xref: /linux/fs/libfs.c (revision 1fbde52bde73e5a4c90577bce935e966a1ba1387)
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/writeback.h>
19  #include <linux/buffer_head.h> /* sync_mapping_buffers */
20  #include <linux/fs_context.h>
21  #include <linux/pseudo_fs.h>
22  #include <linux/fsnotify.h>
23  #include <linux/unicode.h>
24  #include <linux/fscrypt.h>
25  
26  #include <linux/uaccess.h>
27  
28  #include "internal.h"
29  
30  int simple_getattr(struct user_namespace *mnt_userns, const struct path *path,
31  		   struct kstat *stat, u32 request_mask,
32  		   unsigned int query_flags)
33  {
34  	struct inode *inode = d_inode(path->dentry);
35  	generic_fillattr(&init_user_ns, inode, stat);
36  	stat->blocks = inode->i_mapping->nrpages << (PAGE_SHIFT - 9);
37  	return 0;
38  }
39  EXPORT_SYMBOL(simple_getattr);
40  
41  int simple_statfs(struct dentry *dentry, struct kstatfs *buf)
42  {
43  	buf->f_type = dentry->d_sb->s_magic;
44  	buf->f_bsize = PAGE_SIZE;
45  	buf->f_namelen = NAME_MAX;
46  	return 0;
47  }
48  EXPORT_SYMBOL(simple_statfs);
49  
50  /*
51   * Retaining negative dentries for an in-memory filesystem just wastes
52   * memory and lookup time: arrange for them to be deleted immediately.
53   */
54  int always_delete_dentry(const struct dentry *dentry)
55  {
56  	return 1;
57  }
58  EXPORT_SYMBOL(always_delete_dentry);
59  
60  const struct dentry_operations simple_dentry_operations = {
61  	.d_delete = always_delete_dentry,
62  };
63  EXPORT_SYMBOL(simple_dentry_operations);
64  
65  /*
66   * Lookup the data. This is trivial - if the dentry didn't already
67   * exist, we know it is negative.  Set d_op to delete negative dentries.
68   */
69  struct dentry *simple_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
70  {
71  	if (dentry->d_name.len > NAME_MAX)
72  		return ERR_PTR(-ENAMETOOLONG);
73  	if (!dentry->d_sb->s_d_op)
74  		d_set_d_op(dentry, &simple_dentry_operations);
75  	d_add(dentry, NULL);
76  	return NULL;
77  }
78  EXPORT_SYMBOL(simple_lookup);
79  
80  int dcache_dir_open(struct inode *inode, struct file *file)
81  {
82  	file->private_data = d_alloc_cursor(file->f_path.dentry);
83  
84  	return file->private_data ? 0 : -ENOMEM;
85  }
86  EXPORT_SYMBOL(dcache_dir_open);
87  
88  int dcache_dir_close(struct inode *inode, struct file *file)
89  {
90  	dput(file->private_data);
91  	return 0;
92  }
93  EXPORT_SYMBOL(dcache_dir_close);
94  
95  /* parent is locked at least shared */
96  /*
97   * Returns an element of siblings' list.
98   * We are looking for <count>th positive after <p>; if
99   * found, dentry is grabbed and returned to caller.
100   * If no such element exists, NULL is returned.
101   */
102  static struct dentry *scan_positives(struct dentry *cursor,
103  					struct list_head *p,
104  					loff_t count,
105  					struct dentry *last)
106  {
107  	struct dentry *dentry = cursor->d_parent, *found = NULL;
108  
109  	spin_lock(&dentry->d_lock);
110  	while ((p = p->next) != &dentry->d_subdirs) {
111  		struct dentry *d = list_entry(p, struct dentry, d_child);
112  		// we must at least skip cursors, to avoid livelocks
113  		if (d->d_flags & DCACHE_DENTRY_CURSOR)
114  			continue;
115  		if (simple_positive(d) && !--count) {
116  			spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
117  			if (simple_positive(d))
118  				found = dget_dlock(d);
119  			spin_unlock(&d->d_lock);
120  			if (likely(found))
121  				break;
122  			count = 1;
123  		}
124  		if (need_resched()) {
125  			list_move(&cursor->d_child, p);
126  			p = &cursor->d_child;
127  			spin_unlock(&dentry->d_lock);
128  			cond_resched();
129  			spin_lock(&dentry->d_lock);
130  		}
131  	}
132  	spin_unlock(&dentry->d_lock);
133  	dput(last);
134  	return found;
135  }
136  
137  loff_t dcache_dir_lseek(struct file *file, loff_t offset, int whence)
138  {
139  	struct dentry *dentry = file->f_path.dentry;
140  	switch (whence) {
141  		case 1:
142  			offset += file->f_pos;
143  			fallthrough;
144  		case 0:
145  			if (offset >= 0)
146  				break;
147  			fallthrough;
148  		default:
149  			return -EINVAL;
150  	}
151  	if (offset != file->f_pos) {
152  		struct dentry *cursor = file->private_data;
153  		struct dentry *to = NULL;
154  
155  		inode_lock_shared(dentry->d_inode);
156  
157  		if (offset > 2)
158  			to = scan_positives(cursor, &dentry->d_subdirs,
159  					    offset - 2, NULL);
160  		spin_lock(&dentry->d_lock);
161  		if (to)
162  			list_move(&cursor->d_child, &to->d_child);
163  		else
164  			list_del_init(&cursor->d_child);
165  		spin_unlock(&dentry->d_lock);
166  		dput(to);
167  
168  		file->f_pos = offset;
169  
170  		inode_unlock_shared(dentry->d_inode);
171  	}
172  	return offset;
173  }
174  EXPORT_SYMBOL(dcache_dir_lseek);
175  
176  /* Relationship between i_mode and the DT_xxx types */
177  static inline unsigned char dt_type(struct inode *inode)
178  {
179  	return (inode->i_mode >> 12) & 15;
180  }
181  
182  /*
183   * Directory is locked and all positive dentries in it are safe, since
184   * for ramfs-type trees they can't go away without unlink() or rmdir(),
185   * both impossible due to the lock on directory.
