1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	* (C) 1997 Linus Torvalds
4	* (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5	*/
6	#include <linux/export.h>
7	#include <linux/fs.h>
8	#include <linux/filelock.h>
9	#include <linux/mm.h>
10	#include <linux/backing-dev.h>
11	#include <linux/hash.h>
12	#include <linux/swap.h>
13	#include <linux/security.h>
14	#include <linux/cdev.h>
15	#include <linux/memblock.h>
16	#include <linux/fsnotify.h>
17	#include <linux/mount.h>
18	#include <linux/posix_acl.h>
19	#include <linux/buffer_head.h> /* for inode_has_buffers */
20	#include <linux/ratelimit.h>
21	#include <linux/list_lru.h>
22	#include <linux/iversion.h>
23	#include <trace/events/writeback.h>
24	#include "internal.h"
25
26	/*
27	* Inode locking rules:
28	*
29	* inode->i_lock protects:
30	* inode->i_state, inode->i_hash, __iget(), inode->i_io_list
31	* Inode LRU list locks protect:
32	* inode->i_sb->s_inode_lru, inode->i_lru
33	* inode->i_sb->s_inode_list_lock protects:
34	* inode->i_sb->s_inodes, inode->i_sb_list
35	* bdi->wb.list_lock protects:
36	* bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
37	* inode_hash_lock protects:
38	* inode_hashtable, inode->i_hash
39	*
40	* Lock ordering:
41	*
42	* inode->i_sb->s_inode_list_lock
43	* inode->i_lock
44	* Inode LRU list locks
45	*
46	* bdi->wb.list_lock
47	* inode->i_lock
48	*
49	* inode_hash_lock
50	* inode->i_sb->s_inode_list_lock
51	* inode->i_lock
52	*
53	* iunique_lock
54	* inode_hash_lock
55	*/
56
57	static unsigned int i_hash_mask __ro_after_init;
58	static unsigned int i_hash_shift __ro_after_init;
59	static struct hlist_head *inode_hashtable __ro_after_init;
60	static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
61
62	/*
63	* Empty aops. Can be used for the cases where the user does not
64	* define any of the address_space operations.
65	*/
66	const struct address_space_operations empty_aops = {
67	};
68	EXPORT_SYMBOL(empty_aops);
69
70	static DEFINE_PER_CPU(unsigned long, nr_inodes);
71	static DEFINE_PER_CPU(unsigned long, nr_unused);
72
73	static struct kmem_cache *inode_cachep __ro_after_init;
74
75	static long get_nr_inodes(void)
76	{
77	int i;
78	long sum = 0;
79	for_each_possible_cpu(i)
80	sum += per_cpu(nr_inodes, i);
81	return sum < 0 ? 0 : sum;
82	}
83
84	static inline long get_nr_inodes_unused(void)
85	{
86	int i;
87	long sum = 0;
88	for_each_possible_cpu(i)
89	sum += per_cpu(nr_unused, i);
90	return sum < 0 ? 0 : sum;
91	}
92
93	long get_nr_dirty_inodes(void)
94	{
95	/* not actually dirty inodes, but a wild approximation */
96	long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
97	return nr_dirty > 0 ? nr_dirty : 0;
98	}
99
100	/*
101	* Handle nr_inode sysctl
102	*/
103	#ifdef CONFIG_SYSCTL
104	/*
105	* Statistics gathering..
106	*/
107	static struct inodes_stat_t inodes_stat;
108
109	static int proc_nr_inodes(struct ctl_table *table, int write, void *buffer,
110	size_t *lenp, loff_t *ppos)
111	{
112	inodes_stat.nr_inodes = get_nr_inodes();
113	inodes_stat.nr_unused = get_nr_inodes_unused();
114	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
115	}
116
117	static struct ctl_table inodes_sysctls[] = {
118	{
119	.procname = "inode-nr",
120	.data = &inodes_stat,
121	.maxlen = 2*sizeof(long),
122	.mode = 0444,
123	.proc_handler = proc_nr_inodes,
124	},
125	{
126	.procname = "inode-state",
127	.data = &inodes_stat,
128	.maxlen = 7*sizeof(long),
129	.mode = 0444,
130	.proc_handler = proc_nr_inodes,
131	},
132	{ }
133	};
134
135	static int __init init_fs_inode_sysctls(void)
136	{
137	register_sysctl_init("fs", inodes_sysctls);
138	return 0;
139	}
140	early_initcall(init_fs_inode_sysctls);
141	#endif
142
143	static int no_open(struct inode *inode, struct file *file)
144	{
145	return -ENXIO;
146	}
147
148	/**
149	* inode_init_always - perform inode structure initialisation
150	* @sb: superblock inode belongs to
151	* @inode: inode to initialise
152	*
153	* These are initializations that need to be done on every inode
154	* allocation as the fields are not initialised by slab allocation.
155	*/
156	int inode_init_always(struct super_block *sb, struct inode *inode)
157	{
158	static const struct inode_operations empty_iops;
159	static const struct file_operations no_open_fops = {.open = no_open};
160	struct address_space *const mapping = &inode->i_data;
161
162	inode->i_sb = sb;
163	inode->i_blkbits = sb->s_blocksize_bits;
164	inode->i_flags = 0;
165	atomic64_set(v: &inode->i_sequence, i: 0);
166	atomic_set(v: &inode->i_count, i: 1);
167	inode->i_op = &empty_iops;
168	inode->i_fop = &no_open_fops;
169	inode->i_ino = 0;
170	inode->__i_nlink = 1;
171	inode->i_opflags = 0;
172	if (sb->s_xattr)
173	inode->i_opflags |= IOP_XATTR;
174	i_uid_write(inode, uid: 0);
175	i_gid_write(inode, gid: 0);
176	atomic_set(v: &inode->i_writecount, i: 0);
177	inode->i_size = 0;
178	inode->i_write_hint = WRITE_LIFE_NOT_SET;
179	inode->i_blocks = 0;
180	inode->i_bytes = 0;
181	inode->i_generation = 0;
182	inode->i_pipe = NULL;
183	inode->i_cdev = NULL;
184	inode->i_link = NULL;
185	inode->i_dir_seq = 0;
186	inode->i_rdev = 0;
187	inode->dirtied_when = 0;
188
189	#ifdef CONFIG_CGROUP_WRITEBACK
190	inode->i_wb_frn_winner = 0;
191	inode->i_wb_frn_avg_time = 0;
192	inode->i_wb_frn_history = 0;
193	#endif
194
195	spin_lock_init(&inode->i_lock);
196	lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
197
198	init_rwsem(&inode->i_rwsem);
199	lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
200
201	atomic_set(v: &inode->i_dio_count, i: 0);
202
203	mapping->a_ops = &empty_aops;
204	mapping->host = inode;
205	mapping->flags = 0;
206	mapping->wb_err = 0;
207	atomic_set(v: &mapping->i_mmap_writable, i: 0);
208	#ifdef CONFIG_READ_ONLY_THP_FOR_FS
209	atomic_set(v: &mapping->nr_thps, i: 0);
210	#endif
211	mapping_set_gfp_mask(m: mapping, GFP_HIGHUSER_MOVABLE);
212	mapping->private_data = NULL;
213	mapping->writeback_index = 0;
214	init_rwsem(&mapping->invalidate_lock);
215	lockdep_set_class_and_name(&mapping->invalidate_lock,
216	&sb->s_type->invalidate_lock_key,
217	"mapping.invalidate_lock");
218	inode->i_private = NULL;
219	inode->i_mapping = mapping;
220	INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
221	#ifdef CONFIG_FS_POSIX_ACL
222	inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
223	#endif
224
225	#ifdef CONFIG_FSNOTIFY
226	inode->i_fsnotify_mask = 0;
227	#endif
228	inode->i_flctx = NULL;
229
230	if (unlikely(security_inode_alloc(inode)))
231	return -ENOMEM;
232	this_cpu_inc(nr_inodes);
233
234	return 0;
235	}
236	EXPORT_SYMBOL(inode_init_always);
237
238	void free_inode_nonrcu(struct inode *inode)
239	{
240	kmem_cache_free(s: inode_cachep, objp: inode);
241	}
242	EXPORT_SYMBOL(free_inode_nonrcu);
243
244	static void i_callback(struct rcu_head *head)
245	{
246	struct inode *inode = container_of(head, struct inode, i_rcu);
247	if (inode->free_inode)
248	inode->free_inode(inode);
249	else
250	free_inode_nonrcu(inode);
251	}
252
253	static struct inode *alloc_inode(struct super_block *sb)
254	{
255	const struct super_operations *ops = sb->s_op;
256	struct inode *inode;
257
258	if (ops->alloc_inode)
259	inode = ops->alloc_inode(sb);
260	else
261	inode = alloc_inode_sb(sb, cache: inode_cachep, GFP_KERNEL);
262
263	if (!inode)
264	return NULL;
265
266	if (unlikely(inode_init_always(sb, inode))) {
267	if (ops->destroy_inode) {
268	ops->destroy_inode(inode);
269	if (!ops->free_inode)
270	return NULL;
271	}
272	inode->free_inode = ops->free_inode;
273	i_callback(head: &inode->i_rcu);
274	return NULL;
275	}
276
277	return inode;
278	}
279
280	void __destroy_inode(struct inode *inode)
281	{
282	BUG_ON(inode_has_buffers(inode));
283	inode_detach_wb(inode);
284	security_inode_free(inode);
285	fsnotify_inode_delete(inode);
286	locks_free_lock_context(inode);
287	if (!inode->i_nlink) {
288	WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
289	atomic_long_dec(v: &inode->i_sb->s_remove_count);
290	}
291
292	#ifdef CONFIG_FS_POSIX_ACL
293	if (inode->i_acl && !is_uncached_acl(acl: inode->i_acl))
294	posix_acl_release(acl: inode->i_acl);
295	if (inode->i_default_acl && !is_uncached_acl(acl: inode->i_default_acl))
296	posix_acl_release(acl: inode->i_default_acl);
297	#endif
298	this_cpu_dec(nr_inodes);
299	}
300	EXPORT_SYMBOL(__destroy_inode);
301
302	static void destroy_inode(struct inode *inode)
303	{
304	const struct super_operations *ops = inode->i_sb->s_op;
305
306	BUG_ON(!list_empty(&inode->i_lru));
307	__destroy_inode(inode);
308	if (ops->destroy_inode) {
309	ops->destroy_inode(inode);
310	if (!ops->free_inode)
311	return;
312	}
313	inode->free_inode = ops->free_inode;
314	call_rcu(head: &inode->i_rcu, func: i_callback);
315	}
316
317	/**
318	* drop_nlink - directly drop an inode's link count
319	* @inode: inode
320	*
321	* This is a low-level filesystem helper to replace any
322	* direct filesystem manipulation of i_nlink. In cases
323	* where we are attempting to track writes to the
324	* filesystem, a decrement to zero means an imminent
325	* write when the file is truncated and actually unlinked
326	* on the filesystem.
327	*/
328	void drop_nlink(struct inode *inode)
329	{
330	WARN_ON(inode->i_nlink == 0);
331	inode->__i_nlink--;
332	if (!inode->i_nlink)
333	atomic_long_inc(v: &inode->i_sb->s_remove_count);
334	}
335	EXPORT_SYMBOL(drop_nlink);
336
337	/**
338	* clear_nlink - directly zero an inode's link count
339	* @inode: inode
340	*
341	* This is a low-level filesystem helper to replace any
342	* direct filesystem manipulation of i_nlink. See
343	* drop_nlink() for why we care about i_nlink hitting zero.
