xref: /linux/fs/libfs.c (revision ff30564411ffdcee49d579cb15eb13185a36e253)
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 /**
454  * offset_dir_llseek - Advance the read position of a directory descriptor
455  * @file: an open directory whose position is to be updated
456  * @offset: a byte offset
457  * @whence: enumerator describing the starting position for this update
458  *
459  * SEEK_END, SEEK_DATA, and SEEK_HOLE are not supported for directories.
460  *
461  * Returns the updated read position if successful; otherwise a
462  * negative errno is returned and the read position remains unchanged.
463  */
464 static loff_t offset_dir_llseek(struct file *file, loff_t offset, int whence)
465 {
466 	switch (whence) {
467 	case SEEK_CUR:
468 		offset += file->f_pos;
469 		fallthrough;
470 	case SEEK_SET:
471 		if (offset >= 0)
472 			break;
473 		fallthrough;
474 	default:
475 		return -EINVAL;
476 	}
477 
478 	/* In this case, ->private_data is protected by f_pos_lock */
479 	file->private_data = NULL;
480 	return vfs_setpos(file, offset, LONG_MAX);
481 }
482 
483 static struct dentry *offset_find_next(struct offset_ctx *octx, loff_t offset)
484 {
485 	MA_STATE(mas, &octx->mt, offset, offset);
486 	struct dentry *child, *found = NULL;
487 
488 	rcu_read_lock();
489 	child = mas_find(&mas, LONG_MAX);
490 	if (!child)
491 		goto out;
492 	spin_lock(&child->d_lock);
493 	if (simple_positive(child))
494 		found = dget_dlock(child);
495 	spin_unlock(&child->d_lock);
496 out:
497 	rcu_read_unlock();
498 	return found;
499 }
500 
501 static bool offset_dir_emit(struct dir_context *ctx, struct dentry *dentry)
502 {
503 	struct inode *inode = d_inode(dentry);
504 	long offset = dentry2offset(dentry);
505 
506 	return ctx->actor(ctx, dentry->d_name.name, dentry->d_name.len, offset,
507 			  inode->i_ino, fs_umode_to_dtype(inode->i_mode));
508 }
509 
510 static void *offset_iterate_dir(struct inode *inode, struct dir_context *ctx)
511 {
512 	struct offset_ctx *octx = inode->i_op->get_offset_ctx(inode);
513 	struct dentry *dentry;
514 
515 	while (true) {
516 		dentry = offset_find_next(octx, ctx->pos);
517 		if (!dentry)
518 			return ERR_PTR(-ENOENT);
519 
520 		if (!offset_dir_emit(ctx, dentry)) {
521 			dput(dentry);
522 			break;
523 		}
524 
525 		ctx->pos = dentry2offset(dentry) + 1;
526 		dput(dentry);
527 	}
528 	return NULL;
529 }
530 
531 /**
532  * offset_readdir - Emit entries starting at offset @ctx->pos
533  * @file: an open directory to iterate over
534  * @ctx: directory iteration context
535  *
536  * Caller must hold @file's i_rwsem to prevent insertion or removal of
537  * entries during this call.
538  *
539  * On entry, @ctx->pos contains an offset that represents the first entry
540  * to be read from the directory.
541  *
542  * The operation continues until there are no more entries to read, or
543  * until the ctx->actor indicates there is no more space in the caller's
544  * output buffer.
545  *
546  * On return, @ctx->pos contains an offset that will read the next entry
547  * in this directory when offset_readdir() is called again with @ctx.
548  *
549  * Return values:
550  *   %0 - Complete
551  */
552 static int offset_readdir(struct file *file, struct dir_context *ctx)
553 {
554 	struct dentry *dir = file->f_path.dentry;
555 
556 	lockdep_assert_held(&d_inode(dir)->i_rwsem);
557 
558 	if (!dir_emit_dots(file, ctx))
559 		return 0;
560 
561 	/* In this case, ->private_data is protected by f_pos_lock */
562 	if (ctx->pos == DIR_OFFSET_MIN)
563 		file->private_data = NULL;
564 	else if (file->private_data == ERR_PTR(-ENOENT))
565 		return 0;
566 	file->private_data = offset_iterate_dir(d_inode(dir), ctx);
567 	return 0;
568 }
569 
570 const struct file_operations simple_offset_dir_operations = {
571 	.llseek		= offset_dir_llseek,
572 	.iterate_shared	= offset_readdir,
573 	.read		= generic_read_dir,
574 	.fsync		= noop_fsync,
575 };
576 
577 static struct dentry *find_next_child(struct dentry *parent, struct dentry *prev)
578 {
579 	struct dentry *child = NULL, *d;
580 
581 	spin_lock(&parent->d_lock);
582 	d = prev ? d_next_sibling(prev) : d_first_child(parent);
583 	hlist_for_each_entry_from(d, d_sib) {
584 		if (simple_positive(d)) {
585 			spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
586 			if (simple_positive(d))
587 				child = dget_dlock(d);
588 			spin_unlock(&d->d_lock);
589 			if (likely(child))
590 				break;
591 		}
592 	}
593 	spin_unlock(&parent->d_lock);
594 	dput(prev);
595 	return child;
596 }
597 
598 void simple_recursive_removal(struct dentry *dentry,
599                               void (*callback)(struct dentry *))
600 {
601 	struct dentry *this = dget(dentry);
602 	while (true) {
603 		struct dentry *victim = NULL, *child;
604 		struct inode *inode = this->d_inode;
605 
606 		inode_lock(inode);
607 		if (d_is_dir(this))
608 			inode->i_flags |= S_DEAD;
609 		while ((child = find_next_child(this, victim)) == NULL) {
610 			// kill and ascend
611 			// update metadata while it's still locked
612 			inode_set_ctime_current(inode);
613 			clear_nlink(inode);
614 			inode_unlock(inode);
615 			victim = this;
616 			this = this->d_parent;
617 			inode = this->d_inode;
618 			inode_lock(inode);
619 			if (simple_positive(victim)) {
620 				d_invalidate(victim);	// avoid lost mounts
621 				if (d_is_dir(victim))
622 					fsnotify_rmdir(inode, victim);
623 				else
624 					fsnotify_unlink(inode, victim);
625 				if (callback)
626 					callback(victim);
627 				dput(victim);		// unpin it
628 			}
629 			if (victim == dentry) {
630 				inode_set_mtime_to_ts(inode,
631 						      inode_set_ctime_current(inode));
632 				if (d_is_dir(dentry))
633 					drop_nlink(inode);
634 				inode_unlock(inode);
635 				dput(dentry);
636 				return;
637 			}
638 		}
639 		inode_unlock(inode);
640 		this = child;
641 	}
642 }
643 EXPORT_SYMBOL(simple_recursive_removal);
644 
645 static const struct super_operations simple_super_operations = {
646 	.statfs		= simple_statfs,
647 };
648 
649 static int pseudo_fs_fill_super(struct super_block *s, struct fs_context *fc)
650 {
651 	struct pseudo_fs_context *ctx = fc->fs_private;
652 	struct inode *root;
653 
654 	s->s_maxbytes = MAX_LFS_FILESIZE;
655 	s->s_blocksize = PAGE_SIZE;
656 	s->s_blocksize_bits = PAGE_SHIFT;
657 	s->s_magic = ctx->magic;
658 	s->s_op = ctx->ops ?: &simple_super_operations;
659 	s->s_xattr = ctx->xattr;
660 	s->s_time_gran = 1;
661 	root = new_inode(s);
662 	if (!root)
663 		return -ENOMEM;
664 
665 	/*
666 	 * since this is the first inode, make it number 1. New inodes created
667 	 * after this must take care not to collide with it (by passing
668 	 * max_reserved of 1 to iunique).