186   */
187  
188  int dcache_readdir(struct file *file, struct dir_context *ctx)
189  {
190  	struct dentry *dentry = file->f_path.dentry;
191  	struct dentry *cursor = file->private_data;
192  	struct list_head *anchor = &dentry->d_subdirs;
193  	struct dentry *next = NULL;
194  	struct list_head *p;
195  
196  	if (!dir_emit_dots(file, ctx))
197  		return 0;
198  
199  	if (ctx->pos == 2)
200  		p = anchor;
201  	else if (!list_empty(&cursor->d_child))
202  		p = &cursor->d_child;
203  	else
204  		return 0;
205  
206  	while ((next = scan_positives(cursor, p, 1, next)) != NULL) {
207  		if (!dir_emit(ctx, next->d_name.name, next->d_name.len,
208  			      d_inode(next)->i_ino, dt_type(d_inode(next))))
209  			break;
210  		ctx->pos++;
211  		p = &next->d_child;
212  	}
213  	spin_lock(&dentry->d_lock);
214  	if (next)
215  		list_move_tail(&cursor->d_child, &next->d_child);
216  	else
217  		list_del_init(&cursor->d_child);
218  	spin_unlock(&dentry->d_lock);
219  	dput(next);
220  
221  	return 0;
222  }
223  EXPORT_SYMBOL(dcache_readdir);
224  
225  ssize_t generic_read_dir(struct file *filp, char __user *buf, size_t siz, loff_t *ppos)
226  {
227  	return -EISDIR;
228  }
229  EXPORT_SYMBOL(generic_read_dir);
230  
231  const struct file_operations simple_dir_operations = {
232  	.open		= dcache_dir_open,
233  	.release	= dcache_dir_close,
234  	.llseek		= dcache_dir_lseek,
235  	.read		= generic_read_dir,
236  	.iterate_shared	= dcache_readdir,
237  	.fsync		= noop_fsync,
238  };
239  EXPORT_SYMBOL(simple_dir_operations);
240  
241  const struct inode_operations simple_dir_inode_operations = {
242  	.lookup		= simple_lookup,
243  };
244  EXPORT_SYMBOL(simple_dir_inode_operations);
245  
246  static struct dentry *find_next_child(struct dentry *parent, struct dentry *prev)
247  {
248  	struct dentry *child = NULL;
249  	struct list_head *p = prev ? &prev->d_child : &parent->d_subdirs;
250  
251  	spin_lock(&parent->d_lock);
252  	while ((p = p->next) != &parent->d_subdirs) {
253  		struct dentry *d = container_of(p, struct dentry, d_child);
254  		if (simple_positive(d)) {
255  			spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
256  			if (simple_positive(d))
257  				child = dget_dlock(d);
258  			spin_unlock(&d->d_lock);
259  			if (likely(child))
260  				break;
261  		}
262  	}
263  	spin_unlock(&parent->d_lock);
264  	dput(prev);
265  	return child;
266  }
267  
268  void simple_recursive_removal(struct dentry *dentry,
269                                void (*callback)(struct dentry *))
270  {
271  	struct dentry *this = dget(dentry);
272  	while (true) {
273  		struct dentry *victim = NULL, *child;
274  		struct inode *inode = this->d_inode;
275  
276  		inode_lock(inode);
277  		if (d_is_dir(this))
278  			inode->i_flags |= S_DEAD;
279  		while ((child = find_next_child(this, victim)) == NULL) {
280  			// kill and ascend
281  			// update metadata while it's still locked
282  			inode->i_ctime = current_time(inode);
283  			clear_nlink(inode);
284  			inode_unlock(inode);
285  			victim = this;
286  			this = this->d_parent;
287  			inode = this->d_inode;
288  			inode_lock(inode);
289  			if (simple_positive(victim)) {
290  				d_invalidate(victim);	// avoid lost mounts
291  				if (d_is_dir(victim))
292  					fsnotify_rmdir(inode, victim);
293  				else
294  					fsnotify_unlink(inode, victim);
295  				if (callback)
296  					callback(victim);
297  				dput(victim);		// unpin it
298  			}
299  			if (victim == dentry) {
300  				inode->i_ctime = inode->i_mtime =
301  					current_time(inode);
302  				if (d_is_dir(dentry))
303  					drop_nlink(inode);
304  				inode_unlock(inode);
305  				dput(dentry);
306  				return;
307  			}
308  		}
309  		inode_unlock(inode);
310  		this = child;
311  	}
312  }
313  EXPORT_SYMBOL(simple_recursive_removal);
314  
315  static const struct super_operations simple_super_operations = {
316  	.statfs		= simple_statfs,
317  };
318  
319  static int pseudo_fs_fill_super(struct super_block *s, struct fs_context *fc)
320  {
321  	struct pseudo_fs_context *ctx = fc->fs_private;
322  	struct inode *root;
323  
324  	s->s_maxbytes = MAX_LFS_FILESIZE;
325  	s->s_blocksize = PAGE_SIZE;
326  	s->s_blocksize_bits = PAGE_SHIFT;
327  	s->s_magic = ctx->magic;
328  	s->s_op = ctx->ops ?: &simple_super_operations;
329  	s->s_xattr = ctx->xattr;
330  	s->s_time_gran = 1;
331  	root = new_inode(s);
332  	if (!root)
333  		return -ENOMEM;
334  
335  	/*
336  	 * since this is the first inode, make it number 1. New inodes created
337  	 * after this must take care not to collide with it (by passing
338  	 * max_reserved of 1 to iunique).
339  	 */
340  	root->i_ino = 1;
341  	root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
342  	root->i_atime = root->i_mtime = root->i_ctime = current_time(root);
343  	s->s_root = d_make_root(root);
344  	if (!s->s_root)
345  		return -ENOMEM;
346  	s->s_d_op = ctx->dops;
347  	return 0;
348  }
349  
350  static int pseudo_fs_get_tree(struct fs_context *fc)
351  {
352  	return get_tree_nodev(fc, pseudo_fs_fill_super);
353  }
354  
355  static void pseudo_fs_free(struct fs_context *fc)
356  {
357  	kfree(fc->fs_private);
358  }
359  
360  static const struct fs_context_operations pseudo_fs_context_ops = {
361  	.free		= pseudo_fs_free,
362  	.get_tree	= pseudo_fs_get_tree,
363  };
364  
365  /*
366   * Common helper for pseudo-filesystems (sockfs, pipefs, bdev - stuff that
367   * will never be mountable)
368   */
369  struct pseudo_fs_context *init_pseudo(struct fs_context *fc,
370  					unsigned long magic)
371  {
372  	struct pseudo_fs_context *ctx;
373  
374  	ctx = kzalloc(sizeof(struct pseudo_fs_context), GFP_KERNEL);
375  	if (likely(ctx)) {
376  		ctx->magic = magic;
377  		fc->fs_private = ctx;
378  		fc->ops = &pseudo_fs_context_ops;
379  		fc->sb_flags |= SB_NOUSER;
380  		fc->global = true;
381  	}
382  	return ctx;
383  }
384  EXPORT_SYMBOL(init_pseudo);
385  
386  int simple_open(struct inode *inode, struct file *file)
387  {
388  	if (inode->i_private)
389  		file->private_data = inode->i_private;
390  	return 0;
391  }
392  EXPORT_SYMBOL(simple_open);
393  
394  int simple_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
395  {
396  	struct inode *inode = d_inode(old_dentry);
397  
398  	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
399  	inc_nlink(inode);
400  	ihold(inode);
401  	dget(dentry);
402  	d_instantiate(dentry, inode);
403  	return 0;
404  }
405  EXPORT_SYMBOL(simple_link);
406  
407  int simple_empty(struct dentry *dentry)
