
929	* LRU does not have strict ordering. Hence we don't want to reclaim inodes
930	* with this flag set because they are the inodes that are out of order.
931	*/
932	static enum lru_status inode_lru_isolate(struct list_head *item,
933	struct list_lru_one lru, void arg)
934	{
935	struct list_head *freeable = arg;
936	struct inode *inode = container_of(item, struct inode, i_lru);
937
938	/*
939	* We are inverting the lru lock/inode->i_lock here, so use a
940	* trylock. If we fail to get the lock, just skip it.
941	*/
942	if (!spin_trylock(&inode->i_lock))
943	return LRU_SKIP;
944
945	/*
946	* Inodes can get referenced, redirtied, or repopulated while
947	* they're already on the LRU, and this can make them
948	* unreclaimable for a while. Remove them lazily here; iput,
949	* sync, or the last page cache deletion will requeue them.
950	*/
951	if (icount_read(inode) \|\|
952	(inode_state_read(inode) & ~I_REFERENCED) \|\|
953	!mapping_shrinkable(&inode->i_data)) {
954	list_lru_isolate(lru, &inode->i_lru);
955	spin_unlock(&inode->i_lock);
956	this_cpu_dec(nr_unused);
957	return LRU_REMOVED;
958	}
959
960	/* Recently referenced inodes get one more pass */
961	if (inode_state_read(inode) & I_REFERENCED) {
962	inode_state_clear(inode, I_REFERENCED);
963	spin_unlock(&inode->i_lock);
964	return LRU_ROTATE;
965	}
966
967	/*
968	* On highmem systems, mapping_shrinkable() permits dropping
969	* page cache in order to free up struct inodes: lowmem might
970	* be under pressure before the cache inside the highmem zone.
971	*/
972	if (!mapping_empty(&inode->i_data)) {
973	unsigned long reap;
974
975	inode_pin_lru_isolating(inode);
976	spin_unlock(&inode->i_lock);
977	spin_unlock(&lru->lock);
978	reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
979	if (current_is_kswapd())
980	__count_vm_events(KSWAPD_INODESTEAL, reap);
981	else
982	__count_vm_events(PGINODESTEAL, reap);
983	mm_account_reclaimed_pages(reap);
984	inode_unpin_lru_isolating(inode);
985	return LRU_RETRY;
986	}
987
988	WARN_ON(inode_state_read(inode) & I_NEW);
989	inode_state_set(inode, I_FREEING);
990	list_lru_isolate_move(lru, &inode->i_lru, freeable);
991	spin_unlock(&inode->i_lock);
992
993	this_cpu_dec(nr_unused);
994	return LRU_REMOVED;
995	}
996
997	/*
998	* Walk the superblock inode LRU for freeable inodes and attempt to free them.
999	* This is called from the superblock shrinker function with a number of inodes
1000	* to trim from the LRU. Inodes to be freed are moved to a temporary list and
1001	* then are freed outside inode_lock by dispose_list().
1002	*/
1003	long prune_icache_sb(struct super_block sb, struct shrink_control sc)
1004	{
1005	LIST_HEAD(freeable);
1006	long freed;
1007
1008	freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
1009	inode_lru_isolate, &freeable);
1010	dispose_list(&freeable);
1011	return freed;
1012	}
1013
1014	static void __wait_on_freeing_inode(struct inode *inode, bool hash_locked, bool rcu_locked);
1015	static bool igrab_from_hash(struct inode *inode);
1016
1017	/*
1018	* Called with the inode lock held.
1019	*/
1020	static struct inode find_inode(struct super_block sb,
1021	struct hlist_head *head,
1022	int (test)(struct inode , void *),
1023	void *data, bool hash_locked,
1024	bool *isnew)
1025	{
1026	struct inode *inode = NULL;
1027
1028	if (hash_locked)
1029	lockdep_assert_held(&inode_hash_lock);
1030	else
1031	lockdep_assert_not_held(&inode_hash_lock);
1032
1033	rcu_read_lock();
1034	repeat:
1035	hlist_for_each_entry_rcu(inode, head, i_hash) {
1036	if (inode->i_sb != sb)
1037	continue;
1038	if (!test(inode, data))
1039	continue;
1040	if (igrab_from_hash(inode)) {
1041	rcu_read_unlock();
1042	*isnew = false;
1043	return inode;
1044	}
1045	spin_lock(&inode->i_lock);
1046	if (inode_state_read(inode) & (I_FREEING \| I_WILL_FREE)) {
1047	__wait_on_freeing_inode(inode, hash_locked, true);
1048	goto repeat;
1049	}
1050	if (unlikely(inode_state_read(inode) & I_CREATING)) {
1051	spin_unlock(&inode->i_lock);
1052	rcu_read_unlock();
1053	return ERR_PTR(-ESTALE);
1054	}
1055	__iget(inode);
1056	*isnew = !!(inode_state_read(inode) & I_NEW);
1057	spin_unlock(&inode->i_lock);
1058	rcu_read_unlock();
1059	return inode;
1060	}
1061	rcu_read_unlock();
1062	return NULL;
1063	}
1064
1065	/*
1066	* find_inode_fast is the fast path version of find_inode, see the comment at
1067	* iget_locked for details.
1068	*/
1069	static struct inode find_inode_fast(struct super_block sb,
1070	struct hlist_head *head, u64 ino,
1071	bool hash_locked, bool *isnew)
1072	{
1073	struct inode *inode = NULL;
1074
1075	if (hash_locked)
1076	lockdep_assert_held(&inode_hash_lock);
1077	else
1078	lockdep_assert_not_held(&inode_hash_lock);
1079
1080	rcu_read_lock();
1081	repeat:
1082	hlist_for_each_entry_rcu(inode, head, i_hash) {
1083	if (inode->i_ino != ino)
1084	continue;
1085	if (inode->i_sb != sb)
1086	continue;
1087	if (igrab_from_hash(inode)) {
1088	rcu_read_unlock();
1089	*isnew = false;
1090	return inode;
1091	}
1092	spin_lock(&inode->i_lock);
1093	if (inode_state_read(inode) & (I_FREEING \| I_WILL_FREE)) {
1094	__wait_on_freeing_inode(inode, hash_locked, true);
1095	goto repeat;
1096	}
1097	if (unlikely(inode_state_read(inode) & I_CREATING)) {
1098	spin_unlock(&inode->i_lock);
1099	rcu_read_unlock();
1100	return ERR_PTR(-ESTALE);
1101	}
1102	__iget(inode);
1103	*isnew = !!(inode_state_read(inode) & I_NEW);
1104	spin_unlock(&inode->i_lock);
1105	rcu_read_unlock();
1106	return inode;
1107	}
1108	rcu_read_unlock();
1109	return NULL;
1110	}
1111
1112	/*
1113	* Each cpu owns a range of LAST_INO_BATCH numbers.
1114	* 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
1115	* to renew the exhausted range.
1116	*
1117	* This does not significantly increase overflow rate because every CPU can
1118	* consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
1119	* NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
1120	* 2^32 range, and is a worst-case. Even a 50% wastage would only increase
1121	* overflow rate by 2x, which does not seem too significant.
1122	*
1123	* On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1124	* error if st_ino won't fit in target struct field. Use 32bit counter
1125	* here to attempt to avoid that.
1126	*/
1127	#define LAST_INO_BATCH 1024
1128	static DEFINE_PER_CPU(unsigned int, last_ino);
1129
1130	unsigned int get_next_ino(void)
1131	{
1132	unsigned int *p = &get_cpu_var(last_ino);
1133	unsigned int res = *p;
1134
1135	#ifdef CONFIG_SMP
1136	if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
1137	static atomic_t shared_last_ino;
1138	int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
1139
1140	res = next - LAST_INO_BATCH;
1141	}
1142	#endif
1143
1144	res++;
1145	/* get_next_ino should not provide a 0 inode number */
1146	if (unlikely(!res))
1147	res++;
1148	*p = res;
1149	put_cpu_var(last_ino);
1150	return res;
1151	}
1152	EXPORT_SYMBOL(get_next_ino);
1153
1154	/**
1155	* new_inode - obtain an inode
1156	* @sb: superblock
1157	*
1158	* Allocates a new inode for given superblock. The default gfp_mask
1159	* for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1160	* If HIGHMEM pages are unsuitable or it is known that pages allocated
1161	* for the page cache are not reclaimable or migratable,
1162	* mapping_set_gfp_mask() must be called with suitable flags on the
1163	* newly created inode's mapping
1164	*
1165	*/
1166	struct inode new_inode(struct super_block sb)
1167	{
1168	struct inode *inode;
1169
1170	inode = alloc_inode(sb);
1171	if (inode)
1172	inode_sb_list_add(inode);
1173	return inode;
1174	}
1175	EXPORT_SYMBOL(new_inode);
1176
1177	#ifdef CONFIG_DEBUG_LOCK_ALLOC
1178	void lockdep_annotate_inode_mutex_key(struct inode *inode)
1179	{
1180	if (S_ISDIR(inode->i_mode)) {
1181	struct file_system_type *type = inode->i_sb->s_type;
1182
1183	/* Set new key only if filesystem hasn't already changed it */
1184	if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1185	/*
1186	* ensure nobody is actually holding i_rwsem
1187	*/
1188	init_rwsem(&inode->i_rwsem);
1189	lockdep_set_class(&inode->i_rwsem,
1190	&type->i_mutex_dir_key);
1191	}
1192	}
1193	}
1194	EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1195	#endif
1196
1197	/**
1198	* unlock_new_inode - clear the I_NEW state and wake up any waiters
1199	* @inode: new inode to unlock
1200	*
1201	* Called when the inode is fully initialised to clear the new state of the
1202	* inode and wake up anyone waiting for the inode to finish initialisation.
