xref: /linux/fs/libfs.c (revision 4b132aacb0768ac1e652cf517097ea6f237214b9)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *	fs/libfs.c
4  *	Library for filesystems writers.
5  */
6 
7 #include <linux/blkdev.h>
8 #include <linux/export.h>
9 #include <linux/pagemap.h>
10 #include <linux/slab.h>
11 #include <linux/cred.h>
12 #include <linux/mount.h>
13 #include <linux/vfs.h>
14 #include <linux/quotaops.h>
15 #include <linux/mutex.h>
16 #include <linux/namei.h>
17 #include <linux/exportfs.h>
18 #include <linux/iversion.h>
19 #include <linux/writeback.h>
20 #include <linux/buffer_head.h> /* sync_mapping_buffers */
21 #include <linux/fs_context.h>
22 #include <linux/pseudo_fs.h>
23 #include <linux/fsnotify.h>
24 #include <linux/unicode.h>
25 #include <linux/fscrypt.h>
26 #include <linux/pidfs.h>
27 
28 #include <linux/uaccess.h>
29 
30 #include "internal.h"
31 
32 int simple_getattr(struct mnt_idmap *idmap, const struct path *path,
33 		   struct kstat *stat, u32 request_mask,
34 		   unsigned int query_flags)
35 {
36 	struct inode *inode = d_inode(path->dentry);
37 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
38 	stat->blocks = inode->i_mapping->nrpages << (PAGE_SHIFT - 9);
39 	return 0;
40 }
41 EXPORT_SYMBOL(simple_getattr);
42 
43 int simple_statfs(struct dentry *dentry, struct kstatfs *buf)
44 {
45 	u64 id = huge_encode_dev(dentry->d_sb->s_dev);
46 
47 	buf->f_fsid = u64_to_fsid(id);
48 	buf->f_type = dentry->d_sb->s_magic;
49 	buf->f_bsize = PAGE_SIZE;
50 	buf->f_namelen = NAME_MAX;
51 	return 0;
52 }
53 EXPORT_SYMBOL(simple_statfs);
54 
55 /*
56  * Retaining negative dentries for an in-memory filesystem just wastes
57  * memory and lookup time: arrange for them to be deleted immediately.
58  */
59 int always_delete_dentry(const struct dentry *dentry)
60 {
61 	return 1;
62 }
63 EXPORT_SYMBOL(always_delete_dentry);
64 
65 const struct dentry_operations simple_dentry_operations = {
66 	.d_delete = always_delete_dentry,
67 };
68 EXPORT_SYMBOL(simple_dentry_operations);
69 
70 /*
71  * Lookup the data. This is trivial - if the dentry didn't already
72  * exist, we know it is negative.  Set d_op to delete negative dentries.
73  */
74 struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
75 {
76 	if (dentry->d_name.len > NAME_MAX)
77 		return ERR_PTR(-ENAMETOOLONG);
78 	if (!dentry->d_sb->s_d_op)
79 		d_set_d_op(dentry, &simple_dentry_operations);
80 	d_add(dentry, NULL);
81 	return NULL;
82 }
83 EXPORT_SYMBOL(simple_lookup);
84 
85 int dcache_dir_open(struct inode *inode, struct file *file)
86 {
87 	file->private_data = d_alloc_cursor(file->f_path.dentry);
88 
89 	return file->private_data ? 0 : -ENOMEM;
90 }
91 EXPORT_SYMBOL(dcache_dir_open);
92 
93 int dcache_dir_close(struct inode *inode, struct file *file)
94 {
95 	dput(file->private_data);
96 	return 0;
97 }
98 EXPORT_SYMBOL(dcache_dir_close);
99 
100 /* parent is locked at least shared */
101 /*
102  * Returns an element of siblings' list.
103  * We are looking for <count>th positive after <p>; if
104  * found, dentry is grabbed and returned to caller.
105  * If no such element exists, NULL is returned.
106  */
107 static struct dentry *scan_positives(struct dentry *cursor,
108 					struct hlist_node **p,
109 					loff_t count,
110 					struct dentry *last)
111 {
112 	struct dentry *dentry = cursor->d_parent, *found = NULL;
113 
114 	spin_lock(&dentry->d_lock);
115 	while (*p) {
116 		struct dentry *d = hlist_entry(*p, struct dentry, d_sib);
117 		p = &d->d_sib.next;
118 		// we must at least skip cursors, to avoid livelocks
119 		if (d->d_flags & DCACHE_DENTRY_CURSOR)
120 			continue;
121 		if (simple_positive(d) && !--count) {
122 			spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
123 			if (simple_positive(d))
124 				found = dget_dlock(d);
125 			spin_unlock(&d->d_lock);
126 			if (likely(found))
127 				break;
128 			count = 1;
129 		}
130 		if (need_resched()) {
131 			if (!hlist_unhashed(&cursor->d_sib))
132 				__hlist_del(&cursor->d_sib);
133 			hlist_add_behind(&cursor->d_sib, &d->d_sib);
134 			p = &cursor->d_sib.next;
135 			spin_unlock(&dentry->d_lock);
136 			cond_resched();
137 			spin_lock(&dentry->d_lock);
138 		}
139 	}
140 	spin_unlock(&dentry->d_lock);
141 	dput(last);
142 	return found;
143 }
144 
145 loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence)
146 {
147 	struct dentry *dentry = file->f_path.dentry;
148 	switch (whence) {
149 		case 1:
150 			offset += file->f_pos;
151 			fallthrough;
152 		case 0:
153 			if (offset >= 0)
154 				break;
155 			fallthrough;
156 		default:
157 			return -EINVAL;
158 	}
159 	if (offset != file->f_pos) {
160 		struct dentry *cursor = file->private_data;
161 		struct dentry *to = NULL;
162 
163 		inode_lock_shared(dentry->d_inode);
164 
165 		if (offset > 2)
166 			to = scan_positives(cursor, &dentry->d_children.first,
167 					    offset - 2, NULL);
168 		spin_lock(&dentry->d_lock);
169 		hlist_del_init(&cursor->d_sib);
170 		if (to)
171 			hlist_add_behind(&cursor->d_sib, &to->d_sib);
172 		spin_unlock(&dentry->d_lock);
173 		dput(to);
174 
175 		file->f_pos = offset;
176 
177 		inode_unlock_shared(dentry->d_inode);
178 	}
179 	return offset;
180 }
181 EXPORT_SYMBOL(dcache_dir_lseek);
182 
183 /*
184  * Directory is locked and all positive dentries in it are safe, since
185  * for ramfs-type trees they can't go away without unlink() or rmdir(),
186  * both impossible due to the lock on directory.
187  */
188 
189 int dcache_readdir(struct file *file, struct dir_context *ctx)
190 {
191 	struct dentry *dentry = file->f_path.dentry;
192 	struct dentry *cursor = file->private_data;
193 	struct dentry *next = NULL;
194 	struct hlist_node **p;
195 
196 	if (!dir_emit_dots(file, ctx))
197 		return 0;
198 
199 	if (ctx->pos == 2)
200 		p = &dentry->d_children.first;
201 	else
202 		p = &cursor->d_sib.next;
203 
204 	while ((next = scan_positives(cursor, p, 1, next)) != NULL) {
205 		if (!dir_emit(ctx, next->d_name.name, next->d_name.len,
206 			      d_inode(next)->i_ino,
207 			      fs_umode_to_dtype(d_inode(next)->i_mode)))
208 			break;
209 		ctx->pos++;
210 		p = &next->d_sib.next;
211 	}
212 	spin_lock(&dentry->d_lock);
213 	hlist_del_init(&cursor->d_sib);
214 	if (next)
215 		hlist_add_before(&cursor->d_sib, &next->d_sib);
216 	spin_unlock(&dentry->d_lock);
217 	dput(next);
218 
219 	return 0;
220 }
221 EXPORT_SYMBOL(dcache_readdir);
222 
223 ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
224 {
225 	return -EISDIR;
226 }
227 EXPORT_SYMBOL(generic_read_dir);
228 
229 const struct file_operations simple_dir_operations = {
230 	.open		= dcache_dir_open,
231 	.release	= dcache_dir_close,
232 	.llseek		= dcache_dir_lseek,
233 	.read		= generic_read_dir,
234 	.iterate_shared	= dcache_readdir,
235 	.fsync		= noop_fsync,
236 };
237 EXPORT_SYMBOL(simple_dir_operations);
238 
239 const struct inode_operations simple_dir_inode_operations = {
240 	.lookup		= simple_lookup,
241 };
242 EXPORT_SYMBOL(simple_dir_inode_operations);
243 
244 /* 0 is '.', 1 is '..', so always start with offset 2 or more */
245 enum {
246 	DIR_OFFSET_MIN	= 2,
247 };
248 
249 static void offset_set(struct dentry *dentry, long offset)
250 {
251 	dentry->d_fsdata = (void *)offset;
252 }
253 
254 static long dentry2offset(struct dentry *dentry)
255 {
256 	return (long)dentry->d_fsdata;
257 }
258 
259 static struct lock_class_key simple_offset_lock_class;
260 
261 /**
262  * simple_offset_init - initialize an offset_ctx
263  * @octx: directory offset map to be initialized
264  *
265  */
266 void simple_offset_init(struct offset_ctx *octx)
267 {
268 	mt_init_flags(&octx->mt, MT_FLAGS_ALLOC_RANGE);
269 	lockdep_set_class(&octx->mt.ma_lock, &simple_offset_lock_class);
270 	octx->next_offset = DIR_OFFSET_MIN;
271 }
272 
273 /**
274  * simple_offset_add - Add an entry to a directory's offset map
275  * @octx: directory offset ctx to be updated
276  * @dentry: new dentry being added
277  *
278  * Returns zero on success. @octx and the dentry's offset are updated.
279  * Otherwise, a negative errno value is returned.