408  {
409  	struct dentry *child;
410  	int ret = 0;
411  
412  	spin_lock(&dentry->d_lock);
413  	list_for_each_entry(child, &dentry->d_subdirs, d_child) {
414  		spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
415  		if (simple_positive(child)) {
416  			spin_unlock(&child->d_lock);
417  			goto out;
418  		}
419  		spin_unlock(&child->d_lock);
420  	}
421  	ret = 1;
422  out:
423  	spin_unlock(&dentry->d_lock);
424  	return ret;
425  }
426  EXPORT_SYMBOL(simple_empty);
427  
428  int simple_unlink(struct inode *dir, struct dentry *dentry)
429  {
430  	struct inode *inode = d_inode(dentry);
431  
432  	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
433  	drop_nlink(inode);
434  	dput(dentry);
435  	return 0;
436  }
437  EXPORT_SYMBOL(simple_unlink);
438  
439  int simple_rmdir(struct inode *dir, struct dentry *dentry)
440  {
441  	if (!simple_empty(dentry))
442  		return -ENOTEMPTY;
443  
444  	drop_nlink(d_inode(dentry));
445  	simple_unlink(dir, dentry);
446  	drop_nlink(dir);
447  	return 0;
448  }
449  EXPORT_SYMBOL(simple_rmdir);
450  
451  int simple_rename_exchange(struct inode *old_dir, struct dentry *old_dentry,
452  			   struct inode *new_dir, struct dentry *new_dentry)
453  {
454  	bool old_is_dir = d_is_dir(old_dentry);
455  	bool new_is_dir = d_is_dir(new_dentry);
456  
457  	if (old_dir != new_dir && old_is_dir != new_is_dir) {
458  		if (old_is_dir) {
459  			drop_nlink(old_dir);
460  			inc_nlink(new_dir);
461  		} else {
462  			drop_nlink(new_dir);
463  			inc_nlink(old_dir);
464  		}
465  	}
466  	old_dir->i_ctime = old_dir->i_mtime =
467  	new_dir->i_ctime = new_dir->i_mtime =
468  	d_inode(old_dentry)->i_ctime =
469  	d_inode(new_dentry)->i_ctime = current_time(old_dir);
470  
471  	return 0;
472  }
473  EXPORT_SYMBOL_GPL(simple_rename_exchange);
474  
475  int simple_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
476  		  struct dentry *old_dentry, struct inode *new_dir,
477  		  struct dentry *new_dentry, unsigned int flags)
478  {
479  	struct inode *inode = d_inode(old_dentry);
480  	int they_are_dirs = d_is_dir(old_dentry);
481  
482  	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
483  		return -EINVAL;
484  
485  	if (flags & RENAME_EXCHANGE)
486  		return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
487  
488  	if (!simple_empty(new_dentry))
489  		return -ENOTEMPTY;
490  
491  	if (d_really_is_positive(new_dentry)) {
492  		simple_unlink(new_dir, new_dentry);
493  		if (they_are_dirs) {
494  			drop_nlink(d_inode(new_dentry));
495  			drop_nlink(old_dir);
496  		}
497  	} else if (they_are_dirs) {
498  		drop_nlink(old_dir);
499  		inc_nlink(new_dir);
500  	}
501  
502  	old_dir->i_ctime = old_dir->i_mtime = new_dir->i_ctime =
503  		new_dir->i_mtime = inode->i_ctime = current_time(old_dir);
504  
505  	return 0;
506  }
507  EXPORT_SYMBOL(simple_rename);
508  
509  /**
510   * simple_setattr - setattr for simple filesystem
511   * @mnt_userns: user namespace of the target mount
512   * @dentry: dentry
513   * @iattr: iattr structure
514   *
515   * Returns 0 on success, -error on failure.
516   *
517   * simple_setattr is a simple ->setattr implementation without a proper
518   * implementation of size changes.
519   *
520   * It can either be used for in-memory filesystems or special files
521   * on simple regular filesystems.  Anything that needs to change on-disk
522   * or wire state on size changes needs its own setattr method.
523   */
524  int simple_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
525  		   struct iattr *iattr)
526  {
527  	struct inode *inode = d_inode(dentry);
528  	int error;
529  
530  	error = setattr_prepare(mnt_userns, dentry, iattr);
531  	if (error)
532  		return error;
533  
534  	if (iattr->ia_valid & ATTR_SIZE)
535  		truncate_setsize(inode, iattr->ia_size);
536  	setattr_copy(mnt_userns, inode, iattr);
537  	mark_inode_dirty(inode);
538  	return 0;
539  }
540  EXPORT_SYMBOL(simple_setattr);
541  
542  static int simple_read_folio(struct file *file, struct folio *folio)
543  {
544  	folio_zero_range(folio, 0, folio_size(folio));
545  	flush_dcache_folio(folio);
546  	folio_mark_uptodate(folio);
547  	folio_unlock(folio);
548  	return 0;
549  }
550  
551  int simple_write_begin(struct file *file, struct address_space *mapping,
552  			loff_t pos, unsigned len,
553  			struct page **pagep, void **fsdata)
554  {
555  	struct page *page;
556  	pgoff_t index;
557  
558  	index = pos >> PAGE_SHIFT;
559  
560  	page = grab_cache_page_write_begin(mapping, index);
561  	if (!page)
562  		return -ENOMEM;
563  
564  	*pagep = page;
565  
566  	if (!PageUptodate(page) && (len != PAGE_SIZE)) {
567  		unsigned from = pos & (PAGE_SIZE - 1);
568  
569  		zero_user_segments(page, 0, from, from + len, PAGE_SIZE);
570  	}
571  	return 0;
572  }
573  EXPORT_SYMBOL(simple_write_begin);
574  
575  /**
576   * simple_write_end - .write_end helper for non-block-device FSes
577   * @file: See .write_end of address_space_operations
578   * @mapping: 		"
579   * @pos: 		"
580   * @len: 		"
581   * @copied: 		"
582   * @page: 		"
583   * @fsdata: 		"
584   *
585   * simple_write_end does the minimum needed for updating a page after writing is
586   * done. It has the same API signature as the .write_end of
587   * address_space_operations vector. So it can just be set onto .write_end for
588   * FSes that don't need any other processing. i_mutex is assumed to be held.
589   * Block based filesystems should use generic_write_end().
590   * NOTE: Even though i_size might get updated by this function, mark_inode_dirty
591   * is not called, so a filesystem that actually does store data in .write_inode
592   * should extend on what's done here with a call to mark_inode_dirty() in the
593   * case that i_size has changed.
594   *
595   * Use *ONLY* with simple_read_folio()
596   */
597  static int simple_write_end(struct file *file, struct address_space *mapping,
598  			loff_t pos, unsigned len, unsigned copied,
599  			struct page *page, void *fsdata)
600  {
601  	struct inode *inode = page->mapping->host;
602  	loff_t last_pos = pos + copied;
603  
604  	/* zero the stale part of the page if we did a short copy */
605  	if (!PageUptodate(page)) {
606  		if (copied < len) {
607  			unsigned from = pos & (PAGE_SIZE - 1);
608  
609  			zero_user(page, from + copied, len - copied);
610  		}
611  		SetPageUptodate(page);
612  	}
613  	/*
614  	 * No need to use i_size_read() here, the i_size
615  	 * cannot change under us because we hold the i_mutex.