344	*/
345	void clear_nlink(struct inode *inode)
346	{
347	if (inode->i_nlink) {
348	inode->__i_nlink = 0;
349	atomic_long_inc(v: &inode->i_sb->s_remove_count);
350	}
351	}
352	EXPORT_SYMBOL(clear_nlink);
353
354	/**
355	* set_nlink - directly set an inode's link count
356	* @inode: inode
357	* @nlink: new nlink (should be non-zero)
358	*
359	* This is a low-level filesystem helper to replace any
360	* direct filesystem manipulation of i_nlink.
361	*/
362	void set_nlink(struct inode *inode, unsigned int nlink)
363	{
364	if (!nlink) {
365	clear_nlink(inode);
366	} else {
367	/* Yes, some filesystems do change nlink from zero to one */
368	if (inode->i_nlink == 0)
369	atomic_long_dec(v: &inode->i_sb->s_remove_count);
370
371	inode->__i_nlink = nlink;
372	}
373	}
374	EXPORT_SYMBOL(set_nlink);
375
376	/**
377	* inc_nlink - directly increment an inode's link count
378	* @inode: inode
379	*
380	* This is a low-level filesystem helper to replace any
381	* direct filesystem manipulation of i_nlink. Currently,
382	* it is only here for parity with dec_nlink().
383	*/
384	void inc_nlink(struct inode *inode)
385	{
386	if (unlikely(inode->i_nlink == 0)) {
387	WARN_ON(!(inode->i_state & I_LINKABLE));
388	atomic_long_dec(v: &inode->i_sb->s_remove_count);
389	}
390
391	inode->__i_nlink++;
392	}
393	EXPORT_SYMBOL(inc_nlink);
394
395	static void __address_space_init_once(struct address_space *mapping)
396	{
397	xa_init_flags(xa: &mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
398	init_rwsem(&mapping->i_mmap_rwsem);
399	INIT_LIST_HEAD(list: &mapping->private_list);
400	spin_lock_init(&mapping->private_lock);
401	mapping->i_mmap = RB_ROOT_CACHED;
402	}
403
404	void address_space_init_once(struct address_space *mapping)
405	{
406	memset(mapping, 0, sizeof(*mapping));
407	__address_space_init_once(mapping);
408	}
409	EXPORT_SYMBOL(address_space_init_once);
410
411	/*
412	* These are initializations that only need to be done
413	* once, because the fields are idempotent across use
414	* of the inode, so let the slab aware of that.
415	*/
416	void inode_init_once(struct inode *inode)
417	{
418	memset(inode, 0, sizeof(*inode));
419	INIT_HLIST_NODE(h: &inode->i_hash);
420	INIT_LIST_HEAD(list: &inode->i_devices);
421	INIT_LIST_HEAD(list: &inode->i_io_list);
422	INIT_LIST_HEAD(list: &inode->i_wb_list);
423	INIT_LIST_HEAD(list: &inode->i_lru);
424	INIT_LIST_HEAD(list: &inode->i_sb_list);
425	__address_space_init_once(mapping: &inode->i_data);
426	i_size_ordered_init(inode);
427	}
428	EXPORT_SYMBOL(inode_init_once);
429
430	static void init_once(void *foo)
431	{
432	struct inode *inode = (struct inode *) foo;
433
434	inode_init_once(inode);
435	}
436
437	/*
438	* inode->i_lock must be held
439	*/
440	void __iget(struct inode *inode)
441	{
442	atomic_inc(v: &inode->i_count);
443	}
444
445	/*
446	* get additional reference to inode; caller must already hold one.
447	*/
448	void ihold(struct inode *inode)
449	{
450	WARN_ON(atomic_inc_return(&inode->i_count) < 2);
451	}
452	EXPORT_SYMBOL(ihold);
453
454	static void __inode_add_lru(struct inode *inode, bool rotate)
455	{
456	if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
457	return;
458	if (atomic_read(v: &inode->i_count))
459	return;
460	if (!(inode->i_sb->s_flags & SB_ACTIVE))
461	return;
462	if (!mapping_shrinkable(mapping: &inode->i_data))
463	return;
464
465	if (list_lru_add(lru: &inode->i_sb->s_inode_lru, item: &inode->i_lru))
466	this_cpu_inc(nr_unused);
467	else if (rotate)
468	inode->i_state |= I_REFERENCED;
469	}
470
471	/*
472	* Add inode to LRU if needed (inode is unused and clean).
473	*
474	* Needs inode->i_lock held.
475	*/
476	void inode_add_lru(struct inode *inode)
477	{
478	__inode_add_lru(inode, rotate: false);
479	}
480
481	static void inode_lru_list_del(struct inode *inode)
482	{
483	if (list_lru_del(lru: &inode->i_sb->s_inode_lru, item: &inode->i_lru))
484	this_cpu_dec(nr_unused);
485	}
486
487	/**
488	* inode_sb_list_add - add inode to the superblock list of inodes
489	* @inode: inode to add
490	*/
491	void inode_sb_list_add(struct inode *inode)
492	{
493	spin_lock(lock: &inode->i_sb->s_inode_list_lock);
494	list_add(new: &inode->i_sb_list, head: &inode->i_sb->s_inodes);
495	spin_unlock(lock: &inode->i_sb->s_inode_list_lock);
496	}
497	EXPORT_SYMBOL_GPL(inode_sb_list_add);
498
499	static inline void inode_sb_list_del(struct inode *inode)
500	{
501	if (!list_empty(head: &inode->i_sb_list)) {
502	spin_lock(lock: &inode->i_sb->s_inode_list_lock);
503	list_del_init(entry: &inode->i_sb_list);
504	spin_unlock(lock: &inode->i_sb->s_inode_list_lock);
505	}
506	}
507
508	static unsigned long hash(struct super_block *sb, unsigned long hashval)
509	{
510	unsigned long tmp;
511
512	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
513	L1_CACHE_BYTES;
514	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
515	return tmp & i_hash_mask;
516	}
517
518	/**
519	* __insert_inode_hash - hash an inode
520	* @inode: unhashed inode
521	* @hashval: unsigned long value used to locate this object in the
522	* inode_hashtable.
523	*
524	* Add an inode to the inode hash for this superblock.
525	*/
526	void __insert_inode_hash(struct inode *inode, unsigned long hashval)
527	{
528	struct hlist_head *b = inode_hashtable + hash(sb: inode->i_sb, hashval);
529
530	spin_lock(lock: &inode_hash_lock);
531	spin_lock(lock: &inode->i_lock);
532	hlist_add_head_rcu(n: &inode->i_hash, h: b);
533	spin_unlock(lock: &inode->i_lock);
534	spin_unlock(lock: &inode_hash_lock);
535	}
536	EXPORT_SYMBOL(__insert_inode_hash);
537
538	/**
539	* __remove_inode_hash - remove an inode from the hash
540	* @inode: inode to unhash
541	*
542	* Remove an inode from the superblock.
543	*/
544	void __remove_inode_hash(struct inode *inode)
545	{
546	spin_lock(lock: &inode_hash_lock);
547	spin_lock(lock: &inode->i_lock);
548	hlist_del_init_rcu(n: &inode->i_hash);
549	spin_unlock(lock: &inode->i_lock);
550	spin_unlock(lock: &inode_hash_lock);
551	}
552	EXPORT_SYMBOL(__remove_inode_hash);
553
554	void dump_mapping(const struct address_space *mapping)
555	{
556	struct inode *host;
557	const struct address_space_operations *a_ops;
558	struct hlist_node *dentry_first;
559	struct dentry *dentry_ptr;
560	struct dentry dentry;
561	unsigned long ino;
562
563	/*
564	* If mapping is an invalid pointer, we don't want to crash
565	* accessing it, so probe everything depending on it carefully.
566	*/
567	if (get_kernel_nofault(host, &mapping->host) ||
568	get_kernel_nofault(a_ops, &mapping->a_ops)) {
569	pr_warn("invalid mapping:%px\n", mapping);
570	return;
571	}
572
573	if (!host) {
574	pr_warn("aops:%ps\n", a_ops);
575	return;
576	}
577
578	if (get_kernel_nofault(dentry_first, &host->i_dentry.first) ||
579	get_kernel_nofault(ino, &host->i_ino)) {
580	pr_warn("aops:%ps invalid inode:%px\n", a_ops, host);
581	return;
582	}
583
584	if (!dentry_first) {
585	pr_warn("aops:%ps ino:%lx\n", a_ops, ino);
586	return;
587	}
588
589	dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
590	if (get_kernel_nofault(dentry, dentry_ptr)) {
591	pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
592	a_ops, ino, dentry_ptr);
593	return;
594	}
595
596	/*
597	* if dentry is corrupted, the %pd handler may still crash,
598	* but it's unlikely that we reach here with a corrupt mapping
599	*/
600	pr_warn("aops:%ps ino:%lx dentry name:\"%pd\"\n", a_ops, ino, &dentry);
601	}
602
603	void clear_inode(struct inode *inode)
604	{
605	/*
606	* We have to cycle the i_pages lock here because reclaim can be in the
607	* process of removing the last page (in __filemap_remove_folio())
608	* and we must not free the mapping under it.
609	*/
610	xa_lock_irq(&inode->i_data.i_pages);
611	BUG_ON(inode->i_data.nrpages);
612	/*
613	* Almost always, mapping_empty(&inode->i_data) here; but there are
614	* two known and long-standing ways in which nodes may get left behind
615	* (when deep radix-tree node allocation failed partway; or when THP
616	* collapse_file() failed). Until those two known cases are cleaned up,
617	* or a cleanup function is called here, do not BUG_ON(!mapping_empty),
618	* nor even WARN_ON(!mapping_empty).
619	*/
620	xa_unlock_irq(&inode->i_data.i_pages);
621	BUG_ON(!list_empty(&inode->i_data.private_list));
622	BUG_ON(!(inode->i_state & I_FREEING));
623	BUG_ON(inode->i_state & I_CLEAR);
624	BUG_ON(!list_empty(&inode->i_wb_list));
625	/* don't need i_lock here, no concurrent mods to i_state */
626	inode->i_state = I_FREEING | I_CLEAR;
627	}
628	EXPORT_SYMBOL(clear_inode);
629
630	/*
631	* Free the inode passed in, removing it from the lists it is still connected
632	* to. We remove any pages still attached to the inode and wait for any IO that
633	* is still in progress before finally destroying the inode.
634	*
635	* An inode must already be marked I_FREEING so that we avoid the inode being
636	* moved back onto lists if we race with other code that manipulates the lists
637	* (e.g. writeback_single_inode). The caller is responsible for setting this.
638	*
639	* An inode must already be removed from the LRU list before being evicted from
640	* the cache. This should occur atomically with setting the I_FREEING state
641	* flag, so no inodes here should ever be on the LRU when being evicted.
642	*/
643	static void evict(struct inode *inode)
644	{
645	const struct super_operations *op = inode->i_sb->s_op;
646
647	BUG_ON(!(inode->i_state & I_FREEING));
648	BUG_ON(!list_empty(&inode->i_lru));
649
650	if (!list_empty(head: &inode->i_io_list))
651	inode_io_list_del(inode);
652
653	inode_sb_list_del(inode);
654
655	/*
656	* Wait for flusher thread to be done with the inode so that filesystem
657	* does not start destroying it while writeback is still running. Since
658	* the inode has I_FREEING set, flusher thread won't start new work on
659	* the inode. We just have to wait for running writeback to finish.