280  */
281 int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry)
282 {
283 	unsigned long offset;
284 	int ret;
285 
286 	if (dentry2offset(dentry) != 0)
287 		return -EBUSY;
288 
289 	ret = mtree_alloc_cyclic(&octx->mt, &offset, dentry, DIR_OFFSET_MIN,
290 				 LONG_MAX, &octx->next_offset, GFP_KERNEL);
291 	if (ret < 0)
292 		return ret;
293 
294 	offset_set(dentry, offset);
295 	return 0;
296 }
297 
298 static int simple_offset_replace(struct offset_ctx *octx, struct dentry *dentry,
299 				 long offset)
300 {
301 	int ret;
302 
303 	ret = mtree_store(&octx->mt, offset, dentry, GFP_KERNEL);
304 	if (ret)
305 		return ret;
306 	offset_set(dentry, offset);
307 	return 0;
308 }
309 
310 /**
311  * simple_offset_remove - Remove an entry to a directory's offset map
312  * @octx: directory offset ctx to be updated
313  * @dentry: dentry being removed
314  *
315  */
316 void simple_offset_remove(struct offset_ctx *octx, struct dentry *dentry)
317 {
318 	long offset;
319 
320 	offset = dentry2offset(dentry);
321 	if (offset == 0)
322 		return;
323 
324 	mtree_erase(&octx->mt, offset);
325 	offset_set(dentry, 0);
326 }
327 
328 /**
329  * simple_offset_empty - Check if a dentry can be unlinked
330  * @dentry: dentry to be tested
331  *
332  * Returns 0 if @dentry is a non-empty directory; otherwise returns 1.
333  */
334 int simple_offset_empty(struct dentry *dentry)
335 {
336 	struct inode *inode = d_inode(dentry);
337 	struct offset_ctx *octx;
338 	struct dentry *child;
339 	unsigned long index;
340 	int ret = 1;
341 
342 	if (!inode || !S_ISDIR(inode->i_mode))
343 		return ret;
344 
345 	index = DIR_OFFSET_MIN;
346 	octx = inode->i_op->get_offset_ctx(inode);
347 	mt_for_each(&octx->mt, child, index, LONG_MAX) {
348 		spin_lock(&child->d_lock);
349 		if (simple_positive(child)) {
350 			spin_unlock(&child->d_lock);
351 			ret = 0;
352 			break;
353 		}
354 		spin_unlock(&child->d_lock);
355 	}
356 
357 	return ret;
358 }
359 
360 /**
361  * simple_offset_rename - handle directory offsets for rename
362  * @old_dir: parent directory of source entry
363  * @old_dentry: dentry of source entry
364  * @new_dir: parent_directory of destination entry
365  * @new_dentry: dentry of destination
366  *
367  * Caller provides appropriate serialization.
368  *
369  * User space expects the directory offset value of the replaced
370  * (new) directory entry to be unchanged after a rename.
371  *
372  * Returns zero on success, a negative errno value on failure.
373  */
374 int simple_offset_rename(struct inode *old_dir, struct dentry *old_dentry,
375 			 struct inode *new_dir, struct dentry *new_dentry)
376 {
377 	struct offset_ctx *old_ctx = old_dir->i_op->get_offset_ctx(old_dir);
378 	struct offset_ctx *new_ctx = new_dir->i_op->get_offset_ctx(new_dir);
379 	long new_offset = dentry2offset(new_dentry);
380 
381 	simple_offset_remove(old_ctx, old_dentry);
382 
383 	if (new_offset) {
384 		offset_set(new_dentry, 0);
385 		return simple_offset_replace(new_ctx, old_dentry, new_offset);
386 	}
387 	return simple_offset_add(new_ctx, old_dentry);
388 }
389 
390 /**
391  * simple_offset_rename_exchange - exchange rename with directory offsets
392  * @old_dir: parent of dentry being moved
393  * @old_dentry: dentry being moved
394  * @new_dir: destination parent
395  * @new_dentry: destination dentry
396  *
397  * This API preserves the directory offset values. Caller provides
398  * appropriate serialization.
399  *
400  * Returns zero on success. Otherwise a negative errno is returned and the
401  * rename is rolled back.
402  */
403 int simple_offset_rename_exchange(struct inode *old_dir,
404 				  struct dentry *old_dentry,
405 				  struct inode *new_dir,
406 				  struct dentry *new_dentry)
407 {
408 	struct offset_ctx *old_ctx = old_dir->i_op->get_offset_ctx(old_dir);
409 	struct offset_ctx *new_ctx = new_dir->i_op->get_offset_ctx(new_dir);
410 	long old_index = dentry2offset(old_dentry);
411 	long new_index = dentry2offset(new_dentry);
412 	int ret;
413 
414 	simple_offset_remove(old_ctx, old_dentry);
415 	simple_offset_remove(new_ctx, new_dentry);
416 
417 	ret = simple_offset_replace(new_ctx, old_dentry, new_index);
418 	if (ret)
419 		goto out_restore;
420 
421 	ret = simple_offset_replace(old_ctx, new_dentry, old_index);
422 	if (ret) {
423 		simple_offset_remove(new_ctx, old_dentry);
424 		goto out_restore;
425 	}
426 
427 	ret = simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
428 	if (ret) {
429 		simple_offset_remove(new_ctx, old_dentry);
430 		simple_offset_remove(old_ctx, new_dentry);
431 		goto out_restore;
432 	}
433 	return 0;
434 
435 out_restore:
436 	(void)simple_offset_replace(old_ctx, old_dentry, old_index);
437 	(void)simple_offset_replace(new_ctx, new_dentry, new_index);
438 	return ret;
439 }
440 
441 /**
442  * simple_offset_destroy - Release offset map
443  * @octx: directory offset ctx that is about to be destroyed
444  *
445  * During fs teardown (eg. umount), a directory's offset map might still
446  * contain entries. xa_destroy() cleans out anything that remains.
447  */
448 void simple_offset_destroy(struct offset_ctx *octx)
449 {
450 	mtree_destroy(&octx->mt);
451 }
452 
453 static int offset_dir_open(struct inode *inode, struct file *file)
454 {
455 	struct offset_ctx *ctx = inode->i_op->get_offset_ctx(inode);
456 
457 	file->private_data = (void *)ctx->next_offset;
458 	return 0;
459 }
460 
461 /**
462  * offset_dir_llseek - Advance the read position of a directory descriptor
463  * @file: an open directory whose position is to be updated
464  * @offset: a byte offset
465  * @whence: enumerator describing the starting position for this update
466  *
467  * SEEK_END, SEEK_DATA, and SEEK_HOLE are not supported for directories.
468  *
469  * Returns the updated read position if successful; otherwise a
470  * negative errno is returned and the read position remains unchanged.
471  */
472 static loff_t offset_dir_llseek(struct file *file, loff_t offset, int whence)
473 {
474 	struct inode *inode = file->f_inode;
475 	struct offset_ctx *ctx = inode->i_op->get_offset_ctx(inode);
476 
477 	switch (whence) {
478 	case SEEK_CUR:
479 		offset += file->f_pos;
480 		fallthrough;
481 	case SEEK_SET:
482 		if (offset >= 0)
483 			break;
484 		fallthrough;
485 	default:
486 		return -EINVAL;
487 	}
488 
489 	/* In this case, ->private_data is protected by f_pos_lock */
490 	if (!offset)
491 		file->private_data = (void *)ctx->next_offset;
492 	return vfs_setpos(file, offset, LONG_MAX);
493 }
494 
495 static struct dentry *offset_find_next(struct offset_ctx *octx, loff_t offset)
496 {
497 	MA_STATE(mas, &octx->mt, offset, offset);
498 	struct dentry *child, *found = NULL;
499 
500 	rcu_read_lock();
501 	child = mas_find(&mas, LONG_MAX);
502 	if (!child)
503 		goto out;
504 	spin_lock(&child->d_lock);
505 	if (simple_positive(child))
506 		found = dget_dlock(child);
507 	spin_unlock(&child->d_lock);
508 out:
509 	rcu_read_unlock();
510 	return found;
511 }
512 
513 static bool offset_dir_emit(struct dir_context *ctx, struct dentry *dentry)
514 {
515 	struct inode *inode = d_inode(dentry);
516 	long offset = dentry2offset(dentry);
517 
518 	return ctx->actor(ctx, dentry->d_name.name, dentry->d_name.len, offset,
519 			  inode->i_ino, fs_umode_to_dtype(inode->i_mode));
520 }
521 
522 static void offset_iterate_dir(struct inode *inode, struct dir_context *ctx, long last_index)
523 {
524 	struct offset_ctx *octx = inode->i_op->get_offset_ctx(inode);
525 	struct dentry *dentry;
526 
527 	while (true) {
528 		dentry = offset_find_next(octx, ctx->pos);
529 		if (!dentry)
530 			return;
531 
532 		if (dentry2offset(dentry) >= last_index) {
533 			dput(dentry);
534 			return;
535 		}
536 
537 		if (!offset_dir_emit(ctx, dentry)) {
538 			dput(dentry);
539 			return;
540 		}
541 
542 		ctx->pos = dentry2offset(dentry) + 1;
543 		dput(dentry);
544 	}
545 }
546 
547 /**
548  * offset_readdir - Emit entries starting at offset @ctx->pos
549  * @file: an open directory to iterate over
550  * @ctx: directory iteration context
551  *
552  * Caller must hold @file's i_rwsem to prevent insertion or removal of
553  * entries during this call.
554  *
555  * On entry, @ctx->pos contains an offset that represents the first entry
556  * to be read from the directory.
557  *
558  * The operation continues until there are no more entries to read, or
559  * until the ctx->actor indicates there is no more space in the caller's
560  * output buffer.
561  *
562  * On return, @ctx->pos contains an offset that will read the next entry
563  * in this directory when offset_readdir() is called again with @ctx.