616  	 */
617  	if (last_pos > inode->i_size)
618  		i_size_write(inode, last_pos);
619  
620  	set_page_dirty(page);
621  	unlock_page(page);
622  	put_page(page);
623  
624  	return copied;
625  }
626  
627  /*
628   * Provides ramfs-style behavior: data in the pagecache, but no writeback.
629   */
630  const struct address_space_operations ram_aops = {
631  	.read_folio	= simple_read_folio,
632  	.write_begin	= simple_write_begin,
633  	.write_end	= simple_write_end,
634  	.dirty_folio	= noop_dirty_folio,
635  };
636  EXPORT_SYMBOL(ram_aops);
637  
638  /*
639   * the inodes created here are not hashed. If you use iunique to generate
640   * unique inode values later for this filesystem, then you must take care
641   * to pass it an appropriate max_reserved value to avoid collisions.
642   */
643  int simple_fill_super(struct super_block *s, unsigned long magic,
644  		      const struct tree_descr *files)
645  {
646  	struct inode *inode;
647  	struct dentry *root;
648  	struct dentry *dentry;
649  	int i;
650  
651  	s->s_blocksize = PAGE_SIZE;
652  	s->s_blocksize_bits = PAGE_SHIFT;
653  	s->s_magic = magic;
654  	s->s_op = &simple_super_operations;
655  	s->s_time_gran = 1;
656  
657  	inode = new_inode(s);
658  	if (!inode)
659  		return -ENOMEM;
660  	/*
661  	 * because the root inode is 1, the files array must not contain an
662  	 * entry at index 1
663  	 */
664  	inode->i_ino = 1;
665  	inode->i_mode = S_IFDIR | 0755;
666  	inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
667  	inode->i_op = &simple_dir_inode_operations;
668  	inode->i_fop = &simple_dir_operations;
669  	set_nlink(inode, 2);
670  	root = d_make_root(inode);
671  	if (!root)
672  		return -ENOMEM;
673  	for (i = 0; !files->name || files->name[0]; i++, files++) {
674  		if (!files->name)
675  			continue;
676  
677  		/* warn if it tries to conflict with the root inode */
678  		if (unlikely(i == 1))
679  			printk(KERN_WARNING "%s: %s passed in a files array"
680  				"with an index of 1!\n", __func__,
681  				s->s_type->name);
682  
683  		dentry = d_alloc_name(root, files->name);
684  		if (!dentry)
685  			goto out;
686  		inode = new_inode(s);
687  		if (!inode) {
688  			dput(dentry);
689  			goto out;
690  		}
691  		inode->i_mode = S_IFREG | files->mode;
692  		inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
693  		inode->i_fop = files->ops;
694  		inode->i_ino = i;
695  		d_add(dentry, inode);
696  	}
697  	s->s_root = root;
698  	return 0;
699  out:
700  	d_genocide(root);
701  	shrink_dcache_parent(root);
702  	dput(root);
703  	return -ENOMEM;
704  }
705  EXPORT_SYMBOL(simple_fill_super);
706  
707  static DEFINE_SPINLOCK(pin_fs_lock);
708  
709  int simple_pin_fs(struct file_system_type *type, struct vfsmount **mount, int *count)
710  {
711  	struct vfsmount *mnt = NULL;
712  	spin_lock(&pin_fs_lock);
713  	if (unlikely(!*mount)) {
714  		spin_unlock(&pin_fs_lock);
715  		mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
716  		if (IS_ERR(mnt))
717  			return PTR_ERR(mnt);
718  		spin_lock(&pin_fs_lock);
719  		if (!*mount)
720  			*mount = mnt;
721  	}
722  	mntget(*mount);
723  	++*count;
724  	spin_unlock(&pin_fs_lock);
725  	mntput(mnt);
726  	return 0;
727  }
728  EXPORT_SYMBOL(simple_pin_fs);
729  
730  void simple_release_fs(struct vfsmount **mount, int *count)
731  {
732  	struct vfsmount *mnt;
733  	spin_lock(&pin_fs_lock);
734  	mnt = *mount;
735  	if (!--*count)
736  		*mount = NULL;
737  	spin_unlock(&pin_fs_lock);
738  	mntput(mnt);
739  }
740  EXPORT_SYMBOL(simple_release_fs);
741  
742  /**
743   * simple_read_from_buffer - copy data from the buffer to user space
744   * @to: the user space buffer to read to
745   * @count: the maximum number of bytes to read
746   * @ppos: the current position in the buffer
747   * @from: the buffer to read from
748   * @available: the size of the buffer
749   *
750   * The simple_read_from_buffer() function reads up to @count bytes from the
751   * buffer @from at offset @ppos into the user space address starting at @to.
752   *
753   * On success, the number of bytes read is returned and the offset @ppos is
754   * advanced by this number, or negative value is returned on error.
755   **/
756  ssize_t simple_read_from_buffer(void __user *to, size_t count, loff_t *ppos,
757  				const void *from, size_t available)
758  {
759  	loff_t pos = *ppos;
760  	size_t ret;
761  
762  	if (pos < 0)
763  		return -EINVAL;
764  	if (pos >= available || !count)
765  		return 0;
766  	if (count > available - pos)
767  		count = available - pos;
768  	ret = copy_to_user(to, from + pos, count);
769  	if (ret == count)
770  		return -EFAULT;
771  	count -= ret;
772  	*ppos = pos + count;
773  	return count;
774  }
775  EXPORT_SYMBOL(simple_read_from_buffer);
776  
777  /**
778   * simple_write_to_buffer - copy data from user space to the buffer
779   * @to: the buffer to write to
780   * @available: the size of the buffer
781   * @ppos: the current position in the buffer
782   * @from: the user space buffer to read from
783   * @count: the maximum number of bytes to read
784   *
785   * The simple_write_to_buffer() function reads up to @count bytes from the user
786   * space address starting at @from into the buffer @to at offset @ppos.
787   *
788   * On success, the number of bytes written is returned and the offset @ppos is
789   * advanced by this number, or negative value is returned on error.