669 	 */
670 	root->i_ino = 1;
671 	root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
672 	simple_inode_init_ts(root);
673 	s->s_root = d_make_root(root);
674 	if (!s->s_root)
675 		return -ENOMEM;
676 	s->s_d_op = ctx->dops;
677 	return 0;
678 }
679 
680 static int pseudo_fs_get_tree(struct fs_context *fc)
681 {
682 	return get_tree_nodev(fc, pseudo_fs_fill_super);
683 }
684 
685 static void pseudo_fs_free(struct fs_context *fc)
686 {
687 	kfree(fc->fs_private);
688 }
689 
690 static const struct fs_context_operations pseudo_fs_context_ops = {
691 	.free		= pseudo_fs_free,
692 	.get_tree	= pseudo_fs_get_tree,
693 };
694 
695 /*
696  * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
697  * will never be mountable)
698  */
699 struct pseudo_fs_context *init_pseudo(struct fs_context *fc,
700 					unsigned long magic)
701 {
702 	struct pseudo_fs_context *ctx;
703 
704 	ctx = kzalloc(sizeof(struct pseudo_fs_context), GFP_KERNEL);
705 	if (likely(ctx)) {
706 		ctx->magic = magic;
707 		fc->fs_private = ctx;
708 		fc->ops = &pseudo_fs_context_ops;
709 		fc->sb_flags |= SB_NOUSER;
710 		fc->global = true;
711 	}
712 	return ctx;
713 }
714 EXPORT_SYMBOL(init_pseudo);
715 
716 int simple_open(struct inode *inode, struct file *file)
717 {
718 	if (inode->i_private)
719 		file->private_data = inode->i_private;
720 	return 0;
721 }
722 EXPORT_SYMBOL(simple_open);
723 
724 int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
725 {
726 	struct inode *inode = d_inode(old_dentry);
727 
728 	inode_set_mtime_to_ts(dir,
729 			      inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
730 	inc_nlink(inode);
731 	ihold(inode);
732 	dget(dentry);
733 	d_instantiate(dentry, inode);
734 	return 0;
735 }
736 EXPORT_SYMBOL(simple_link);
737 
738 int simple_empty(struct dentry *dentry)
739 {
740 	struct dentry *child;
741 	int ret = 0;
742 
743 	spin_lock(&dentry->d_lock);
744 	hlist_for_each_entry(child, &dentry->d_children, d_sib) {
745 		spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
746 		if (simple_positive(child)) {
747 			spin_unlock(&child->d_lock);
748 			goto out;
749 		}
750 		spin_unlock(&child->d_lock);
751 	}
752 	ret = 1;
753 out:
754 	spin_unlock(&dentry->d_lock);
755 	return ret;
756 }
757 EXPORT_SYMBOL(simple_empty);
758 
759 int simple_unlink(struct inode *dir, struct dentry *dentry)
760 {
761 	struct inode *inode = d_inode(dentry);
762 
763 	inode_set_mtime_to_ts(dir,
764 			      inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
765 	drop_nlink(inode);
766 	dput(dentry);
767 	return 0;
768 }
769 EXPORT_SYMBOL(simple_unlink);
770 
771 int simple_rmdir(struct inode *dir, struct dentry *dentry)
772 {
773 	if (!simple_empty(dentry))
774 		return -ENOTEMPTY;
775 
776 	drop_nlink(d_inode(dentry));
777 	simple_unlink(dir, dentry);
778 	drop_nlink(dir);
779 	return 0;
780 }
781 EXPORT_SYMBOL(simple_rmdir);
782 
783 /**
784  * simple_rename_timestamp - update the various inode timestamps for rename
785  * @old_dir: old parent directory
786  * @old_dentry: dentry that is being renamed
787  * @new_dir: new parent directory
788  * @new_dentry: target for rename
789  *
790  * POSIX mandates that the old and new parent directories have their ctime and
791  * mtime updated, and that inodes of @old_dentry and @new_dentry (if any), have
792  * their ctime updated.
793  */
794 void simple_rename_timestamp(struct inode *old_dir, struct dentry *old_dentry,
795 			     struct inode *new_dir, struct dentry *new_dentry)
796 {
797 	struct inode *newino = d_inode(new_dentry);
798 
799 	inode_set_mtime_to_ts(old_dir, inode_set_ctime_current(old_dir));
800 	if (new_dir != old_dir)
801 		inode_set_mtime_to_ts(new_dir,
802 				      inode_set_ctime_current(new_dir));
803 	inode_set_ctime_current(d_inode(old_dentry));
804 	if (newino)
805 		inode_set_ctime_current(newino);
806 }
807 EXPORT_SYMBOL_GPL(simple_rename_timestamp);
808 
809 int simple_rename_exchange(struct inode *old_dir, struct dentry *old_dentry,
810 			   struct inode *new_dir, struct dentry *new_dentry)
811 {
812 	bool old_is_dir = d_is_dir(old_dentry);
813 	bool new_is_dir = d_is_dir(new_dentry);
814 
815 	if (old_dir != new_dir && old_is_dir != new_is_dir) {
816 		if (old_is_dir) {
817 			drop_nlink(old_dir);
818 			inc_nlink(new_dir);
819 		} else {
820 			drop_nlink(new_dir);
821 			inc_nlink(old_dir);
822 		}
823 	}
824 	simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
825 	return 0;
826 }
827 EXPORT_SYMBOL_GPL(simple_rename_exchange);
828 
829 int simple_rename(struct mnt_idmap *idmap, struct inode *old_dir,
830 		  struct dentry *old_dentry, struct inode *new_dir,
831 		  struct dentry *new_dentry, unsigned int flags)
832 {
833 	int they_are_dirs = d_is_dir(old_dentry);
834 
835 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
836 		return -EINVAL;
837 
838 	if (flags & RENAME_EXCHANGE)
839 		return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
840 
841 	if (!simple_empty(new_dentry))
842 		return -ENOTEMPTY;
843 
844 	if (d_really_is_positive(new_dentry)) {
845 		simple_unlink(new_dir, new_dentry);
846 		if (they_are_dirs) {
847 			drop_nlink(d_inode(new_dentry));
848 			drop_nlink(old_dir);
849 		}
850 	} else if (they_are_dirs) {
851 		drop_nlink(old_dir);
852 		inc_nlink(new_dir);
853 	}
854 
855 	simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
856 	return 0;
857 }
858 EXPORT_SYMBOL(simple_rename);
859 
860 /**
861  * simple_setattr - setattr for simple filesystem
862  * @idmap: idmap of the target mount
863  * @dentry: dentry
864  * @iattr: iattr structure
865  *
866  * Returns 0 on success, -error on failure.
867  *
868  * simple_setattr is a simple ->setattr implementation without a proper
869  * implementation of size changes.
870  *
871  * It can either be used for in-memory filesystems or special files
872  * on simple regular filesystems.  Anything that needs to change on-disk
873  * or wire state on size changes needs its own setattr method.
874  */
875 int simple_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
876 		   struct iattr *iattr)
877 {
878 	struct inode *inode = d_inode(dentry);
879 	int error;
880 
881 	error = setattr_prepare(idmap, dentry, iattr);
882 	if (error)
883 		return error;
884 
885 	if (iattr->ia_valid & ATTR_SIZE)
886 		truncate_setsize(inode, iattr->ia_size);
887 	setattr_copy(idmap, inode, iattr);
888 	mark_inode_dirty(inode);
889 	return 0;
890 }
891 EXPORT_SYMBOL(simple_setattr);
892 
893 static int simple_read_folio(struct file *file, struct folio *folio)
894 {
895 	folio_zero_range(folio, 0, folio_size(folio));
896 	flush_dcache_folio(folio);
897 	folio_mark_uptodate(folio);
898 	folio_unlock(folio);
899 	return 0;
900 }
901 
902 int simple_write_begin(struct file *file, struct address_space *mapping,
903 			loff_t pos, unsigned len,
904 			struct page **pagep, void **fsdata)
905 {
906 	struct folio *folio;
907 
908 	folio = __filemap_get_folio(mapping, pos / PAGE_SIZE, FGP_WRITEBEGIN,
909 			mapping_gfp_mask(mapping));
910 	if (IS_ERR(folio))
911 		return PTR_ERR(folio);
912 
913 	*pagep = &folio->page;
914 
915 	if (!folio_test_uptodate(folio) && (len != folio_size(folio))) {
916 		size_t from = offset_in_folio(folio, pos);
917 
918 		folio_zero_segments(folio, 0, from,
919 				from + len, folio_size(folio));
920 	}
921 	return 0;
922 }
923 EXPORT_SYMBOL(simple_write_begin);
924 
925 /**
926  * simple_write_end - .write_end helper for non-block-device FSes
927  * @file: See .write_end of address_space_operations
928  * @mapping: 		"
929  * @pos: 		"
930  * @len: 		"
931  * @copied: 		"
932  * @page: 		"
933  * @fsdata: 		"
934  *
935  * simple_write_end does the minimum needed for updating a page after writing is
936  * done. It has the same API signature as the .write_end of
937  * address_space_operations vector. So it can just be set onto .write_end for
938  * FSes that don't need any other processing. i_mutex is assumed to be held.