1203	*/
1204	void unlock_new_inode(struct inode *inode)
1205	{
1206	lockdep_annotate_inode_mutex_key(inode);
1207	spin_lock(&inode->i_lock);
1208	WARN_ON(!(inode_state_read(inode) & I_NEW));
1209	/*
1210	* Paired with igrab_from_hash()
1211	*/
1212	smp_wmb();
1213	inode_state_clear(inode, I_NEW \| I_CREATING);
1214	inode_wake_up_bit(inode, __I_NEW);
1215	spin_unlock(&inode->i_lock);
1216	}
1217	EXPORT_SYMBOL(unlock_new_inode);
1218
1219	void discard_new_inode(struct inode *inode)
1220	{
1221	lockdep_annotate_inode_mutex_key(inode);
1222	spin_lock(&inode->i_lock);
1223	WARN_ON(!(inode_state_read(inode) & I_NEW));
1224	/*
1225	* Paired with igrab_from_hash()
1226	*/
1227	smp_wmb();
1228	inode_state_clear(inode, I_NEW);
1229	inode_wake_up_bit(inode, __I_NEW);
1230	spin_unlock(&inode->i_lock);
1231	iput(inode);
1232	}
1233	EXPORT_SYMBOL(discard_new_inode);
1234
1235	/**
1236	* lock_two_nondirectories - take two i_mutexes on non-directory objects
1237	*
1238	* Lock any non-NULL argument. Passed objects must not be directories.
1239	* Zero, one or two objects may be locked by this function.
1240	*
1241	* @inode1: first inode to lock
1242	* @inode2: second inode to lock
1243	*/
1244	void lock_two_nondirectories(struct inode inode1, struct inode inode2)
1245	{
1246	if (inode1)
1247	WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1248	if (inode2)
1249	WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1250	if (inode1 > inode2)
1251	swap(inode1, inode2);
1252	if (inode1)
1253	inode_lock(inode1);
1254	if (inode2 && inode2 != inode1)
1255	inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1256	}
1257	EXPORT_SYMBOL(lock_two_nondirectories);
1258
1259	/**
1260	* unlock_two_nondirectories - release locks from lock_two_nondirectories()
1261	* @inode1: first inode to unlock
1262	* @inode2: second inode to unlock
1263	*/
1264	void unlock_two_nondirectories(struct inode inode1, struct inode inode2)
1265	{
1266	if (inode1) {
1267	WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1268	inode_unlock(inode1);
1269	}
1270	if (inode2 && inode2 != inode1) {
1271	WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1272	inode_unlock(inode2);
1273	}
1274	}
1275	EXPORT_SYMBOL(unlock_two_nondirectories);
1276
1277	/**
1278	* inode_insert5 - obtain an inode from a mounted file system
1279	* @inode: pre-allocated inode to use for insert to cache
1280	* @hashval: hash value (usually inode number) to get
1281	* @test: callback used for comparisons between inodes
1282	* @set: callback used to initialize a new struct inode
1283	* @data: opaque data pointer to pass to @test and @set
1284	*
1285	* Search for the inode specified by @hashval and @data in the inode cache,
1286	* and if present return it with an increased reference count. This is a
1287	* variant of iget5_locked() that doesn't allocate an inode.
1288	*
1289	* If the inode is not present in the cache, insert the pre-allocated inode and
1290	* return it locked, hashed, and with the I_NEW flag set. The file system gets
1291	* to fill it in before unlocking it via unlock_new_inode().
1292	*
1293	* Note that both @test and @set are called with the inode_hash_lock held, so
1294	* they can't sleep.
1295	*/
1296	struct inode inode_insert5(struct inode inode, u64 hashval,
1297	int (test)(struct inode , void *),
1298	int (set)(struct inode , void ), void data)
1299	{
1300	struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1301	struct inode *old;
1302	bool isnew;
1303
1304	might_sleep();
1305
1306	again:
1307	spin_lock(&inode_hash_lock);
1308	old = find_inode(inode->i_sb, head, test, data, true, &isnew);
1309	if (unlikely(old)) {
1310	/*
1311	* Uhhuh, somebody else created the same inode under us.
1312	* Use the old inode instead of the preallocated one.
1313	*/
1314	spin_unlock(&inode_hash_lock);
1315	if (IS_ERR(old))
1316	return NULL;
1317	if (unlikely(isnew))
1318	wait_on_new_inode(old);
1319	if (unlikely(inode_unhashed(old))) {
1320	iput(old);
1321	goto again;
1322	}
1323	return old;
1324	}
1325
1326	if (set && unlikely(set(inode, data))) {
1327	spin_unlock(&inode_hash_lock);
1328	return NULL;
1329	}
1330
1331	/*
1332	* Return the locked inode with I_NEW set, the
1333	* caller is responsible for filling in the contents
1334	*/
1335	spin_lock(&inode->i_lock);
1336	inode_state_set(inode, I_NEW);
1337	hlist_add_head_rcu(&inode->i_hash, head);
1338	spin_unlock(&inode->i_lock);
1339
1340	spin_unlock(&inode_hash_lock);
1341
1342	/*
1343	* Add inode to the sb list if it's not already. It has I_NEW at this
1344	* point, so it should be safe to test i_sb_list locklessly.
1345	*/
1346	if (list_empty(&inode->i_sb_list))
1347	inode_sb_list_add(inode);
1348
1349	return inode;
1350	}
1351	EXPORT_SYMBOL(inode_insert5);
1352
1353	/**
1354	* iget5_locked - obtain an inode from a mounted file system
1355	* @sb: super block of file system
1356	* @hashval: hash value (usually inode number) to get
1357	* @test: callback used for comparisons between inodes
1358	* @set: callback used to initialize a new struct inode
1359	* @data: opaque data pointer to pass to @test and @set
1360	*
1361	* Search for the inode specified by @hashval and @data in the inode cache,
1362	* and if present return it with an increased reference count. This is a
1363	* generalized version of iget_locked() for file systems where the inode
1364	* number is not sufficient for unique identification of an inode.
1365	*
1366	* If the inode is not present in the cache, allocate and insert a new inode
1367	* and return it locked, hashed, and with the I_NEW flag set. The file system
1368	* gets to fill it in before unlocking it via unlock_new_inode().
1369	*
1370	* Note that both @test and @set are called with the inode_hash_lock held, so
1371	* they can't sleep.
1372	*/
1373	struct inode iget5_locked(struct super_block sb, u64 hashval,
1374	int (test)(struct inode , void *),
1375	int (set)(struct inode , void ), void data)
1376	{
1377	struct inode *inode = ilookup5(sb, hashval, test, data);
1378
1379	if (!inode) {
1380	struct inode *new = alloc_inode(sb);
1381
1382	if (new) {
1383	inode = inode_insert5(new, hashval, test, set, data);
1384	if (unlikely(inode != new))
1385	destroy_inode(new);
1386	}
1387	}
1388	return inode;
1389	}
1390	EXPORT_SYMBOL(iget5_locked);
1391
1392	/**
1393	* iget5_locked_rcu - obtain an inode from a mounted file system
1394	* @sb: super block of file system
1395	* @hashval: hash value (usually inode number) to get
1396	* @test: callback used for comparisons between inodes
1397	* @set: callback used to initialize a new struct inode
1398	* @data: opaque data pointer to pass to @test and @set
1399	*
1400	* This is equivalent to iget5_locked, except the @test callback must
1401	* tolerate the inode not being stable, including being mid-teardown.
1402	*/
1403	struct inode iget5_locked_rcu(struct super_block sb, u64 hashval,
1404	int (test)(struct inode , void *),
1405	int (set)(struct inode , void ), void data)
1406	{
1407	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1408	struct inode inode, new;
1409	bool isnew;
1410
1411	might_sleep();
1412
1413	again:
1414	inode = find_inode(sb, head, test, data, false, &isnew);
1415	if (inode) {
1416	if (IS_ERR(inode))
1417	return NULL;
1418	if (unlikely(isnew))
1419	wait_on_new_inode(inode);
1420	if (unlikely(inode_unhashed(inode))) {
1421	iput(inode);
1422	goto again;
1423	}
1424	return inode;
1425	}
1426
1427	new = alloc_inode(sb);
1428	if (new) {
1429	inode = inode_insert5(new, hashval, test, set, data);
1430	if (unlikely(inode != new))
1431	destroy_inode(new);
1432	}
1433	return inode;
1434	}
1435	EXPORT_SYMBOL_GPL(iget5_locked_rcu);
1436
1437	/**
1438	* iget_locked - obtain an inode from a mounted file system
1439	* @sb: super block of file system
1440	* @ino: inode number to get
1441	*
1442	* Search for the inode specified by @ino in the inode cache and if present
1443	* return it with an increased reference count. This is for file systems
1444	* where the inode number is sufficient for unique identification of an inode.
1445	*
1446	* If the inode is not in cache, allocate a new inode and return it locked,
1447	* hashed, and with the I_NEW flag set. The file system gets to fill it in
1448	* before unlocking it via unlock_new_inode().
1449	*/
1450	struct inode iget_locked(struct super_block sb, u64 ino)
1451	{
1452	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1453	struct inode *inode;
1454	bool isnew;
1455
1456	might_sleep();
1457
1458	again:
1459	inode = find_inode_fast(sb, head, ino, false, &isnew);
1460	if (inode) {
1461	if (IS_ERR(inode))
1462	return NULL;
1463	if (unlikely(isnew))
1464	wait_on_new_inode(inode);
1465	if (unlikely(inode_unhashed(inode))) {
1466	iput(inode);
1467	goto again;
1468	}
1469	return inode;
1470	}
1471
1472	inode = alloc_inode(sb);
1473	if (inode) {
1474	struct inode *old;
1475
1476	spin_lock(&inode_hash_lock);
1477	/* We released the lock, so.. */
1478	old = find_inode_fast(sb, head, ino, true, &isnew);
1479	if (!old) {
1480	inode->i_ino = ino;
1481	spin_lock(&inode->i_lock);
1482	inode_state_assign(inode, I_NEW);
1483	hlist_add_head_rcu(&inode->i_hash, head);
1484	spin_unlock(&inode->i_lock);
1485	spin_unlock(&inode_hash_lock);
1486	inode_sb_list_add(inode);
1487
1488	/* Return the locked inode with I_NEW set, the
1489	* caller is responsible for filling in the contents
1490	*/
1491	return inode;
1492	}
1493
1494	/*
1495	* Uhhuh, somebody else created the same inode under
1496	* us. Use the old inode instead of the one we just
1497	* allocated.