660	*/
661	inode_wait_for_writeback(inode);
662
663	if (op->evict_inode) {
664	op->evict_inode(inode);
665	} else {
666	truncate_inode_pages_final(&inode->i_data);
667	clear_inode(inode);
668	}
669	if (S_ISCHR(inode->i_mode) && inode->i_cdev)
670	cd_forget(inode);
671
672	remove_inode_hash(inode);
673
674	spin_lock(lock: &inode->i_lock);
675	wake_up_bit(word: &inode->i_state, __I_NEW);
676	BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
677	spin_unlock(lock: &inode->i_lock);
678
679	destroy_inode(inode);
680	}
681
682	/*
683	* dispose_list - dispose of the contents of a local list
684	* @head: the head of the list to free
685	*
686	* Dispose-list gets a local list with local inodes in it, so it doesn't
687	* need to worry about list corruption and SMP locks.
688	*/
689	static void dispose_list(struct list_head *head)
690	{
691	while (!list_empty(head)) {
692	struct inode *inode;
693
694	inode = list_first_entry(head, struct inode, i_lru);
695	list_del_init(entry: &inode->i_lru);
696
697	evict(inode);
698	cond_resched();
699	}
700	}
701
702	/**
703	* evict_inodes - evict all evictable inodes for a superblock
704	* @sb: superblock to operate on
705	*
706	* Make sure that no inodes with zero refcount are retained. This is
707	* called by superblock shutdown after having SB_ACTIVE flag removed,
708	* so any inode reaching zero refcount during or after that call will
709	* be immediately evicted.
710	*/
711	void evict_inodes(struct super_block *sb)
712	{
713	struct inode *inode, *next;
714	LIST_HEAD(dispose);
715
716	again:
717	spin_lock(lock: &sb->s_inode_list_lock);
718	list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
719	if (atomic_read(v: &inode->i_count))
720	continue;
721
722	spin_lock(lock: &inode->i_lock);
723	if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
724	spin_unlock(lock: &inode->i_lock);
725	continue;
726	}
727
728	inode->i_state |= I_FREEING;
729	inode_lru_list_del(inode);
730	spin_unlock(lock: &inode->i_lock);
731	list_add(new: &inode->i_lru, head: &dispose);
732
733	/*
734	* We can have a ton of inodes to evict at unmount time given
735	* enough memory, check to see if we need to go to sleep for a
736	* bit so we don't livelock.
737	*/
738	if (need_resched()) {
739	spin_unlock(lock: &sb->s_inode_list_lock);
740	cond_resched();
741	dispose_list(head: &dispose);
742	goto again;
743	}
744	}
745	spin_unlock(lock: &sb->s_inode_list_lock);
746
747	dispose_list(head: &dispose);
748	}
749	EXPORT_SYMBOL_GPL(evict_inodes);
750
751	/**
752	* invalidate_inodes - attempt to free all inodes on a superblock
753	* @sb: superblock to operate on
754	*
755	* Attempts to free all inodes (including dirty inodes) for a given superblock.
756	*/
757	void invalidate_inodes(struct super_block *sb)
758	{
759	struct inode *inode, *next;
760	LIST_HEAD(dispose);
761
762	again:
763	spin_lock(lock: &sb->s_inode_list_lock);
764	list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
765	spin_lock(lock: &inode->i_lock);
766	if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
767	spin_unlock(lock: &inode->i_lock);
768	continue;
769	}
770	if (atomic_read(v: &inode->i_count)) {
771	spin_unlock(lock: &inode->i_lock);
772	continue;
773	}
774
775	inode->i_state |= I_FREEING;
776	inode_lru_list_del(inode);
777	spin_unlock(lock: &inode->i_lock);
778	list_add(new: &inode->i_lru, head: &dispose);
779	if (need_resched()) {
780	spin_unlock(lock: &sb->s_inode_list_lock);
781	cond_resched();
782	dispose_list(head: &dispose);
783	goto again;
784	}
785	}
786	spin_unlock(lock: &sb->s_inode_list_lock);
787
788	dispose_list(head: &dispose);
789	}
790
791	/*
792	* Isolate the inode from the LRU in preparation for freeing it.
793	*
794	* If the inode has the I_REFERENCED flag set, then it means that it has been
795	* used recently - the flag is set in iput_final(). When we encounter such an
796	* inode, clear the flag and move it to the back of the LRU so it gets another
797	* pass through the LRU before it gets reclaimed. This is necessary because of
798	* the fact we are doing lazy LRU updates to minimise lock contention so the
799	* LRU does not have strict ordering. Hence we don't want to reclaim inodes
800	* with this flag set because they are the inodes that are out of order.
801	*/
802	static enum lru_status inode_lru_isolate(struct list_head *item,
803	struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
804	{
805	struct list_head *freeable = arg;
806	struct inode *inode = container_of(item, struct inode, i_lru);
807
808	/*
809	* We are inverting the lru lock/inode->i_lock here, so use a
810	* trylock. If we fail to get the lock, just skip it.
811	*/
812	if (!spin_trylock(lock: &inode->i_lock))
813	return LRU_SKIP;
814
815	/*
816	* Inodes can get referenced, redirtied, or repopulated while
817	* they're already on the LRU, and this can make them
818	* unreclaimable for a while. Remove them lazily here; iput,
819	* sync, or the last page cache deletion will requeue them.
820	*/
821	if (atomic_read(v: &inode->i_count) ||
822	(inode->i_state & ~I_REFERENCED) ||
823	!mapping_shrinkable(mapping: &inode->i_data)) {
824	list_lru_isolate(list: lru, item: &inode->i_lru);
825	spin_unlock(lock: &inode->i_lock);
826	this_cpu_dec(nr_unused);
827	return LRU_REMOVED;
828	}
829
830	/* Recently referenced inodes get one more pass */
831	if (inode->i_state & I_REFERENCED) {
832	inode->i_state &= ~I_REFERENCED;
833	spin_unlock(lock: &inode->i_lock);
834	return LRU_ROTATE;
835	}
836
837	/*
838	* On highmem systems, mapping_shrinkable() permits dropping
839	* page cache in order to free up struct inodes: lowmem might
840	* be under pressure before the cache inside the highmem zone.
841	*/
842	if (inode_has_buffers(inode) || !mapping_empty(mapping: &inode->i_data)) {
843	__iget(inode);
844	spin_unlock(lock: &inode->i_lock);
845	spin_unlock(lock: lru_lock);
846	if (remove_inode_buffers(inode)) {
847	unsigned long reap;
848	reap = invalidate_mapping_pages(mapping: &inode->i_data, start: 0, end: -1);
849	if (current_is_kswapd())
850	__count_vm_events(item: KSWAPD_INODESTEAL, delta: reap);
851	else
852	__count_vm_events(item: PGINODESTEAL, delta: reap);
853	mm_account_reclaimed_pages(pages: reap);
854	}
855	iput(inode);
856	spin_lock(lock: lru_lock);
857	return LRU_RETRY;
858	}
859
860	WARN_ON(inode->i_state & I_NEW);
861	inode->i_state |= I_FREEING;
862	list_lru_isolate_move(list: lru, item: &inode->i_lru, head: freeable);
863	spin_unlock(lock: &inode->i_lock);
864
865	this_cpu_dec(nr_unused);
866	return LRU_REMOVED;
867	}
868
869	/*
870	* Walk the superblock inode LRU for freeable inodes and attempt to free them.
871	* This is called from the superblock shrinker function with a number of inodes
872	* to trim from the LRU. Inodes to be freed are moved to a temporary list and
873	* then are freed outside inode_lock by dispose_list().
874	*/
875	long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
876	{
877	LIST_HEAD(freeable);
878	long freed;
879
880	freed = list_lru_shrink_walk(lru: &sb->s_inode_lru, sc,
881	isolate: inode_lru_isolate, cb_arg: &freeable);
882	dispose_list(head: &freeable);
883	return freed;
884	}
885
886	static void __wait_on_freeing_inode(struct inode *inode);
887	/*
888	* Called with the inode lock held.
889	*/
890	static struct inode *find_inode(struct super_block *sb,
891	struct hlist_head *head,
892	int (*test)(struct inode *, void *),
893	void *data)
894	{
895	struct inode *inode = NULL;
896
897	repeat:
898	hlist_for_each_entry(inode, head, i_hash) {
899	if (inode->i_sb != sb)
900	continue;
901	if (!test(inode, data))
902	continue;
903	spin_lock(lock: &inode->i_lock);
904	if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
905	__wait_on_freeing_inode(inode);
906	goto repeat;
907	}
908	if (unlikely(inode->i_state & I_CREATING)) {
909	spin_unlock(lock: &inode->i_lock);
910	return ERR_PTR(error: -ESTALE);
911	}
912	__iget(inode);
913	spin_unlock(lock: &inode->i_lock);
914	return inode;
915	}
916	return NULL;
917	}
918
919	/*
920	* find_inode_fast is the fast path version of find_inode, see the comment at
921	* iget_locked for details.
922	*/
923	static struct inode *find_inode_fast(struct super_block *sb,
924	struct hlist_head *head, unsigned long ino)
925	{
926	struct inode *inode = NULL;
927
928	repeat:
929	hlist_for_each_entry(inode, head, i_hash) {
930	if (inode->i_ino != ino)
931	continue;
932	if (inode->i_sb != sb)
933	continue;
934	spin_lock(lock: &inode->i_lock);
935	if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
936	__wait_on_freeing_inode(inode);
937	goto repeat;
938	}
939	if (unlikely(inode->i_state & I_CREATING)) {
940	spin_unlock(lock: &inode->i_lock);
941	return ERR_PTR(error: -ESTALE);
942	}
943	__iget(inode);
944	spin_unlock(lock: &inode->i_lock);
945	return inode;
946	}
947	return NULL;
948	}
949
950	/*
951	* Each cpu owns a range of LAST_INO_BATCH numbers.
952	* 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
953	* to renew the exhausted range.
954	*
955	* This does not significantly increase overflow rate because every CPU can
956	* consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
957	* NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
958	* 2^32 range, and is a worst-case. Even a 50% wastage would only increase
959	* overflow rate by 2x, which does not seem too significant.
960	*
961	* On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
962	* error if st_ino won't fit in target struct field. Use 32bit counter
963	* here to attempt to avoid that.
964	*/
965	#define LAST_INO_BATCH 1024
966	static DEFINE_PER_CPU(unsigned int, last_ino);
967
968	unsigned int get_next_ino(void)
969	{
970	unsigned int *p = &get_cpu_var(last_ino);
971	unsigned int res = *p;
972
973	#ifdef CONFIG_SMP
974	if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
975	static atomic_t shared_last_ino;
976	int next = atomic_add_return(LAST_INO_BATCH, v: &shared_last_ino);
977
978	res = next - LAST_INO_BATCH;
979	}
980	#endif
981
982	res++;
983	/* get_next_ino should not provide a 0 inode number */
984	if (unlikely(!res))
985	res++;
986	*p = res;
987	put_cpu_var(last_ino);
988	return res;
989	}
990	EXPORT_SYMBOL(get_next_ino);
991
992	/**
993	* new_inode_pseudo - obtain an inode
994	* @sb: superblock
995	*
996	* Allocates a new inode for given superblock.