564  *
565  * Return values:
566  *   %0 - Complete
567  */
568 static int offset_readdir(struct file *file, struct dir_context *ctx)
569 {
570 	struct dentry *dir = file->f_path.dentry;
571 	long last_index = (long)file->private_data;
572 
573 	lockdep_assert_held(&d_inode(dir)->i_rwsem);
574 
575 	if (!dir_emit_dots(file, ctx))
576 		return 0;
577 
578 	offset_iterate_dir(d_inode(dir), ctx, last_index);
579 	return 0;
580 }
581 
582 const struct file_operations simple_offset_dir_operations = {
583 	.open		= offset_dir_open,
584 	.llseek		= offset_dir_llseek,
585 	.iterate_shared	= offset_readdir,
586 	.read		= generic_read_dir,
587 	.fsync		= noop_fsync,
588 };
589 
590 static struct dentry *find_next_child(struct dentry *parent, struct dentry *prev)
591 {
592 	struct dentry *child = NULL, *d;
593 
594 	spin_lock(&parent->d_lock);
595 	d = prev ? d_next_sibling(prev) : d_first_child(parent);
596 	hlist_for_each_entry_from(d, d_sib) {
597 		if (simple_positive(d)) {
598 			spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
599 			if (simple_positive(d))
600 				child = dget_dlock(d);
601 			spin_unlock(&d->d_lock);
602 			if (likely(child))
603 				break;
604 		}
605 	}
606 	spin_unlock(&parent->d_lock);
607 	dput(prev);
608 	return child;
609 }
610 
611 void simple_recursive_removal(struct dentry *dentry,
612                               void (*callback)(struct dentry *))
613 {
614 	struct dentry *this = dget(dentry);
615 	while (true) {
616 		struct dentry *victim = NULL, *child;
617 		struct inode *inode = this->d_inode;
618 
619 		inode_lock(inode);
620 		if (d_is_dir(this))
621 			inode->i_flags |= S_DEAD;
622 		while ((child = find_next_child(this, victim)) == NULL) {
623 			// kill and ascend
624 			// update metadata while it's still locked
625 			inode_set_ctime_current(inode);
626 			clear_nlink(inode);
627 			inode_unlock(inode);
628 			victim = this;
629 			this = this->d_parent;
630 			inode = this->d_inode;
631 			inode_lock(inode);
632 			if (simple_positive(victim)) {
633 				d_invalidate(victim);	// avoid lost mounts
634 				if (d_is_dir(victim))
635 					fsnotify_rmdir(inode, victim);
636 				else
637 					fsnotify_unlink(inode, victim);
638 				if (callback)
639 					callback(victim);
640 				dput(victim);		// unpin it
641 			}
642 			if (victim == dentry) {
643 				inode_set_mtime_to_ts(inode,
644 						      inode_set_ctime_current(inode));
645 				if (d_is_dir(dentry))
646 					drop_nlink(inode);
647 				inode_unlock(inode);
648 				dput(dentry);
649 				return;
650 			}
651 		}
652 		inode_unlock(inode);
653 		this = child;
654 	}
655 }
656 EXPORT_SYMBOL(simple_recursive_removal);
657 
658 static const struct super_operations simple_super_operations = {
659 	.statfs		= simple_statfs,
660 };
661 
662 static int pseudo_fs_fill_super(struct super_block *s, struct fs_context *fc)
663 {
664 	struct pseudo_fs_context *ctx = fc->fs_private;
665 	struct inode *root;
666 
667 	s->s_maxbytes = MAX_LFS_FILESIZE;
668 	s->s_blocksize = PAGE_SIZE;
669 	s->s_blocksize_bits = PAGE_SHIFT;
670 	s->s_magic = ctx->magic;
671 	s->s_op = ctx->ops ?: &simple_super_operations;
672 	s->s_xattr = ctx->xattr;
673 	s->s_time_gran = 1;
674 	root = new_inode(s);
675 	if (!root)
676 		return -ENOMEM;
677 
678 	/*
679 	 * since this is the first inode, make it number 1. New inodes created
680 	 * after this must take care not to collide with it (by passing
681 	 * max_reserved of 1 to iunique).
682 	 */
683 	root->i_ino = 1;
684 	root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
685 	simple_inode_init_ts(root);
686 	s->s_root = d_make_root(root);
687 	if (!s->s_root)
688 		return -ENOMEM;
689 	s->s_d_op = ctx->dops;
690 	return 0;
691 }
692 
693 static int pseudo_fs_get_tree(struct fs_context *fc)
694 {
695 	return get_tree_nodev(fc, pseudo_fs_fill_super);
696 }
697 
698 static void pseudo_fs_free(struct fs_context *fc)
699 {
700 	kfree(fc->fs_private);
701 }
702 
703 static const struct fs_context_operations pseudo_fs_context_ops = {
704 	.free		= pseudo_fs_free,
705 	.get_tree	= pseudo_fs_get_tree,
706 };
707 
708 /*
709  * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
710  * will never be mountable)
711  */
712 struct pseudo_fs_context *init_pseudo(struct fs_context *fc,
713 					unsigned long magic)
714 {
715 	struct pseudo_fs_context *ctx;
716 
717 	ctx = kzalloc(sizeof(struct pseudo_fs_context), GFP_KERNEL);
718 	if (likely(ctx)) {
719 		ctx->magic = magic;
720 		fc->fs_private = ctx;
721 		fc->ops = &pseudo_fs_context_ops;
722 		fc->sb_flags |= SB_NOUSER;
723 		fc->global = true;
724 	}
725 	return ctx;
726 }
727 EXPORT_SYMBOL(init_pseudo);
728 
729 int simple_open(struct inode *inode, struct file *file)
730 {
731 	if (inode->i_private)
732 		file->private_data = inode->i_private;
733 	return 0;
734 }
735 EXPORT_SYMBOL(simple_open);
736 
737 int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
738 {
739 	struct inode *inode = d_inode(old_dentry);
740 
741 	inode_set_mtime_to_ts(dir,
742 			      inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
743 	inc_nlink(inode);
744 	ihold(inode);
745 	dget(dentry);
746 	d_instantiate(dentry, inode);
747 	return 0;
748 }
749 EXPORT_SYMBOL(simple_link);
750 
751 int simple_empty(struct dentry *dentry)
752 {
753 	struct dentry *child;
754 	int ret = 0;
755 
756 	spin_lock(&dentry->d_lock);
757 	hlist_for_each_entry(child, &dentry->d_children, d_sib) {
758 		spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
759 		if (simple_positive(child)) {
760 			spin_unlock(&child->d_lock);
761 			goto out;
762 		}
763 		spin_unlock(&child->d_lock);
764 	}
765 	ret = 1;
766 out:
767 	spin_unlock(&dentry->d_lock);
768 	return ret;
769 }
770 EXPORT_SYMBOL(simple_empty);
771 
772 int simple_unlink(struct inode *dir, struct dentry *dentry)
773 {
774 	struct inode *inode = d_inode(dentry);
775 
776 	inode_set_mtime_to_ts(dir,
777 			      inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
778 	drop_nlink(inode);
779 	dput(dentry);
780 	return 0;
781 }
782 EXPORT_SYMBOL(simple_unlink);
783 
784 int simple_rmdir(struct inode *dir, struct dentry *dentry)
785 {
786 	if (!simple_empty(dentry))
787 		return -ENOTEMPTY;
788 
789 	drop_nlink(d_inode(dentry));
790 	simple_unlink(dir, dentry);
791 	drop_nlink(dir);
792 	return 0;
793 }
794 EXPORT_SYMBOL(simple_rmdir);
795 
796 /**
797  * simple_rename_timestamp - update the various inode timestamps for rename
798  * @old_dir: old parent directory
799  * @old_dentry: dentry that is being renamed
800  * @new_dir: new parent directory
801  * @new_dentry: target for rename
802  *
803  * POSIX mandates that the old and new parent directories have their ctime and
804  * mtime updated, and that inodes of @old_dentry and @new_dentry (if any), have
805  * their ctime updated.
806  */
807 void simple_rename_timestamp(struct inode *old_dir, struct dentry *old_dentry,
808 			     struct inode *new_dir, struct dentry *new_dentry)
809 {
810 	struct inode *newino = d_inode(new_dentry);
811 
812 	inode_set_mtime_to_ts(old_dir, inode_set_ctime_current(old_dir));
813 	if (new_dir != old_dir)
814 		inode_set_mtime_to_ts(new_dir,
815 				      inode_set_ctime_current(new_dir));
816 	inode_set_ctime_current(d_inode(old_dentry));
817 	if (newino)
818 		inode_set_ctime_current(newino);
819 }
820 EXPORT_SYMBOL_GPL(simple_rename_timestamp);
821 
822 int simple_rename_exchange(struct inode *old_dir, struct dentry *old_dentry,
823 			   struct inode *new_dir, struct dentry *new_dentry)
824 {
825 	bool old_is_dir = d_is_dir(old_dentry);
826 	bool new_is_dir = d_is_dir(new_dentry);
827 
828 	if (old_dir != new_dir && old_is_dir != new_is_dir) {
829 		if (old_is_dir) {
830 			drop_nlink(old_dir);
831 			inc_nlink(new_dir);
832 		} else {
833 			drop_nlink(new_dir);
834 			inc_nlink(old_dir);
835 		}
836 	}
837 	simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
838 	return 0;
839 }
840 EXPORT_SYMBOL_GPL(simple_rename_exchange);
841 
842 int simple_rename(struct mnt_idmap *idmap, struct inode *old_dir,
843 		  struct dentry *old_dentry, struct inode *new_dir,
844 		  struct dentry *new_dentry, unsigned int flags)
845 {
846 	int they_are_dirs = d_is_dir(old_dentry);
847 
848 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
849 		return -EINVAL;
850 
851 	if (flags & RENAME_EXCHANGE)
852 		return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
853 
854 	if (!simple_empty(new_dentry))
855 		return -ENOTEMPTY;
856 
857 	if (d_really_is_positive(new_dentry)) {
858 		simple_unlink(new_dir, new_dentry);
859 		if (they_are_dirs) {
860 			drop_nlink(d_inode(new_dentry));
861 			drop_nlink(old_dir);
862 		}
863 	} else if (they_are_dirs) {
864 		drop_nlink(old_dir);
865 		inc_nlink(new_dir);
866 	}
867 
868 	simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
869 	return 0;
870 }
871 EXPORT_SYMBOL(simple_rename);
872 
873 /**
874  * simple_setattr - setattr for simple filesystem
875  * @idmap: idmap of the target mount
876  * @dentry: dentry
877  * @iattr: iattr structure
878  *
879  * Returns 0 on success, -error on failure.
880  *
881  * simple_setattr is a simple ->setattr implementation without a proper
882  * implementation of size changes.
883  *
884  * It can either be used for in-memory filesystems or special files
885  * on simple regular filesystems.  Anything that needs to change on-disk
886  * or wire state on size changes needs its own setattr method.