1498	*/
1499	spin_unlock(&inode_hash_lock);
1500	destroy_inode(inode);
1501	if (IS_ERR(old))
1502	return NULL;
1503	inode = old;
1504	if (unlikely(isnew))
1505	wait_on_new_inode(inode);
1506	if (unlikely(inode_unhashed(inode))) {
1507	iput(inode);
1508	goto again;
1509	}
1510	}
1511	return inode;
1512	}
1513	EXPORT_SYMBOL(iget_locked);
1514
1515	/*
1516	* search the inode cache for a matching inode number.
1517	* If we find one, then the inode number we are trying to
1518	* allocate is not unique and so we should not use it.
1519	*
1520	* Returns 1 if the inode number is unique, 0 if it is not.
1521	*/
1522	static int test_inode_iunique(struct super_block *sb, u64 ino)
1523	{
1524	struct hlist_head *b = inode_hashtable + hash(sb, ino);
1525	struct inode *inode;
1526
1527	hlist_for_each_entry_rcu(inode, b, i_hash) {
1528	if (inode->i_ino == ino && inode->i_sb == sb)
1529	return 0;
1530	}
1531	return 1;
1532	}
1533
1534	/**
1535	* iunique - get a unique inode number
1536	* @sb: superblock
1537	* @max_reserved: highest reserved inode number
1538	*
1539	* Obtain an inode number that is unique on the system for a given
1540	* superblock. This is used by file systems that have no natural
1541	* permanent inode numbering system. An inode number is returned that
1542	* is higher than the reserved limit but unique.
1543	*
1544	* BUGS:
1545	* With a large number of inodes live on the file system this function
1546	* currently becomes quite slow.
1547	*/
1548	ino_t iunique(struct super_block *sb, ino_t max_reserved)
1549	{
1550	/*
1551	* On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1552	* error if st_ino won't fit in target struct field. Use 32bit counter
1553	* here to attempt to avoid that.
1554	*/
1555	static DEFINE_SPINLOCK(iunique_lock);
1556	static unsigned int counter;
1557	ino_t res;
1558
1559	rcu_read_lock();
1560	spin_lock(&iunique_lock);
1561	do {
1562	if (counter <= max_reserved)
1563	counter = max_reserved + 1;
1564	res = counter++;
1565	} while (!test_inode_iunique(sb, res));
1566	spin_unlock(&iunique_lock);
1567	rcu_read_unlock();
1568
1569	return res;
1570	}
1571	EXPORT_SYMBOL(iunique);
1572
1573	/**
1574	* ihold - get a reference on the inode, provided you already have one
1575	* @inode: inode to operate on
1576	*/
1577	void ihold(struct inode *inode)
1578	{
1579	VFS_BUG_ON_INODE(icount_read_once(inode) < 1, inode);
1580	WARN_ON(atomic_inc_return(&inode->i_count) < 2);
1581	}
1582	EXPORT_SYMBOL(ihold);
1583
1584	struct inode igrab(struct inode inode)
1585	{
1586	/*
1587	* Read commentary above igrab_from_hash() for an explanation why this works.
1588	*/
1589	if (atomic_add_unless(&inode->i_count, 1, 0)) {
1590	VFS_BUG_ON_INODE(inode_state_read_once(inode) & (I_FREEING \| I_WILL_FREE), inode);
1591	return inode;
1592	}
1593
1594	spin_lock(&inode->i_lock);
1595	if (!(inode_state_read(inode) & (I_FREEING \| I_WILL_FREE))) {
1596	__iget(inode);
1597	spin_unlock(&inode->i_lock);
1598	} else {
1599	spin_unlock(&inode->i_lock);
1600	/*
1601	* Handle the case where s_op->clear_inode is not been
1602	* called yet, and somebody is calling igrab
1603	* while the inode is getting freed.
1604	*/
1605	inode = NULL;
1606	}
1607	return inode;
1608	}
1609	EXPORT_SYMBOL(igrab);
1610
1611	/*
1612	* igrab_from_hash - special inode refcount acquire primitive for the inode hash
1613	*
1614	* It provides lockless refcount acquire in the common case of no problematic
1615	* flags being set and the count being > 0.
1616	*
1617	* There are 4 state flags to worry about and the routine makes sure to not bump the
1618	* ref if any of them is present.
1619	*
1620	* I_NEW and I_CREATING can only legally get set before the inode becomes visible
1621	* during lookup. Thus if the flags are not spotted, they are guaranteed to not be
1622	* a factor. However, we need an acquire fence before returning the inode just
1623	* in case we raced against clearing the state to make sure our consumer picks up
1624	* any other changes made prior. atomic_add_unless provides a full fence, which
1625	* takes care of it.
1626	*
1627	* I_FREEING and I_WILL_FREE can only legally get set if ->i_count == 0 and it is
1628	* illegal to bump the ref if either is present. Consequently if atomic_add_unless
1629	* managed to replace a non-0 value with a bigger one, we have a guarantee neither
1630	* of these flags is set. Note this means explicitly checking of these flags below
1631	* is not necessary, it is only done because it does not cost anything on top of the
1632	* load which already needs to be done to handle the other flags.
1633	*/
1634	static bool igrab_from_hash(struct inode *inode)
1635	{
1636	if (inode_state_read_once(inode) & (I_NEW \| I_CREATING \| I_FREEING \| I_WILL_FREE))
1637	return false;
1638	/*
1639	* Paired with routines clearing I_NEW
1640	*/
1641	if (atomic_add_unless(&inode->i_count, 1, 0)) {
1642	VFS_BUG_ON_INODE(inode_state_read_once(inode) & (I_FREEING \| I_WILL_FREE), inode);
1643	return true;
1644	}
1645	return false;
1646	}
1647
1648	/**
1649	* ilookup5_nowait - search for an inode in the inode cache
1650	* @sb: super block of file system to search
1651	* @hashval: hash value (usually inode number) to search for
1652	* @test: callback used for comparisons between inodes
1653	* @data: opaque data pointer to pass to @test
1654	* @isnew: return argument telling whether I_NEW was set when
1655	* the inode was found in hash (the caller needs to
1656	* wait for I_NEW to clear)
1657	*
1658	* Search for the inode specified by @hashval and @data in the inode cache.
1659	* If the inode is in the cache, the inode is returned with an incremented
1660	* reference count.
1661	*
1662	* Note: I_NEW is not waited upon so you have to be very careful what you do
1663	* with the returned inode. You probably should be using ilookup5() instead.
1664	*
1665	* Note2: @test is called with the inode_hash_lock held, so can't sleep.
1666	*/
1667	struct inode ilookup5_nowait(struct super_block sb, u64 hashval,
1668	int (test)(struct inode , void ), void data, bool *isnew)
1669	{
1670	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1671	struct inode *inode;
1672
1673	spin_lock(&inode_hash_lock);
1674	inode = find_inode(sb, head, test, data, true, isnew);
1675	spin_unlock(&inode_hash_lock);
1676
1677	return IS_ERR(inode) ? NULL : inode;
1678	}
1679	EXPORT_SYMBOL(ilookup5_nowait);
1680
1681	/**
1682	* ilookup5 - search for an inode in the inode cache
1683	* @sb: super block of file system to search
1684	* @hashval: hash value (usually inode number) to search for
1685	* @test: callback used for comparisons between inodes
1686	* @data: opaque data pointer to pass to @test
1687	*
1688	* Search for the inode specified by @hashval and @data in the inode cache,
1689	* and if the inode is in the cache, return the inode with an incremented
1690	* reference count. Waits on I_NEW before returning the inode.
1691	* returned with an incremented reference count.
1692	*
1693	* This is a generalized version of ilookup() for file systems where the
1694	* inode number is not sufficient for unique identification of an inode.
1695	*
1696	* Note: @test is called with the inode_hash_lock held, so can't sleep.
1697	*/
1698	struct inode ilookup5(struct super_block sb, u64 hashval,
1699	int (test)(struct inode , void ), void data)
1700	{
1701	struct inode *inode;
1702	bool isnew;
1703
1704	might_sleep();
1705
1706	again:
1707	inode = ilookup5_nowait(sb, hashval, test, data, &isnew);
1708	if (inode) {
1709	if (unlikely(isnew))
1710	wait_on_new_inode(inode);
1711	if (unlikely(inode_unhashed(inode))) {
1712	iput(inode);
1713	goto again;
1714	}
1715	}
1716	return inode;
1717	}
1718	EXPORT_SYMBOL(ilookup5);
1719
1720	/**
1721	* ilookup - search for an inode in the inode cache
1722	* @sb: super block of file system to search
1723	* @ino: inode number to search for
1724	*
1725	* Search for the inode @ino in the inode cache, and if the inode is in the
1726	* cache, the inode is returned with an incremented reference count.
1727	*/
1728	struct inode ilookup(struct super_block sb, u64 ino)
1729	{
1730	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1731	struct inode *inode;
1732	bool isnew;
1733
1734	might_sleep();
1735
1736	again:
1737	inode = find_inode_fast(sb, head, ino, false, &isnew);
1738
1739	if (inode) {
1740	if (IS_ERR(inode))
1741	return NULL;
1742	if (unlikely(isnew))
1743	wait_on_new_inode(inode);
1744	if (unlikely(inode_unhashed(inode))) {
1745	iput(inode);
1746	goto again;
1747	}
1748	}
1749	return inode;
1750	}
1751	EXPORT_SYMBOL(ilookup);
1752
1753	/**
1754	* find_inode_nowait - find an inode in the inode cache
1755	* @sb: super block of file system to search
1756	* @hashval: hash value (usually inode number) to search for
1757	* @match: callback used for comparisons between inodes
1758	* @data: opaque data pointer to pass to @match
1759	*
1760	* Search for the inode specified by @hashval and @data in the inode
1761	* cache, where the helper function @match will return 0 if the inode
1762	* does not match, 1 if the inode does match, and -1 if the search
1763	* should be stopped. The @match function must be responsible for
1764	* taking the i_lock spin_lock and checking i_state for an inode being
1765	* freed or being initialized, and incrementing the reference count
1766	* before returning 1. It also must not sleep, since it is called with
1767	* the inode_hash_lock spinlock held.
