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