887  */
888 int simple_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
889 		   struct iattr *iattr)
890 {
891 	struct inode *inode = d_inode(dentry);
892 	int error;
893 
894 	error = setattr_prepare(idmap, dentry, iattr);
895 	if (error)
896 		return error;
897 
898 	if (iattr->ia_valid & ATTR_SIZE)
899 		truncate_setsize(inode, iattr->ia_size);
900 	setattr_copy(idmap, inode, iattr);
901 	mark_inode_dirty(inode);
902 	return 0;
903 }
904 EXPORT_SYMBOL(simple_setattr);
905 
906 static int simple_read_folio(struct file *file, struct folio *folio)
907 {
908 	folio_zero_range(folio, 0, folio_size(folio));
909 	flush_dcache_folio(folio);
910 	folio_mark_uptodate(folio);
911 	folio_unlock(folio);
912 	return 0;
913 }
914 
915 int simple_write_begin(struct file *file, struct address_space *mapping,
916 			loff_t pos, unsigned len,
917 			struct page **pagep, void **fsdata)
918 {
919 	struct folio *folio;
920 
921 	folio = __filemap_get_folio(mapping, pos / PAGE_SIZE, FGP_WRITEBEGIN,
922 			mapping_gfp_mask(mapping));
923 	if (IS_ERR(folio))
924 		return PTR_ERR(folio);
925 
926 	*pagep = &folio->page;
927 
928 	if (!folio_test_uptodate(folio) && (len != folio_size(folio))) {
929 		size_t from = offset_in_folio(folio, pos);
930 
931 		folio_zero_segments(folio, 0, from,
932 				from + len, folio_size(folio));
933 	}
934 	return 0;
935 }
936 EXPORT_SYMBOL(simple_write_begin);
937 
938 /**
939  * simple_write_end - .write_end helper for non-block-device FSes
940  * @file: See .write_end of address_space_operations
941  * @mapping: 		"
942  * @pos: 		"
943  * @len: 		"
944  * @copied: 		"
945  * @page: 		"
946  * @fsdata: 		"
947  *
948  * simple_write_end does the minimum needed for updating a page after writing is
949  * done. It has the same API signature as the .write_end of
950  * address_space_operations vector. So it can just be set onto .write_end for
951  * FSes that don't need any other processing. i_mutex is assumed to be held.
952  * Block based filesystems should use generic_write_end().
953  * NOTE: Even though i_size might get updated by this function, mark_inode_dirty
954  * is not called, so a filesystem that actually does store data in .write_inode
955  * should extend on what's done here with a call to mark_inode_dirty() in the
956  * case that i_size has changed.
957  *
958  * Use *ONLY* with simple_read_folio()
959  */
960 static int simple_write_end(struct file *file, struct address_space *mapping,
961 			loff_t pos, unsigned len, unsigned copied,
962 			struct page *page, void *fsdata)
963 {
964 	struct folio *folio = page_folio(page);
965 	struct inode *inode = folio->mapping->host;
966 	loff_t last_pos = pos + copied;
967 
968 	/* zero the stale part of the folio if we did a short copy */
969 	if (!folio_test_uptodate(folio)) {
970 		if (copied < len) {
971 			size_t from = offset_in_folio(folio, pos);
972 
973 			folio_zero_range(folio, from + copied, len - copied);
974 		}
975 		folio_mark_uptodate(folio);
976 	}
977 	/*
978 	 * No need to use i_size_read() here, the i_size
979 	 * cannot change under us because we hold the i_mutex.
980 	 */
981 	if (last_pos > inode->i_size)
982 		i_size_write(inode, last_pos);
983 
984 	folio_mark_dirty(folio);
985 	folio_unlock(folio);
986 	folio_put(folio);
987 
988 	return copied;
989 }
990 
991 /*
992  * Provides ramfs-style behavior: data in the pagecache, but no writeback.
993  */
994 const struct address_space_operations ram_aops = {
995 	.read_folio	= simple_read_folio,
996 	.write_begin	= simple_write_begin,
997 	.write_end	= simple_write_end,
998 	.dirty_folio	= noop_dirty_folio,
999 };
1000 EXPORT_SYMBOL(ram_aops);
1001 
1002 /*
1003  * the inodes created here are not hashed. If you use iunique to generate
1004  * unique inode values later for this filesystem, then you must take care
1005  * to pass it an appropriate max_reserved value to avoid collisions.
1006  */
1007 int simple_fill_super(struct super_block *s, unsigned long magic,
1008 		      const struct tree_descr *files)
1009 {
1010 	struct inode *inode;
1011 	struct dentry *dentry;
1012 	int i;
1013 
1014 	s->s_blocksize = PAGE_SIZE;
1015 	s->s_blocksize_bits = PAGE_SHIFT;
1016 	s->s_magic = magic;
1017 	s->s_op = &simple_super_operations;
1018 	s->s_time_gran = 1;
1019 
1020 	inode = new_inode(s);
1021 	if (!inode)
1022 		return -ENOMEM;
1023 	/*
1024 	 * because the root inode is 1, the files array must not contain an
1025 	 * entry at index 1
1026 	 */
1027 	inode->i_ino = 1;
1028 	inode->i_mode = S_IFDIR | 0755;
1029 	simple_inode_init_ts(inode);
1030 	inode->i_op = &simple_dir_inode_operations;
1031 	inode->i_fop = &simple_dir_operations;
1032 	set_nlink(inode, 2);
1033 	s->s_root = d_make_root(inode);
1034 	if (!s->s_root)
1035 		return -ENOMEM;
1036 	for (i = 0; !files->name || files->name[0]; i++, files++) {
1037 		if (!files->name)
1038 			continue;
1039 
1040 		/* warn if it tries to conflict with the root inode */
1041 		if (unlikely(i == 1))
1042 			printk(KERN_WARNING "%s: %s passed in a files array"
1043 				"with an index of 1!\n", __func__,
1044 				s->s_type->name);
1045 
1046 		dentry = d_alloc_name(s->s_root, files->name);
1047 		if (!dentry)
1048 			return -ENOMEM;
1049 		inode = new_inode(s);
1050 		if (!inode) {
1051 			dput(dentry);
1052 			return -ENOMEM;
1053 		}
1054 		inode->i_mode = S_IFREG | files->mode;
1055 		simple_inode_init_ts(inode);
1056 		inode->i_fop = files->ops;
1057 		inode->i_ino = i;
1058 		d_add(dentry, inode);
1059 	}
1060 	return 0;
1061 }
1062 EXPORT_SYMBOL(simple_fill_super);
1063 
1064 static DEFINE_SPINLOCK(pin_fs_lock);
1065 
1066 int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
1067 {
1068 	struct vfsmount *mnt = NULL;
1069 	spin_lock(&pin_fs_lock);
1070 	if (unlikely(!*mount)) {
1071 		spin_unlock(&pin_fs_lock);
1072 		mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
1073 		if (IS_ERR(mnt))
1074 			return PTR_ERR(mnt);
1075 		spin_lock(&pin_fs_lock);
1076 		if (!*mount)
1077 			*mount = mnt;
1078 	}
1079 	mntget(*mount);
1080 	++*count;
1081 	spin_unlock(&pin_fs_lock);
1082 	mntput(mnt);
1083 	return 0;
1084 }
1085 EXPORT_SYMBOL(simple_pin_fs);
1086 
1087 void simple_release_fs(struct vfsmount **mount, int *count)
1088 {
1089 	struct vfsmount *mnt;
1090 	spin_lock(&pin_fs_lock);
1091 	mnt = *mount;
1092 	if (!--*count)
1093 		*mount = NULL;
1094 	spin_unlock(&pin_fs_lock);
1095 	mntput(mnt);
1096 }
1097 EXPORT_SYMBOL(simple_release_fs);
1098 
1099 /**
1100  * simple_read_from_buffer - copy data from the buffer to user space
1101  * @to: the user space buffer to read to
1102  * @count: the maximum number of bytes to read
1103  * @ppos: the current position in the buffer
1104  * @from: the buffer to read from
1105  * @available: the size of the buffer
1106  *
1107  * The simple_read_from_buffer() function reads up to @count bytes from the
1108  * buffer @from at offset @ppos into the user space address starting at @to.
1109  *
1110  * On success, the number of bytes read is returned and the offset @ppos is
1111  * advanced by this number, or negative value is returned on error.
1112  **/
1113 ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
1114 				const void *from, size_t available)
1115 {
1116 	loff_t pos = *ppos;
1117 	size_t ret;
1118 
1119 	if (pos < 0)
1120 		return -EINVAL;
1121 	if (pos >= available || !count)
1122 		return 0;
1123 	if (count > available - pos)
1124 		count = available - pos;
1125 	ret = copy_to_user(to, from + pos, count);
1126 	if (ret == count)
1127 		return -EFAULT;
1128 	count -= ret;
1129 	*ppos = pos + count;
1130 	return count;
1131 }
1132 EXPORT_SYMBOL(simple_read_from_buffer);
1133 
1134 /**
1135  * simple_write_to_buffer - copy data from user space to the buffer
1136  * @to: the buffer to write to
1137  * @available: the size of the buffer
1138  * @ppos: the current position in the buffer
1139  * @from: the user space buffer to read from
1140  * @count: the maximum number of bytes to read
1141  *
1142  * The simple_write_to_buffer() function reads up to @count bytes from the user
1143  * space address starting at @from into the buffer @to at offset @ppos.
1144  *
1145  * On success, the number of bytes written is returned and the offset @ppos is
1146  * advanced by this number, or negative value is returned on error.
1147  **/
1148 ssize_t simple_write_to_buffer(void *to, size_t available, loff_t *ppos,
1149 		const void __user *from, size_t count)
1150 {
1151 	loff_t pos = *ppos;
1152 	size_t res;
1153 
1154 	if (pos < 0)
1155 		return -EINVAL;
1156 	if (pos >= available || !count)
1157 		return 0;
1158 	if (count > available - pos)
1159 		count = available - pos;
1160 	res = copy_from_user(to + pos, from, count);
1161 	if (res == count)
1162 		return -EFAULT;
1163 	count -= res;
1164 	*ppos = pos + count;
1165 	return count;
1166 }
1167 EXPORT_SYMBOL(simple_write_to_buffer);
1168 
1169 /**
1170  * memory_read_from_buffer - copy data from the buffer
1171  * @to: the kernel space buffer to read to
1172  * @count: the maximum number of bytes to read
1173  * @ppos: the current position in the buffer
1174  * @from: the buffer to read from
1175  * @available: the size of the buffer
1176  *
1177  * The memory_read_from_buffer() function reads up to @count bytes from the
1178  * buffer @from at offset @ppos into the kernel space address starting at @to.