790   **/
791  ssize_t simple_write_to_buffer(void *to, size_t available, loff_t *ppos,
792  		const void __user *from, size_t count)
793  {
794  	loff_t pos = *ppos;
795  	size_t res;
796  
797  	if (pos < 0)
798  		return -EINVAL;
799  	if (pos >= available || !count)
800  		return 0;
801  	if (count > available - pos)
802  		count = available - pos;
803  	res = copy_from_user(to + pos, from, count);
804  	if (res == count)
805  		return -EFAULT;
806  	count -= res;
807  	*ppos = pos + count;
808  	return count;
809  }
810  EXPORT_SYMBOL(simple_write_to_buffer);
811  
812  /**
813   * memory_read_from_buffer - copy data from the buffer
814   * @to: the kernel space buffer to read to
815   * @count: the maximum number of bytes to read
816   * @ppos: the current position in the buffer
817   * @from: the buffer to read from
818   * @available: the size of the buffer
819   *
820   * The memory_read_from_buffer() function reads up to @count bytes from the
821   * buffer @from at offset @ppos into the kernel space address starting at @to.
822   *
823   * On success, the number of bytes read is returned and the offset @ppos is
824   * advanced by this number, or negative value is returned on error.
825   **/
826  ssize_t memory_read_from_buffer(void *to, size_t count, loff_t *ppos,
827  				const void *from, size_t available)
828  {
829  	loff_t pos = *ppos;
830  
831  	if (pos < 0)
832  		return -EINVAL;
833  	if (pos >= available)
834  		return 0;
835  	if (count > available - pos)
836  		count = available - pos;
837  	memcpy(to, from + pos, count);
838  	*ppos = pos + count;
839  
840  	return count;
841  }
842  EXPORT_SYMBOL(memory_read_from_buffer);
843  
844  /*
845   * Transaction based IO.
846   * The file expects a single write which triggers the transaction, and then
847   * possibly a read which collects the result - which is stored in a
848   * file-local buffer.
849   */
850  
851  void simple_transaction_set(struct file *file, size_t n)
852  {
853  	struct simple_transaction_argresp *ar = file->private_data;
854  
855  	BUG_ON(n > SIMPLE_TRANSACTION_LIMIT);
856  
857  	/*
858  	 * The barrier ensures that ar->size will really remain zero until
859  	 * ar->data is ready for reading.
860  	 */
861  	smp_mb();
862  	ar->size = n;
863  }
864  EXPORT_SYMBOL(simple_transaction_set);
865  
866  char *simple_transaction_get(struct file *file, const char __user *buf, size_t size)
867  {
868  	struct simple_transaction_argresp *ar;
869  	static DEFINE_SPINLOCK(simple_transaction_lock);
870  
871  	if (size > SIMPLE_TRANSACTION_LIMIT - 1)
872  		return ERR_PTR(-EFBIG);
873  
874  	ar = (struct simple_transaction_argresp *)get_zeroed_page(GFP_KERNEL);
875  	if (!ar)
876  		return ERR_PTR(-ENOMEM);
877  
878  	spin_lock(&simple_transaction_lock);
879  
880  	/* only one write allowed per open */
881  	if (file->private_data) {
882  		spin_unlock(&simple_transaction_lock);
883  		free_page((unsigned long)ar);
884  		return ERR_PTR(-EBUSY);
885  	}
886  
887  	file->private_data = ar;
888  
889  	spin_unlock(&simple_transaction_lock);
890  
891  	if (copy_from_user(ar->data, buf, size))
892  		return ERR_PTR(-EFAULT);
893  
894  	return ar->data;
895  }
896  EXPORT_SYMBOL(simple_transaction_get);
897  
898  ssize_t simple_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
899  {
900  	struct simple_transaction_argresp *ar = file->private_data;
901  
902  	if (!ar)
903  		return 0;
904  	return simple_read_from_buffer(buf, size, pos, ar->data, ar->size);
905  }
906  EXPORT_SYMBOL(simple_transaction_read);
907  
908  int simple_transaction_release(struct inode *inode, struct file *file)
909  {
910  	free_page((unsigned long)file->private_data);
911  	return 0;
912  }
913  EXPORT_SYMBOL(simple_transaction_release);
914  
915  /* Simple attribute files */
916  
917  struct simple_attr {
918  	int (*get)(void *, u64 *);
919  	int (*set)(void *, u64);
920  	char get_buf[24];	/* enough to store a u64 and "\n\0" */
921  	char set_buf[24];
922  	void *data;
923  	const char *fmt;	/* format for read operation */
924  	struct mutex mutex;	/* protects access to these buffers */
925  };
926  
927  /* simple_attr_open is called by an actual attribute open file operation
928   * to set the attribute specific access operations. */
929  int simple_attr_open(struct inode *inode, struct file *file,
930  		     int (*get)(void *, u64 *), int (*set)(void *, u64),
931  		     const char *fmt)
932  {
933  	struct simple_attr *attr;
934  
935  	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
936  	if (!attr)
937  		return -ENOMEM;
938  
939  	attr->get = get;
940  	attr->set = set;
941  	attr->data = inode->i_private;
942  	attr->fmt = fmt;
943  	mutex_init(&attr->mutex);
944  
945  	file->private_data = attr;
946  
947  	return nonseekable_open(inode, file);
948  }
949  EXPORT_SYMBOL_GPL(simple_attr_open);
950  
951  int simple_attr_release(struct inode *inode, struct file *file)
952  {
953  	kfree(file->private_data);
954  	return 0;
955  }
956  EXPORT_SYMBOL_GPL(simple_attr_release);	/* GPL-only?  This?  Really? */
957  
958  /* read from the buffer that is filled with the get function */
959  ssize_t simple_attr_read(struct file *file, char __user *buf,
960  			 size_t len, loff_t *ppos)
961  {
962  	struct simple_attr *attr;
963  	size_t size;
964  	ssize_t ret;
965  
966  	attr = file->private_data;
967  
968  	if (!attr->get)
969  		return -EACCES;
970  
971  	ret = mutex_lock_interruptible(&attr->mutex);
972  	if (ret)
973  		return ret;
974  
975  	if (*ppos && attr->get_buf[0]) {
976  		/* continued read */
977  		size = strlen(attr->get_buf);
978  	} else {
979  		/* first read */
980  		u64 val;
981  		ret = attr->get(attr->data, &val);
982  		if (ret)
983  			goto out;
984  
985  		size = scnprintf(attr->get_buf, sizeof(attr->get_buf),
986  				 attr->fmt, (unsigned long long)val);
987  	}
988  
989  	ret = simple_read_from_buffer(buf, len, ppos, attr->get_buf, size);
990  out:
991  	mutex_unlock(&attr->mutex);
992  	return ret;
993  }
994  EXPORT_SYMBOL_GPL(simple_attr_read);
995  
996  /* interpret the buffer as a number to call the set function with */
997  ssize_t simple_attr_write(struct file *file, const char __user *buf,
998  			  size_t len, loff_t *ppos)
999  {
1000  	struct simple_attr *attr;
1001  	unsigned long long val;
1002  	size_t size;
1003  	ssize_t ret;
1004  
1005  	attr = file->private_data;
1006  	if (!attr->set)
1007  		return -EACCES;
1008  
1009  	ret = mutex_lock_interruptible(&attr->mutex);
1010  	if (ret)
1011  		return ret;
1012  
1013  	ret = -EFAULT;
1014  	size = min(sizeof(attr->set_buf) - 1, len);
1015  	if (copy_from_user(attr->set_buf, buf, size))
1016  		goto out;
1017  
1018  	attr->set_buf[size] = '\0';
1019  	ret = kstrtoull(attr->set_buf, 0, &val);
1020  	if (ret)
1021  		goto out;
1022  	ret = attr->set(attr->data, val);
1023  	if (ret == 0)
1024  		ret = len; /* on success, claim we got the whole input */
1025  out:
1026  	mutex_unlock(&attr->mutex);
1027  	return ret;
1028  }
1029  EXPORT_SYMBOL_GPL(simple_attr_write);
1030  
1031  /**
1032   * generic_fh_to_dentry - generic helper for the fh_to_dentry export operation
1033   * @sb:		filesystem to do the file handle conversion on
1034   * @fid:	file handle to convert
1035   * @fh_len:	length of the file handle in bytes
1036   * @fh_type:	type of file handle
1037   * @get_inode:	filesystem callback to retrieve inode
1038   *
1039   * This function decodes @fid as long as it has one of the well-known
1040   * Linux filehandle types and calls @get_inode on it to retrieve the
1041   * inode for the object specified in the file handle.