1768	*
1769	* This is a even more generalized version of ilookup5() when the
1770	* function must never block --- find_inode() can block in
1771	* __wait_on_freeing_inode() --- or when the caller can not increment
1772	* the reference count because the resulting iput() might cause an
1773	* inode eviction. The tradeoff is that the @match function must be
1774	* very carefully implemented.
1775	*/
1776	struct inode find_inode_nowait(struct super_block sb,
1777	u64 hashval,
1778	int (match)(struct inode , u64,
1779	void *),
1780	void *data)
1781	{
1782	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1783	struct inode inode, ret_inode = NULL;
1784	int mval;
1785
1786	spin_lock(&inode_hash_lock);
1787	hlist_for_each_entry(inode, head, i_hash) {
1788	if (inode->i_sb != sb)
1789	continue;
1790	mval = match(inode, hashval, data);
1791	if (mval == 0)
1792	continue;
1793	if (mval == 1)
1794	ret_inode = inode;
1795	goto out;
1796	}
1797	out:
1798	spin_unlock(&inode_hash_lock);
1799	return ret_inode;
1800	}
1801	EXPORT_SYMBOL(find_inode_nowait);
1802
1803	/**
1804	* find_inode_rcu - find an inode in the inode cache
1805	* @sb: Super block of file system to search
1806	* @hashval: Key to hash
1807	* @test: Function to test match on an inode
1808	* @data: Data for test function
1809	*
1810	* Search for the inode specified by @hashval and @data in the inode cache,
1811	* where the helper function @test will return 0 if the inode does not match
1812	* and 1 if it does. The @test function must be responsible for taking the
1813	* i_lock spin_lock and checking i_state for an inode being freed or being
1814	* initialized.
1815	*
1816	* If successful, this will return the inode for which the @test function
1817	* returned 1 and NULL otherwise.
1818	*
1819	* The @test function is not permitted to take a ref on any inode presented.
1820	* It is also not permitted to sleep.
1821	*
1822	* The caller must hold the RCU read lock.
1823	*/
1824	struct inode find_inode_rcu(struct super_block sb, u64 hashval,
1825	int (test)(struct inode , void ), void data)
1826	{
1827	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1828	struct inode *inode;
1829
1830	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1831	"suspicious find_inode_rcu() usage");
1832
1833	hlist_for_each_entry_rcu(inode, head, i_hash) {
1834	if (inode->i_sb == sb &&
1835	!(inode_state_read_once(inode) & (I_FREEING \| I_WILL_FREE)) &&
1836	test(inode, data))
1837	return inode;
1838	}
1839	return NULL;
1840	}
1841	EXPORT_SYMBOL(find_inode_rcu);
1842
1843	/**
1844	* find_inode_by_ino_rcu - Find an inode in the inode cache
1845	* @sb: Super block of file system to search
1846	* @ino: The inode number to match
1847	*
1848	* Search for the inode specified by @hashval and @data in the inode cache,
1849	* where the helper function @test will return 0 if the inode does not match
1850	* and 1 if it does. The @test function must be responsible for taking the
1851	* i_lock spin_lock and checking i_state for an inode being freed or being
1852	* initialized.
1853	*
1854	* If successful, this will return the inode for which the @test function
1855	* returned 1 and NULL otherwise.
1856	*
1857	* The @test function is not permitted to take a ref on any inode presented.
1858	* It is also not permitted to sleep.
1859	*
1860	* The caller must hold the RCU read lock.
1861	*/
1862	struct inode find_inode_by_ino_rcu(struct super_block sb,
1863	u64 ino)
1864	{
1865	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1866	struct inode *inode;
1867
1868	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1869	"suspicious find_inode_by_ino_rcu() usage");
1870
1871	hlist_for_each_entry_rcu(inode, head, i_hash) {
1872	if (inode->i_ino == ino &&
1873	inode->i_sb == sb &&
1874	!(inode_state_read_once(inode) & (I_FREEING \| I_WILL_FREE)))
1875	return inode;
1876	}
1877	return NULL;
1878	}
1879	EXPORT_SYMBOL(find_inode_by_ino_rcu);
1880
1881	int insert_inode_locked(struct inode *inode)
1882	{
1883	struct super_block *sb = inode->i_sb;
1884	u64 ino = inode->i_ino;
1885	struct hlist_head *head = inode_hashtable + hash(sb, ino);
1886	bool isnew;
1887
1888	might_sleep();
1889
1890	while (1) {
1891	struct inode *old = NULL;
1892	spin_lock(&inode_hash_lock);
1893	repeat:
1894	hlist_for_each_entry(old, head, i_hash) {
1895	if (old->i_ino != ino)
1896	continue;
1897	if (old->i_sb != sb)
1898	continue;
1899	spin_lock(&old->i_lock);
1900	break;
1901	}
1902	if (likely(!old)) {
1903	spin_lock(&inode->i_lock);
1904	inode_state_set(inode, I_NEW \| I_CREATING);
1905	hlist_add_head_rcu(&inode->i_hash, head);
1906	spin_unlock(&inode->i_lock);
1907	spin_unlock(&inode_hash_lock);
1908	return 0;
1909	}
1910	if (inode_state_read(old) & (I_FREEING \| I_WILL_FREE)) {
1911	__wait_on_freeing_inode(old, true, false);
1912	old = NULL;
1913	goto repeat;
1914	}
1915	if (unlikely(inode_state_read(old) & I_CREATING)) {
1916	spin_unlock(&old->i_lock);
1917	spin_unlock(&inode_hash_lock);
1918	return -EBUSY;
1919	}
1920	__iget(old);
1921	isnew = !!(inode_state_read(old) & I_NEW);
1922	spin_unlock(&old->i_lock);
1923	spin_unlock(&inode_hash_lock);
1924	if (isnew)
1925	wait_on_new_inode(old);
1926	if (unlikely(!inode_unhashed(old))) {
1927	iput(old);
1928	return -EBUSY;
1929	}
1930	iput(old);
1931	}
1932	}
1933	EXPORT_SYMBOL(insert_inode_locked);
1934
1935	int insert_inode_locked4(struct inode *inode, u64 hashval,
1936	int (test)(struct inode , void ), void data)
1937	{
1938	struct inode *old;
1939
1940	might_sleep();
1941
1942	inode_state_set_raw(inode, I_CREATING);
1943	old = inode_insert5(inode, hashval, test, NULL, data);
1944
1945	if (old != inode) {
1946	iput(old);
1947	return -EBUSY;
1948	}
1949	return 0;
1950	}
1951	EXPORT_SYMBOL(insert_inode_locked4);
1952
1953
1954	int inode_just_drop(struct inode *inode)
1955	{
1956	return 1;
1957	}
1958	EXPORT_SYMBOL(inode_just_drop);
1959
1960	/*
1961	* Called when we're dropping the last reference
1962	* to an inode.
1963	*
1964	* Call the FS "drop_inode()" function, defaulting to
1965	* the legacy UNIX filesystem behaviour. If it tells
1966	* us to evict inode, do so. Otherwise, retain inode
1967	* in cache if fs is alive, sync and evict if fs is
1968	* shutting down.
1969	*/
1970	static void iput_final(struct inode *inode)
1971	{
1972	struct super_block *sb = inode->i_sb;
1973	const struct super_operations *op = inode->i_sb->s_op;
1974	int drop;
1975
1976	WARN_ON(inode_state_read(inode) & I_NEW);
1977	VFS_BUG_ON_INODE(icount_read(inode) != 0, inode);
1978
1979	if (op->drop_inode)
1980	drop = op->drop_inode(inode);
1981	else
1982	drop = inode_generic_drop(inode);
1983
1984	if (!drop &&
1985	!(inode_state_read(inode) & I_DONTCACHE) &&
1986	(sb->s_flags & SB_ACTIVE)) {
1987	__inode_lru_list_add(inode, true);
1988	spin_unlock(&inode->i_lock);
1989	return;
1990	}
1991
1992	/*
1993	* Re-check ->i_count in case the ->drop_inode() hooks played games.
1994	* Note we only execute this if the verdict was to drop the inode.
1995	*/
1996	VFS_BUG_ON_INODE(icount_read(inode) != 0, inode);
1997
1998	if (drop) {
1999	inode_state_set(inode, I_FREEING);
2000	} else {
2001	inode_state_set(inode, I_WILL_FREE);
2002	spin_unlock(&inode->i_lock);
2003
2004	write_inode_now(inode, 1);
2005
2006	spin_lock(&inode->i_lock);
2007	WARN_ON(inode_state_read(inode) & I_NEW);
2008	inode_state_replace(inode, I_WILL_FREE, I_FREEING);
2009	}
2010
2011	inode_lru_list_del(inode);
2012	spin_unlock(&inode->i_lock);
2013
2014	evict(inode);
2015	}
2016
2017	/**
2018	* iput - put an inode
2019	* @inode: inode to put
2020	*
2021	* Puts an inode, dropping its usage count. If the inode use count hits
2022	* zero, the inode is then freed and may also be destroyed.
2023	*
2024	* Consequently, iput() can sleep.
2025	*/
2026	void iput(struct inode *inode)
2027	{
2028	might_sleep();
2029	if (unlikely(!inode))
2030	return;
2031
2032	retry:
2033	lockdep_assert_not_held(&inode->i_lock);
2034	VFS_BUG_ON_INODE(inode_state_read_once(inode) & (I_FREEING \| I_CLEAR), inode);
2035	/*
2036	* Note this assert is technically racy as if the count is bogusly
2037	* equal to one, then two CPUs racing to further drop it can both
2038	* conclude it's fine.