1179  *
1180  * On success, the number of bytes read is returned and the offset @ppos is
1181  * advanced by this number, or negative value is returned on error.
1182  **/
1183 ssize_t memory_read_from_buffer(void *to, size_t count, loff_t *ppos,
1184 				const void *from, size_t available)
1185 {
1186 	loff_t pos = *ppos;
1187 
1188 	if (pos < 0)
1189 		return -EINVAL;
1190 	if (pos >= available)
1191 		return 0;
1192 	if (count > available - pos)
1193 		count = available - pos;
1194 	memcpy(to, from + pos, count);
1195 	*ppos = pos + count;
1196 
1197 	return count;
1198 }
1199 EXPORT_SYMBOL(memory_read_from_buffer);
1200 
1201 /*
1202  * Transaction based IO.
1203  * The file expects a single write which triggers the transaction, and then
1204  * possibly a read which collects the result - which is stored in a
1205  * file-local buffer.
1206  */
1207 
1208 void simple_transaction_set(struct file *file, size_t n)
1209 {
1210 	struct simple_transaction_argresp *ar = file->private_data;
1211 
1212 	BUG_ON(n > SIMPLE_TRANSACTION_LIMIT);
1213 
1214 	/*
1215 	 * The barrier ensures that ar->size will really remain zero until
1216 	 * ar->data is ready for reading.
1217 	 */
1218 	smp_mb();
1219 	ar->size = n;
1220 }
1221 EXPORT_SYMBOL(simple_transaction_set);
1222 
1223 char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
1224 {
1225 	struct simple_transaction_argresp *ar;
1226 	static DEFINE_SPINLOCK(simple_transaction_lock);
1227 
1228 	if (size > SIMPLE_TRANSACTION_LIMIT - 1)
1229 		return ERR_PTR(-EFBIG);
1230 
1231 	ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
1232 	if (!ar)
1233 		return ERR_PTR(-ENOMEM);
1234 
1235 	spin_lock(&simple_transaction_lock);
1236 
1237 	/* only one write allowed per open */
1238 	if (file->private_data) {
1239 		spin_unlock(&simple_transaction_lock);
1240 		free_page((unsigned long)ar);
1241 		return ERR_PTR(-EBUSY);
1242 	}
1243 
1244 	file->private_data = ar;
1245 
1246 	spin_unlock(&simple_transaction_lock);
1247 
1248 	if (copy_from_user(ar->data, buf, size))
1249 		return ERR_PTR(-EFAULT);
1250 
1251 	return ar->data;
1252 }
1253 EXPORT_SYMBOL(simple_transaction_get);
1254 
1255 ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
1256 {
1257 	struct simple_transaction_argresp *ar = file->private_data;
1258 
1259 	if (!ar)
1260 		return 0;
1261 	return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
1262 }
1263 EXPORT_SYMBOL(simple_transaction_read);
1264 
1265 int simple_transaction_release(struct inode *inode, struct file *file)
1266 {
1267 	free_page((unsigned long)file->private_data);
1268 	return 0;
1269 }
1270 EXPORT_SYMBOL(simple_transaction_release);
1271 
1272 /* Simple attribute files */
1273 
1274 struct simple_attr {
1275 	int (*get)(void *, u64 *);
1276 	int (*set)(void *, u64);
1277 	char get_buf[24];	/* enough to store a u64 and "\n\0" */
1278 	char set_buf[24];
1279 	void *data;
1280 	const char *fmt;	/* format for read operation */
1281 	struct mutex mutex;	/* protects access to these buffers */
1282 };
1283 
1284 /* simple_attr_open is called by an actual attribute open file operation
1285  * to set the attribute specific access operations. */
1286 int simple_attr_open(struct inode *inode, struct file *file,
1287 		     int (*get)(void *, u64 *), int (*set)(void *, u64),
1288 		     const char *fmt)
1289 {
1290 	struct simple_attr *attr;
1291 
1292 	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
1293 	if (!attr)
1294 		return -ENOMEM;
1295 
1296 	attr->get = get;
1297 	attr->set = set;
1298 	attr->data = inode->i_private;
1299 	attr->fmt = fmt;
1300 	mutex_init(&attr->mutex);
1301 
1302 	file->private_data = attr;
1303 
1304 	return nonseekable_open(inode, file);
1305 }
1306 EXPORT_SYMBOL_GPL(simple_attr_open);
1307 
1308 int simple_attr_release(struct inode *inode, struct file *file)
1309 {
1310 	kfree(file->private_data);
1311 	return 0;
1312 }
1313 EXPORT_SYMBOL_GPL(simple_attr_release);	/* GPL-only?  This?  Really? */
1314 
1315 /* read from the buffer that is filled with the get function */
1316 ssize_t simple_attr_read(struct file *file, char __user *buf,
1317 			 size_t len, loff_t *ppos)
1318 {
1319 	struct simple_attr *attr;
1320 	size_t size;
1321 	ssize_t ret;
1322 
1323 	attr = file->private_data;
1324 
1325 	if (!attr->get)
1326 		return -EACCES;
1327 
1328 	ret = mutex_lock_interruptible(&attr->mutex);
1329 	if (ret)
1330 		return ret;
1331 
1332 	if (*ppos && attr->get_buf[0]) {
1333 		/* continued read */
1334 		size = strlen(attr->get_buf);
1335 	} else {
1336 		/* first read */
1337 		u64 val;
1338 		ret = attr->get(attr->data, &val);
1339 		if (ret)
1340 			goto out;
1341 
1342 		size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
1343 				 attr->fmt, (unsigned long long)val);
1344 	}
1345 
1346 	ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
1347 out:
1348 	mutex_unlock(&attr->mutex);
1349 	return ret;
1350 }
1351 EXPORT_SYMBOL_GPL(simple_attr_read);
1352 
1353 /* interpret the buffer as a number to call the set function with */
1354 static ssize_t simple_attr_write_xsigned(struct file *file, const char __user *buf,
1355 			  size_t len, loff_t *ppos, bool is_signed)
1356 {
1357 	struct simple_attr *attr;
1358 	unsigned long long val;
1359 	size_t size;
1360 	ssize_t ret;
1361 
1362 	attr = file->private_data;
1363 	if (!attr->set)
1364 		return -EACCES;
1365 
1366 	ret = mutex_lock_interruptible(&attr->mutex);
1367 	if (ret)
1368 		return ret;
1369 
1370 	ret = -EFAULT;
1371 	size = min(sizeof(attr->set_buf) - 1, len);
1372 	if (copy_from_user(attr->set_buf, buf, size))
1373 		goto out;
1374 
1375 	attr->set_buf[size] = '\0';
1376 	if (is_signed)
1377 		ret = kstrtoll(attr->set_buf, 0, &val);
1378 	else
1379 		ret = kstrtoull(attr->set_buf, 0, &val);
1380 	if (ret)
1381 		goto out;
1382 	ret = attr->set(attr->data, val);
1383 	if (ret == 0)
1384 		ret = len; /* on success, claim we got the whole input */
1385 out:
1386 	mutex_unlock(&attr->mutex);
1387 	return ret;
1388 }
1389 
1390 ssize_t simple_attr_write(struct file *file, const char __user *buf,
1391 			  size_t len, loff_t *ppos)
1392 {
1393 	return simple_attr_write_xsigned(file, buf, len, ppos, false);
1394 }
1395 EXPORT_SYMBOL_GPL(simple_attr_write);
1396 
1397 ssize_t simple_attr_write_signed(struct file *file, const char __user *buf,
1398 			  size_t len, loff_t *ppos)
1399 {
1400 	return simple_attr_write_xsigned(file, buf, len, ppos, true);
1401 }
1402 EXPORT_SYMBOL_GPL(simple_attr_write_signed);
1403 
1404 /**
1405  * generic_encode_ino32_fh - generic export_operations->encode_fh function
1406  * @inode:   the object to encode
1407  * @fh:      where to store the file handle fragment
1408  * @max_len: maximum length to store there (in 4 byte units)
1409  * @parent:  parent directory inode, if wanted
1410  *
1411  * This generic encode_fh function assumes that the 32 inode number
1412  * is suitable for locating an inode, and that the generation number
1413  * can be used to check that it is still valid.  It places them in the
1414  * filehandle fragment where export_decode_fh expects to find them.
1415  */
1416 int generic_encode_ino32_fh(struct inode *inode, __u32 *fh, int *max_len,
1417 			    struct inode *parent)
1418 {
1419 	struct fid *fid = (void *)fh;
1420 	int len = *max_len;
1421 	int type = FILEID_INO32_GEN;
1422 
1423 	if (parent && (len < 4)) {
1424 		*max_len = 4;
1425 		return FILEID_INVALID;
1426 	} else if (len < 2) {
1427 		*max_len = 2;
1428 		return FILEID_INVALID;
1429 	}
1430 
1431 	len = 2;
1432 	fid->i32.ino = inode->i_ino;
1433 	fid->i32.gen = inode->i_generation;
1434 	if (parent) {
1435 		fid->i32.parent_ino = parent->i_ino;
1436 		fid->i32.parent_gen = parent->i_generation;
1437 		len = 4;
1438 		type = FILEID_INO32_GEN_PARENT;
1439 	}
1440 	*max_len = len;
1441 	return type;
1442 }
1443 EXPORT_SYMBOL_GPL(generic_encode_ino32_fh);
1444 
1445 /**
1446  * generic_fh_to_dentry - generic helper for the fh_to_dentry export operation
1447  * @sb:		filesystem to do the file handle conversion on
1448  * @fid:	file handle to convert
1449  * @fh_len:	length of the file handle in bytes
1450  * @fh_type:	type of file handle
1451  * @get_inode:	filesystem callback to retrieve inode
1452  *
1453  * This function decodes @fid as long as it has one of the well-known
1454  * Linux filehandle types and calls @get_inode on it to retrieve the
1455  * inode for the object specified in the file handle.