997	* Inode wont be chained in superblock s_inodes list
998	* This means :
999	* - fs can't be unmount
1000	* - quotas, fsnotify, writeback can't work
1001	*/
1002	struct inode *new_inode_pseudo(struct super_block *sb)
1003	{
1004	struct inode *inode = alloc_inode(sb);
1005
1006	if (inode) {
1007	spin_lock(lock: &inode->i_lock);
1008	inode->i_state = 0;
1009	spin_unlock(lock: &inode->i_lock);
1010	}
1011	return inode;
1012	}
1013
1014	/**
1015	* new_inode - obtain an inode
1016	* @sb: superblock
1017	*
1018	* Allocates a new inode for given superblock. The default gfp_mask
1019	* for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1020	* If HIGHMEM pages are unsuitable or it is known that pages allocated
1021	* for the page cache are not reclaimable or migratable,
1022	* mapping_set_gfp_mask() must be called with suitable flags on the
1023	* newly created inode's mapping
1024	*
1025	*/
1026	struct inode *new_inode(struct super_block *sb)
1027	{
1028	struct inode *inode;
1029
1030	inode = new_inode_pseudo(sb);
1031	if (inode)
1032	inode_sb_list_add(inode);
1033	return inode;
1034	}
1035	EXPORT_SYMBOL(new_inode);
1036
1037	#ifdef CONFIG_DEBUG_LOCK_ALLOC
1038	void lockdep_annotate_inode_mutex_key(struct inode *inode)
1039	{
1040	if (S_ISDIR(inode->i_mode)) {
1041	struct file_system_type *type = inode->i_sb->s_type;
1042
1043	/* Set new key only if filesystem hasn't already changed it */
1044	if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1045	/*
1046	* ensure nobody is actually holding i_mutex
1047	*/
1048	// mutex_destroy(&inode->i_mutex);
1049	init_rwsem(&inode->i_rwsem);
1050	lockdep_set_class(&inode->i_rwsem,
1051	&type->i_mutex_dir_key);
1052	}
1053	}
1054	}
1055	EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1056	#endif
1057
1058	/**
1059	* unlock_new_inode - clear the I_NEW state and wake up any waiters
1060	* @inode: new inode to unlock
1061	*
1062	* Called when the inode is fully initialised to clear the new state of the
1063	* inode and wake up anyone waiting for the inode to finish initialisation.
1064	*/
1065	void unlock_new_inode(struct inode *inode)
1066	{
1067	lockdep_annotate_inode_mutex_key(inode);
1068	spin_lock(lock: &inode->i_lock);
1069	WARN_ON(!(inode->i_state & I_NEW));
1070	inode->i_state &= ~I_NEW & ~I_CREATING;
1071	smp_mb();
1072	wake_up_bit(word: &inode->i_state, __I_NEW);
1073	spin_unlock(lock: &inode->i_lock);
1074	}
1075	EXPORT_SYMBOL(unlock_new_inode);
1076
1077	void discard_new_inode(struct inode *inode)
1078	{
1079	lockdep_annotate_inode_mutex_key(inode);
1080	spin_lock(lock: &inode->i_lock);
1081	WARN_ON(!(inode->i_state & I_NEW));
1082	inode->i_state &= ~I_NEW;
1083	smp_mb();
1084	wake_up_bit(word: &inode->i_state, __I_NEW);
1085	spin_unlock(lock: &inode->i_lock);
1086	iput(inode);
1087	}
1088	EXPORT_SYMBOL(discard_new_inode);
1089
1090	/**
1091	* lock_two_inodes - lock two inodes (may be regular files but also dirs)
1092	*
1093	* Lock any non-NULL argument. The caller must make sure that if he is passing
1094	* in two directories, one is not ancestor of the other. Zero, one or two
1095	* objects may be locked by this function.
1096	*
1097	* @inode1: first inode to lock
1098	* @inode2: second inode to lock
1099	* @subclass1: inode lock subclass for the first lock obtained
1100	* @subclass2: inode lock subclass for the second lock obtained
1101	*/
1102	void lock_two_inodes(struct inode *inode1, struct inode *inode2,
1103	unsigned subclass1, unsigned subclass2)
1104	{
1105	if (!inode1 || !inode2) {
1106	/*
1107	* Make sure @subclass1 will be used for the acquired lock.
1108	* This is not strictly necessary (no current caller cares) but
1109	* let's keep things consistent.
1110	*/
1111	if (!inode1)
1112	swap(inode1, inode2);
1113	goto lock;
1114	}
1115
1116	/*
1117	* If one object is directory and the other is not, we must make sure
1118	* to lock directory first as the other object may be its child.
1119	*/
1120	if (S_ISDIR(inode2->i_mode) == S_ISDIR(inode1->i_mode)) {
1121	if (inode1 > inode2)
1122	swap(inode1, inode2);
1123	} else if (!S_ISDIR(inode1->i_mode))
1124	swap(inode1, inode2);
1125	lock:
1126	if (inode1)
1127	inode_lock_nested(inode: inode1, subclass: subclass1);
1128	if (inode2 && inode2 != inode1)
1129	inode_lock_nested(inode: inode2, subclass: subclass2);
1130	}
1131
1132	/**
1133	* lock_two_nondirectories - take two i_mutexes on non-directory objects
1134	*
1135	* Lock any non-NULL argument. Passed objects must not be directories.
1136	* Zero, one or two objects may be locked by this function.
1137	*
1138	* @inode1: first inode to lock
1139	* @inode2: second inode to lock
1140	*/
1141	void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1142	{
1143	if (inode1)
1144	WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1145	if (inode2)
1146	WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1147	lock_two_inodes(inode1, inode2, subclass1: I_MUTEX_NORMAL, subclass2: I_MUTEX_NONDIR2);
1148	}
1149	EXPORT_SYMBOL(lock_two_nondirectories);
1150
1151	/**
1152	* unlock_two_nondirectories - release locks from lock_two_nondirectories()
1153	* @inode1: first inode to unlock
1154	* @inode2: second inode to unlock
1155	*/
1156	void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1157	{
1158	if (inode1) {
1159	WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1160	inode_unlock(inode: inode1);
1161	}
1162	if (inode2 && inode2 != inode1) {
1163	WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1164	inode_unlock(inode: inode2);
1165	}
1166	}
1167	EXPORT_SYMBOL(unlock_two_nondirectories);
1168
1169	/**
1170	* inode_insert5 - obtain an inode from a mounted file system
1171	* @inode: pre-allocated inode to use for insert to cache
1172	* @hashval: hash value (usually inode number) to get
1173	* @test: callback used for comparisons between inodes
1174	* @set: callback used to initialize a new struct inode
1175	* @data: opaque data pointer to pass to @test and @set
1176	*
1177	* Search for the inode specified by @hashval and @data in the inode cache,
1178	* and if present it is return it with an increased reference count. This is
1179	* a variant of iget5_locked() for callers that don't want to fail on memory
1180	* allocation of inode.
1181	*
1182	* If the inode is not in cache, insert the pre-allocated inode to cache and
1183	* return it locked, hashed, and with the I_NEW flag set. The file system gets
1184	* to fill it in before unlocking it via unlock_new_inode().
1185	*
1186	* Note both @test and @set are called with the inode_hash_lock held, so can't
1187	* sleep.
1188	*/
1189	struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1190	int (*test)(struct inode *, void *),
1191	int (*set)(struct inode *, void *), void *data)
1192	{
1193	struct hlist_head *head = inode_hashtable + hash(sb: inode->i_sb, hashval);
1194	struct inode *old;
1195
1196	again:
1197	spin_lock(lock: &inode_hash_lock);
1198	old = find_inode(sb: inode->i_sb, head, test, data);
1199	if (unlikely(old)) {
1200	/*
1201	* Uhhuh, somebody else created the same inode under us.
1202	* Use the old inode instead of the preallocated one.
1203	*/
1204	spin_unlock(lock: &inode_hash_lock);
1205	if (IS_ERR(ptr: old))
1206	return NULL;
1207	wait_on_inode(inode: old);
1208	if (unlikely(inode_unhashed(old))) {
1209	iput(old);
1210	goto again;
1211	}
1212	return old;
1213	}
1214
1215	if (set && unlikely(set(inode, data))) {
1216	inode = NULL;
1217	goto unlock;
1218	}
1219
1220	/*
1221	* Return the locked inode with I_NEW set, the
1222	* caller is responsible for filling in the contents
1223	*/
1224	spin_lock(lock: &inode->i_lock);
1225	inode->i_state |= I_NEW;
1226	hlist_add_head_rcu(n: &inode->i_hash, h: head);
1227	spin_unlock(lock: &inode->i_lock);
1228
1229	/*
1230	* Add inode to the sb list if it's not already. It has I_NEW at this
1231	* point, so it should be safe to test i_sb_list locklessly.
1232	*/
1233	if (list_empty(head: &inode->i_sb_list))
1234	inode_sb_list_add(inode);
1235	unlock:
1236	spin_unlock(lock: &inode_hash_lock);
1237
1238	return inode;
1239	}
1240	EXPORT_SYMBOL(inode_insert5);
1241
1242	/**
1243	* iget5_locked - obtain an inode from a mounted file system
1244	* @sb: super block of file system
1245	* @hashval: hash value (usually inode number) to get
1246	* @test: callback used for comparisons between inodes
1247	* @set: callback used to initialize a new struct inode
1248	* @data: opaque data pointer to pass to @test and @set
1249	*
1250	* Search for the inode specified by @hashval and @data in the inode cache,
1251	* and if present it is return it with an increased reference count. This is
1252	* a generalized version of iget_locked() for file systems where the inode
1253	* number is not sufficient for unique identification of an inode.
1254	*
1255	* If the inode is not in cache, allocate a new inode and return it locked,
1256	* hashed, and with the I_NEW flag set. The file system gets to fill it in
1257	* before unlocking it via unlock_new_inode().
1258	*
1259	* Note both @test and @set are called with the inode_hash_lock held, so can't
1260	* sleep.
1261	*/
1262	struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1263	int (*test)(struct inode *, void *),
1264	int (*set)(struct inode *, void *), void *data)
1265	{
1266	struct inode *inode = ilookup5(sb, hashval, test, data);
1267
1268	if (!inode) {
1269	struct inode *new = alloc_inode(sb);
1270
1271	if (new) {
1272	new->i_state = 0;
1273	inode = inode_insert5(new, hashval, test, set, data);
1274	if (unlikely(inode != new))
1275	destroy_inode(inode: new);
1276	}
1277	}
1278	return inode;
1279	}
1280	EXPORT_SYMBOL(iget5_locked);
1281
1282	/**
1283	* iget_locked - obtain an inode from a mounted file system
1284	* @sb: super block of file system
1285	* @ino: inode number to get
1286	*
1287	* Search for the inode specified by @ino in the inode cache and if present
1288	* return it with an increased reference count. This is for file systems
1289	* where the inode number is sufficient for unique identification of an inode.
1290	*
1291	* If the inode is not in cache, allocate a new inode and return it locked,
1292	* hashed, and with the I_NEW flag set. The file system gets to fill it in
1293	* before unlocking it via unlock_new_inode().