2039	*/
2040	VFS_BUG_ON_INODE(icount_read_once(inode) < 1, inode);
2041
2042	if (atomic_add_unless(&inode->i_count, -1, 1))
2043	return;
2044
2045	if (inode->i_nlink && sync_lazytime(inode))
2046	goto retry;
2047
2048	spin_lock(&inode->i_lock);
2049	if (unlikely((inode_state_read(inode) & I_DIRTY_TIME) && inode->i_nlink)) {
2050	spin_unlock(&inode->i_lock);
2051	goto retry;
2052	}
2053
2054	if (!atomic_dec_and_test(&inode->i_count)) {
2055	spin_unlock(&inode->i_lock);
2056	return;
2057	}
2058
2059	/*
2060	* iput_final() drops ->i_lock, we can't assert on it as the inode may
2061	* be deallocated by the time the call returns.
2062	*/
2063	iput_final(inode);
2064	}
2065	EXPORT_SYMBOL(iput);
2066
2067	/**
2068	* iput_not_last - put an inode assuming this is not the last reference
2069	* @inode: inode to put
2070	*/
2071	void iput_not_last(struct inode *inode)
2072	{
2073	VFS_BUG_ON_INODE(inode_state_read_once(inode) & (I_FREEING \| I_CLEAR), inode);
2074	VFS_BUG_ON_INODE(icount_read_once(inode) < 2, inode);
2075
2076	WARN_ON(atomic_sub_return(1, &inode->i_count) == 0);
2077	}
2078	EXPORT_SYMBOL(iput_not_last);
2079
2080	#ifdef CONFIG_BLOCK
2081	/**
2082	* bmap - find a block number in a file
2083	* @inode: inode owning the block number being requested
2084	* @block: pointer containing the block to find
2085	*
2086	* Replaces the value in ``*block`` with the block number on the device holding
2087	* corresponding to the requested block number in the file.
2088	* That is, asked for block 4 of inode 1 the function will replace the
2089	* 4 in ``*block``, with disk block relative to the disk start that holds that
2090	* block of the file.
2091	*
2092	* Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
2093	* hole, returns 0 and ``*block`` is also set to 0.
2094	*/
2095	int bmap(struct inode inode, sector_t block)
2096	{
2097	if (!inode->i_mapping->a_ops->bmap)
2098	return -EINVAL;
2099
2100	block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
2101	return 0;
2102	}
2103	EXPORT_SYMBOL(bmap);
2104	#endif
2105
2106	/*
2107	* With relative atime, only update atime if the previous atime is
2108	* earlier than or equal to either the ctime or mtime,
2109	* or if at least a day has passed since the last atime update.
2110	*/
2111	static bool relatime_need_update(struct vfsmount mnt, struct inode inode,
2112	struct timespec64 now)
2113	{
2114	struct timespec64 atime, mtime, ctime;
2115
2116	if (!(mnt->mnt_flags & MNT_RELATIME))
2117	return true;
2118	/*
2119	* Is mtime younger than or equal to atime? If yes, update atime:
2120	*/
2121	atime = inode_get_atime(inode);
2122	mtime = inode_get_mtime(inode);
2123	if (timespec64_compare(&mtime, &atime) >= 0)
2124	return true;
2125	/*
2126	* Is ctime younger than or equal to atime? If yes, update atime:
2127	*/
2128	ctime = inode_get_ctime(inode);
2129	if (timespec64_compare(&ctime, &atime) >= 0)
2130	return true;
2131
2132	/*
2133	* Is the previous atime value older than a day? If yes,
2134	* update atime:
2135	*/
2136	if ((long)(now.tv_sec - atime.tv_sec) >= 246060)
2137	return true;
2138	/*
2139	* Good, we can skip the atime update:
2140	*/
2141	return false;
2142	}
2143
2144	static int inode_update_atime(struct inode *inode)
2145	{
2146	struct timespec64 atime = inode_get_atime(inode);
2147	struct timespec64 now = current_time(inode);
2148
2149	if (timespec64_equal(&now, &atime))
2150	return 0;
2151
2152	inode_set_atime_to_ts(inode, now);
2153	return inode_time_dirty_flag(inode);
2154	}
2155
2156	static int inode_update_cmtime(struct inode *inode, unsigned int flags)
2157	{
2158	struct timespec64 ctime = inode_get_ctime(inode);
2159	struct timespec64 mtime = inode_get_mtime(inode);
2160	struct timespec64 now = inode_set_ctime_current(inode);
2161	unsigned int dirty = 0;
2162	bool mtime_changed;
2163
2164	mtime_changed = !timespec64_equal(&now, &mtime);
2165	if (mtime_changed \|\| !timespec64_equal(&now, &ctime))
2166	dirty = inode_time_dirty_flag(inode);
2167
2168	/*
2169	* Pure timestamp updates can be recorded in the inode without blocking
2170	* by not dirtying the inode. But when the file system requires
2171	* i_version updates, the update of i_version can still block.
2172	* Error out if we'd actually have to update i_version or don't support
2173	* lazytime.
2174	*/
2175	if (IS_I_VERSION(inode)) {
2176	if (flags & IOCB_NOWAIT) {
2177	if (!(inode->i_sb->s_flags & SB_LAZYTIME) \|\|
2178	inode_iversion_need_inc(inode))
2179	return -EAGAIN;
2180	} else {
2181	/*
2182	* Don't force iversion increment for pure lazytime
2183	* updates (I_DIRTY_TIME only), let I_VERSION_QUERIED
2184	* dictate whether the increment is needed.
2185	*/
2186	if (inode_maybe_inc_iversion(inode,
2187	dirty != I_DIRTY_TIME))
2188	dirty \|= I_DIRTY_SYNC;
2189	}
2190	}
2191
2192	if (mtime_changed)
2193	inode_set_mtime_to_ts(inode, now);
2194	return dirty;
2195	}
2196
2197	/**
2198	* inode_update_time - update either atime or c/mtime and i_version on the inode
2199	* @inode: inode to be updated
2200	* @type: timestamp to be updated
2201	* @flags: flags for the update
2202	*
2203	* Update either atime or c/mtime and version in a inode if needed for a file
2204	* access or modification. It is up to the caller to mark the inode dirty
2205	* appropriately.
2206	*
2207	* Returns the positive I_DIRTY_* flags for __mark_inode_dirty() if the inode
2208	* needs to be marked dirty, 0 if it did not, or a negative errno if an error
2209	* happened.
2210	*/
2211	int inode_update_time(struct inode *inode, enum fs_update_time type,
2212	unsigned int flags)
2213	{
2214	switch (type) {
2215	case FS_UPD_ATIME:
2216	return inode_update_atime(inode);
2217	case FS_UPD_CMTIME:
2218	return inode_update_cmtime(inode, flags);
2219	default:
2220	WARN_ON_ONCE(1);
2221	return -EIO;
2222	}
2223	}
2224	EXPORT_SYMBOL(inode_update_time);
2225
2226	/**
2227	* generic_update_time - update the timestamps on the inode
2228	* @inode: inode to be updated
2229	* @type: timestamp to be updated
2230	* @flags: flags for the update
2231	*
2232	* Returns a negative error value on error, else 0.
2233	*/
2234	int generic_update_time(struct inode *inode, enum fs_update_time type,
2235	unsigned int flags)
2236	{
2237	int dirty;
2238
2239	/*
2240	* ->dirty_inode is what could make generic timestamp updates block.
2241	* Don't support non-blocking timestamp updates here if it is set.
2242	* File systems that implement ->dirty_inode but want to support
2243	* non-blocking timestamp updates should call inode_update_time
2244	* directly.
2245	*/
2246	if ((flags & IOCB_NOWAIT) && inode->i_sb->s_op->dirty_inode)
2247	return -EAGAIN;
2248
2249	dirty = inode_update_time(inode, type, flags);
2250	if (dirty <= 0)
2251	return dirty;
2252	__mark_inode_dirty(inode, dirty);
2253	return 0;
2254	}
2255	EXPORT_SYMBOL(generic_update_time);
2256
2257	/**
2258	* atime_needs_update - update the access time
2259	* @path: the &struct path to update
2260	* @inode: inode to update
2261	*
2262	* Update the accessed time on an inode and mark it for writeback.
2263	* This function automatically handles read only file systems and media,
2264	* as well as the "noatime" flag and inode specific "noatime" markers.
2265	*/
2266	bool atime_needs_update(const struct path path, struct inode inode)
2267	{
2268	struct vfsmount *mnt = path->mnt;
2269	struct timespec64 now, atime;
2270
2271	if (inode->i_flags & S_NOATIME)
2272	return false;
2273
2274	/* Atime updates will likely cause i_uid and i_gid to be written
2275	* back improprely if their true value is unknown to the vfs.
2276	*/
2277	if (HAS_UNMAPPED_ID(mnt_idmap(mnt), inode))
2278	return false;
2279
2280	if (IS_NOATIME(inode))
2281	return false;
2282	if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
2283	return false;
2284
2285	if (mnt->mnt_flags & MNT_NOATIME)
2286	return false;
2287	if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
2288	return false;
2289
2290	now = current_time(inode);
2291
2292	if (!relatime_need_update(mnt, inode, now))
2293	return false;
2294
2295	atime = inode_get_atime(inode);
2296	if (timespec64_equal(&atime, &now))
2297	return false;
2298
2299	return true;
2300	}
2301
2302	void touch_atime(const struct path *path)
2303	{
2304	struct vfsmount *mnt = path->mnt;
2305	struct inode *inode = d_inode(path->dentry);
2306
2307	if (!atime_needs_update(path, inode))
2308	return;
2309
2310	if (!sb_start_write_trylock(inode->i_sb))
2311	return;
2312
2313	if (mnt_get_write_access(mnt) != 0)
2314	goto skip_update;
2315	/*
2316	* File systems can error out when updating inodes if they need to
2317	* allocate new space to modify an inode (such is the case for
2318	* Btrfs), but since we touch atime while walking down the path we
2319	* really don't care if we failed to update the atime of the file,
2320	* so just ignore the return value.
2321	* We may also fail on filesystems that have the ability to make parts
2322	* of the fs read only, e.g. subvolumes in Btrfs.
2323	*/
2324	if (inode->i_op->update_time)
2325	inode->i_op->update_time(inode, FS_UPD_ATIME, 0);
2326	else
2327	generic_update_time(inode, FS_UPD_ATIME, 0);
2328	mnt_put_write_access(mnt);
2329	skip_update:
2330	sb_end_write(inode->i_sb);
2331	}
2332	EXPORT_SYMBOL(touch_atime);
2333
2334	/*
2335	* Return mask of changes for notify_change() that need to be done as a
2336	* response to write or truncate. Return 0 if nothing has to be changed.