1042   */
1043  struct dentry *generic_fh_to_dentry(struct super_block *sb, struct fid *fid,
1044  		int fh_len, int fh_type, struct inode *(*get_inode)
1045  			(struct super_block *sb, u64 ino, u32 gen))
1046  {
1047  	struct inode *inode = NULL;
1048  
1049  	if (fh_len < 2)
1050  		return NULL;
1051  
1052  	switch (fh_type) {
1053  	case FILEID_INO32_GEN:
1054  	case FILEID_INO32_GEN_PARENT:
1055  		inode = get_inode(sb, fid->i32.ino, fid->i32.gen);
1056  		break;
1057  	}
1058  
1059  	return d_obtain_alias(inode);
1060  }
1061  EXPORT_SYMBOL_GPL(generic_fh_to_dentry);
1062  
1063  /**
1064   * generic_fh_to_parent - generic helper for the fh_to_parent export operation
1065   * @sb:		filesystem to do the file handle conversion on
1066   * @fid:	file handle to convert
1067   * @fh_len:	length of the file handle in bytes
1068   * @fh_type:	type of file handle
1069   * @get_inode:	filesystem callback to retrieve inode
1070   *
1071   * This function decodes @fid as long as it has one of the well-known
1072   * Linux filehandle types and calls @get_inode on it to retrieve the
1073   * inode for the _parent_ object specified in the file handle if it
1074   * is specified in the file handle, or NULL otherwise.
1075   */
1076  struct dentry *generic_fh_to_parent(struct super_block *sb, struct fid *fid,
1077  		int fh_len, int fh_type, struct inode *(*get_inode)
1078  			(struct super_block *sb, u64 ino, u32 gen))
1079  {
1080  	struct inode *inode = NULL;
1081  
1082  	if (fh_len <= 2)
1083  		return NULL;
1084  
1085  	switch (fh_type) {
1086  	case FILEID_INO32_GEN_PARENT:
1087  		inode = get_inode(sb, fid->i32.parent_ino,
1088  				  (fh_len > 3 ? fid->i32.parent_gen : 0));
1089  		break;
1090  	}
1091  
1092  	return d_obtain_alias(inode);
1093  }
1094  EXPORT_SYMBOL_GPL(generic_fh_to_parent);
1095  
1096  /**
1097   * __generic_file_fsync - generic fsync implementation for simple filesystems
1098   *
1099   * @file:	file to synchronize
1100   * @start:	start offset in bytes
1101   * @end:	end offset in bytes (inclusive)
1102   * @datasync:	only synchronize essential metadata if true
1103   *
1104   * This is a generic implementation of the fsync method for simple
1105   * filesystems which track all non-inode metadata in the buffers list
1106   * hanging off the address_space structure.
1107   */
1108  int __generic_file_fsync(struct file *file, loff_t start, loff_t end,
1109  				 int datasync)
1110  {
1111  	struct inode *inode = file->f_mapping->host;
1112  	int err;
1113  	int ret;
1114  
1115  	err = file_write_and_wait_range(file, start, end);
1116  	if (err)
1117  		return err;
1118  
1119  	inode_lock(inode);
1120  	ret = sync_mapping_buffers(inode->i_mapping);
1121  	if (!(inode->i_state & I_DIRTY_ALL))
1122  		goto out;
1123  	if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
1124  		goto out;
1125  
1126  	err = sync_inode_metadata(inode, 1);
1127  	if (ret == 0)
1128  		ret = err;
1129  
1130  out:
1131  	inode_unlock(inode);
1132  	/* check and advance again to catch errors after syncing out buffers */
1133  	err = file_check_and_advance_wb_err(file);
1134  	if (ret == 0)
1135  		ret = err;
1136  	return ret;
1137  }
1138  EXPORT_SYMBOL(__generic_file_fsync);
1139  
1140  /**
1141   * generic_file_fsync - generic fsync implementation for simple filesystems
1142   *			with flush
1143   * @file:	file to synchronize
1144   * @start:	start offset in bytes
1145   * @end:	end offset in bytes (inclusive)
1146   * @datasync:	only synchronize essential metadata if true
1147   *
1148   */
1149  
1150  int generic_file_fsync(struct file *file, loff_t start, loff_t end,
1151  		       int datasync)
1152  {
1153  	struct inode *inode = file->f_mapping->host;
1154  	int err;
1155  
1156  	err = __generic_file_fsync(file, start, end, datasync);
1157  	if (err)
1158  		return err;
1159  	return blkdev_issue_flush(inode->i_sb->s_bdev);
1160  }
1161  EXPORT_SYMBOL(generic_file_fsync);
1162  
1163  /**
1164   * generic_check_addressable - Check addressability of file system
1165   * @blocksize_bits:	log of file system block size
1166   * @num_blocks:		number of blocks in file system
1167   *
1168   * Determine whether a file system with @num_blocks blocks (and a
1169   * block size of 2**@blocksize_bits) is addressable by the sector_t
1170   * and page cache of the system.  Return 0 if so and -EFBIG otherwise.
1171   */
1172  int generic_check_addressable(unsigned blocksize_bits, u64 num_blocks)
1173  {
1174  	u64 last_fs_block = num_blocks - 1;
1175  	u64 last_fs_page =
1176  		last_fs_block >> (PAGE_SHIFT - blocksize_bits);
1177  
1178  	if (unlikely(num_blocks == 0))
1179  		return 0;
1180  
1181  	if ((blocksize_bits < 9) || (blocksize_bits > PAGE_SHIFT))
1182  		return -EINVAL;
1183  
1184  	if ((last_fs_block > (sector_t)(~0ULL) >> (blocksize_bits - 9)) ||
1185  	    (last_fs_page > (pgoff_t)(~0ULL))) {
1186  		return -EFBIG;
1187  	}
1188  	return 0;
1189  }
1190  EXPORT_SYMBOL(generic_check_addressable);
1191  
1192  /*
1193   * No-op implementation of ->fsync for in-memory filesystems.