939  * Block based filesystems should use generic_write_end().
940  * NOTE: Even though i_size might get updated by this function, mark_inode_dirty
941  * is not called, so a filesystem that actually does store data in .write_inode
942  * should extend on what's done here with a call to mark_inode_dirty() in the
943  * case that i_size has changed.
944  *
945  * Use *ONLY* with simple_read_folio()
946  */
947 static int simple_write_end(struct file *file, struct address_space *mapping,
948 			loff_t pos, unsigned len, unsigned copied,
949 			struct page *page, void *fsdata)
950 {
951 	struct folio *folio = page_folio(page);
952 	struct inode *inode = folio->mapping->host;
953 	loff_t last_pos = pos + copied;
954 
955 	/* zero the stale part of the folio if we did a short copy */
956 	if (!folio_test_uptodate(folio)) {
957 		if (copied < len) {
958 			size_t from = offset_in_folio(folio, pos);
959 
960 			folio_zero_range(folio, from + copied, len - copied);
961 		}
962 		folio_mark_uptodate(folio);
963 	}
964 	/*
965 	 * No need to use i_size_read() here, the i_size
966 	 * cannot change under us because we hold the i_mutex.
967 	 */
968 	if (last_pos > inode->i_size)
969 		i_size_write(inode, last_pos);
970 
971 	folio_mark_dirty(folio);
972 	folio_unlock(folio);
973 	folio_put(folio);
974 
975 	return copied;
976 }
977 
978 /*
979  * Provides ramfs-style behavior: data in the pagecache, but no writeback.
980  */
981 const struct address_space_operations ram_aops = {
982 	.read_folio	= simple_read_folio,
983 	.write_begin	= simple_write_begin,
984 	.write_end	= simple_write_end,
985 	.dirty_folio	= noop_dirty_folio,
986 };
987 EXPORT_SYMBOL(ram_aops);
988 
989 /*
990  * the inodes created here are not hashed. If you use iunique to generate
991  * unique inode values later for this filesystem, then you must take care
992  * to pass it an appropriate max_reserved value to avoid collisions.
993  */
994 int simple_fill_super(struct super_block *s, unsigned long magic,
995 		      const struct tree_descr *files)
996 {
997 	struct inode *inode;
998 	struct dentry *dentry;
999 	int i;
1000 
1001 	s->s_blocksize = PAGE_SIZE;
1002 	s->s_blocksize_bits = PAGE_SHIFT;
1003 	s->s_magic = magic;
1004 	s->s_op = &simple_super_operations;
1005 	s->s_time_gran = 1;
1006 
1007 	inode = new_inode(s);
1008 	if (!inode)
1009 		return -ENOMEM;
1010 	/*
1011 	 * because the root inode is 1, the files array must not contain an
1012 	 * entry at index 1
1013 	 */
1014 	inode->i_ino = 1;
1015 	inode->i_mode = S_IFDIR | 0755;
1016 	simple_inode_init_ts(inode);
1017 	inode->i_op = &simple_dir_inode_operations;
1018 	inode->i_fop = &simple_dir_operations;
1019 	set_nlink(inode, 2);
1020 	s->s_root = d_make_root(inode);
1021 	if (!s->s_root)
1022 		return -ENOMEM;
1023 	for (i = 0; !files->name || files->name[0]; i++, files++) {
1024 		if (!files->name)
1025 			continue;
1026 
1027 		/* warn if it tries to conflict with the root inode */
1028 		if (unlikely(i == 1))
1029 			printk(KERN_WARNING "%s: %s passed in a files array"
1030 				"with an index of 1!\n", __func__,
1031 				s->s_type->name);
1032 
1033 		dentry = d_alloc_name(s->s_root, files->name);
1034 		if (!dentry)
1035 			return -ENOMEM;
1036 		inode = new_inode(s);
1037 		if (!inode) {
1038 			dput(dentry);
1039 			return -ENOMEM;
1040 		}
1041 		inode->i_mode = S_IFREG | files->mode;
1042 		simple_inode_init_ts(inode);
1043 		inode->i_fop = files->ops;
1044 		inode->i_ino = i;
1045 		d_add(dentry, inode);
1046 	}
1047 	return 0;
1048 }
1049 EXPORT_SYMBOL(simple_fill_super);
1050 
1051 static DEFINE_SPINLOCK(pin_fs_lock);
1052 
1053 int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
1054 {
1055 	struct vfsmount *mnt = NULL;
1056 	spin_lock(&pin_fs_lock);
1057 	if (unlikely(!*mount)) {
1058 		spin_unlock(&pin_fs_lock);
1059 		mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
1060 		if (IS_ERR(mnt))
1061 			return PTR_ERR(mnt);
1062 		spin_lock(&pin_fs_lock);
1063 		if (!*mount)
1064 			*mount = mnt;
1065 	}
1066 	mntget(*mount);
1067 	++*count;
1068 	spin_unlock(&pin_fs_lock);
1069 	mntput(mnt);
1070 	return 0;
1071 }
1072 EXPORT_SYMBOL(simple_pin_fs);
1073 
1074 void simple_release_fs(struct vfsmount **mount, int *count)
1075 {
1076 	struct vfsmount *mnt;
1077 	spin_lock(&pin_fs_lock);
1078 	mnt = *mount;
1079 	if (!--*count)
1080 		*mount = NULL;
1081 	spin_unlock(&pin_fs_lock);
1082 	mntput(mnt);
1083 }
1084 EXPORT_SYMBOL(simple_release_fs);
1085 
1086 /**
1087  * simple_read_from_buffer - copy data from the buffer to user space
1088  * @to: the user space buffer to read to
1089  * @count: the maximum number of bytes to read
1090  * @ppos: the current position in the buffer
1091  * @from: the buffer to read from
1092  * @available: the size of the buffer
1093  *
1094  * The simple_read_from_buffer() function reads up to @count bytes from the
1095  * buffer @from at offset @ppos into the user space address starting at @to.
1096  *
1097  * On success, the number of bytes read is returned and the offset @ppos is
1098  * advanced by this number, or negative value is returned on error.
1099  **/
1100 ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
1101 				const void *from, size_t available)
1102 {
1103 	loff_t pos = *ppos;
1104 	size_t ret;
1105 
1106 	if (pos < 0)
1107 		return -EINVAL;
1108 	if (pos >= available || !count)
1109 		return 0;
1110 	if (count > available - pos)
1111 		count = available - pos;
1112 	ret = copy_to_user(to, from + pos, count);
1113 	if (ret == count)
1114 		return -EFAULT;
1115 	count -= ret;
1116 	*ppos = pos + count;
1117 	return count;
1118 }
1119 EXPORT_SYMBOL(simple_read_from_buffer);
1120 
1121 /**
1122  * simple_write_to_buffer - copy data from user space to the buffer
1123  * @to: the buffer to write to
1124  * @available: the size of the buffer
1125  * @ppos: the current position in the buffer
1126  * @from: the user space buffer to read from
1127  * @count: the maximum number of bytes to read
1128  *
1129  * The simple_write_to_buffer() function reads up to @count bytes from the user
1130  * space address starting at @from into the buffer @to at offset @ppos.
1131  *
1132  * On success, the number of bytes written is returned and the offset @ppos is
1133  * advanced by this number, or negative value is returned on error.
1134  **/
1135 ssize_t simple_write_to_buffer(void *to, size_t available, loff_t *ppos,
1136 		const void __user *from, size_t count)
1137 {
1138 	loff_t pos = *ppos;
1139 	size_t res;
1140 
1141 	if (pos < 0)
1142 		return -EINVAL;
1143 	if (pos >= available || !count)
1144 		return 0;
1145 	if (count > available - pos)
1146 		count = available - pos;
1147 	res = copy_from_user(to + pos, from, count);
1148 	if (res == count)
1149 		return -EFAULT;
1150 	count -= res;
1151 	*ppos = pos + count;
1152 	return count;
1153 }
1154 EXPORT_SYMBOL(simple_write_to_buffer);
1155 
1156 /**
1157  * memory_read_from_buffer - copy data from the buffer
1158  * @to: the kernel space buffer to read to
1159  * @count: the maximum number of bytes to read
1160  * @ppos: the current position in the buffer
1161  * @from: the buffer to read from
1162  * @available: the size of the buffer
1163  *
1164  * The memory_read_from_buffer() function reads up to @count bytes from the
1165  * buffer @from at offset @ppos into the kernel space address starting at @to.