2337	* Negative value on error (change should be denied).
2338	*/
2339	int dentry_needs_remove_privs(const struct mnt_idmap *idmap,
2340	struct dentry *dentry)
2341	{
2342	struct inode *inode = d_inode(dentry);
2343	int mask = 0;
2344	int ret;
2345
2346	if (IS_NOSEC(inode))
2347	return 0;
2348
2349	mask = setattr_should_drop_suidgid(idmap, inode);
2350	ret = security_inode_need_killpriv(dentry);
2351	if (ret < 0)
2352	return ret;
2353	if (ret)
2354	mask \|= ATTR_KILL_PRIV;
2355	return mask;
2356	}
2357
2358	static int __remove_privs(const struct mnt_idmap *idmap,
2359	struct dentry *dentry, int kill)
2360	{
2361	struct iattr newattrs;
2362
2363	newattrs.ia_valid = ATTR_FORCE \| kill;
2364	/*
2365	* Note we call this on write, so notify_change will not
2366	* encounter any conflicting delegations:
2367	*/
2368	return notify_change(idmap, dentry, &newattrs, NULL);
2369	}
2370
2371	static int file_remove_privs_flags(struct file *file, unsigned int flags)
2372	{
2373	struct dentry *dentry = file_dentry(file);
2374	struct inode *inode = file_inode(file);
2375	int error = 0;
2376	int kill;
2377
2378	if (IS_NOSEC(inode) \|\| !S_ISREG(inode->i_mode))
2379	return 0;
2380
2381	kill = dentry_needs_remove_privs(file_mnt_idmap(file), dentry);
2382	if (kill < 0)
2383	return kill;
2384
2385	if (kill) {
2386	if (flags & IOCB_NOWAIT)
2387	return -EAGAIN;
2388
2389	error = __remove_privs(file_mnt_idmap(file), dentry, kill);
2390	}
2391
2392	if (!error)
2393	inode_has_no_xattr(inode);
2394	return error;
2395	}
2396
2397	/**
2398	* file_remove_privs - remove special file privileges (suid, capabilities)
2399	* @file: file to remove privileges from
2400	*
2401	* When file is modified by a write or truncation ensure that special
2402	* file privileges are removed.
2403	*
2404	* Return: 0 on success, negative errno on failure.
2405	*/
2406	int file_remove_privs(struct file *file)
2407	{
2408	return file_remove_privs_flags(file, 0);
2409	}
2410	EXPORT_SYMBOL(file_remove_privs);
2411
2412	/**
2413	* current_time - Return FS time (possibly fine-grained)
2414	* @inode: inode.
2415	*
2416	* Return the current time truncated to the time granularity supported by
2417	* the fs, as suitable for a ctime/mtime change. If the ctime is flagged
2418	* as having been QUERIED, get a fine-grained timestamp, but don't update
2419	* the floor.
2420	*
2421	* For a multigrain inode, this is effectively an estimate of the timestamp
2422	* that a file would receive. An actual update must go through
2423	* inode_set_ctime_current().
2424	*/
2425	struct timespec64 current_time(struct inode *inode)
2426	{
2427	struct timespec64 now;
2428	u32 cns;
2429
2430	ktime_get_coarse_real_ts64_mg(&now);
2431
2432	if (!is_mgtime(inode))
2433	goto out;
2434
2435	/* If nothing has queried it, then coarse time is fine */
2436	cns = smp_load_acquire(&inode->i_ctime_nsec);
2437	if (cns & I_CTIME_QUERIED) {
2438	/*
2439	* If there is no apparent change, then get a fine-grained
2440	* timestamp.
2441	*/
2442	if (now.tv_nsec == (cns & ~I_CTIME_QUERIED))
2443	ktime_get_real_ts64(&now);
2444	}
2445	out:
2446	return timestamp_truncate(now, inode);
2447	}
2448	EXPORT_SYMBOL(current_time);
2449
2450	static inline bool need_cmtime_update(struct inode *inode)
2451	{
2452	struct timespec64 now = current_time(inode), ts;
2453
2454	ts = inode_get_mtime(inode);
2455	if (!timespec64_equal(&ts, &now))
2456	return true;
2457	ts = inode_get_ctime(inode);
2458	if (!timespec64_equal(&ts, &now))
2459	return true;
2460	return IS_I_VERSION(inode) && inode_iversion_need_inc(inode);
2461	}
2462
2463	static int file_update_time_flags(struct file *file, unsigned int flags)
2464	{
2465	struct inode *inode = file_inode(file);
2466	int ret;
2467
2468	/* First try to exhaust all avenues to not sync */
2469	if (IS_NOCMTIME(inode))
2470	return 0;
2471	if (unlikely(file->f_mode & FMODE_NOCMTIME))
2472	return 0;
2473	if (!need_cmtime_update(inode))
2474	return 0;
2475
2476	flags &= IOCB_NOWAIT;
2477	if (mnt_get_write_access_file(file))
2478	return 0;
2479	if (inode->i_op->update_time)
2480	ret = inode->i_op->update_time(inode, FS_UPD_CMTIME, flags);
2481	else
2482	ret = generic_update_time(inode, FS_UPD_CMTIME, flags);
2483	mnt_put_write_access_file(file);
2484	return ret;
2485	}
2486
2487	/**
2488	* file_update_time - update mtime and ctime time
2489	* @file: file accessed
2490	*
2491	* Update the mtime and ctime members of an inode and mark the inode for
2492	* writeback. Note that this function is meant exclusively for usage in
2493	* the file write path of filesystems, and filesystems may choose to
2494	* explicitly ignore updates via this function with the _NOCMTIME inode
2495	* flag, e.g. for network filesystem where these imestamps are handled
2496	* by the server. This can return an error for file systems who need to
2497	* allocate space in order to update an inode.
2498	*
2499	* Return: 0 on success, negative errno on failure.
2500	*/
2501	int file_update_time(struct file *file)
2502	{
2503	return file_update_time_flags(file, 0);
2504	}
2505	EXPORT_SYMBOL(file_update_time);
2506
2507	/**
2508	* file_modified_flags - handle mandated vfs changes when modifying a file
2509	* @file: file that was modified
2510	* @flags: kiocb flags
2511	*
2512	* When file has been modified ensure that special
2513	* file privileges are removed and time settings are updated.
2514	*
2515	* If IOCB_NOWAIT is set, special file privileges will not be removed and
2516	* time settings will not be updated. It will return -EAGAIN.
2517	*
2518	* Context: Caller must hold the file's inode lock.
2519	*
2520	* Return: 0 on success, negative errno on failure.
2521	*/
2522	static int file_modified_flags(struct file *file, int flags)
2523	{
2524	int ret;
2525
2526	/*
2527	* Clear the security bits if the process is not being run by root.
2528	* This keeps people from modifying setuid and setgid binaries.
2529	*/
2530	ret = file_remove_privs_flags(file, flags);
2531	if (ret)
2532	return ret;
2533	return file_update_time_flags(file, flags);
2534	}
2535
2536	/**
2537	* file_modified - handle mandated vfs changes when modifying a file
2538	* @file: file that was modified
2539	*
2540	* When file has been modified ensure that special
2541	* file privileges are removed and time settings are updated.
2542	*
2543	* Context: Caller must hold the file's inode lock.
2544	*
2545	* Return: 0 on success, negative errno on failure.
2546	*/
2547	int file_modified(struct file *file)
2548	{
2549	return file_modified_flags(file, 0);
2550	}
2551	EXPORT_SYMBOL(file_modified);
2552
2553	/**
2554	* kiocb_modified - handle mandated vfs changes when modifying a file
2555	* @iocb: iocb that was modified
2556	*
2557	* When file has been modified ensure that special
2558	* file privileges are removed and time settings are updated.
2559	*
2560	* Context: Caller must hold the file's inode lock.
2561	*
2562	* Return: 0 on success, negative errno on failure.
2563	*/
2564	int kiocb_modified(struct kiocb *iocb)
2565	{
2566	return file_modified_flags(iocb->ki_filp, iocb->ki_flags);
2567	}
2568	EXPORT_SYMBOL_GPL(kiocb_modified);
2569
2570	int inode_needs_sync(struct inode *inode)
2571	{
2572	if (IS_SYNC(inode))
2573	return 1;
2574	if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2575	return 1;
2576	return 0;
2577	}
2578	EXPORT_SYMBOL(inode_needs_sync);
2579
2580	/*
2581	* If we try to find an inode in the inode hash while it is being
2582	* deleted, we have to wait until the filesystem completes its
2583	* deletion before reporting that it isn't found. This function waits
2584	* until the deletion _might_ have completed. Callers are responsible
2585	* to recheck inode state.
2586	*
2587	* It doesn't matter if I_NEW is not set initially, a call to
2588	* wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2589	* will DTRT.
2590	*/
2591	static void __wait_on_freeing_inode(struct inode *inode, bool hash_locked, bool rcu_locked)
2592	{
2593	struct wait_bit_queue_entry wqe;
2594	struct wait_queue_head *wq_head;
2595
2596	VFS_BUG_ON(!hash_locked && !rcu_locked);
2597
2598	/*
2599	* Handle racing against evict(), see that routine for more details.
2600	*/
2601	if (unlikely(inode_unhashed(inode))) {
2602	WARN_ON(hash_locked);
2603	spin_unlock(&inode->i_lock);
2604	return;
2605	}
2606
2607	wq_head = inode_bit_waitqueue(&wqe, inode, __I_NEW);
2608	prepare_to_wait_event(wq_head, &wqe.wq_entry, TASK_UNINTERRUPTIBLE);
2609	spin_unlock(&inode->i_lock);
2610	if (rcu_locked)
2611	rcu_read_unlock();
2612	if (hash_locked)
2613	spin_unlock(&inode_hash_lock);
2614	schedule();
2615	finish_wait(wq_head, &wqe.wq_entry);
2616	if (hash_locked)
2617	spin_lock(&inode_hash_lock);
2618	if (rcu_locked)
2619	rcu_read_lock();
2620	}
2621
2622	static __initdata unsigned long ihash_entries;
2623	static int __init set_ihash_entries(char *str)
2624	{
2625	return kstrtoul(str, 0, &ihash_entries) == 0;
2626	}
2627	__setup("ihash_entries=", set_ihash_entries);
2628
2629	/*
2630	* Initialize the waitqueues and inode hash table.