1194   */
1195  int noop_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1196  {
1197  	return 0;
1198  }
1199  EXPORT_SYMBOL(noop_fsync);
1200  
1201  ssize_t noop_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1202  {
1203  	/*
1204  	 * iomap based filesystems support direct I/O without need for
1205  	 * this callback. However, it still needs to be set in
1206  	 * inode->a_ops so that open/fcntl know that direct I/O is
1207  	 * generally supported.
1208  	 */
1209  	return -EINVAL;
1210  }
1211  EXPORT_SYMBOL_GPL(noop_direct_IO);
1212  
1213  /* Because kfree isn't assignment-compatible with void(void*) ;-/ */
1214  void kfree_link(void *p)
1215  {
1216  	kfree(p);
1217  }
1218  EXPORT_SYMBOL(kfree_link);
1219  
1220  struct inode *alloc_anon_inode(struct super_block *s)
1221  {
1222  	static const struct address_space_operations anon_aops = {
1223  		.dirty_folio	= noop_dirty_folio,
1224  	};
1225  	struct inode *inode = new_inode_pseudo(s);
1226  
1227  	if (!inode)
1228  		return ERR_PTR(-ENOMEM);
1229  
1230  	inode->i_ino = get_next_ino();
1231  	inode->i_mapping->a_ops = &anon_aops;
1232  
1233  	/*
1234  	 * Mark the inode dirty from the very beginning,
1235  	 * that way it will never be moved to the dirty
1236  	 * list because mark_inode_dirty() will think
1237  	 * that it already _is_ on the dirty list.
1238  	 */
1239  	inode->i_state = I_DIRTY;
1240  	inode->i_mode = S_IRUSR | S_IWUSR;
1241  	inode->i_uid = current_fsuid();
1242  	inode->i_gid = current_fsgid();
1243  	inode->i_flags |= S_PRIVATE;
1244  	inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
1245  	return inode;
1246  }
1247  EXPORT_SYMBOL(alloc_anon_inode);
1248  
1249  /**
1250   * simple_nosetlease - generic helper for prohibiting leases
1251   * @filp: file pointer
1252   * @arg: type of lease to obtain
1253   * @flp: new lease supplied for insertion
1254   * @priv: private data for lm_setup operation
1255   *
1256   * Generic helper for filesystems that do not wish to allow leases to be set.
1257   * All arguments are ignored and it just returns -EINVAL.
1258   */
1259  int
1260  simple_nosetlease(struct file *filp, long arg, struct file_lock **flp,
1261  		  void **priv)
1262  {
1263  	return -EINVAL;
1264  }
1265  EXPORT_SYMBOL(simple_nosetlease);
1266  
1267  /**
1268   * simple_get_link - generic helper to get the target of "fast" symlinks
1269   * @dentry: not used here
1270   * @inode: the symlink inode
1271   * @done: not used here
1272   *
1273   * Generic helper for filesystems to use for symlink inodes where a pointer to
1274   * the symlink target is stored in ->i_link.  NOTE: this isn't normally called,
1275   * since as an optimization the path lookup code uses any non-NULL ->i_link
1276   * directly, without calling ->get_link().  But ->get_link() still must be set,
1277   * to mark the inode_operations as being for a symlink.
1278   *
1279   * Return: the symlink target
1280   */
1281  const char *simple_get_link(struct dentry *dentry, struct inode *inode,
1282  			    struct delayed_call *done)
1283  {
1284  	return inode->i_link;
1285  }
1286  EXPORT_SYMBOL(simple_get_link);
1287  
1288  const struct inode_operations simple_symlink_inode_operations = {
1289  	.get_link = simple_get_link,
1290  };
1291  EXPORT_SYMBOL(simple_symlink_inode_operations);
1292  
1293  /*
1294   * Operations for a permanently empty directory.
1295   */
1296  static struct dentry *empty_dir_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1297  {
1298  	return ERR_PTR(-ENOENT);
1299  }
1300  
1301  static int empty_dir_getattr(struct user_namespace *mnt_userns,
1302  			     const struct path *path, struct kstat *stat,
1303  			     u32 request_mask, unsigned int query_flags)
1304  {
1305  	struct inode *inode = d_inode(path->dentry);
1306  	generic_fillattr(&init_user_ns, inode, stat);
1307  	return 0;
1308  }
1309  
1310  static int empty_dir_setattr(struct user_namespace *mnt_userns,
1311  			     struct dentry *dentry, struct iattr *attr)
1312  {
1313  	return -EPERM;
1314  }
1315  
1316  static ssize_t empty_dir_listxattr(struct dentry *dentry, char *list, size_t size)
1317  {
1318  	return -EOPNOTSUPP;
1319  }
1320  
1321  static const struct inode_operations empty_dir_inode_operations = {
1322  	.lookup		= empty_dir_lookup,
1323  	.permission	= generic_permission,
1324  	.setattr	= empty_dir_setattr,
1325  	.getattr	= empty_dir_getattr,
1326  	.listxattr	= empty_dir_listxattr,
1327  };
1328  
1329  static loff_t empty_dir_llseek(struct file *file, loff_t offset, int whence)
1330  {
1331  	/* An empty directory has two entries . and .. at offsets 0 and 1 */
1332  	return generic_file_llseek_size(file, offset, whence, 2, 2);
1333  }
1334  
1335  static int empty_dir_readdir(struct file *file, struct dir_context *ctx)
1336  {
1337  	dir_emit_dots(file, ctx);
1338  	return 0;
1339  }
1340  
1341  static const struct file_operations empty_dir_operations = {
1342  	.llseek		= empty_dir_llseek,
1343  	.read		= generic_read_dir,
1344  	.iterate_shared	= empty_dir_readdir,
1345  	.fsync		= noop_fsync,
1346  };
1347  
1348  
1349  void make_empty_dir_inode(struct inode *inode)
1350  {
1351  	set_nlink(inode, 2);
1352  	inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO;
1353  	inode->i_uid = GLOBAL_ROOT_UID;
1354  	inode->i_gid = GLOBAL_ROOT_GID;
1355  	inode->i_rdev = 0;
1356  	inode->i_size = 0;
1357  	inode->i_blkbits = PAGE_SHIFT;
1358  	inode->i_blocks = 0;
1359  
1360  	inode->i_op = &empty_dir_inode_operations;
1361  	inode->i_opflags &= ~IOP_XATTR;
1362  	inode->i_fop = &empty_dir_operations;
1363  }
1364  
1365  bool is_empty_dir_inode(struct inode *inode)
1366  {
1367  	return (inode->i_fop == &empty_dir_operations) &&
1368  		(inode->i_op == &empty_dir_inode_operations);
1369  }
1370  
1371  #if IS_ENABLED(CONFIG_UNICODE)
1372  /*
1373   * Determine if the name of a dentry should be casefolded.