1456  */
1457 struct dentry *generic_fh_to_dentry(struct super_block *sb, struct fid *fid,
1458 		int fh_len, int fh_type, struct inode *(*get_inode)
1459 			(struct super_block *sb, u64 ino, u32 gen))
1460 {
1461 	struct inode *inode = NULL;
1462 
1463 	if (fh_len < 2)
1464 		return NULL;
1465 
1466 	switch (fh_type) {
1467 	case FILEID_INO32_GEN:
1468 	case FILEID_INO32_GEN_PARENT:
1469 		inode = get_inode(sb, fid->i32.ino, fid->i32.gen);
1470 		break;
1471 	}
1472 
1473 	return d_obtain_alias(inode);
1474 }
1475 EXPORT_SYMBOL_GPL(generic_fh_to_dentry);
1476 
1477 /**
1478  * generic_fh_to_parent - generic helper for the fh_to_parent export operation
1479  * @sb:		filesystem to do the file handle conversion on
1480  * @fid:	file handle to convert
1481  * @fh_len:	length of the file handle in bytes
1482  * @fh_type:	type of file handle
1483  * @get_inode:	filesystem callback to retrieve inode
1484  *
1485  * This function decodes @fid as long as it has one of the well-known
1486  * Linux filehandle types and calls @get_inode on it to retrieve the
1487  * inode for the _parent_ object specified in the file handle if it
1488  * is specified in the file handle, or NULL otherwise.
1489  */
1490 struct dentry *generic_fh_to_parent(struct super_block *sb, struct fid *fid,
1491 		int fh_len, int fh_type, struct inode *(*get_inode)
1492 			(struct super_block *sb, u64 ino, u32 gen))
1493 {
1494 	struct inode *inode = NULL;
1495 
1496 	if (fh_len <= 2)
1497 		return NULL;
1498 
1499 	switch (fh_type) {
1500 	case FILEID_INO32_GEN_PARENT:
1501 		inode = get_inode(sb, fid->i32.parent_ino,
1502 				  (fh_len > 3 ? fid->i32.parent_gen : 0));
1503 		break;
1504 	}
1505 
1506 	return d_obtain_alias(inode);
1507 }
1508 EXPORT_SYMBOL_GPL(generic_fh_to_parent);
1509 
1510 /**
1511  * __generic_file_fsync - generic fsync implementation for simple filesystems
1512  *
1513  * @file:	file to synchronize
1514  * @start:	start offset in bytes
1515  * @end:	end offset in bytes (inclusive)
1516  * @datasync:	only synchronize essential metadata if true
1517  *
1518  * This is a generic implementation of the fsync method for simple
1519  * filesystems which track all non-inode metadata in the buffers list
1520  * hanging off the address_space structure.
1521  */
1522 int __generic_file_fsync(struct file *file, loff_t start, loff_t end,
1523 				 int datasync)
1524 {
1525 	struct inode *inode = file->f_mapping->host;
1526 	int err;
1527 	int ret;
1528 
1529 	err = file_write_and_wait_range(file, start, end);
1530 	if (err)
1531 		return err;
1532 
1533 	inode_lock(inode);
1534 	ret = sync_mapping_buffers(inode->i_mapping);
1535 	if (!(inode->i_state & I_DIRTY_ALL))
1536 		goto out;
1537 	if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
1538 		goto out;
1539 
1540 	err = sync_inode_metadata(inode, 1);
1541 	if (ret == 0)
1542 		ret = err;
1543 
1544 out:
1545 	inode_unlock(inode);
1546 	/* check and advance again to catch errors after syncing out buffers */
1547 	err = file_check_and_advance_wb_err(file);
1548 	if (ret == 0)
1549 		ret = err;
1550 	return ret;
1551 }
1552 EXPORT_SYMBOL(__generic_file_fsync);
1553 
1554 /**
1555  * generic_file_fsync - generic fsync implementation for simple filesystems
1556  *			with flush
1557  * @file:	file to synchronize
1558  * @start:	start offset in bytes
1559  * @end:	end offset in bytes (inclusive)
1560  * @datasync:	only synchronize essential metadata if true
1561  *
1562  */
1563 
1564 int generic_file_fsync(struct file *file, loff_t start, loff_t end,
1565 		       int datasync)
1566 {
1567 	struct inode *inode = file->f_mapping->host;
1568 	int err;
1569 
1570 	err = __generic_file_fsync(file, start, end, datasync);
1571 	if (err)
1572 		return err;
1573 	return blkdev_issue_flush(inode->i_sb->s_bdev);
1574 }
1575 EXPORT_SYMBOL(generic_file_fsync);
1576 
1577 /**
1578  * generic_check_addressable - Check addressability of file system
1579  * @blocksize_bits:	log of file system block size
1580  * @num_blocks:		number of blocks in file system
1581  *
1582  * Determine whether a file system with @num_blocks blocks (and a
1583  * block size of 2**@blocksize_bits) is addressable by the sector_t
1584  * and page cache of the system.  Return 0 if so and -EFBIG otherwise.
1585  */
1586 int generic_check_addressable(unsigned blocksize_bits, u64 num_blocks)
1587 {
1588 	u64 last_fs_block = num_blocks - 1;
1589 	u64 last_fs_page =
1590 		last_fs_block >> (PAGE_SHIFT - blocksize_bits);
1591 
1592 	if (unlikely(num_blocks == 0))
1593 		return 0;
1594 
1595 	if ((blocksize_bits < 9) || (blocksize_bits > PAGE_SHIFT))
1596 		return -EINVAL;
1597 
1598 	if ((last_fs_block > (sector_t)(~0ULL) >> (blocksize_bits - 9)) ||
1599 	    (last_fs_page > (pgoff_t)(~0ULL))) {
1600 		return -EFBIG;
1601 	}
1602 	return 0;
1603 }
1604 EXPORT_SYMBOL(generic_check_addressable);
1605 
1606 /*
1607  * No-op implementation of ->fsync for in-memory filesystems.
1608  */
1609 int noop_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1610 {
1611 	return 0;
1612 }
1613 EXPORT_SYMBOL(noop_fsync);
1614 
1615 ssize_t noop_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1616 {
1617 	/*
1618 	 * iomap based filesystems support direct I/O without need for
1619 	 * this callback. However, it still needs to be set in
1620 	 * inode->a_ops so that open/fcntl know that direct I/O is
1621 	 * generally supported.
1622 	 */
1623 	return -EINVAL;
1624 }
1625 EXPORT_SYMBOL_GPL(noop_direct_IO);
1626 
1627 /* Because kfree isn't assignment-compatible with void(void*) ;-/ */
1628 void kfree_link(void *p)
1629 {
1630 	kfree(p);
1631 }
1632 EXPORT_SYMBOL(kfree_link);
1633 
1634 struct inode *alloc_anon_inode(struct super_block *s)
1635 {
1636 	static const struct address_space_operations anon_aops = {
1637 		.dirty_folio	= noop_dirty_folio,
1638 	};
1639 	struct inode *inode = new_inode_pseudo(s);
1640 
1641 	if (!inode)
1642 		return ERR_PTR(-ENOMEM);
1643 
1644 	inode->i_ino = get_next_ino();
1645 	inode->i_mapping->a_ops = &anon_aops;
1646 
1647 	/*
1648 	 * Mark the inode dirty from the very beginning,
1649 	 * that way it will never be moved to the dirty
1650 	 * list because mark_inode_dirty() will think
1651 	 * that it already _is_ on the dirty list.
1652 	 */
1653 	inode->i_state = I_DIRTY;
1654 	inode->i_mode = S_IRUSR | S_IWUSR;
1655 	inode->i_uid = current_fsuid();
1656 	inode->i_gid = current_fsgid();
1657 	inode->i_flags |= S_PRIVATE;
1658 	simple_inode_init_ts(inode);
1659 	return inode;
1660 }
1661 EXPORT_SYMBOL(alloc_anon_inode);
1662 
1663 /**
1664  * simple_nosetlease - generic helper for prohibiting leases
1665  * @filp: file pointer
1666  * @arg: type of lease to obtain
1667  * @flp: new lease supplied for insertion
1668  * @priv: private data for lm_setup operation
1669  *
1670  * Generic helper for filesystems that do not wish to allow leases to be set.
1671  * All arguments are ignored and it just returns -EINVAL.
1672  */
1673 int
1674 simple_nosetlease(struct file *filp, int arg, struct file_lease **flp,
1675 		  void **priv)
1676 {
1677 	return -EINVAL;
1678 }
1679 EXPORT_SYMBOL(simple_nosetlease);
1680 
1681 /**
1682  * simple_get_link - generic helper to get the target of "fast" symlinks
1683  * @dentry: not used here
1684  * @inode: the symlink inode
1685  * @done: not used here
1686  *
1687  * Generic helper for filesystems to use for symlink inodes where a pointer to
1688  * the symlink target is stored in ->i_link.  NOTE: this isn't normally called,
1689  * since as an optimization the path lookup code uses any non-NULL ->i_link
1690  * directly, without calling ->get_link().  But ->get_link() still must be set,
1691  * to mark the inode_operations as being for a symlink.
1692  *
1693  * Return: the symlink target
1694  */
1695 const char *simple_get_link(struct dentry *dentry, struct inode *inode,
1696 			    struct delayed_call *done)
1697 {
1698 	return inode->i_link;
1699 }
1700 EXPORT_SYMBOL(simple_get_link);
1701 
1702 const struct inode_operations simple_symlink_inode_operations = {
1703 	.get_link = simple_get_link,
1704 };
1705 EXPORT_SYMBOL(simple_symlink_inode_operations);
1706 
1707 /*
1708  * Operations for a permanently empty directory.