2631	*/
2632	void __init inode_init_early(void)
2633	{
2634	/* If hashes are distributed across NUMA nodes, defer
2635	* hash allocation until vmalloc space is available.
2636	*/
2637	if (hashdist)
2638	return;
2639
2640	inode_hashtable =
2641	alloc_large_system_hash("Inode-cache",
2642	sizeof(struct hlist_head),
2643	ihash_entries,
2644	14,
2645	HASH_EARLY \| HASH_ZERO,
2646	&i_hash_shift,
2647	&i_hash_mask,
2648	0,
2649	0);
2650	}
2651
2652	void __init inode_init(void)
2653	{
2654	/* inode slab cache */
2655	inode_cachep = kmem_cache_create("inode_cache",
2656	sizeof(struct inode),
2657	0,
2658	(SLAB_RECLAIM_ACCOUNT\|SLAB_PANIC\|
2659	SLAB_ACCOUNT),
2660	init_once);
2661
2662	/* Hash may have been set up in inode_init_early */
2663	if (!hashdist)
2664	return;
2665
2666	inode_hashtable =
2667	alloc_large_system_hash("Inode-cache",
2668	sizeof(struct hlist_head),
2669	ihash_entries,
2670	14,
2671	HASH_ZERO,
2672	&i_hash_shift,
2673	&i_hash_mask,
2674	0,
2675	0);
2676	}
2677
2678	void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2679	{
2680	inode->i_mode = mode;
2681	switch (inode->i_mode & S_IFMT) {
2682	case S_IFCHR:
2683	inode->i_fop = &def_chr_fops;
2684	inode->i_rdev = rdev;
2685	break;
2686	case S_IFBLK:
2687	if (IS_ENABLED(CONFIG_BLOCK))
2688	inode->i_fop = &def_blk_fops;
2689	inode->i_rdev = rdev;
2690	break;
2691	case S_IFIFO:
2692	inode->i_fop = &pipefifo_fops;
2693	break;
2694	case S_IFSOCK:
2695	/* leave it no_open_fops */
2696	break;
2697	default:
2698	pr_debug("init_special_inode: bogus i_mode (%o) for inode %s:%llu\n",
2699	mode, inode->i_sb->s_id, inode->i_ino);
2700	break;
2701	}
2702	}
2703	EXPORT_SYMBOL(init_special_inode);
2704
2705	/**
2706	* inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2707	* @idmap: idmap of the mount the inode was created from
2708	* @inode: New inode
2709	* @dir: Directory inode
2710	* @mode: mode of the new inode
2711	*
2712	* If the inode has been created through an idmapped mount the idmap of
2713	* the vfsmount must be passed through @idmap. This function will then take
2714	* care to map the inode according to @idmap before checking permissions
2715	* and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2716	* checking is to be performed on the raw inode simply pass @nop_mnt_idmap.
2717	*/
2718	void inode_init_owner(const struct mnt_idmap idmap, struct inode inode,
2719	const struct inode *dir, umode_t mode)
2720	{
2721	inode_fsuid_set(inode, idmap);
2722	if (dir && dir->i_mode & S_ISGID) {
2723	inode->i_gid = dir->i_gid;
2724
2725	/* Directories are special, and always inherit S_ISGID */
2726	if (S_ISDIR(mode))
2727	mode \|= S_ISGID;
2728	} else
2729	inode_fsgid_set(inode, idmap);
2730	inode->i_mode = mode;
2731	}
2732	EXPORT_SYMBOL(inode_init_owner);
2733
2734	/**
2735	* inode_owner_or_capable - check current task permissions to inode
2736	* @idmap: idmap of the mount the inode was found from
2737	* @inode: inode being checked
2738	*
2739	* Return true if current either has CAP_FOWNER in a namespace with the
2740	* inode owner uid mapped, or owns the file.
2741	*
2742	* If the inode has been found through an idmapped mount the idmap of
2743	* the vfsmount must be passed through @idmap. This function will then take
2744	* care to map the inode according to @idmap before checking permissions.
2745	* On non-idmapped mounts or if permission checking is to be performed on the
2746	* raw inode simply pass @nop_mnt_idmap.
2747	*/
2748	bool inode_owner_or_capable(const struct mnt_idmap *idmap,
2749	const struct inode *inode)
2750	{
2751	vfsuid_t vfsuid;
2752	struct user_namespace *ns;
2753
2754	vfsuid = i_uid_into_vfsuid(idmap, inode);
2755	if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
2756	return true;
2757
2758	ns = current_user_ns();
2759	if (vfsuid_has_mapping(ns, vfsuid) && ns_capable(ns, CAP_FOWNER))
2760	return true;
2761	return false;
2762	}
2763	EXPORT_SYMBOL(inode_owner_or_capable);
2764
2765	/*
2766	* Direct i/o helper functions
2767	*/
2768	bool inode_dio_finished(const struct inode *inode)
2769	{
2770	return atomic_read(&inode->i_dio_count) == 0;
2771	}
2772	EXPORT_SYMBOL(inode_dio_finished);
2773
2774	/**
2775	* inode_dio_wait - wait for outstanding DIO requests to finish
2776	* @inode: inode to wait for
2777	*
2778	* Waits for all pending direct I/O requests to finish so that we can
2779	* proceed with a truncate or equivalent operation.
2780	*
2781	* Must be called under a lock that serializes taking new references
2782	* to i_dio_count, usually by inode->i_rwsem.
2783	*/
2784	void inode_dio_wait(struct inode *inode)
2785	{
2786	wait_var_event(&inode->i_dio_count, inode_dio_finished(inode));
2787	}
2788	EXPORT_SYMBOL(inode_dio_wait);
2789
2790	void inode_dio_wait_interruptible(struct inode *inode)
2791	{
2792	wait_var_event_interruptible(&inode->i_dio_count,
2793	inode_dio_finished(inode));
2794	}
2795	EXPORT_SYMBOL(inode_dio_wait_interruptible);
2796
2797	/*
2798	* inode_set_flags - atomically set some inode flags
2799	*
2800	* Note: the caller should be holding i_rwsem exclusively, or else be sure that
2801	* they have exclusive access to the inode structure (i.e., while the
2802	* inode is being instantiated). The reason for the cmpxchg() loop
2803	* --- which wouldn't be necessary if all code paths which modify
2804	* i_flags actually followed this rule, is that there is at least one
2805	* code path which doesn't today so we use cmpxchg() out of an abundance
2806	* of caution.
2807	*
2808	* In the long run, i_rwsem is overkill, and we should probably look
2809	* at using the i_lock spinlock to protect i_flags, and then make sure
2810	* it is so documented in include/linux/fs.h and that all code follows
2811	* the locking convention!!
2812	*/
2813	void inode_set_flags(struct inode *inode, unsigned int flags,
2814	unsigned int mask)
2815	{
2816	WARN_ON_ONCE(flags & ~mask);
2817	set_mask_bits(&inode->i_flags, mask, flags);
2818	}
2819	EXPORT_SYMBOL(inode_set_flags);
2820
2821	void inode_nohighmem(struct inode *inode)
2822	{
2823	mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2824	}
2825	EXPORT_SYMBOL(inode_nohighmem);
2826
2827	struct timespec64 inode_set_ctime_to_ts(struct inode *inode, struct timespec64 ts)
2828	{
2829	trace_inode_set_ctime_to_ts(inode, &ts);
2830	set_normalized_timespec64(&ts, ts.tv_sec, ts.tv_nsec);
2831	WRITE_ONCE(inode->i_ctime_sec, ts.tv_sec);
2832	WRITE_ONCE(inode->i_ctime_nsec, ts.tv_nsec);
2833	return ts;
2834	}
2835	EXPORT_SYMBOL(inode_set_ctime_to_ts);
2836
2837	/**
2838	* timestamp_truncate - Truncate timespec to a granularity
2839	* @t: Timespec
2840	* @inode: inode being updated
2841	*
2842	* Truncate a timespec to the granularity supported by the fs
2843	* containing the inode. Always rounds down. gran must
2844	* not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2845	*/
2846	struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2847	{
2848	struct super_block *sb = inode->i_sb;
2849	unsigned int gran = sb->s_time_gran;
2850
2851	t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2852	if (unlikely(t.tv_sec == sb->s_time_max \|\| t.tv_sec == sb->s_time_min))
2853	t.tv_nsec = 0;
2854
2855	/* Avoid division in the common cases 1 ns and 1 s. */
2856	if (gran == 1)
2857	; /* nothing */
2858	else if (gran == NSEC_PER_SEC)
2859	t.tv_nsec = 0;
2860	else if (gran > 1 && gran < NSEC_PER_SEC)
2861	t.tv_nsec -= t.tv_nsec % gran;
2862	else
2863	WARN(1, "invalid file time granularity: %u", gran);
2864	return t;
2865	}
2866	EXPORT_SYMBOL(timestamp_truncate);
2867
2868	/**
2869	* inode_set_ctime_current - set the ctime to current_time
2870	* @inode: inode
2871	*
2872	* Set the inode's ctime to the current value for the inode. Returns the
2873	* current value that was assigned. If this is not a multigrain inode, then we
2874	* set it to the later of the coarse time and floor value.
2875	*
2876	* If it is multigrain, then we first see if the coarse-grained timestamp is
2877	* distinct from what is already there. If so, then use that. Otherwise, get a
2878	* fine-grained timestamp.
2879	*
2880	* After that, try to swap the new value into i_ctime_nsec. Accept the
2881	* resulting ctime, regardless of the outcome of the swap. If it has
2882	* already been replaced, then that timestamp is later than the earlier
2883	* unacceptable one, and is thus acceptable.