1294	*/
1295	struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1296	{
1297	struct hlist_head *head = inode_hashtable + hash(sb, hashval: ino);
1298	struct inode *inode;
1299	again:
1300	spin_lock(lock: &inode_hash_lock);
1301	inode = find_inode_fast(sb, head, ino);
1302	spin_unlock(lock: &inode_hash_lock);
1303	if (inode) {
1304	if (IS_ERR(ptr: inode))
1305	return NULL;
1306	wait_on_inode(inode);
1307	if (unlikely(inode_unhashed(inode))) {
1308	iput(inode);
1309	goto again;
1310	}
1311	return inode;
1312	}
1313
1314	inode = alloc_inode(sb);
1315	if (inode) {
1316	struct inode *old;
1317
1318	spin_lock(lock: &inode_hash_lock);
1319	/* We released the lock, so.. */
1320	old = find_inode_fast(sb, head, ino);
1321	if (!old) {
1322	inode->i_ino = ino;
1323	spin_lock(lock: &inode->i_lock);
1324	inode->i_state = I_NEW;
1325	hlist_add_head_rcu(n: &inode->i_hash, h: head);
1326	spin_unlock(lock: &inode->i_lock);
1327	inode_sb_list_add(inode);
1328	spin_unlock(lock: &inode_hash_lock);
1329
1330	/* Return the locked inode with I_NEW set, the
1331	* caller is responsible for filling in the contents
1332	*/
1333	return inode;
1334	}
1335
1336	/*
1337	* Uhhuh, somebody else created the same inode under
1338	* us. Use the old inode instead of the one we just
1339	* allocated.
1340	*/
1341	spin_unlock(lock: &inode_hash_lock);
1342	destroy_inode(inode);
1343	if (IS_ERR(ptr: old))
1344	return NULL;
1345	inode = old;
1346	wait_on_inode(inode);
1347	if (unlikely(inode_unhashed(inode))) {
1348	iput(inode);
1349	goto again;
1350	}
1351	}
1352	return inode;
1353	}
1354	EXPORT_SYMBOL(iget_locked);
1355
1356	/*
1357	* search the inode cache for a matching inode number.
1358	* If we find one, then the inode number we are trying to
1359	* allocate is not unique and so we should not use it.
1360	*
1361	* Returns 1 if the inode number is unique, 0 if it is not.
1362	*/
1363	static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1364	{
1365	struct hlist_head *b = inode_hashtable + hash(sb, hashval: ino);
1366	struct inode *inode;
1367
1368	hlist_for_each_entry_rcu(inode, b, i_hash) {
1369	if (inode->i_ino == ino && inode->i_sb == sb)
1370	return 0;
1371	}
1372	return 1;
1373	}
1374
1375	/**
1376	* iunique - get a unique inode number
1377	* @sb: superblock
1378	* @max_reserved: highest reserved inode number
1379	*
1380	* Obtain an inode number that is unique on the system for a given
1381	* superblock. This is used by file systems that have no natural
1382	* permanent inode numbering system. An inode number is returned that
1383	* is higher than the reserved limit but unique.
1384	*
1385	* BUGS:
1386	* With a large number of inodes live on the file system this function
1387	* currently becomes quite slow.
1388	*/
1389	ino_t iunique(struct super_block *sb, ino_t max_reserved)
1390	{
1391	/*
1392	* On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1393	* error if st_ino won't fit in target struct field. Use 32bit counter
1394	* here to attempt to avoid that.
1395	*/
1396	static DEFINE_SPINLOCK(iunique_lock);
1397	static unsigned int counter;
1398	ino_t res;
1399
1400	rcu_read_lock();
1401	spin_lock(lock: &iunique_lock);
1402	do {
1403	if (counter <= max_reserved)
1404	counter = max_reserved + 1;
1405	res = counter++;
1406	} while (!test_inode_iunique(sb, ino: res));
1407	spin_unlock(lock: &iunique_lock);
1408	rcu_read_unlock();
1409
1410	return res;
1411	}
1412	EXPORT_SYMBOL(iunique);
1413
1414	struct inode *igrab(struct inode *inode)
1415	{
1416	spin_lock(lock: &inode->i_lock);
1417	if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1418	__iget(inode);
1419	spin_unlock(lock: &inode->i_lock);
1420	} else {
1421	spin_unlock(lock: &inode->i_lock);
1422	/*
1423	* Handle the case where s_op->clear_inode is not been
1424	* called yet, and somebody is calling igrab
1425	* while the inode is getting freed.
1426	*/
1427	inode = NULL;
1428	}
1429	return inode;
1430	}
1431	EXPORT_SYMBOL(igrab);
1432
1433	/**
1434	* ilookup5_nowait - search for an inode in the inode cache
1435	* @sb: super block of file system to search
1436	* @hashval: hash value (usually inode number) to search for
1437	* @test: callback used for comparisons between inodes
1438	* @data: opaque data pointer to pass to @test
1439	*
1440	* Search for the inode specified by @hashval and @data in the inode cache.
1441	* If the inode is in the cache, the inode is returned with an incremented
1442	* reference count.
1443	*
1444	* Note: I_NEW is not waited upon so you have to be very careful what you do
1445	* with the returned inode. You probably should be using ilookup5() instead.
1446	*
1447	* Note2: @test is called with the inode_hash_lock held, so can't sleep.
1448	*/
1449	struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1450	int (*test)(struct inode *, void *), void *data)
1451	{
1452	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1453	struct inode *inode;
1454
1455	spin_lock(lock: &inode_hash_lock);
1456	inode = find_inode(sb, head, test, data);
1457	spin_unlock(lock: &inode_hash_lock);
1458
1459	return IS_ERR(ptr: inode) ? NULL : inode;
1460	}
1461	EXPORT_SYMBOL(ilookup5_nowait);
1462
1463	/**
1464	* ilookup5 - search for an inode in the inode cache
1465	* @sb: super block of file system to search
1466	* @hashval: hash value (usually inode number) to search for
1467	* @test: callback used for comparisons between inodes
1468	* @data: opaque data pointer to pass to @test
1469	*
1470	* Search for the inode specified by @hashval and @data in the inode cache,
1471	* and if the inode is in the cache, return the inode with an incremented
1472	* reference count. Waits on I_NEW before returning the inode.
1473	* returned with an incremented reference count.
1474	*
1475	* This is a generalized version of ilookup() for file systems where the
1476	* inode number is not sufficient for unique identification of an inode.
1477	*
1478	* Note: @test is called with the inode_hash_lock held, so can't sleep.
1479	*/
1480	struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1481	int (*test)(struct inode *, void *), void *data)
1482	{
1483	struct inode *inode;
1484	again:
1485	inode = ilookup5_nowait(sb, hashval, test, data);
1486	if (inode) {
1487	wait_on_inode(inode);
1488	if (unlikely(inode_unhashed(inode))) {
1489	iput(inode);
1490	goto again;
1491	}
1492	}
1493	return inode;
1494	}
1495	EXPORT_SYMBOL(ilookup5);
1496
1497	/**
1498	* ilookup - search for an inode in the inode cache
1499	* @sb: super block of file system to search
1500	* @ino: inode number to search for
1501	*
1502	* Search for the inode @ino in the inode cache, and if the inode is in the
1503	* cache, the inode is returned with an incremented reference count.
1504	*/
1505	struct inode *ilookup(struct super_block *sb, unsigned long ino)
1506	{
1507	struct hlist_head *head = inode_hashtable + hash(sb, hashval: ino);
1508	struct inode *inode;
1509	again:
1510	spin_lock(lock: &inode_hash_lock);
1511	inode = find_inode_fast(sb, head, ino);
1512	spin_unlock(lock: &inode_hash_lock);
1513
1514	if (inode) {
1515	if (IS_ERR(ptr: inode))
1516	return NULL;
1517	wait_on_inode(inode);
1518	if (unlikely(inode_unhashed(inode))) {
1519	iput(inode);
1520	goto again;
1521	}
1522	}
1523	return inode;
1524	}
1525	EXPORT_SYMBOL(ilookup);
1526
1527	/**
1528	* find_inode_nowait - find an inode in the inode cache
1529	* @sb: super block of file system to search
1530	* @hashval: hash value (usually inode number) to search for
1531	* @match: callback used for comparisons between inodes
1532	* @data: opaque data pointer to pass to @match
1533	*
1534	* Search for the inode specified by @hashval and @data in the inode
1535	* cache, where the helper function @match will return 0 if the inode
1536	* does not match, 1 if the inode does match, and -1 if the search
1537	* should be stopped. The @match function must be responsible for
1538	* taking the i_lock spin_lock and checking i_state for an inode being
1539	* freed or being initialized, and incrementing the reference count
1540	* before returning 1. It also must not sleep, since it is called with
1541	* the inode_hash_lock spinlock held.
1542	*
1543	* This is a even more generalized version of ilookup5() when the
1544	* function must never block --- find_inode() can block in
1545	* __wait_on_freeing_inode() --- or when the caller can not increment
1546	* the reference count because the resulting iput() might cause an
1547	* inode eviction. The tradeoff is that the @match funtion must be
1548	* very carefully implemented.
1549	*/
1550	struct inode *find_inode_nowait(struct super_block *sb,
1551	unsigned long hashval,
1552	int (*match)(struct inode *, unsigned long,
1553	void *),
1554	void *data)
1555	{
1556	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1557	struct inode *inode, *ret_inode = NULL;
1558	int mval;
1559
1560	spin_lock(lock: &inode_hash_lock);
1561	hlist_for_each_entry(inode, head, i_hash) {
1562	if (inode->i_sb != sb)
1563	continue;
1564	mval = match(inode, hashval, data);
1565	if (mval == 0)
1566	continue;
1567	if (mval == 1)
1568	ret_inode = inode;
1569	goto out;
1570	}
1571	out:
1572	spin_unlock(lock: &inode_hash_lock);
1573	return ret_inode;
1574	}
1575	EXPORT_SYMBOL(find_inode_nowait);
1576
1577	/**
1578	* find_inode_rcu - find an inode in the inode cache
1579	* @sb: Super block of file system to search
1580	* @hashval: Key to hash
1581	* @test: Function to test match on an inode
1582	* @data: Data for test function
1583	*
1584	* Search for the inode specified by @hashval and @data in the inode cache,
1585	* where the helper function @test will return 0 if the inode does not match
1586	* and 1 if it does. The @test function must be responsible for taking the
1587	* i_lock spin_lock and checking i_state for an inode being freed or being
1588	* initialized.
1589	*
1590	* If successful, this will return the inode for which the @test function
1591	* returned 1 and NULL otherwise.
1592	*
1593	* The @test function is not permitted to take a ref on any inode presented.
1594	* It is also not permitted to sleep.
1595	*
1596	* The caller must hold the RCU read lock.
1597	*/
1598	struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1599	int (*test)(struct inode *, void *), void *data)
1600	{
1601	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1602	struct inode *inode;
1603
1604	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1605	"suspicious find_inode_rcu() usage");
1606
1607	hlist_for_each_entry_rcu(inode, head, i_hash) {
1608	if (inode->i_sb == sb &&
1609	!(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1610	test(inode, data))
1611	return inode;
1612	}
1613	return NULL;
1614	}
1615	EXPORT_SYMBOL(find_inode_rcu);
1616
1617	/**
1618	* find_inode_by_ino_rcu - Find an inode in the inode cache
1619	* @sb: Super block of file system to search
1620	* @ino: The inode number to match
1621	*
1622	* Search for the inode specified by @hashval and @data in the inode cache,
1623	* where the helper function @test will return 0 if the inode does not match
1624	* and 1 if it does. The @test function must be responsible for taking the
1625	* i_lock spin_lock and checking i_state for an inode being freed or being
1626	* initialized.
1627	*
1628	* If successful, this will return the inode for which the @test function
1629	* returned 1 and NULL otherwise.
1630	*
1631	* The @test function is not permitted to take a ref on any inode presented.