1166  *
1167  * On success, the number of bytes read is returned and the offset @ppos is
1168  * advanced by this number, or negative value is returned on error.
1169  **/
1170 ssize_t memory_read_from_buffer(void *to, size_t count, loff_t *ppos,
1171 				const void *from, size_t available)
1172 {
1173 	loff_t pos = *ppos;
1174 
1175 	if (pos < 0)
1176 		return -EINVAL;
1177 	if (pos >= available)
1178 		return 0;
1179 	if (count > available - pos)
1180 		count = available - pos;
1181 	memcpy(to, from + pos, count);
1182 	*ppos = pos + count;
1183 
1184 	return count;
1185 }
1186 EXPORT_SYMBOL(memory_read_from_buffer);
1187 
1188 /*
1189  * Transaction based IO.
1190  * The file expects a single write which triggers the transaction, and then
1191  * possibly a read which collects the result - which is stored in a
1192  * file-local buffer.
1193  */
1194 
1195 void simple_transaction_set(struct file *file, size_t n)
1196 {
1197 	struct simple_transaction_argresp *ar = file->private_data;
1198 
1199 	BUG_ON(n > SIMPLE_TRANSACTION_LIMIT);
1200 
1201 	/*
1202 	 * The barrier ensures that ar->size will really remain zero until
1203 	 * ar->data is ready for reading.
1204 	 */
1205 	smp_mb();
1206 	ar->size = n;
1207 }
1208 EXPORT_SYMBOL(simple_transaction_set);
1209 
1210 char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
1211 {
1212 	struct simple_transaction_argresp *ar;
1213 	static DEFINE_SPINLOCK(simple_transaction_lock);
1214 
1215 	if (size > SIMPLE_TRANSACTION_LIMIT - 1)
1216 		return ERR_PTR(-EFBIG);
1217 
1218 	ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
1219 	if (!ar)
1220 		return ERR_PTR(-ENOMEM);
1221 
1222 	spin_lock(&simple_transaction_lock);
1223 
1224 	/* only one write allowed per open */
1225 	if (file->private_data) {
1226 		spin_unlock(&simple_transaction_lock);
1227 		free_page((unsigned long)ar);
1228 		return ERR_PTR(-EBUSY);
1229 	}
1230 
1231 	file->private_data = ar;
1232 
1233 	spin_unlock(&simple_transaction_lock);
1234 
1235 	if (copy_from_user(ar->data, buf, size))
1236 		return ERR_PTR(-EFAULT);
1237 
1238 	return ar->data;
1239 }
1240 EXPORT_SYMBOL(simple_transaction_get);
1241 
1242 ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
1243 {
1244 	struct simple_transaction_argresp *ar = file->private_data;
1245 
1246 	if (!ar)
1247 		return 0;
1248 	return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
1249 }
1250 EXPORT_SYMBOL(simple_transaction_read);
1251 
1252 int simple_transaction_release(struct inode *inode, struct file *file)
1253 {
1254 	free_page((unsigned long)file->private_data);
1255 	return 0;
1256 }
1257 EXPORT_SYMBOL(simple_transaction_release);
1258 
1259 /* Simple attribute files */
1260 
1261 struct simple_attr {
1262 	int (*get)(void *, u64 *);
1263 	int (*set)(void *, u64);
1264 	char get_buf[24];	/* enough to store a u64 and "\n\0" */
1265 	char set_buf[24];
1266 	void *data;
1267 	const char *fmt;	/* format for read operation */
1268 	struct mutex mutex;	/* protects access to these buffers */
1269 };
1270 
1271 /* simple_attr_open is called by an actual attribute open file operation
1272  * to set the attribute specific access operations. */
1273 int simple_attr_open(struct inode *inode, struct file *file,
1274 		     int (*get)(void *, u64 *), int (*set)(void *, u64),
1275 		     const char *fmt)
1276 {
1277 	struct simple_attr *attr;
1278 
1279 	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
1280 	if (!attr)
1281 		return -ENOMEM;
1282 
1283 	attr->get = get;
1284 	attr->set = set;
1285 	attr->data = inode->i_private;
1286 	attr->fmt = fmt;
1287 	mutex_init(&attr->mutex);
1288 
1289 	file->private_data = attr;
1290 
1291 	return nonseekable_open(inode, file);
1292 }
1293 EXPORT_SYMBOL_GPL(simple_attr_open);
1294 
1295 int simple_attr_release(struct inode *inode, struct file *file)
1296 {
1297 	kfree(file->private_data);
1298 	return 0;
1299 }
1300 EXPORT_SYMBOL_GPL(simple_attr_release);	/* GPL-only?  This?  Really? */
1301 
1302 /* read from the buffer that is filled with the get function */
1303 ssize_t simple_attr_read(struct file *file, char __user *buf,
1304 			 size_t len, loff_t *ppos)
1305 {
1306 	struct simple_attr *attr;
1307 	size_t size;
1308 	ssize_t ret;
1309 
1310 	attr = file->private_data;
1311 
1312 	if (!attr->get)
1313 		return -EACCES;
1314 
1315 	ret = mutex_lock_interruptible(&attr->mutex);
1316 	if (ret)
1317 		return ret;
1318 
1319 	if (*ppos && attr->get_buf[0]) {
1320 		/* continued read */
1321 		size = strlen(attr->get_buf);
1322 	} else {
1323 		/* first read */
1324 		u64 val;
1325 		ret = attr->get(attr->data, &val);
1326 		if (ret)
1327 			goto out;
1328 
1329 		size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
1330 				 attr->fmt, (unsigned long long)val);
1331 	}
1332 
1333 	ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
1334 out:
1335 	mutex_unlock(&attr->mutex);
1336 	return ret;
1337 }
1338 EXPORT_SYMBOL_GPL(simple_attr_read);
1339 
1340 /* interpret the buffer as a number to call the set function with */
1341 static ssize_t simple_attr_write_xsigned(struct file *file, const char __user *buf,
1342 			  size_t len, loff_t *ppos, bool is_signed)
1343 {
1344 	struct simple_attr *attr;
1345 	unsigned long long val;
1346 	size_t size;
1347 	ssize_t ret;
1348 
1349 	attr = file->private_data;
1350 	if (!attr->set)
1351 		return -EACCES;
1352 
1353 	ret = mutex_lock_interruptible(&attr->mutex);
1354 	if (ret)
1355 		return ret;
1356 
1357 	ret = -EFAULT;
1358 	size = min(sizeof(attr->set_buf) - 1, len);
1359 	if (copy_from_user(attr->set_buf, buf, size))
1360 		goto out;
1361 
1362 	attr->set_buf[size] = '\0';
1363 	if (is_signed)
1364 		ret = kstrtoll(attr->set_buf, 0, &val);
1365 	else
1366 		ret = kstrtoull(attr->set_buf, 0, &val);
1367 	if (ret)
1368 		goto out;
1369 	ret = attr->set(attr->data, val);
1370 	if (ret == 0)
1371 		ret = len; /* on success, claim we got the whole input */
1372 out:
1373 	mutex_unlock(&attr->mutex);
1374 	return ret;
1375 }
1376 
1377 ssize_t simple_attr_write(struct file *file, const char __user *buf,
1378 			  size_t len, loff_t *ppos)
1379 {
1380 	return simple_attr_write_xsigned(file, buf, len, ppos, false);
1381 }
1382 EXPORT_SYMBOL_GPL(simple_attr_write);
1383 
1384 ssize_t simple_attr_write_signed(struct file *file, const char __user *buf,
1385 			  size_t len, loff_t *ppos)
1386 {
1387 	return simple_attr_write_xsigned(file, buf, len, ppos, true);
1388 }
1389 EXPORT_SYMBOL_GPL(simple_attr_write_signed);
1390 
1391 /**
1392  * generic_encode_ino32_fh - generic export_operations->encode_fh function
1393  * @inode:   the object to encode
1394  * @fh:      where to store the file handle fragment
1395  * @max_len: maximum length to store there (in 4 byte units)
1396  * @parent:  parent directory inode, if wanted
1397  *
1398  * This generic encode_fh function assumes that the 32 inode number
1399  * is suitable for locating an inode, and that the generation number
1400  * can be used to check that it is still valid.  It places them in the
1401  * filehandle fragment where export_decode_fh expects to find them.