1709  */
1710 static struct dentry *empty_dir_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1711 {
1712 	return ERR_PTR(-ENOENT);
1713 }
1714 
1715 static int empty_dir_getattr(struct mnt_idmap *idmap,
1716 			     const struct path *path, struct kstat *stat,
1717 			     u32 request_mask, unsigned int query_flags)
1718 {
1719 	struct inode *inode = d_inode(path->dentry);
1720 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1721 	return 0;
1722 }
1723 
1724 static int empty_dir_setattr(struct mnt_idmap *idmap,
1725 			     struct dentry *dentry, struct iattr *attr)
1726 {
1727 	return -EPERM;
1728 }
1729 
1730 static ssize_t empty_dir_listxattr(struct dentry *dentry, char *list, size_t size)
1731 {
1732 	return -EOPNOTSUPP;
1733 }
1734 
1735 static const struct inode_operations empty_dir_inode_operations = {
1736 	.lookup		= empty_dir_lookup,
1737 	.permission	= generic_permission,
1738 	.setattr	= empty_dir_setattr,
1739 	.getattr	= empty_dir_getattr,
1740 	.listxattr	= empty_dir_listxattr,
1741 };
1742 
1743 static loff_t empty_dir_llseek(struct file *file, loff_t offset, int whence)
1744 {
1745 	/* An empty directory has two entries . and .. at offsets 0 and 1 */
1746 	return generic_file_llseek_size(file, offset, whence, 2, 2);
1747 }
1748 
1749 static int empty_dir_readdir(struct file *file, struct dir_context *ctx)
1750 {
1751 	dir_emit_dots(file, ctx);
1752 	return 0;
1753 }
1754 
1755 static const struct file_operations empty_dir_operations = {
1756 	.llseek		= empty_dir_llseek,
1757 	.read		= generic_read_dir,
1758 	.iterate_shared	= empty_dir_readdir,
1759 	.fsync		= noop_fsync,
1760 };
1761 
1762 
1763 void make_empty_dir_inode(struct inode *inode)
1764 {
1765 	set_nlink(inode, 2);
1766 	inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO;
1767 	inode->i_uid = GLOBAL_ROOT_UID;
1768 	inode->i_gid = GLOBAL_ROOT_GID;
1769 	inode->i_rdev = 0;
1770 	inode->i_size = 0;
1771 	inode->i_blkbits = PAGE_SHIFT;
1772 	inode->i_blocks = 0;
1773 
1774 	inode->i_op = &empty_dir_inode_operations;
1775 	inode->i_opflags &= ~IOP_XATTR;
1776 	inode->i_fop = &empty_dir_operations;
1777 }
1778 
1779 bool is_empty_dir_inode(struct inode *inode)
1780 {
1781 	return (inode->i_fop == &empty_dir_operations) &&
1782 		(inode->i_op == &empty_dir_inode_operations);
1783 }
1784 
1785 #if IS_ENABLED(CONFIG_UNICODE)
1786 /**
1787  * generic_ci_d_compare - generic d_compare implementation for casefolding filesystems
1788  * @dentry:	dentry whose name we are checking against
1789  * @len:	len of name of dentry
1790  * @str:	str pointer to name of dentry
1791  * @name:	Name to compare against
1792  *
1793  * Return: 0 if names match, 1 if mismatch, or -ERRNO
1794  */
1795 static int generic_ci_d_compare(const struct dentry *dentry, unsigned int len,
1796 				const char *str, const struct qstr *name)
1797 {
1798 	const struct dentry *parent;
1799 	const struct inode *dir;
1800 	char strbuf[DNAME_INLINE_LEN];
1801 	struct qstr qstr;
1802 
1803 	/*
1804 	 * Attempt a case-sensitive match first. It is cheaper and
1805 	 * should cover most lookups, including all the sane
1806 	 * applications that expect a case-sensitive filesystem.
1807 	 *
1808 	 * This comparison is safe under RCU because the caller
1809 	 * guarantees the consistency between str and len. See
1810 	 * __d_lookup_rcu_op_compare() for details.
1811 	 */
1812 	if (len == name->len && !memcmp(str, name->name, len))
1813 		return 0;
1814 
1815 	parent = READ_ONCE(dentry->d_parent);
1816 	dir = READ_ONCE(parent->d_inode);
1817 	if (!dir || !IS_CASEFOLDED(dir))
1818 		return 1;
1819 
1820 	/*
1821 	 * If the dentry name is stored in-line, then it may be concurrently
1822 	 * modified by a rename.  If this happens, the VFS will eventually retry
1823 	 * the lookup, so it doesn't matter what ->d_compare() returns.
1824 	 * However, it's unsafe to call utf8_strncasecmp() with an unstable
1825 	 * string.  Therefore, we have to copy the name into a temporary buffer.
1826 	 */
1827 	if (len <= DNAME_INLINE_LEN - 1) {
1828 		memcpy(strbuf, str, len);
1829 		strbuf[len] = 0;
1830 		str = strbuf;
1831 		/* prevent compiler from optimizing out the temporary buffer */
1832 		barrier();
1833 	}
1834 	qstr.len = len;
1835 	qstr.name = str;
1836 
1837 	return utf8_strncasecmp(dentry->d_sb->s_encoding, name, &qstr);
1838 }
1839 
1840 /**
1841  * generic_ci_d_hash - generic d_hash implementation for casefolding filesystems
1842  * @dentry:	dentry of the parent directory
1843  * @str:	qstr of name whose hash we should fill in
1844  *
1845  * Return: 0 if hash was successful or unchanged, and -EINVAL on error
1846  */
1847 static int generic_ci_d_hash(const struct dentry *dentry, struct qstr *str)
1848 {
1849 	const struct inode *dir = READ_ONCE(dentry->d_inode);
1850 	struct super_block *sb = dentry->d_sb;
1851 	const struct unicode_map *um = sb->s_encoding;
1852 	int ret;
1853 
1854 	if (!dir || !IS_CASEFOLDED(dir))
1855 		return 0;
1856 
1857 	ret = utf8_casefold_hash(um, dentry, str);
1858 	if (ret < 0 && sb_has_strict_encoding(sb))
1859 		return -EINVAL;
1860 	return 0;
1861 }
1862 
1863 static const struct dentry_operations generic_ci_dentry_ops = {
1864 	.d_hash = generic_ci_d_hash,
1865 	.d_compare = generic_ci_d_compare,
1866 #ifdef CONFIG_FS_ENCRYPTION
1867 	.d_revalidate = fscrypt_d_revalidate,
1868 #endif
1869 };
1870 
1871 /**
1872  * generic_ci_match() - Match a name (case-insensitively) with a dirent.
1873  * This is a filesystem helper for comparison with directory entries.
1874  * generic_ci_d_compare should be used in VFS' ->d_compare instead.
1875  *
1876  * @parent: Inode of the parent of the dirent under comparison
1877  * @name: name under lookup.
1878  * @folded_name: Optional pre-folded name under lookup
1879  * @de_name: Dirent name.
1880  * @de_name_len: dirent name length.
1881  *
1882  * Test whether a case-insensitive directory entry matches the filename
1883  * being searched.  If @folded_name is provided, it is used instead of
1884  * recalculating the casefold of @name.
1885  *
1886  * Return: > 0 if the directory entry matches, 0 if it doesn't match, or
1887  * < 0 on error.
1888  */
1889 int generic_ci_match(const struct inode *parent,
1890 		     const struct qstr *name,
1891 		     const struct qstr *folded_name,
1892 		     const u8 *de_name, u32 de_name_len)
1893 {
1894 	const struct super_block *sb = parent->i_sb;
1895 	const struct unicode_map *um = sb->s_encoding;
1896 	struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1897 	struct qstr dirent = QSTR_INIT(de_name, de_name_len);
1898 	int res = 0;
1899 
1900 	if (IS_ENCRYPTED(parent)) {
1901 		const struct fscrypt_str encrypted_name =
1902 			FSTR_INIT((u8 *) de_name, de_name_len);
1903 
1904 		if (WARN_ON_ONCE(!fscrypt_has_encryption_key(parent)))
1905 			return -EINVAL;
1906 
1907 		decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1908 		if (!decrypted_name.name)
1909 			return -ENOMEM;
1910 		res = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1911 						&decrypted_name);
1912 		if (res < 0) {
1913 			kfree(decrypted_name.name);
1914 			return res;
1915 		}
1916 		dirent.name = decrypted_name.name;
1917 		dirent.len = decrypted_name.len;
1918 	}
1919 
1920 	/*
1921 	 * Attempt a case-sensitive match first. It is cheaper and
1922 	 * should cover most lookups, including all the sane
1923 	 * applications that expect a case-sensitive filesystem.
1924 	 */
1925 
1926 	if (dirent.len == name->len &&
1927 	    !memcmp(name->name, dirent.name, dirent.len))
1928 		goto out;
1929 
1930 	if (folded_name->name)
1931 		res = utf8_strncasecmp_folded(um, folded_name, &dirent);
1932 	else
1933 		res = utf8_strncasecmp(um, name, &dirent);
1934 
1935 out:
1936 	kfree(decrypted_name.name);
1937 	if (res < 0 && sb_has_strict_encoding(sb)) {
1938 		pr_err_ratelimited("Directory contains filename that is invalid UTF-8");
1939 		return 0;
1940 	}
1941 	return !res;
1942 }
1943 EXPORT_SYMBOL(generic_ci_match);
1944 #endif
1945 
1946 #ifdef CONFIG_FS_ENCRYPTION
1947 static const struct dentry_operations generic_encrypted_dentry_ops = {
1948 	.d_revalidate = fscrypt_d_revalidate,
1949 };
1950 #endif
1951 
1952 /**
1953  * generic_set_sb_d_ops - helper for choosing the set of
1954  * filesystem-wide dentry operations for the enabled features
1955  * @sb: superblock to be configured
1956  *
1957  * Filesystems supporting casefolding and/or fscrypt can call this
1958  * helper at mount-time to configure sb->s_d_op to best set of dentry
1959  * operations required for the enabled features. The helper must be
1960  * called after these have been configured, but before the root dentry
1961  * is created.
1962  */
1963 void generic_set_sb_d_ops(struct super_block *sb)
1964 {
1965 #if IS_ENABLED(CONFIG_UNICODE)
1966 	if (sb->s_encoding) {
1967 		sb->s_d_op = &generic_ci_dentry_ops;
1968 		return;
1969 	}
1970 #endif
1971 #ifdef CONFIG_FS_ENCRYPTION
1972 	if (sb->s_cop) {
1973 		sb->s_d_op = &generic_encrypted_dentry_ops;
1974 		return;
1975 	}
1976 #endif
1977 }
1978 EXPORT_SYMBOL(generic_set_sb_d_ops);
1979 
1980 /**
1981  * inode_maybe_inc_iversion - increments i_version
1982  * @inode: inode with the i_version that should be updated
1983  * @force: increment the counter even if it's not necessary?
1984  *
1985  * Every time the inode is modified, the i_version field must be seen to have
1986  * changed by any observer.
1987  *
1988  * If "force" is set or the QUERIED flag is set, then ensure that we increment
1989  * the value, and clear the queried flag.
1990  *
1991  * In the common case where neither is set, then we can return "false" without
1992  * updating i_version.