1632	* It is also not permitted to sleep.
1633	*
1634	* The caller must hold the RCU read lock.
1635	*/
1636	struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1637	unsigned long ino)
1638	{
1639	struct hlist_head *head = inode_hashtable + hash(sb, hashval: ino);
1640	struct inode *inode;
1641
1642	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1643	"suspicious find_inode_by_ino_rcu() usage");
1644
1645	hlist_for_each_entry_rcu(inode, head, i_hash) {
1646	if (inode->i_ino == ino &&
1647	inode->i_sb == sb &&
1648	!(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1649	return inode;
1650	}
1651	return NULL;
1652	}
1653	EXPORT_SYMBOL(find_inode_by_ino_rcu);
1654
1655	int insert_inode_locked(struct inode *inode)
1656	{
1657	struct super_block *sb = inode->i_sb;
1658	ino_t ino = inode->i_ino;
1659	struct hlist_head *head = inode_hashtable + hash(sb, hashval: ino);
1660
1661	while (1) {
1662	struct inode *old = NULL;
1663	spin_lock(lock: &inode_hash_lock);
1664	hlist_for_each_entry(old, head, i_hash) {
1665	if (old->i_ino != ino)
1666	continue;
1667	if (old->i_sb != sb)
1668	continue;
1669	spin_lock(lock: &old->i_lock);
1670	if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1671	spin_unlock(lock: &old->i_lock);
1672	continue;
1673	}
1674	break;
1675	}
1676	if (likely(!old)) {
1677	spin_lock(lock: &inode->i_lock);
1678	inode->i_state |= I_NEW | I_CREATING;
1679	hlist_add_head_rcu(n: &inode->i_hash, h: head);
1680	spin_unlock(lock: &inode->i_lock);
1681	spin_unlock(lock: &inode_hash_lock);
1682	return 0;
1683	}
1684	if (unlikely(old->i_state & I_CREATING)) {
1685	spin_unlock(lock: &old->i_lock);
1686	spin_unlock(lock: &inode_hash_lock);
1687	return -EBUSY;
1688	}
1689	__iget(inode: old);
1690	spin_unlock(lock: &old->i_lock);
1691	spin_unlock(lock: &inode_hash_lock);
1692	wait_on_inode(inode: old);
1693	if (unlikely(!inode_unhashed(old))) {
1694	iput(old);
1695	return -EBUSY;
1696	}
1697	iput(old);
1698	}
1699	}
1700	EXPORT_SYMBOL(insert_inode_locked);
1701
1702	int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1703	int (*test)(struct inode *, void *), void *data)
1704	{
1705	struct inode *old;
1706
1707	inode->i_state |= I_CREATING;
1708	old = inode_insert5(inode, hashval, test, NULL, data);
1709
1710	if (old != inode) {
1711	iput(old);
1712	return -EBUSY;
1713	}
1714	return 0;
1715	}
1716	EXPORT_SYMBOL(insert_inode_locked4);
1717
1718
1719	int generic_delete_inode(struct inode *inode)
1720	{
1721	return 1;
1722	}
1723	EXPORT_SYMBOL(generic_delete_inode);
1724
1725	/*
1726	* Called when we're dropping the last reference
1727	* to an inode.
1728	*
1729	* Call the FS "drop_inode()" function, defaulting to
1730	* the legacy UNIX filesystem behaviour. If it tells
1731	* us to evict inode, do so. Otherwise, retain inode
1732	* in cache if fs is alive, sync and evict if fs is
1733	* shutting down.
1734	*/
1735	static void iput_final(struct inode *inode)
1736	{
1737	struct super_block *sb = inode->i_sb;
1738	const struct super_operations *op = inode->i_sb->s_op;
1739	unsigned long state;
1740	int drop;
1741
1742	WARN_ON(inode->i_state & I_NEW);
1743
1744	if (op->drop_inode)
1745	drop = op->drop_inode(inode);
1746	else
1747	drop = generic_drop_inode(inode);
1748
1749	if (!drop &&
1750	!(inode->i_state & I_DONTCACHE) &&
1751	(sb->s_flags & SB_ACTIVE)) {
1752	__inode_add_lru(inode, rotate: true);
1753	spin_unlock(lock: &inode->i_lock);
1754	return;
1755	}
1756
1757	state = inode->i_state;
1758	if (!drop) {
1759	WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1760	spin_unlock(lock: &inode->i_lock);
1761
1762	write_inode_now(inode, sync: 1);
1763
1764	spin_lock(lock: &inode->i_lock);
1765	state = inode->i_state;
1766	WARN_ON(state & I_NEW);
1767	state &= ~I_WILL_FREE;
1768	}
1769
1770	WRITE_ONCE(inode->i_state, state | I_FREEING);
1771	if (!list_empty(head: &inode->i_lru))
1772	inode_lru_list_del(inode);
1773	spin_unlock(lock: &inode->i_lock);
1774
1775	evict(inode);
1776	}
1777
1778	/**
1779	* iput - put an inode
1780	* @inode: inode to put
1781	*
1782	* Puts an inode, dropping its usage count. If the inode use count hits
1783	* zero, the inode is then freed and may also be destroyed.
1784	*
1785	* Consequently, iput() can sleep.
1786	*/
1787	void iput(struct inode *inode)
1788	{
1789	if (!inode)
1790	return;
1791	BUG_ON(inode->i_state & I_CLEAR);
1792	retry:
1793	if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1794	if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1795	atomic_inc(v: &inode->i_count);
1796	spin_unlock(lock: &inode->i_lock);
1797	trace_writeback_lazytime_iput(inode);
1798	mark_inode_dirty_sync(inode);
1799	goto retry;
1800	}
1801	iput_final(inode);
1802	}
1803	}
1804	EXPORT_SYMBOL(iput);
1805
1806	#ifdef CONFIG_BLOCK
1807	/**
1808	* bmap - find a block number in a file
1809	* @inode: inode owning the block number being requested
1810	* @block: pointer containing the block to find
1811	*
1812	* Replaces the value in ``*block`` with the block number on the device holding
1813	* corresponding to the requested block number in the file.
1814	* That is, asked for block 4 of inode 1 the function will replace the
1815	* 4 in ``*block``, with disk block relative to the disk start that holds that
1816	* block of the file.
1817	*
1818	* Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1819	* hole, returns 0 and ``*block`` is also set to 0.
1820	*/
1821	int bmap(struct inode *inode, sector_t *block)
1822	{
1823	if (!inode->i_mapping->a_ops->bmap)
1824	return -EINVAL;
1825
1826	*block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1827	return 0;
1828	}
1829	EXPORT_SYMBOL(bmap);
1830	#endif
1831
1832	/*
1833	* With relative atime, only update atime if the previous atime is
1834	* earlier than or equal to either the ctime or mtime,
1835	* or if at least a day has passed since the last atime update.
1836	*/
1837	static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1838	struct timespec64 now)
1839	{
1840	struct timespec64 atime, mtime, ctime;
1841
1842	if (!(mnt->mnt_flags & MNT_RELATIME))
1843	return 1;
1844	/*
1845	* Is mtime younger than or equal to atime? If yes, update atime:
1846	*/
1847	atime = inode_get_atime(inode);
1848	mtime = inode_get_mtime(inode);
1849	if (timespec64_compare(lhs: &mtime, rhs: &atime) >= 0)
1850	return 1;
1851	/*
1852	* Is ctime younger than or equal to atime? If yes, update atime:
1853	*/
1854	ctime = inode_get_ctime(inode);
1855	if (timespec64_compare(lhs: &ctime, rhs: &atime) >= 0)
1856	return 1;
1857
1858	/*
1859	* Is the previous atime value older than a day? If yes,
1860	* update atime:
1861	*/
1862	if ((long)(now.tv_sec - atime.tv_sec) >= 24*60*60)
1863	return 1;
1864	/*
1865	* Good, we can skip the atime update:
1866	*/
1867	return 0;
1868	}
1869
1870	/**
1871	* inode_update_timestamps - update the timestamps on the inode
1872	* @inode: inode to be updated
1873	* @flags: S_* flags that needed to be updated
1874	*
1875	* The update_time function is called when an inode's timestamps need to be
1876	* updated for a read or write operation. This function handles updating the
1877	* actual timestamps. It's up to the caller to ensure that the inode is marked
1878	* dirty appropriately.
1879	*
1880	* In the case where any of S_MTIME, S_CTIME, or S_VERSION need to be updated,
1881	* attempt to update all three of them. S_ATIME updates can be handled
1882	* independently of the rest.
1883	*
1884	* Returns a set of S_* flags indicating which values changed.
1885	*/
1886	int inode_update_timestamps(struct inode *inode, int flags)
1887	{
1888	int updated = 0;
1889	struct timespec64 now;
1890
1891	if (flags & (S_MTIME|S_CTIME|S_VERSION)) {
1892	struct timespec64 ctime = inode_get_ctime(inode);
1893	struct timespec64 mtime = inode_get_mtime(inode);
1894
1895	now = inode_set_ctime_current(inode);
1896	if (!timespec64_equal(a: &now, b: &ctime))
1897	updated |= S_CTIME;
1898	if (!timespec64_equal(a: &now, b: &mtime)) {
1899	inode_set_mtime_to_ts(inode, ts: now);
1900	updated |= S_MTIME;
1901	}
1902	if (IS_I_VERSION(inode) && inode_maybe_inc_iversion(inode, force: updated))
1903	updated |= S_VERSION;
1904	} else {
1905	now = current_time(inode);
1906	}
1907
1908	if (flags & S_ATIME) {
1909	struct timespec64 atime = inode_get_atime(inode);
1910
1911	if (!timespec64_equal(a: &now, b: &atime)) {
1912	inode_set_atime_to_ts(inode, ts: now);
1913	updated |= S_ATIME;
1914	}
1915	}
1916	return updated;
1917	}
1918	EXPORT_SYMBOL(inode_update_timestamps);
1919
1920	/**
1921	* generic_update_time - update the timestamps on the inode
1922	* @inode: inode to be updated
1923	* @flags: S_* flags that needed to be updated
1924	*
1925	* The update_time function is called when an inode's timestamps need to be
1926	* updated for a read or write operation. In the case where any of S_MTIME, S_CTIME,
1927	* or S_VERSION need to be updated we attempt to update all three of them. S_ATIME
1928	* updates can be handled done independently of the rest.
1929	*
1930	* Returns a S_* mask indicating which fields were updated.
1931	*/
1932	int generic_update_time(struct inode *inode, int flags)
1933	{
1934	int updated = inode_update_timestamps(inode, flags);
1935	int dirty_flags = 0;
1936
1937	if (updated & (S_ATIME|S_MTIME|S_CTIME))
1938	dirty_flags = inode->i_sb->s_flags & SB_LAZYTIME ? I_DIRTY_TIME : I_DIRTY_SYNC;
1939	if (updated & S_VERSION)
1940	dirty_flags |= I_DIRTY_SYNC;
1941	__mark_inode_dirty(inode, dirty_flags);
1942	return updated;
1943	}
1944	EXPORT_SYMBOL(generic_update_time);
1945
1946	/*
1947	* This does the actual work of updating an inodes time or version. Must have
1948	* had called mnt_want_write() before calling this.