1402  */
1403 int generic_encode_ino32_fh(struct inode *inode, __u32 *fh, int *max_len,
1404 			    struct inode *parent)
1405 {
1406 	struct fid *fid = (void *)fh;
1407 	int len = *max_len;
1408 	int type = FILEID_INO32_GEN;
1409 
1410 	if (parent && (len < 4)) {
1411 		*max_len = 4;
1412 		return FILEID_INVALID;
1413 	} else if (len < 2) {
1414 		*max_len = 2;
1415 		return FILEID_INVALID;
1416 	}
1417 
1418 	len = 2;
1419 	fid->i32.ino = inode->i_ino;
1420 	fid->i32.gen = inode->i_generation;
1421 	if (parent) {
1422 		fid->i32.parent_ino = parent->i_ino;
1423 		fid->i32.parent_gen = parent->i_generation;
1424 		len = 4;
1425 		type = FILEID_INO32_GEN_PARENT;
1426 	}
1427 	*max_len = len;
1428 	return type;
1429 }
1430 EXPORT_SYMBOL_GPL(generic_encode_ino32_fh);
1431 
1432 /**
1433  * generic_fh_to_dentry - generic helper for the fh_to_dentry export operation
1434  * @sb:		filesystem to do the file handle conversion on
1435  * @fid:	file handle to convert
1436  * @fh_len:	length of the file handle in bytes
1437  * @fh_type:	type of file handle
1438  * @get_inode:	filesystem callback to retrieve inode
1439  *
1440  * This function decodes @fid as long as it has one of the well-known
1441  * Linux filehandle types and calls @get_inode on it to retrieve the
1442  * inode for the object specified in the file handle.
1443  */
1444 struct dentry *generic_fh_to_dentry(struct super_block *sb, struct fid *fid,
1445 		int fh_len, int fh_type, struct inode *(*get_inode)
1446 			(struct super_block *sb, u64 ino, u32 gen))
1447 {
1448 	struct inode *inode = NULL;
1449 
1450 	if (fh_len < 2)
1451 		return NULL;
1452 
1453 	switch (fh_type) {
1454 	case FILEID_INO32_GEN:
1455 	case FILEID_INO32_GEN_PARENT:
1456 		inode = get_inode(sb, fid->i32.ino, fid->i32.gen);
1457 		break;
1458 	}
1459 
1460 	return d_obtain_alias(inode);
1461 }
1462 EXPORT_SYMBOL_GPL(generic_fh_to_dentry);
1463 
1464 /**
1465  * generic_fh_to_parent - generic helper for the fh_to_parent export operation
1466  * @sb:		filesystem to do the file handle conversion on
1467  * @fid:	file handle to convert
1468  * @fh_len:	length of the file handle in bytes
1469  * @fh_type:	type of file handle
1470  * @get_inode:	filesystem callback to retrieve inode
1471  *
1472  * This function decodes @fid as long as it has one of the well-known
1473  * Linux filehandle types and calls @get_inode on it to retrieve the
1474  * inode for the _parent_ object specified in the file handle if it
1475  * is specified in the file handle, or NULL otherwise.
1476  */
1477 struct dentry *generic_fh_to_parent(struct super_block *sb, struct fid *fid,
1478 		int fh_len, int fh_type, struct inode *(*get_inode)
1479 			(struct super_block *sb, u64 ino, u32 gen))
1480 {
1481 	struct inode *inode = NULL;
1482 
1483 	if (fh_len <= 2)
1484 		return NULL;
1485 
1486 	switch (fh_type) {
1487 	case FILEID_INO32_GEN_PARENT:
1488 		inode = get_inode(sb, fid->i32.parent_ino,
1489 				  (fh_len > 3 ? fid->i32.parent_gen : 0));
1490 		break;
1491 	}
1492 
1493 	return d_obtain_alias(inode);
1494 }
1495 EXPORT_SYMBOL_GPL(generic_fh_to_parent);
1496 
1497 /**
1498  * __generic_file_fsync - generic fsync implementation for simple filesystems
1499  *
1500  * @file:	file to synchronize
1501  * @start:	start offset in bytes
1502  * @end:	end offset in bytes (inclusive)
1503  * @datasync:	only synchronize essential metadata if true
1504  *
1505  * This is a generic implementation of the fsync method for simple
1506  * filesystems which track all non-inode metadata in the buffers list
1507  * hanging off the address_space structure.
1508  */
1509 int __generic_file_fsync(struct file *file, loff_t start, loff_t end,
1510 				 int datasync)
1511 {
1512 	struct inode *inode = file->f_mapping->host;
1513 	int err;
1514 	int ret;
1515 
1516 	err = file_write_and_wait_range(file, start, end);
1517 	if (err)
1518 		return err;
1519 
1520 	inode_lock(inode);
1521 	ret = sync_mapping_buffers(inode->i_mapping);
1522 	if (!(inode->i_state & I_DIRTY_ALL))
1523 		goto out;
1524 	if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
1525 		goto out;
1526 
1527 	err = sync_inode_metadata(inode, 1);
1528 	if (ret == 0)
1529 		ret = err;
1530 
1531 out:
1532 	inode_unlock(inode);
1533 	/* check and advance again to catch errors after syncing out buffers */
1534 	err = file_check_and_advance_wb_err(file);
1535 	if (ret == 0)
1536 		ret = err;
1537 	return ret;
1538 }
1539 EXPORT_SYMBOL(__generic_file_fsync);
1540 
1541 /**
1542  * generic_file_fsync - generic fsync implementation for simple filesystems
1543  *			with flush
1544  * @file:	file to synchronize
1545  * @start:	start offset in bytes
1546  * @end:	end offset in bytes (inclusive)
1547  * @datasync:	only synchronize essential metadata if true
1548  *
1549  */
1550 
1551 int generic_file_fsync(struct file *file, loff_t start, loff_t end,
1552 		       int datasync)
1553 {
1554 	struct inode *inode = file->f_mapping->host;
1555 	int err;
1556 
1557 	err = __generic_file_fsync(file, start, end, datasync);
1558 	if (err)
1559 		return err;
1560 	return blkdev_issue_flush(inode->i_sb->s_bdev);
1561 }
1562 EXPORT_SYMBOL(generic_file_fsync);
1563 
1564 /**
1565  * generic_check_addressable - Check addressability of file system
1566  * @blocksize_bits:	log of file system block size
1567  * @num_blocks:		number of blocks in file system
1568  *
1569  * Determine whether a file system with @num_blocks blocks (and a
1570  * block size of 2**@blocksize_bits) is addressable by the sector_t
1571  * and page cache of the system.  Return 0 if so and -EFBIG otherwise.
1572  */
1573 int generic_check_addressable(unsigned blocksize_bits, u64 num_blocks)
1574 {
1575 	u64 last_fs_block = num_blocks - 1;
1576 	u64 last_fs_page =
1577 		last_fs_block >> (PAGE_SHIFT - blocksize_bits);
1578 
1579 	if (unlikely(num_blocks == 0))
1580 		return 0;
1581 
1582 	if ((blocksize_bits < 9) || (blocksize_bits > PAGE_SHIFT))
1583 		return -EINVAL;
1584 
1585 	if ((last_fs_block > (sector_t)(~0ULL) >> (blocksize_bits - 9)) ||
1586 	    (last_fs_page > (pgoff_t)(~0ULL))) {
1587 		return -EFBIG;
1588 	}
1589 	return 0;
1590 }
1591 EXPORT_SYMBOL(generic_check_addressable);
1592 
1593 /*
1594  * No-op implementation of ->fsync for in-memory filesystems.
1595  */
1596 int noop_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1597 {
1598 	return 0;
1599 }
1600 EXPORT_SYMBOL(noop_fsync);
1601 
1602 ssize_t noop_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1603 {
1604 	/*
1605 	 * iomap based filesystems support direct I/O without need for
1606 	 * this callback. However, it still needs to be set in
1607 	 * inode->a_ops so that open/fcntl know that direct I/O is
1608 	 * generally supported.