1993  *
1994  * If this function returns false, and no other metadata has changed, then we
1995  * can avoid logging the metadata.
1996  */
1997 bool inode_maybe_inc_iversion(struct inode *inode, bool force)
1998 {
1999 	u64 cur, new;
2000 
2001 	/*
2002 	 * The i_version field is not strictly ordered with any other inode
2003 	 * information, but the legacy inode_inc_iversion code used a spinlock
2004 	 * to serialize increments.
2005 	 *
2006 	 * Here, we add full memory barriers to ensure that any de-facto
2007 	 * ordering with other info is preserved.
2008 	 *
2009 	 * This barrier pairs with the barrier in inode_query_iversion()
2010 	 */
2011 	smp_mb();
2012 	cur = inode_peek_iversion_raw(inode);
2013 	do {
2014 		/* If flag is clear then we needn't do anything */
2015 		if (!force && !(cur & I_VERSION_QUERIED))
2016 			return false;
2017 
2018 		/* Since lowest bit is flag, add 2 to avoid it */
2019 		new = (cur & ~I_VERSION_QUERIED) + I_VERSION_INCREMENT;
2020 	} while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
2021 	return true;
2022 }
2023 EXPORT_SYMBOL(inode_maybe_inc_iversion);
2024 
2025 /**
2026  * inode_query_iversion - read i_version for later use
2027  * @inode: inode from which i_version should be read
2028  *
2029  * Read the inode i_version counter. This should be used by callers that wish
2030  * to store the returned i_version for later comparison. This will guarantee
2031  * that a later query of the i_version will result in a different value if
2032  * anything has changed.
2033  *
2034  * In this implementation, we fetch the current value, set the QUERIED flag and
2035  * then try to swap it into place with a cmpxchg, if it wasn't already set. If
2036  * that fails, we try again with the newly fetched value from the cmpxchg.
2037  */
2038 u64 inode_query_iversion(struct inode *inode)
2039 {
2040 	u64 cur, new;
2041 
2042 	cur = inode_peek_iversion_raw(inode);
2043 	do {
2044 		/* If flag is already set, then no need to swap */
2045 		if (cur & I_VERSION_QUERIED) {
2046 			/*
2047 			 * This barrier (and the implicit barrier in the
2048 			 * cmpxchg below) pairs with the barrier in
2049 			 * inode_maybe_inc_iversion().
2050 			 */
2051 			smp_mb();
2052 			break;
2053 		}
2054 
2055 		new = cur | I_VERSION_QUERIED;
2056 	} while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
2057 	return cur >> I_VERSION_QUERIED_SHIFT;
2058 }
2059 EXPORT_SYMBOL(inode_query_iversion);
2060 
2061 ssize_t direct_write_fallback(struct kiocb *iocb, struct iov_iter *iter,
2062 		ssize_t direct_written, ssize_t buffered_written)
2063 {
2064 	struct address_space *mapping = iocb->ki_filp->f_mapping;
2065 	loff_t pos = iocb->ki_pos - buffered_written;
2066 	loff_t end = iocb->ki_pos - 1;
2067 	int err;
2068 
2069 	/*
2070 	 * If the buffered write fallback returned an error, we want to return
2071 	 * the number of bytes which were written by direct I/O, or the error
2072 	 * code if that was zero.
2073 	 *
2074 	 * Note that this differs from normal direct-io semantics, which will
2075 	 * return -EFOO even if some bytes were written.
2076 	 */
2077 	if (unlikely(buffered_written < 0)) {
2078 		if (direct_written)
2079 			return direct_written;
2080 		return buffered_written;
2081 	}
2082 
2083 	/*
2084 	 * We need to ensure that the page cache pages are written to disk and
2085 	 * invalidated to preserve the expected O_DIRECT semantics.
2086 	 */
2087 	err = filemap_write_and_wait_range(mapping, pos, end);
2088 	if (err < 0) {
2089 		/*
2090 		 * We don't know how much we wrote, so just return the number of
2091 		 * bytes which were direct-written
2092 		 */
2093 		iocb->ki_pos -= buffered_written;
2094 		if (direct_written)
2095 			return direct_written;
2096 		return err;
2097 	}
2098 	invalidate_mapping_pages(mapping, pos >> PAGE_SHIFT, end >> PAGE_SHIFT);
2099 	return direct_written + buffered_written;
2100 }
2101 EXPORT_SYMBOL_GPL(direct_write_fallback);
2102 
2103 /**
2104  * simple_inode_init_ts - initialize the timestamps for a new inode
2105  * @inode: inode to be initialized
2106  *
2107  * When a new inode is created, most filesystems set the timestamps to the
2108  * current time. Add a helper to do this.
2109  */
2110 struct timespec64 simple_inode_init_ts(struct inode *inode)
2111 {
2112 	struct timespec64 ts = inode_set_ctime_current(inode);
2113 
2114 	inode_set_atime_to_ts(inode, ts);
2115 	inode_set_mtime_to_ts(inode, ts);
2116 	return ts;
2117 }
2118 EXPORT_SYMBOL(simple_inode_init_ts);
2119 
2120 static inline struct dentry *get_stashed_dentry(struct dentry *stashed)
2121 {
2122 	struct dentry *dentry;
2123 
2124 	guard(rcu)();
2125 	dentry = READ_ONCE(stashed);
2126 	if (!dentry)
2127 		return NULL;
2128 	if (!lockref_get_not_dead(&dentry->d_lockref))
2129 		return NULL;
2130 	return dentry;
2131 }
2132 
2133 static struct dentry *prepare_anon_dentry(struct dentry **stashed,
2134 					  struct super_block *sb,
2135 					  void *data)
2136 {
2137 	struct dentry *dentry;
2138 	struct inode *inode;
2139 	const struct stashed_operations *sops = sb->s_fs_info;
2140 	int ret;
2141 
2142 	inode = new_inode_pseudo(sb);
2143 	if (!inode) {
2144 		sops->put_data(data);
2145 		return ERR_PTR(-ENOMEM);
2146 	}
2147 
2148 	inode->i_flags |= S_IMMUTABLE;
2149 	inode->i_mode = S_IFREG;
2150 	simple_inode_init_ts(inode);
2151 
2152 	ret = sops->init_inode(inode, data);
2153 	if (ret < 0) {
2154 		iput(inode);
2155 		return ERR_PTR(ret);
2156 	}
2157 
2158 	/* Notice when this is changed. */
2159 	WARN_ON_ONCE(!S_ISREG(inode->i_mode));
2160 	WARN_ON_ONCE(!IS_IMMUTABLE(inode));
2161 
2162 	dentry = d_alloc_anon(sb);
2163 	if (!dentry) {
2164 		iput(inode);
2165 		return ERR_PTR(-ENOMEM);
2166 	}
2167 
2168 	/* Store address of location where dentry's supposed to be stashed. */
2169 	dentry->d_fsdata = stashed;
2170 
2171 	/* @data is now owned by the fs */
2172 	d_instantiate(dentry, inode);
2173 	return dentry;
2174 }
2175 
2176 static struct dentry *stash_dentry(struct dentry **stashed,
2177 				   struct dentry *dentry)
2178 {
2179 	guard(rcu)();
2180 	for (;;) {
2181 		struct dentry *old;
2182 
2183 		/* Assume any old dentry was cleared out. */
2184 		old = cmpxchg(stashed, NULL, dentry);
2185 		if (likely(!old))
2186 			return dentry;
2187 
2188 		/* Check if somebody else installed a reusable dentry. */
2189 		if (lockref_get_not_dead(&old->d_lockref))
2190 			return old;
2191 
2192 		/* There's an old dead dentry there, try to take it over. */
2193 		if (likely(try_cmpxchg(stashed, &old, dentry)))
2194 			return dentry;
2195 	}
2196 }
2197 
2198 /**
2199  * path_from_stashed - create path from stashed or new dentry
2200  * @stashed:    where to retrieve or stash dentry
2201  * @mnt:        mnt of the filesystems to use
2202  * @data:       data to store in inode->i_private
2203  * @path:       path to create
2204  *
2205  * The function tries to retrieve a stashed dentry from @stashed. If the dentry
2206  * is still valid then it will be reused. If the dentry isn't able the function
2207  * will allocate a new dentry and inode. It will then check again whether it
2208  * can reuse an existing dentry in case one has been added in the meantime or
2209  * update @stashed with the newly added dentry.
2210  *
2211  * Special-purpose helper for nsfs and pidfs.
2212  *
2213  * Return: On success zero and on failure a negative error is returned.
2214  */
2215 int path_from_stashed(struct dentry **stashed, struct vfsmount *mnt, void *data,
2216 		      struct path *path)
2217 {
2218 	struct dentry *dentry;
2219 	const struct stashed_operations *sops = mnt->mnt_sb->s_fs_info;
2220 
2221 	/* See if dentry can be reused. */
2222 	path->dentry = get_stashed_dentry(*stashed);
2223 	if (path->dentry) {
2224 		sops->put_data(data);
2225 		goto out_path;
2226 	}
2227 
2228 	/* Allocate a new dentry. */
2229 	dentry = prepare_anon_dentry(stashed, mnt->mnt_sb, data);
2230 	if (IS_ERR(dentry))
2231 		return PTR_ERR(dentry);
2232 
2233 	/* Added a new dentry. @data is now owned by the filesystem. */
2234 	path->dentry = stash_dentry(stashed, dentry);
2235 	if (path->dentry != dentry)
2236 		dput(dentry);
2237 
2238 out_path:
2239 	WARN_ON_ONCE(path->dentry->d_fsdata != stashed);
2240 	WARN_ON_ONCE(d_inode(path->dentry)->i_private != data);
2241 	path->mnt = mntget(mnt);
2242 	return 0;
2243 }
2244 
2245 void stashed_dentry_prune(struct dentry *dentry)
2246 {
2247 	struct dentry **stashed = dentry->d_fsdata;
2248 	struct inode *inode = d_inode(dentry);
2249 
2250 	if (WARN_ON_ONCE(!stashed))
2251 		return;
2252 
2253 	if (!inode)
2254 		return;
2255 
2256 	/*
2257 	 * Only replace our own @dentry as someone else might've
2258 	 * already cleared out @dentry and stashed their own
2259 	 * dentry in there.
2260 	 */
2261 	cmpxchg(stashed, dentry, NULL);
2262 }
2263