1949	*/
1950	int inode_update_time(struct inode *inode, int flags)
1951	{
1952	if (inode->i_op->update_time)
1953	return inode->i_op->update_time(inode, flags);
1954	generic_update_time(inode, flags);
1955	return 0;
1956	}
1957	EXPORT_SYMBOL(inode_update_time);
1958
1959	/**
1960	* atime_needs_update - update the access time
1961	* @path: the &struct path to update
1962	* @inode: inode to update
1963	*
1964	* Update the accessed time on an inode and mark it for writeback.
1965	* This function automatically handles read only file systems and media,
1966	* as well as the "noatime" flag and inode specific "noatime" markers.
1967	*/
1968	bool atime_needs_update(const struct path *path, struct inode *inode)
1969	{
1970	struct vfsmount *mnt = path->mnt;
1971	struct timespec64 now, atime;
1972
1973	if (inode->i_flags & S_NOATIME)
1974	return false;
1975
1976	/* Atime updates will likely cause i_uid and i_gid to be written
1977	* back improprely if their true value is unknown to the vfs.
1978	*/
1979	if (HAS_UNMAPPED_ID(idmap: mnt_idmap(mnt), inode))
1980	return false;
1981
1982	if (IS_NOATIME(inode))
1983	return false;
1984	if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1985	return false;
1986
1987	if (mnt->mnt_flags & MNT_NOATIME)
1988	return false;
1989	if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1990	return false;
1991
1992	now = current_time(inode);
1993
1994	if (!relatime_need_update(mnt, inode, now))
1995	return false;
1996
1997	atime = inode_get_atime(inode);
1998	if (timespec64_equal(a: &atime, b: &now))
1999	return false;
2000
2001	return true;
2002	}
2003
2004	void touch_atime(const struct path *path)
2005	{
2006	struct vfsmount *mnt = path->mnt;
2007	struct inode *inode = d_inode(dentry: path->dentry);
2008
2009	if (!atime_needs_update(path, inode))
2010	return;
2011
2012	if (!sb_start_write_trylock(sb: inode->i_sb))
2013	return;
2014
2015	if (mnt_get_write_access(mnt) != 0)
2016	goto skip_update;
2017	/*
2018	* File systems can error out when updating inodes if they need to
2019	* allocate new space to modify an inode (such is the case for
2020	* Btrfs), but since we touch atime while walking down the path we
2021	* really don't care if we failed to update the atime of the file,
2022	* so just ignore the return value.
2023	* We may also fail on filesystems that have the ability to make parts
2024	* of the fs read only, e.g. subvolumes in Btrfs.
2025	*/
2026	inode_update_time(inode, S_ATIME);
2027	mnt_put_write_access(mnt);
2028	skip_update:
2029	sb_end_write(sb: inode->i_sb);
2030	}
2031	EXPORT_SYMBOL(touch_atime);
2032
2033	/*
2034	* Return mask of changes for notify_change() that need to be done as a
2035	* response to write or truncate. Return 0 if nothing has to be changed.
2036	* Negative value on error (change should be denied).
2037	*/
2038	int dentry_needs_remove_privs(struct mnt_idmap *idmap,
2039	struct dentry *dentry)
2040	{
2041	struct inode *inode = d_inode(dentry);
2042	int mask = 0;
2043	int ret;
2044
2045	if (IS_NOSEC(inode))
2046	return 0;
2047
2048	mask = setattr_should_drop_suidgid(idmap, inode);
2049	ret = security_inode_need_killpriv(dentry);
2050	if (ret < 0)
2051	return ret;
2052	if (ret)
2053	mask |= ATTR_KILL_PRIV;
2054	return mask;
2055	}
2056
2057	static int __remove_privs(struct mnt_idmap *idmap,
2058	struct dentry *dentry, int kill)
2059	{
2060	struct iattr newattrs;
2061
2062	newattrs.ia_valid = ATTR_FORCE | kill;
2063	/*
2064	* Note we call this on write, so notify_change will not
2065	* encounter any conflicting delegations:
2066	*/
2067	return notify_change(idmap, dentry, &newattrs, NULL);
2068	}
2069
2070	static int __file_remove_privs(struct file *file, unsigned int flags)
2071	{
2072	struct dentry *dentry = file_dentry(file);
2073	struct inode *inode = file_inode(f: file);
2074	int error = 0;
2075	int kill;
2076
2077	if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
2078	return 0;
2079
2080	kill = dentry_needs_remove_privs(idmap: file_mnt_idmap(file), dentry);
2081	if (kill < 0)
2082	return kill;
2083
2084	if (kill) {
2085	if (flags & IOCB_NOWAIT)
2086	return -EAGAIN;
2087
2088	error = __remove_privs(idmap: file_mnt_idmap(file), dentry, kill);
2089	}
2090
2091	if (!error)
2092	inode_has_no_xattr(inode);
2093	return error;
2094	}
2095
2096	/**
2097	* file_remove_privs - remove special file privileges (suid, capabilities)
2098	* @file: file to remove privileges from
2099	*
2100	* When file is modified by a write or truncation ensure that special
2101	* file privileges are removed.
2102	*
2103	* Return: 0 on success, negative errno on failure.
2104	*/
2105	int file_remove_privs(struct file *file)
2106	{
2107	return __file_remove_privs(file, flags: 0);
2108	}
2109	EXPORT_SYMBOL(file_remove_privs);
2110
2111	static int inode_needs_update_time(struct inode *inode)
2112	{
2113	int sync_it = 0;
2114	struct timespec64 now = current_time(inode);
2115	struct timespec64 ts;
2116
2117	/* First try to exhaust all avenues to not sync */
2118	if (IS_NOCMTIME(inode))
2119	return 0;
2120
2121	ts = inode_get_mtime(inode);
2122	if (!timespec64_equal(a: &ts, b: &now))
2123	sync_it = S_MTIME;
2124
2125	ts = inode_get_ctime(inode);
2126	if (!timespec64_equal(a: &ts, b: &now))
2127	sync_it |= S_CTIME;
2128
2129	if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2130	sync_it |= S_VERSION;
2131
2132	return sync_it;
2133	}
2134
2135	static int __file_update_time(struct file *file, int sync_mode)
2136	{
2137	int ret = 0;
2138	struct inode *inode = file_inode(f: file);
2139
2140	/* try to update time settings */
2141	if (!mnt_get_write_access_file(file)) {
2142	ret = inode_update_time(inode, sync_mode);
2143	mnt_put_write_access_file(file);
2144	}
2145
2146	return ret;
2147	}
2148
2149	/**
2150	* file_update_time - update mtime and ctime time
2151	* @file: file accessed
2152	*
2153	* Update the mtime and ctime members of an inode and mark the inode for
2154	* writeback. Note that this function is meant exclusively for usage in
2155	* the file write path of filesystems, and filesystems may choose to
2156	* explicitly ignore updates via this function with the _NOCMTIME inode
2157	* flag, e.g. for network filesystem where these imestamps are handled
2158	* by the server. This can return an error for file systems who need to
2159	* allocate space in order to update an inode.
2160	*
2161	* Return: 0 on success, negative errno on failure.
2162	*/
2163	int file_update_time(struct file *file)
2164	{
2165	int ret;
2166	struct inode *inode = file_inode(f: file);
2167
2168	ret = inode_needs_update_time(inode);
2169	if (ret <= 0)
2170	return ret;
2171
2172	return __file_update_time(file, sync_mode: ret);
2173	}
2174	EXPORT_SYMBOL(file_update_time);
2175
2176	/**
2177	* file_modified_flags - handle mandated vfs changes when modifying a file
2178	* @file: file that was modified
2179	* @flags: kiocb flags
2180	*
2181	* When file has been modified ensure that special
2182	* file privileges are removed and time settings are updated.
2183	*
2184	* If IOCB_NOWAIT is set, special file privileges will not be removed and
2185	* time settings will not be updated. It will return -EAGAIN.
2186	*
2187	* Context: Caller must hold the file's inode lock.
2188	*
2189	* Return: 0 on success, negative errno on failure.
2190	*/
2191	static int file_modified_flags(struct file *file, int flags)
2192	{
2193	int ret;
2194	struct inode *inode = file_inode(f: file);
2195
2196	/*
2197	* Clear the security bits if the process is not being run by root.
2198	* This keeps people from modifying setuid and setgid binaries.
2199	*/
2200	ret = __file_remove_privs(file, flags);
2201	if (ret)
2202	return ret;
2203
2204	if (unlikely(file->f_mode & FMODE_NOCMTIME))
2205	return 0;
2206
2207	ret = inode_needs_update_time(inode);
2208	if (ret <= 0)
2209	return ret;
2210	if (flags & IOCB_NOWAIT)
2211	return -EAGAIN;
2212
2213	return __file_update_time(file, sync_mode: ret);
2214	}
2215
2216	/**
2217	* file_modified - handle mandated vfs changes when modifying a file
2218	* @file: file that was modified
2219	*
2220	* When file has been modified ensure that special
2221	* file privileges are removed and time settings are updated.
2222	*
2223	* Context: Caller must hold the file's inode lock.
2224	*
2225	* Return: 0 on success, negative errno on failure.
2226	*/
2227	int file_modified(struct file *file)
2228	{
2229	return file_modified_flags(file, flags: 0);
2230	}
2231	EXPORT_SYMBOL(file_modified);
2232
2233	/**
2234	* kiocb_modified - handle mandated vfs changes when modifying a file
2235	* @iocb: iocb that was modified
2236	*
2237	* When file has been modified ensure that special
2238	* file privileges are removed and time settings are updated.
2239	*
2240	* Context: Caller must hold the file's inode lock.
2241	*
2242	* Return: 0 on success, negative errno on failure.
2243	*/
2244	int kiocb_modified(struct kiocb *iocb)
2245	{
2246	return file_modified_flags(file: iocb->ki_filp, flags: iocb->ki_flags);
2247	}
2248	EXPORT_SYMBOL_GPL(kiocb_modified);
2249
2250	int inode_needs_sync(struct inode *inode)
2251	{
2252	if (IS_SYNC(inode))
2253	return 1;
2254	if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2255	return 1;
2256	return 0;
2257	}
2258	EXPORT_SYMBOL(inode_needs_sync);
2259
2260	/*
2261	* If we try to find an inode in the inode hash while it is being
2262	* deleted, we have to wait until the filesystem completes its
2263	* deletion before reporting that it isn't found. This function waits
2264	* until the deletion _might_ have completed. Callers are responsible
2265	* to recheck inode state.
2266	*
2267	* It doesn't matter if I_NEW is not set initially, a call to
2268	* wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2269	* will DTRT.
2270	*/
2271	static void __wait_on_freeing_inode(struct inode *inode)
2272	{
2273	wait_queue_head_t *wq;
2274	DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2275	wq = bit_waitqueue(word: &inode->i_state, __I_NEW);
2276	prepare_to_wait(wq_head: wq, wq_entry: &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2277	spin_unlock(lock: &inode->i_lock);
2278	spin_unlock(lock: &inode_hash_lock);
2279	schedule();
2280	finish_wait(wq_head: wq, wq_entry: &wait.wq_entry);
2281	spin_lock(lock: &inode_hash_lock);
2282	}
2283
2284	static __initdata unsigned long ihash_entries;
2285	static int __init set_ihash_entries(char *str)
2286	{
2287	if (!str)
2288	return 0;
2289	ihash_entries = simple_strtoul(str, &str, 0);
2290	return 1;
2291	}
2292	__setup("ihash_entries=", set_ihash_entries);
2293
2294	/*
2295	* Initialize the waitqueues and inode hash table.
2296	*/
2297	void __init inode_init_early(void)
2298	{
2299	/* If hashes are distributed across NUMA nodes, defer
2300	* hash allocation until vmalloc space is available.