2884	*/
2885	struct timespec64 inode_set_ctime_current(struct inode *inode)
2886	{
2887	struct timespec64 now;
2888	u32 cns, cur;
2889
2890	ktime_get_coarse_real_ts64_mg(&now);
2891	now = timestamp_truncate(now, inode);
2892
2893	/* Just return that if this is not a multigrain fs */
2894	if (!is_mgtime(inode)) {
2895	inode_set_ctime_to_ts(inode, now);
2896	goto out;
2897	}
2898
2899	/*
2900	* A fine-grained time is only needed if someone has queried
2901	* for timestamps, and the current coarse grained time isn't
2902	* later than what's already there.
2903	*/
2904	cns = smp_load_acquire(&inode->i_ctime_nsec);
2905	if (cns & I_CTIME_QUERIED) {
2906	struct timespec64 ctime = { .tv_sec = inode_get_ctime_sec(inode),
2907	.tv_nsec = cns & ~I_CTIME_QUERIED };
2908
2909	if (timespec64_compare(&now, &ctime) <= 0) {
2910	ktime_get_real_ts64_mg(&now);
2911	now = timestamp_truncate(now, inode);
2912	mgtime_counter_inc(mg_fine_stamps);
2913	}
2914	}
2915	mgtime_counter_inc(mg_ctime_updates);
2916
2917	/* No need to cmpxchg if it's exactly the same */
2918	if (cns == now.tv_nsec && inode_get_ctime_sec(inode) == now.tv_sec) {
2919	trace_ctime_xchg_skip(inode, &now);
2920	goto out;
2921	}
2922	cur = cns;
2923	retry:
2924	/* Try to swap the nsec value into place. */
2925	if (try_cmpxchg(&inode->i_ctime_nsec, &cur, now.tv_nsec)) {
2926	/* If swap occurred, then we're (mostly) done */
2927	WRITE_ONCE(inode->i_ctime_sec, now.tv_sec);
2928	trace_ctime_ns_xchg(inode, cns, now.tv_nsec, cur);
2929	mgtime_counter_inc(mg_ctime_swaps);
2930	} else {
2931	/*
2932	* Was the change due to someone marking the old ctime QUERIED?
2933	* If so then retry the swap. This can only happen once since
2934	* the only way to clear I_CTIME_QUERIED is to stamp the inode
2935	* with a new ctime.
2936	*/
2937	if (!(cns & I_CTIME_QUERIED) && (cns \| I_CTIME_QUERIED) == cur) {
2938	cns = cur;
2939	goto retry;
2940	}
2941	/* Otherwise, keep the existing ctime */
2942	now.tv_sec = inode_get_ctime_sec(inode);
2943	now.tv_nsec = cur & ~I_CTIME_QUERIED;
2944	}
2945	out:
2946	return now;
2947	}
2948	EXPORT_SYMBOL(inode_set_ctime_current);
2949
2950	/**
2951	* inode_set_ctime_deleg - try to update the ctime on a delegated inode
2952	* @inode: inode to update
2953	* @update: timespec64 to set the ctime
2954	*
2955	* Attempt to atomically update the ctime on behalf of a delegation holder.
2956	*
2957	* The nfs server can call back the holder of a delegation to get updated
2958	* inode attributes, including the mtime. When updating the mtime, update
2959	* the ctime to a value at least equal to that.
2960	*
2961	* This can race with concurrent updates to the inode, in which
2962	* case the update is skipped.
2963	*
2964	* Note that this works even when multigrain timestamps are not enabled,
2965	* so it is used in either case.
2966	*/
2967	struct timespec64 inode_set_ctime_deleg(struct inode *inode, struct timespec64 update)
2968	{
2969	struct timespec64 now, cur_ts;
2970	u32 cur, old;
2971
2972	/* pairs with try_cmpxchg below */
2973	cur = smp_load_acquire(&inode->i_ctime_nsec);
2974	cur_ts.tv_nsec = cur & ~I_CTIME_QUERIED;
2975	cur_ts.tv_sec = inode_get_ctime_sec(inode);
2976
2977	/* If the update is older than the existing value, skip it. */
2978	if (timespec64_compare(&update, &cur_ts) <= 0)
2979	return cur_ts;
2980
2981	ktime_get_coarse_real_ts64_mg(&now);
2982
2983	/* Clamp the update to "now" if it's in the future */
2984	if (timespec64_compare(&update, &now) > 0)
2985	update = now;
2986
2987	update = timestamp_truncate(update, inode);
2988
2989	/* No need to update if the values are already the same */
2990	if (timespec64_equal(&update, &cur_ts))
2991	return cur_ts;
2992
2993	/*
2994	* Try to swap the nsec value into place. If it fails, that means
2995	* it raced with an update due to a write or similar activity. That
2996	* stamp takes precedence, so just skip the update.
2997	*/
2998	retry:
2999	old = cur;
3000	if (try_cmpxchg(&inode->i_ctime_nsec, &cur, update.tv_nsec)) {
3001	WRITE_ONCE(inode->i_ctime_sec, update.tv_sec);
3002	mgtime_counter_inc(mg_ctime_swaps);
3003	return update;
3004	}
3005
3006	/*
3007	* Was the change due to another task marking the old ctime QUERIED?
3008	*
3009	* If so, then retry the swap. This can only happen once since
3010	* the only way to clear I_CTIME_QUERIED is to stamp the inode
3011	* with a new ctime.
3012	*/
3013	if (!(old & I_CTIME_QUERIED) && (cur == (old \| I_CTIME_QUERIED)))
3014	goto retry;
3015
3016	/* Otherwise, it was a new timestamp. */
3017	cur_ts.tv_sec = inode_get_ctime_sec(inode);
3018	cur_ts.tv_nsec = cur & ~I_CTIME_QUERIED;
3019	return cur_ts;
3020	}
3021	EXPORT_SYMBOL(inode_set_ctime_deleg);
3022
3023	/**
3024	* in_group_or_capable - check whether caller is CAP_FSETID privileged
3025	* @idmap: idmap of the mount @inode was found from
3026	* @inode: inode to check
3027	* @vfsgid: the new/current vfsgid of @inode
3028	*
3029	* Check whether @vfsgid is in the caller's group list or if the caller is
3030	* privileged with CAP_FSETID over @inode. This can be used to determine
3031	* whether the setgid bit can be kept or must be dropped.
3032	*
3033	* Return: true if the caller is sufficiently privileged, false if not.
3034	*/
3035	bool in_group_or_capable(const struct mnt_idmap *idmap,
3036	const struct inode *inode, vfsgid_t vfsgid)
3037	{
3038	if (vfsgid_in_group_p(vfsgid))
3039	return true;
3040	if (capable_wrt_inode_uidgid(idmap, inode, CAP_FSETID))
3041	return true;
3042	return false;
3043	}
3044	EXPORT_SYMBOL(in_group_or_capable);
3045
3046	/**
3047	* mode_strip_sgid - handle the sgid bit for non-directories
3048	* @idmap: idmap of the mount the inode was created from
3049	* @dir: parent directory inode
3050	* @mode: mode of the file to be created in @dir
3051	*
3052	* If the @mode of the new file has both the S_ISGID and S_IXGRP bit
3053	* raised and @dir has the S_ISGID bit raised ensure that the caller is
3054	* either in the group of the parent directory or they have CAP_FSETID
3055	* in their user namespace and are privileged over the parent directory.
3056	* In all other cases, strip the S_ISGID bit from @mode.
3057	*
3058	* Return: the new mode to use for the file
3059	*/
3060	umode_t mode_strip_sgid(const struct mnt_idmap *idmap,
3061	const struct inode *dir, umode_t mode)
3062	{
3063	if ((mode & (S_ISGID \| S_IXGRP)) != (S_ISGID \| S_IXGRP))
3064	return mode;
3065	if (S_ISDIR(mode) \|\| !dir \|\| !(dir->i_mode & S_ISGID))
3066	return mode;
3067	if (in_group_or_capable(idmap, dir, i_gid_into_vfsgid(idmap, dir)))
3068	return mode;
3069	return mode & ~S_ISGID;
3070	}
3071	EXPORT_SYMBOL(mode_strip_sgid);
3072
3073	#ifdef CONFIG_DEBUG_VFS
3074	/**
3075	* dump_inode - dump an inode.
3076	* @inode: inode to dump
3077	* @reason: reason for dumping
3078	*
3079	* If inode is an invalid pointer, we don't want to crash accessing it,
3080	* so probe everything depending on it carefully with get_kernel_nofault().
3081	*/
3082	void dump_inode(struct inode inode, const char reason)
3083	{
3084	struct super_block *sb;
3085	struct file_system_type *s_type;
3086	const char *fs_name_ptr;
3087	char fs_name[32] = {};
3088	umode_t mode;
3089	unsigned short opflags;
3090	unsigned int flags;
3091	unsigned int state;
3092	int count;
3093
3094	if (get_kernel_nofault(sb, &inode->i_sb) \|\|
3095	get_kernel_nofault(mode, &inode->i_mode) \|\|
3096	get_kernel_nofault(opflags, &inode->i_opflags) \|\|
3097	get_kernel_nofault(flags, &inode->i_flags)) {
3098	pr_warn("%s: unreadable inode:%px\n", reason, inode);
3099	return;
3100	}
3101
3102	state = inode_state_read_once(inode);
3103	count = icount_read_once(inode);
3104
3105	if (!sb \|\|
3106	get_kernel_nofault(s_type, &sb->s_type) \|\| !s_type \|\|
3107	get_kernel_nofault(fs_name_ptr, &s_type->name) \|\| !fs_name_ptr \|\|
3108	strncpy_from_kernel_nofault(fs_name, fs_name_ptr, sizeof(fs_name) - 1) < 0)
3109	strscpy(fs_name, "<unknown, sb unreadable>");
3110
3111	pr_warn("%s: inode:%px fs:%s mode:%ho opflags:%#x flags:%#x state:%#x count:%d\n",
3112	reason, inode, fs_name, mode, opflags, flags, state, count);
3113	}
3114	EXPORT_SYMBOL(dump_inode);
3115	#endif