1609 	 */
1610 	return -EINVAL;
1611 }
1612 EXPORT_SYMBOL_GPL(noop_direct_IO);
1613 
1614 /* Because kfree isn't assignment-compatible with void(void*) ;-/ */
1615 void kfree_link(void *p)
1616 {
1617 	kfree(p);
1618 }
1619 EXPORT_SYMBOL(kfree_link);
1620 
1621 struct inode *alloc_anon_inode(struct super_block *s)
1622 {
1623 	static const struct address_space_operations anon_aops = {
1624 		.dirty_folio	= noop_dirty_folio,
1625 	};
1626 	struct inode *inode = new_inode_pseudo(s);
1627 
1628 	if (!inode)
1629 		return ERR_PTR(-ENOMEM);
1630 
1631 	inode->i_ino = get_next_ino();
1632 	inode->i_mapping->a_ops = &anon_aops;
1633 
1634 	/*
1635 	 * Mark the inode dirty from the very beginning,
1636 	 * that way it will never be moved to the dirty
1637 	 * list because mark_inode_dirty() will think
1638 	 * that it already _is_ on the dirty list.
1639 	 */
1640 	inode->i_state = I_DIRTY;
1641 	inode->i_mode = S_IRUSR | S_IWUSR;
1642 	inode->i_uid = current_fsuid();
1643 	inode->i_gid = current_fsgid();
1644 	inode->i_flags |= S_PRIVATE;
1645 	simple_inode_init_ts(inode);
1646 	return inode;
1647 }
1648 EXPORT_SYMBOL(alloc_anon_inode);
1649 
1650 /**
1651  * simple_nosetlease - generic helper for prohibiting leases
1652  * @filp: file pointer
1653  * @arg: type of lease to obtain
1654  * @flp: new lease supplied for insertion
1655  * @priv: private data for lm_setup operation
1656  *
1657  * Generic helper for filesystems that do not wish to allow leases to be set.
1658  * All arguments are ignored and it just returns -EINVAL.
1659  */
1660 int
1661 simple_nosetlease(struct file *filp, int arg, struct file_lease **flp,
1662 		  void **priv)
1663 {
1664 	return -EINVAL;
1665 }
1666 EXPORT_SYMBOL(simple_nosetlease);
1667 
1668 /**
1669  * simple_get_link - generic helper to get the target of "fast" symlinks
1670  * @dentry: not used here
1671  * @inode: the symlink inode
1672  * @done: not used here
1673  *
1674  * Generic helper for filesystems to use for symlink inodes where a pointer to
1675  * the symlink target is stored in ->i_link.  NOTE: this isn't normally called,
1676  * since as an optimization the path lookup code uses any non-NULL ->i_link
1677  * directly, without calling ->get_link().  But ->get_link() still must be set,
1678  * to mark the inode_operations as being for a symlink.
1679  *
1680  * Return: the symlink target
1681  */
1682 const char *simple_get_link(struct dentry *dentry, struct inode *inode,
1683 			    struct delayed_call *done)
1684 {
1685 	return inode->i_link;
1686 }
1687 EXPORT_SYMBOL(simple_get_link);
1688 
1689 const struct inode_operations simple_symlink_inode_operations = {
1690 	.get_link = simple_get_link,
1691 };
1692 EXPORT_SYMBOL(simple_symlink_inode_operations);
1693 
1694 /*
1695  * Operations for a permanently empty directory.
1696  */
1697 static struct dentry *empty_dir_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1698 {
1699 	return ERR_PTR(-ENOENT);
1700 }
1701 
1702 static int empty_dir_getattr(struct mnt_idmap *idmap,
1703 			     const struct path *path, struct kstat *stat,
1704 			     u32 request_mask, unsigned int query_flags)
1705 {
1706 	struct inode *inode = d_inode(path->dentry);
1707 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1708 	return 0;
1709 }
1710 
1711 static int empty_dir_setattr(struct mnt_idmap *idmap,
1712 			     struct dentry *dentry, struct iattr *attr)
1713 {
1714 	return -EPERM;
1715 }
1716 
1717 static ssize_t empty_dir_listxattr(struct dentry *dentry, char *list, size_t size)
1718 {
1719 	return -EOPNOTSUPP;
1720 }
1721 
1722 static const struct inode_operations empty_dir_inode_operations = {
1723 	.lookup		= empty_dir_lookup,
1724 	.permission	= generic_permission,
1725 	.setattr	= empty_dir_setattr,
1726 	.getattr	= empty_dir_getattr,
1727 	.listxattr	= empty_dir_listxattr,
1728 };
1729 
1730 static loff_t empty_dir_llseek(struct file *file, loff_t offset, int whence)
1731 {
1732 	/* An empty directory has two entries . and .. at offsets 0 and 1 */
1733 	return generic_file_llseek_size(file, offset, whence, 2, 2);
1734 }
1735 
1736 static int empty_dir_readdir(struct file *file, struct dir_context *ctx)
1737 {
1738 	dir_emit_dots(file, ctx);
1739 	return 0;
1740 }
1741 
1742 static const struct file_operations empty_dir_operations = {
1743 	.llseek		= empty_dir_llseek,
1744 	.read		= generic_read_dir,
1745 	.iterate_shared	= empty_dir_readdir,
1746 	.fsync		= noop_fsync,
1747 };
1748 
1749 
1750 void make_empty_dir_inode(struct inode *inode)
1751 {
1752 	set_nlink(inode, 2);
1753 	inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO;
1754 	inode->i_uid = GLOBAL_ROOT_UID;
1755 	inode->i_gid = GLOBAL_ROOT_GID;
1756 	inode->i_rdev = 0;
1757 	inode->i_size = 0;
1758 	inode->i_blkbits = PAGE_SHIFT;
1759 	inode->i_blocks = 0;
1760 
1761 	inode->i_op = &empty_dir_inode_operations;
1762 	inode->i_opflags &= ~IOP_XATTR;
1763 	inode->i_fop = &empty_dir_operations;
1764 }
1765 
1766 bool is_empty_dir_inode(struct inode *inode)
1767 {
1768 	return (inode->i_fop == &empty_dir_operations) &&
1769 		(inode->i_op == &empty_dir_inode_operations);
1770 }
1771 
1772 #if IS_ENABLED(CONFIG_UNICODE)
1773 /**
1774  * generic_ci_d_compare - generic d_compare implementation for casefolding filesystems
1775  * @dentry:	dentry whose name we are checking against
1776  * @len:	len of name of dentry
1777  * @str:	str pointer to name of dentry
1778  * @name:	Name to compare against
1779  *
1780  * Return: 0 if names match, 1 if mismatch, or -ERRNO
1781  */
1782 static int generic_ci_d_compare(const struct dentry *dentry, unsigned int len,
1783 				const char *str, const struct qstr *name)
1784 {
1785 	const struct dentry *parent;
1786 	const struct inode *dir;
1787 	char strbuf[DNAME_INLINE_LEN];
1788 	struct qstr qstr;
1789 
1790 	/*
1791 	 * Attempt a case-sensitive match first. It is cheaper and
1792 	 * should cover most lookups, including all the sane
1793 	 * applications that expect a case-sensitive filesystem.
1794 	 *
1795 	 * This comparison is safe under RCU because the caller
1796 	 * guarantees the consistency between str and len. See
1797 	 * __d_lookup_rcu_op_compare() for details.
1798 	 */
1799 	if (len == name->len && !memcmp(str, name->name, len))
1800 		return 0;
1801 
1802 	parent = READ_ONCE(dentry->d_parent);
1803 	dir = READ_ONCE(parent->d_inode);
1804 	if (!dir || !IS_CASEFOLDED(dir))
1805 		return 1;
1806 
1807 	/*
1808 	 * If the dentry name is stored in-line, then it may be concurrently
1809 	 * modified by a rename.  If this happens, the VFS will eventually retry
1810 	 * the lookup, so it doesn't matter what ->d_compare() returns.
1811 	 * However, it's unsafe to call utf8_strncasecmp() with an unstable
1812 	 * string.  Therefore, we have to copy the name into a temporary buffer.