2301	*/
2302	if (hashdist)
2303	return;
2304
2305	inode_hashtable =
2306	alloc_large_system_hash(tablename: "Inode-cache",
2307	bucketsize: sizeof(struct hlist_head),
2308	numentries: ihash_entries,
2309	scale: 14,
2310	HASH_EARLY | HASH_ZERO,
2311	hash_shift: &i_hash_shift,
2312	hash_mask: &i_hash_mask,
2313	low_limit: 0,
2314	high_limit: 0);
2315	}
2316
2317	void __init inode_init(void)
2318	{
2319	/* inode slab cache */
2320	inode_cachep = kmem_cache_create(name: "inode_cache",
2321	size: sizeof(struct inode),
2322	align: 0,
2323	flags: (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2324	SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2325	ctor: init_once);
2326
2327	/* Hash may have been set up in inode_init_early */
2328	if (!hashdist)
2329	return;
2330
2331	inode_hashtable =
2332	alloc_large_system_hash(tablename: "Inode-cache",
2333	bucketsize: sizeof(struct hlist_head),
2334	numentries: ihash_entries,
2335	scale: 14,
2336	HASH_ZERO,
2337	hash_shift: &i_hash_shift,
2338	hash_mask: &i_hash_mask,
2339	low_limit: 0,
2340	high_limit: 0);
2341	}
2342
2343	void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2344	{
2345	inode->i_mode = mode;
2346	if (S_ISCHR(mode)) {
2347	inode->i_fop = &def_chr_fops;
2348	inode->i_rdev = rdev;
2349	} else if (S_ISBLK(mode)) {
2350	if (IS_ENABLED(CONFIG_BLOCK))
2351	inode->i_fop = &def_blk_fops;
2352	inode->i_rdev = rdev;
2353	} else if (S_ISFIFO(mode))
2354	inode->i_fop = &pipefifo_fops;
2355	else if (S_ISSOCK(mode))
2356	; /* leave it no_open_fops */
2357	else
2358	printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2359	" inode %s:%lu\n", mode, inode->i_sb->s_id,
2360	inode->i_ino);
2361	}
2362	EXPORT_SYMBOL(init_special_inode);
2363
2364	/**
2365	* inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2366	* @idmap: idmap of the mount the inode was created from
2367	* @inode: New inode
2368	* @dir: Directory inode
2369	* @mode: mode of the new inode
2370	*
2371	* If the inode has been created through an idmapped mount the idmap of
2372	* the vfsmount must be passed through @idmap. This function will then take
2373	* care to map the inode according to @idmap before checking permissions
2374	* and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2375	* checking is to be performed on the raw inode simply pass @nop_mnt_idmap.
2376	*/
2377	void inode_init_owner(struct mnt_idmap *idmap, struct inode *inode,
2378	const struct inode *dir, umode_t mode)
2379	{
2380	inode_fsuid_set(inode, idmap);
2381	if (dir && dir->i_mode & S_ISGID) {
2382	inode->i_gid = dir->i_gid;
2383
2384	/* Directories are special, and always inherit S_ISGID */
2385	if (S_ISDIR(mode))
2386	mode |= S_ISGID;
2387	} else
2388	inode_fsgid_set(inode, idmap);
2389	inode->i_mode = mode;
2390	}
2391	EXPORT_SYMBOL(inode_init_owner);
2392
2393	/**
2394	* inode_owner_or_capable - check current task permissions to inode
2395	* @idmap: idmap of the mount the inode was found from
2396	* @inode: inode being checked
2397	*
2398	* Return true if current either has CAP_FOWNER in a namespace with the
2399	* inode owner uid mapped, or owns the file.
2400	*
2401	* If the inode has been found through an idmapped mount the idmap of
2402	* the vfsmount must be passed through @idmap. This function will then take
2403	* care to map the inode according to @idmap before checking permissions.
2404	* On non-idmapped mounts or if permission checking is to be performed on the
2405	* raw inode simply passs @nop_mnt_idmap.
2406	*/
2407	bool inode_owner_or_capable(struct mnt_idmap *idmap,
2408	const struct inode *inode)
2409	{
2410	vfsuid_t vfsuid;
2411	struct user_namespace *ns;
2412
2413	vfsuid = i_uid_into_vfsuid(idmap, inode);
2414	if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
2415	return true;
2416
2417	ns = current_user_ns();
2418	if (vfsuid_has_mapping(userns: ns, vfsuid) && ns_capable(ns, CAP_FOWNER))
2419	return true;
2420	return false;
2421	}
2422	EXPORT_SYMBOL(inode_owner_or_capable);
2423
2424	/*
2425	* Direct i/o helper functions
2426	*/
2427	static void __inode_dio_wait(struct inode *inode)
2428	{
2429	wait_queue_head_t *wq = bit_waitqueue(word: &inode->i_state, __I_DIO_WAKEUP);
2430	DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2431
2432	do {
2433	prepare_to_wait(wq_head: wq, wq_entry: &q.wq_entry, TASK_UNINTERRUPTIBLE);
2434	if (atomic_read(v: &inode->i_dio_count))
2435	schedule();
2436	} while (atomic_read(v: &inode->i_dio_count));
2437	finish_wait(wq_head: wq, wq_entry: &q.wq_entry);
2438	}
2439
2440	/**
2441	* inode_dio_wait - wait for outstanding DIO requests to finish
2442	* @inode: inode to wait for
2443	*
2444	* Waits for all pending direct I/O requests to finish so that we can
2445	* proceed with a truncate or equivalent operation.
2446	*
2447	* Must be called under a lock that serializes taking new references
2448	* to i_dio_count, usually by inode->i_mutex.
2449	*/
2450	void inode_dio_wait(struct inode *inode)
2451	{
2452	if (atomic_read(v: &inode->i_dio_count))
2453	__inode_dio_wait(inode);
2454	}
2455	EXPORT_SYMBOL(inode_dio_wait);
2456
2457	/*
2458	* inode_set_flags - atomically set some inode flags
2459	*
2460	* Note: the caller should be holding i_mutex, or else be sure that
2461	* they have exclusive access to the inode structure (i.e., while the
2462	* inode is being instantiated). The reason for the cmpxchg() loop
2463	* --- which wouldn't be necessary if all code paths which modify
2464	* i_flags actually followed this rule, is that there is at least one
2465	* code path which doesn't today so we use cmpxchg() out of an abundance
2466	* of caution.
2467	*
2468	* In the long run, i_mutex is overkill, and we should probably look
2469	* at using the i_lock spinlock to protect i_flags, and then make sure
2470	* it is so documented in include/linux/fs.h and that all code follows
2471	* the locking convention!!
2472	*/
2473	void inode_set_flags(struct inode *inode, unsigned int flags,
2474	unsigned int mask)
2475	{
2476	WARN_ON_ONCE(flags & ~mask);
2477	set_mask_bits(&inode->i_flags, mask, flags);
2478	}
2479	EXPORT_SYMBOL(inode_set_flags);
2480
2481	void inode_nohighmem(struct inode *inode)
2482	{
2483	mapping_set_gfp_mask(m: inode->i_mapping, GFP_USER);
2484	}
2485	EXPORT_SYMBOL(inode_nohighmem);
2486
2487	/**
2488	* timestamp_truncate - Truncate timespec to a granularity
2489	* @t: Timespec
2490	* @inode: inode being updated
2491	*
2492	* Truncate a timespec to the granularity supported by the fs
2493	* containing the inode. Always rounds down. gran must
2494	* not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2495	*/
2496	struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2497	{
2498	struct super_block *sb = inode->i_sb;
2499	unsigned int gran = sb->s_time_gran;
2500
2501	t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2502	if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2503	t.tv_nsec = 0;
2504
2505	/* Avoid division in the common cases 1 ns and 1 s. */
2506	if (gran == 1)
2507	; /* nothing */
2508	else if (gran == NSEC_PER_SEC)
2509	t.tv_nsec = 0;
2510	else if (gran > 1 && gran < NSEC_PER_SEC)
2511	t.tv_nsec -= t.tv_nsec % gran;
2512	else
2513	WARN(1, "invalid file time granularity: %u", gran);
2514	return t;
2515	}
2516	EXPORT_SYMBOL(timestamp_truncate);
2517
2518	/**
2519	* current_time - Return FS time
2520	* @inode: inode.
2521	*
2522	* Return the current time truncated to the time granularity supported by
2523	* the fs.
2524	*
2525	* Note that inode and inode->sb cannot be NULL.
2526	* Otherwise, the function warns and returns time without truncation.
2527	*/
2528	struct timespec64 current_time(struct inode *inode)
2529	{
2530	struct timespec64 now;
2531
2532	ktime_get_coarse_real_ts64(ts: &now);
2533	return timestamp_truncate(now, inode);
2534	}
2535	EXPORT_SYMBOL(current_time);
2536
2537	/**
2538	* inode_set_ctime_current - set the ctime to current_time
2539	* @inode: inode
2540	*
2541	* Set the inode->i_ctime to the current value for the inode. Returns
2542	* the current value that was assigned to i_ctime.
2543	*/
2544	struct timespec64 inode_set_ctime_current(struct inode *inode)
2545	{
2546	struct timespec64 now = current_time(inode);
2547
2548	inode_set_ctime(inode, sec: now.tv_sec, nsec: now.tv_nsec);
2549	return now;
2550	}
2551	EXPORT_SYMBOL(inode_set_ctime_current);
2552
2553	/**
2554	* in_group_or_capable - check whether caller is CAP_FSETID privileged
2555	* @idmap: idmap of the mount @inode was found from
2556	* @inode: inode to check
2557	* @vfsgid: the new/current vfsgid of @inode
2558	*
2559	* Check wether @vfsgid is in the caller's group list or if the caller is
2560	* privileged with CAP_FSETID over @inode. This can be used to determine
2561	* whether the setgid bit can be kept or must be dropped.
2562	*
2563	* Return: true if the caller is sufficiently privileged, false if not.
2564	*/
2565	bool in_group_or_capable(struct mnt_idmap *idmap,
2566	const struct inode *inode, vfsgid_t vfsgid)
2567	{
2568	if (vfsgid_in_group_p(vfsgid))
2569	return true;
2570	if (capable_wrt_inode_uidgid(idmap, inode, CAP_FSETID))
2571	return true;
2572	return false;
2573	}
2574
2575	/**
2576	* mode_strip_sgid - handle the sgid bit for non-directories
2577	* @idmap: idmap of the mount the inode was created from
2578	* @dir: parent directory inode
2579	* @mode: mode of the file to be created in @dir
2580	*
2581	* If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2582	* raised and @dir has the S_ISGID bit raised ensure that the caller is
2583	* either in the group of the parent directory or they have CAP_FSETID
2584	* in their user namespace and are privileged over the parent directory.
2585	* In all other cases, strip the S_ISGID bit from @mode.
2586	*
2587	* Return: the new mode to use for the file
2588	*/
2589	umode_t mode_strip_sgid(struct mnt_idmap *idmap,
2590	const struct inode *dir, umode_t mode)
2591	{
2592	if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2593	return mode;
2594	if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2595	return mode;
2596	if (in_group_or_capable(idmap, inode: dir, vfsgid: i_gid_into_vfsgid(idmap, inode: dir)))
2597	return mode;
2598	return mode & ~S_ISGID;
2599	}
2600	EXPORT_SYMBOL(mode_strip_sgid);
2601