1374   *
1375   * Return: if names will need casefolding
1376   */
1377  static bool needs_casefold(const struct inode *dir)
1378  {
1379  	return IS_CASEFOLDED(dir) && dir->i_sb->s_encoding;
1380  }
1381  
1382  /**
1383   * generic_ci_d_compare - generic d_compare implementation for casefolding filesystems
1384   * @dentry:	dentry whose name we are checking against
1385   * @len:	len of name of dentry
1386   * @str:	str pointer to name of dentry
1387   * @name:	Name to compare against
1388   *
1389   * Return: 0 if names match, 1 if mismatch, or -ERRNO
1390   */
1391  static int generic_ci_d_compare(const struct dentry *dentry, unsigned int len,
1392  				const char *str, const struct qstr *name)
1393  {
1394  	const struct dentry *parent = READ_ONCE(dentry->d_parent);
1395  	const struct inode *dir = READ_ONCE(parent->d_inode);
1396  	const struct super_block *sb = dentry->d_sb;
1397  	const struct unicode_map *um = sb->s_encoding;
1398  	struct qstr qstr = QSTR_INIT(str, len);
1399  	char strbuf[DNAME_INLINE_LEN];
1400  	int ret;
1401  
1402  	if (!dir || !needs_casefold(dir))
1403  		goto fallback;
1404  	/*
1405  	 * If the dentry name is stored in-line, then it may be concurrently
1406  	 * modified by a rename.  If this happens, the VFS will eventually retry
1407  	 * the lookup, so it doesn't matter what ->d_compare() returns.
1408  	 * However, it's unsafe to call utf8_strncasecmp() with an unstable
1409  	 * string.  Therefore, we have to copy the name into a temporary buffer.
1410  	 */
1411  	if (len <= DNAME_INLINE_LEN - 1) {
1412  		memcpy(strbuf, str, len);
1413  		strbuf[len] = 0;
1414  		qstr.name = strbuf;
1415  		/* prevent compiler from optimizing out the temporary buffer */
1416  		barrier();
1417  	}
1418  	ret = utf8_strncasecmp(um, name, &qstr);
1419  	if (ret >= 0)
1420  		return ret;
1421  
1422  	if (sb_has_strict_encoding(sb))
1423  		return -EINVAL;
1424  fallback:
1425  	if (len != name->len)
1426  		return 1;
1427  	return !!memcmp(str, name->name, len);
1428  }
1429  
1430  /**
1431   * generic_ci_d_hash - generic d_hash implementation for casefolding filesystems
1432   * @dentry:	dentry of the parent directory
1433   * @str:	qstr of name whose hash we should fill in
1434   *
1435   * Return: 0 if hash was successful or unchanged, and -EINVAL on error
1436   */
1437  static int generic_ci_d_hash(const struct dentry *dentry, struct qstr *str)
1438  {
1439  	const struct inode *dir = READ_ONCE(dentry->d_inode);
1440  	struct super_block *sb = dentry->d_sb;
1441  	const struct unicode_map *um = sb->s_encoding;
1442  	int ret = 0;
1443  
1444  	if (!dir || !needs_casefold(dir))
1445  		return 0;
1446  
1447  	ret = utf8_casefold_hash(um, dentry, str);
1448  	if (ret < 0 && sb_has_strict_encoding(sb))
1449  		return -EINVAL;
1450  	return 0;
1451  }
1452  
1453  static const struct dentry_operations generic_ci_dentry_ops = {
1454  	.d_hash = generic_ci_d_hash,
1455  	.d_compare = generic_ci_d_compare,
1456  };
1457  #endif
1458  
1459  #ifdef CONFIG_FS_ENCRYPTION
1460  static const struct dentry_operations generic_encrypted_dentry_ops = {
1461  	.d_revalidate = fscrypt_d_revalidate,
1462  };
1463  #endif
1464  
1465  #if defined(CONFIG_FS_ENCRYPTION) && IS_ENABLED(CONFIG_UNICODE)
1466  static const struct dentry_operations generic_encrypted_ci_dentry_ops = {
1467  	.d_hash = generic_ci_d_hash,
1468  	.d_compare = generic_ci_d_compare,
1469  	.d_revalidate = fscrypt_d_revalidate,
1470  };
1471  #endif
1472  
1473  /**
1474   * generic_set_encrypted_ci_d_ops - helper for setting d_ops for given dentry
1475   * @dentry:	dentry to set ops on
1476   *
1477   * Casefolded directories need d_hash and d_compare set, so that the dentries
1478   * contained in them are handled case-insensitively.  Note that these operations
1479   * are needed on the parent directory rather than on the dentries in it, and
1480   * while the casefolding flag can be toggled on and off on an empty directory,
1481   * dentry_operations can't be changed later.  As a result, if the filesystem has
1482   * casefolding support enabled at all, we have to give all dentries the
1483   * casefolding operations even if their inode doesn't have the casefolding flag
1484   * currently (and thus the casefolding ops would be no-ops for now).
1485   *
1486   * Encryption works differently in that the only dentry operation it needs is
1487   * d_revalidate, which it only needs on dentries that have the no-key name flag.
1488   * The no-key flag can't be set "later", so we don't have to worry about that.
1489   *
1490   * Finally, to maximize compatibility with overlayfs (which isn't compatible
1491   * with certain dentry operations) and to avoid taking an unnecessary
1492   * performance hit, we use custom dentry_operations for each possible
1493   * combination rather than always installing all operations.
1494   */
1495  void generic_set_encrypted_ci_d_ops(struct dentry *dentry)
1496  {
1497  #ifdef CONFIG_FS_ENCRYPTION
1498  	bool needs_encrypt_ops = dentry->d_flags & DCACHE_NOKEY_NAME;
1499  #endif
1500  #if IS_ENABLED(CONFIG_UNICODE)
1501  	bool needs_ci_ops = dentry->d_sb->s_encoding;
1502  #endif
1503  #if defined(CONFIG_FS_ENCRYPTION) && IS_ENABLED(CONFIG_UNICODE)
1504  	if (needs_encrypt_ops && needs_ci_ops) {
1505  		d_set_d_op(dentry, &generic_encrypted_ci_dentry_ops);
1506  		return;
1507  	}
1508  #endif
1509  #ifdef CONFIG_FS_ENCRYPTION
1510  	if (needs_encrypt_ops) {
1511  		d_set_d_op(dentry, &generic_encrypted_dentry_ops);
1512  		return;
1513  	}
1514  #endif
1515  #if IS_ENABLED(CONFIG_UNICODE)
1516  	if (needs_ci_ops) {
1517  		d_set_d_op(dentry, &generic_ci_dentry_ops);
1518  		return;
1519  	}
1520  #endif
1521  }
1522  EXPORT_SYMBOL(generic_set_encrypted_ci_d_ops);
1523