1813 	 */
1814 	if (len <= DNAME_INLINE_LEN - 1) {
1815 		memcpy(strbuf, str, len);
1816 		strbuf[len] = 0;
1817 		str = strbuf;
1818 		/* prevent compiler from optimizing out the temporary buffer */
1819 		barrier();
1820 	}
1821 	qstr.len = len;
1822 	qstr.name = str;
1823 
1824 	return utf8_strncasecmp(dentry->d_sb->s_encoding, name, &qstr);
1825 }
1826 
1827 /**
1828  * generic_ci_d_hash - generic d_hash implementation for casefolding filesystems
1829  * @dentry:	dentry of the parent directory
1830  * @str:	qstr of name whose hash we should fill in
1831  *
1832  * Return: 0 if hash was successful or unchanged, and -EINVAL on error
1833  */
1834 static int generic_ci_d_hash(const struct dentry *dentry, struct qstr *str)
1835 {
1836 	const struct inode *dir = READ_ONCE(dentry->d_inode);
1837 	struct super_block *sb = dentry->d_sb;
1838 	const struct unicode_map *um = sb->s_encoding;
1839 	int ret;
1840 
1841 	if (!dir || !IS_CASEFOLDED(dir))
1842 		return 0;
1843 
1844 	ret = utf8_casefold_hash(um, dentry, str);
1845 	if (ret < 0 && sb_has_strict_encoding(sb))
1846 		return -EINVAL;
1847 	return 0;
1848 }
1849 
1850 static const struct dentry_operations generic_ci_dentry_ops = {
1851 	.d_hash = generic_ci_d_hash,
1852 	.d_compare = generic_ci_d_compare,
1853 #ifdef CONFIG_FS_ENCRYPTION
1854 	.d_revalidate = fscrypt_d_revalidate,
1855 #endif
1856 };
1857 
1858 /**
1859  * generic_ci_match() - Match a name (case-insensitively) with a dirent.
1860  * This is a filesystem helper for comparison with directory entries.
1861  * generic_ci_d_compare should be used in VFS' ->d_compare instead.
1862  *
1863  * @parent: Inode of the parent of the dirent under comparison
1864  * @name: name under lookup.
1865  * @folded_name: Optional pre-folded name under lookup
1866  * @de_name: Dirent name.
1867  * @de_name_len: dirent name length.
1868  *
1869  * Test whether a case-insensitive directory entry matches the filename
1870  * being searched.  If @folded_name is provided, it is used instead of
1871  * recalculating the casefold of @name.
1872  *
1873  * Return: > 0 if the directory entry matches, 0 if it doesn't match, or
1874  * < 0 on error.
1875  */
1876 int generic_ci_match(const struct inode *parent,
1877 		     const struct qstr *name,
1878 		     const struct qstr *folded_name,
1879 		     const u8 *de_name, u32 de_name_len)
1880 {
1881 	const struct super_block *sb = parent->i_sb;
1882 	const struct unicode_map *um = sb->s_encoding;
1883 	struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1884 	struct qstr dirent = QSTR_INIT(de_name, de_name_len);
1885 	int res = 0;
1886 
1887 	if (IS_ENCRYPTED(parent)) {
1888 		const struct fscrypt_str encrypted_name =
1889 			FSTR_INIT((u8 *) de_name, de_name_len);
1890 
1891 		if (WARN_ON_ONCE(!fscrypt_has_encryption_key(parent)))
1892 			return -EINVAL;
1893 
1894 		decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1895 		if (!decrypted_name.name)
1896 			return -ENOMEM;
1897 		res = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1898 						&decrypted_name);
1899 		if (res < 0) {
1900 			kfree(decrypted_name.name);
1901 			return res;
1902 		}
1903 		dirent.name = decrypted_name.name;
1904 		dirent.len = decrypted_name.len;
1905 	}
1906 
1907 	/*
1908 	 * Attempt a case-sensitive match first. It is cheaper and
1909 	 * should cover most lookups, including all the sane
1910 	 * applications that expect a case-sensitive filesystem.
1911 	 */
1912 
1913 	if (dirent.len == name->len &&
1914 	    !memcmp(name->name, dirent.name, dirent.len))
1915 		goto out;
1916 
1917 	if (folded_name->name)
1918 		res = utf8_strncasecmp_folded(um, folded_name, &dirent);
1919 	else
1920 		res = utf8_strncasecmp(um, name, &dirent);
1921 
1922 out:
1923 	kfree(decrypted_name.name);
1924 	if (res < 0 && sb_has_strict_encoding(sb)) {
1925 		pr_err_ratelimited("Directory contains filename that is invalid UTF-8");
1926 		return 0;
1927 	}
1928 	return !res;
1929 }
1930 EXPORT_SYMBOL(generic_ci_match);
1931 #endif
1932 
1933 #ifdef CONFIG_FS_ENCRYPTION
1934 static const struct dentry_operations generic_encrypted_dentry_ops = {
1935 	.d_revalidate = fscrypt_d_revalidate,
1936 };
1937 #endif
1938 
1939 /**
1940  * generic_set_sb_d_ops - helper for choosing the set of
1941  * filesystem-wide dentry operations for the enabled features
1942  * @sb: superblock to be configured
1943  *
1944  * Filesystems supporting casefolding and/or fscrypt can call this
1945  * helper at mount-time to configure sb->s_d_op to best set of dentry
1946  * operations required for the enabled features. The helper must be
1947  * called after these have been configured, but before the root dentry
1948  * is created.
1949  */
1950 void generic_set_sb_d_ops(struct super_block *sb)
1951 {
1952 #if IS_ENABLED(CONFIG_UNICODE)
1953 	if (sb->s_encoding) {
1954 		sb->s_d_op = &generic_ci_dentry_ops;
1955 		return;
1956 	}
1957 #endif
1958 #ifdef CONFIG_FS_ENCRYPTION
1959 	if (sb->s_cop) {
1960 		sb->s_d_op = &generic_encrypted_dentry_ops;
1961 		return;
1962 	}
1963 #endif
1964 }
1965 EXPORT_SYMBOL(generic_set_sb_d_ops);
1966 
1967 /**
1968  * inode_maybe_inc_iversion - increments i_version
1969  * @inode: inode with the i_version that should be updated
1970  * @force: increment the counter even if it's not necessary?
1971  *
1972  * Every time the inode is modified, the i_version field must be seen to have
1973  * changed by any observer.
1974  *
1975  * If "force" is set or the QUERIED flag is set, then ensure that we increment
1976  * the value, and clear the queried flag.
1977  *
1978  * In the common case where neither is set, then we can return "false" without
1979  * updating i_version.
1980  *
1981  * If this function returns false, and no other metadata has changed, then we
1982  * can avoid logging the metadata.
1983  */
1984 bool inode_maybe_inc_iversion(struct inode *inode, bool force)
1985 {
1986 	u64 cur, new;
1987 
1988 	/*
1989 	 * The i_version field is not strictly ordered with any other inode
1990 	 * information, but the legacy inode_inc_iversion code used a spinlock
1991 	 * to serialize increments.
1992 	 *
1993 	 * Here, we add full memory barriers to ensure that any de-facto
1994 	 * ordering with other info is preserved.
1995 	 *
1996 	 * This barrier pairs with the barrier in inode_query_iversion()
1997 	 */
1998 	smp_mb();
1999 	cur = inode_peek_iversion_raw(inode);
2000 	do {
2001 		/* If flag is clear then we needn't do anything */
2002 		if (!force && !(cur & I_VERSION_QUERIED))
2003 			return false;
2004 
2005 		/* Since lowest bit is flag, add 2 to avoid it */
2006 		new = (cur & ~I_VERSION_QUERIED) + I_VERSION_INCREMENT;
2007 	} while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
2008 	return true;
2009 }
2010 EXPORT_SYMBOL(inode_maybe_inc_iversion);
2011 
2012 /**
2013  * inode_query_iversion - read i_version for later use
2014  * @inode: inode from which i_version should be read
2015  *
2016  * Read the inode i_version counter. This should be used by callers that wish
2017  * to store the returned i_version for later comparison. This will guarantee
2018  * that a later query of the i_version will result in a different value if
2019  * anything has changed.
2020  *
2021  * In this implementation, we fetch the current value, set the QUERIED flag and
2022  * then try to swap it into place with a cmpxchg, if it wasn't already set. If
2023  * that fails, we try again with the newly fetched value from the cmpxchg.
2024  */
2025 u64 inode_query_iversion(struct inode *inode)
2026 {
2027 	u64 cur, new;
2028 
2029 	cur = inode_peek_iversion_raw(inode);
2030 	do {
2031 		/* If flag is already set, then no need to swap */
2032 		if (cur & I_VERSION_QUERIED) {
2033 			/*
2034 			 * This barrier (and the implicit barrier in the
2035 			 * cmpxchg below) pairs with the barrier in
2036 			 * inode_maybe_inc_iversion().
2037 			 */
2038 			smp_mb();
2039 			break;
2040 		}
2041 
2042 		new = cur | I_VERSION_QUERIED;
2043 	} while (!atomic64_try_cmpxchg(&inode->i_version, &cur, new));
2044 	return cur >> I_VERSION_QUERIED_SHIFT;
2045 }
2046 EXPORT_SYMBOL(inode_query_iversion);
2047 
2048 ssize_t direct_write_fallback(struct kiocb *iocb, struct iov_iter *iter,
2049 		ssize_t direct_written, ssize_t buffered_written)
2050 {
2051 	struct address_space *mapping = iocb->ki_filp->f_mapping;
2052 	loff_t pos = iocb->ki_pos - buffered_written;
2053 	loff_t end = iocb->ki_pos - 1;
2054 	int err;
2055 
2056 	/*
2057 	 * If the buffered write fallback returned an error, we want to return
2058 	 * the number of bytes which were written by direct I/O, or the error
2059 	 * code if that was zero.
2060 	 *
2061 	 * Note that this differs from normal direct-io semantics, which will
2062 	 * return -EFOO even if some bytes were written.
2063 	 */
2064 	if (unlikely(buffered_written < 0)) {
2065 		if (direct_written)
2066 			return direct_written;
2067 		return buffered_written;
2068 	}
2069 
2070 	/*
2071 	 * We need to ensure that the page cache pages are written to disk and
2072 	 * invalidated to preserve the expected O_DIRECT semantics.
2073 	 */
2074 	err = filemap_write_and_wait_range(mapping, pos, end);
2075 	if (err < 0) {
2076 		/*
2077 		 * We don't know how much we wrote, so just return the number of
2078 		 * bytes which were direct-written
2079 		 */
2080 		iocb->ki_pos -= buffered_written;
2081 		if (direct_written)
2082 			return direct_written;
2083 		return err;
2084 	}
2085 	invalidate_mapping_pages(mapping, pos >> PAGE_SHIFT, end >> PAGE_SHIFT);
2086 	return direct_written + buffered_written;
2087 }
2088 EXPORT_SYMBOL_GPL(direct_write_fallback);
2089 
2090 /**
2091  * simple_inode_init_ts - initialize the timestamps for a new inode
2092  * @inode: inode to be initialized
2093  *
2094  * When a new inode is created, most filesystems set the timestamps to the
2095  * current time. Add a helper to do this.
2096  */
2097 struct timespec64 simple_inode_init_ts(struct inode *inode)
2098 {
2099 	struct timespec64 ts = inode_set_ctime_current(inode);
2100 
2101 	inode_set_atime_to_ts(inode, ts);
2102 	inode_set_mtime_to_ts(inode, ts);
2103 	return ts;
2104 }
2105 EXPORT_SYMBOL(simple_inode_init_ts);
2106 
2107 static inline struct dentry *get_stashed_dentry(struct dentry *stashed)
2108 {
2109 	struct dentry *dentry;
2110 
2111 	guard(rcu)();
2112 	dentry = READ_ONCE(stashed);
2113 	if (!dentry)
2114 		return NULL;
2115 	if (!lockref_get_not_dead(&dentry->d_lockref))
2116 		return NULL;
2117 	return dentry;
2118 }
2119 
2120 static struct dentry *prepare_anon_dentry(struct dentry **stashed,
2121 					  struct super_block *sb,
2122 					  void *data)
2123 {
2124 	struct dentry *dentry;
2125 	struct inode *inode;
2126 	const struct stashed_operations *sops = sb->s_fs_info;
2127 	int ret;
2128 
2129 	inode = new_inode_pseudo(sb);
2130 	if (!inode) {
2131 		sops->put_data(data);
2132 		return ERR_PTR(-ENOMEM);
2133 	}
2134 
2135 	inode->i_flags |= S_IMMUTABLE;
2136 	inode->i_mode = S_IFREG;
2137 	simple_inode_init_ts(inode);
2138 
2139 	ret = sops->init_inode(inode, data);
2140 	if (ret < 0) {
2141 		iput(inode);
2142 		return ERR_PTR(ret);
2143 	}
2144 
2145 	/* Notice when this is changed. */
2146 	WARN_ON_ONCE(!S_ISREG(inode->i_mode));
2147 	WARN_ON_ONCE(!IS_IMMUTABLE(inode));
2148 
2149 	dentry = d_alloc_anon(sb);
2150 	if (!dentry) {
2151 		iput(inode);
2152 		return ERR_PTR(-ENOMEM);
2153 	}
2154 
2155 	/* Store address of location where dentry's supposed to be stashed. */
2156 	dentry->d_fsdata = stashed;
2157 
2158 	/* @data is now owned by the fs */
2159 	d_instantiate(dentry, inode);
2160 	return dentry;
2161 }
2162 
2163 static struct dentry *stash_dentry(struct dentry **stashed,
2164 				   struct dentry *dentry)
2165 {
2166 	guard(rcu)();
2167 	for (;;) {
2168 		struct dentry *old;
2169 
2170 		/* Assume any old dentry was cleared out. */
2171 		old = cmpxchg(stashed, NULL, dentry);
2172 		if (likely(!old))
2173 			return dentry;
2174 
2175 		/* Check if somebody else installed a reusable dentry. */
2176 		if (lockref_get_not_dead(&old->d_lockref))
2177 			return old;
2178 
2179 		/* There's an old dead dentry there, try to take it over. */
2180 		if (likely(try_cmpxchg(stashed, &old, dentry)))
2181 			return dentry;
2182 	}
2183 }
2184 
2185 /**
2186  * path_from_stashed - create path from stashed or new dentry
2187  * @stashed:    where to retrieve or stash dentry
2188  * @mnt:        mnt of the filesystems to use
2189  * @data:       data to store in inode->i_private
2190  * @path:       path to create
2191  *
2192  * The function tries to retrieve a stashed dentry from @stashed. If the dentry
2193  * is still valid then it will be reused. If the dentry isn't able the function
2194  * will allocate a new dentry and inode. It will then check again whether it
2195  * can reuse an existing dentry in case one has been added in the meantime or
2196  * update @stashed with the newly added dentry.
2197  *
2198  * Special-purpose helper for nsfs and pidfs.
2199  *
2200  * Return: On success zero and on failure a negative error is returned.
2201  */
2202 int path_from_stashed(struct dentry **stashed, struct vfsmount *mnt, void *data,
2203 		      struct path *path)
2204 {
2205 	struct dentry *dentry;
2206 	const struct stashed_operations *sops = mnt->mnt_sb->s_fs_info;
2207 
2208 	/* See if dentry can be reused. */
2209 	path->dentry = get_stashed_dentry(*stashed);
2210 	if (path->dentry) {
2211 		sops->put_data(data);
2212 		goto out_path;
2213 	}
2214 
2215 	/* Allocate a new dentry. */
2216 	dentry = prepare_anon_dentry(stashed, mnt->mnt_sb, data);
2217 	if (IS_ERR(dentry))
2218 		return PTR_ERR(dentry);
2219 
2220 	/* Added a new dentry. @data is now owned by the filesystem. */
2221 	path->dentry = stash_dentry(stashed, dentry);
2222 	if (path->dentry != dentry)
2223 		dput(dentry);
2224 
2225 out_path:
2226 	WARN_ON_ONCE(path->dentry->d_fsdata != stashed);
2227 	WARN_ON_ONCE(d_inode(path->dentry)->i_private != data);
2228 	path->mnt = mntget(mnt);
2229 	return 0;
2230 }
2231 
2232 void stashed_dentry_prune(struct dentry *dentry)
2233 {
2234 	struct dentry **stashed = dentry->d_fsdata;
2235 	struct inode *inode = d_inode(dentry);
2236 
2237 	if (WARN_ON_ONCE(!stashed))
2238 		return;
2239 
2240 	if (!inode)
2241 		return;
2242 
2243 	/*
2244 	 * Only replace our own @dentry as someone else might've
2245 	 * already cleared out @dentry and stashed their own
2246 	 * dentry in there.
2247 	 */
2248 	cmpxchg(stashed, dentry, NULL);
2249 }
2250