xref: /openbmc/linux/fs/debugfs/inode.c (revision 1d1997db)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  inode.c - part of debugfs, a tiny little debug file system
4  *
5  *  Copyright (C) 2004,2019 Greg Kroah-Hartman <greg@kroah.com>
6  *  Copyright (C) 2004 IBM Inc.
7  *  Copyright (C) 2019 Linux Foundation <gregkh@linuxfoundation.org>
8  *
9  *  debugfs is for people to use instead of /proc or /sys.
10  *  See ./Documentation/core-api/kernel-api.rst for more details.
11  */
12 
13 #define pr_fmt(fmt)	"debugfs: " fmt
14 
15 #include <linux/module.h>
16 #include <linux/fs.h>
17 #include <linux/mount.h>
18 #include <linux/pagemap.h>
19 #include <linux/init.h>
20 #include <linux/kobject.h>
21 #include <linux/namei.h>
22 #include <linux/debugfs.h>
23 #include <linux/fsnotify.h>
24 #include <linux/string.h>
25 #include <linux/seq_file.h>
26 #include <linux/parser.h>
27 #include <linux/magic.h>
28 #include <linux/slab.h>
29 #include <linux/security.h>
30 
31 #include "internal.h"
32 
33 #define DEBUGFS_DEFAULT_MODE	0700
34 
35 static struct vfsmount *debugfs_mount;
36 static int debugfs_mount_count;
37 static bool debugfs_registered;
38 
39 /*
40  * Don't allow access attributes to be changed whilst the kernel is locked down
41  * so that we can use the file mode as part of a heuristic to determine whether
42  * to lock down individual files.
43  */
44 static int debugfs_setattr(struct dentry *dentry, struct iattr *ia)
45 {
46 	int ret = security_locked_down(LOCKDOWN_DEBUGFS);
47 
48 	if (ret && (ia->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID)))
49 		return ret;
50 	return simple_setattr(dentry, ia);
51 }
52 
53 static const struct inode_operations debugfs_file_inode_operations = {
54 	.setattr	= debugfs_setattr,
55 };
56 static const struct inode_operations debugfs_dir_inode_operations = {
57 	.lookup		= simple_lookup,
58 	.setattr	= debugfs_setattr,
59 };
60 static const struct inode_operations debugfs_symlink_inode_operations = {
61 	.get_link	= simple_get_link,
62 	.setattr	= debugfs_setattr,
63 };
64 
65 static struct inode *debugfs_get_inode(struct super_block *sb)
66 {
67 	struct inode *inode = new_inode(sb);
68 	if (inode) {
69 		inode->i_ino = get_next_ino();
70 		inode->i_atime = inode->i_mtime =
71 			inode->i_ctime = current_time(inode);
72 	}
73 	return inode;
74 }
75 
76 struct debugfs_mount_opts {
77 	kuid_t uid;
78 	kgid_t gid;
79 	umode_t mode;
80 };
81 
82 enum {
83 	Opt_uid,
84 	Opt_gid,
85 	Opt_mode,
86 	Opt_err
87 };
88 
89 static const match_table_t tokens = {
90 	{Opt_uid, "uid=%u"},
91 	{Opt_gid, "gid=%u"},
92 	{Opt_mode, "mode=%o"},
93 	{Opt_err, NULL}
94 };
95 
96 struct debugfs_fs_info {
97 	struct debugfs_mount_opts mount_opts;
98 };
99 
100 static int debugfs_parse_options(char *data, struct debugfs_mount_opts *opts)
101 {
102 	substring_t args[MAX_OPT_ARGS];
103 	int option;
104 	int token;
105 	kuid_t uid;
106 	kgid_t gid;
107 	char *p;
108 
109 	opts->mode = DEBUGFS_DEFAULT_MODE;
110 
111 	while ((p = strsep(&data, ",")) != NULL) {
112 		if (!*p)
113 			continue;
114 
115 		token = match_token(p, tokens, args);
116 		switch (token) {
117 		case Opt_uid:
118 			if (match_int(&args[0], &option))
119 				return -EINVAL;
120 			uid = make_kuid(current_user_ns(), option);
121 			if (!uid_valid(uid))
122 				return -EINVAL;
123 			opts->uid = uid;
124 			break;
125 		case Opt_gid:
126 			if (match_int(&args[0], &option))
127 				return -EINVAL;
128 			gid = make_kgid(current_user_ns(), option);
129 			if (!gid_valid(gid))
130 				return -EINVAL;
131 			opts->gid = gid;
132 			break;
133 		case Opt_mode:
134 			if (match_octal(&args[0], &option))
135 				return -EINVAL;
136 			opts->mode = option & S_IALLUGO;
137 			break;
138 		/*
139 		 * We might like to report bad mount options here;
140 		 * but traditionally debugfs has ignored all mount options
141 		 */
142 		}
143 	}
144 
145 	return 0;
146 }
147 
148 static int debugfs_apply_options(struct super_block *sb)
149 {
150 	struct debugfs_fs_info *fsi = sb->s_fs_info;
151 	struct inode *inode = d_inode(sb->s_root);
152 	struct debugfs_mount_opts *opts = &fsi->mount_opts;
153 
154 	inode->i_mode &= ~S_IALLUGO;
155 	inode->i_mode |= opts->mode;
156 
157 	inode->i_uid = opts->uid;
158 	inode->i_gid = opts->gid;
159 
160 	return 0;
161 }
162 
163 static int debugfs_remount(struct super_block *sb, int *flags, char *data)
164 {
165 	int err;
166 	struct debugfs_fs_info *fsi = sb->s_fs_info;
167 
168 	sync_filesystem(sb);
169 	err = debugfs_parse_options(data, &fsi->mount_opts);
170 	if (err)
171 		goto fail;
172 
173 	debugfs_apply_options(sb);
174 
175 fail:
176 	return err;
177 }
178 
179 static int debugfs_show_options(struct seq_file *m, struct dentry *root)
180 {
181 	struct debugfs_fs_info *fsi = root->d_sb->s_fs_info;
182 	struct debugfs_mount_opts *opts = &fsi->mount_opts;
183 
184 	if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
185 		seq_printf(m, ",uid=%u",
186 			   from_kuid_munged(&init_user_ns, opts->uid));
187 	if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
188 		seq_printf(m, ",gid=%u",
189 			   from_kgid_munged(&init_user_ns, opts->gid));
190 	if (opts->mode != DEBUGFS_DEFAULT_MODE)
191 		seq_printf(m, ",mode=%o", opts->mode);
192 
193 	return 0;
194 }
195 
196 static void debugfs_free_inode(struct inode *inode)
197 {
198 	if (S_ISLNK(inode->i_mode))
199 		kfree(inode->i_link);
200 	free_inode_nonrcu(inode);
201 }
202 
203 static const struct super_operations debugfs_super_operations = {
204 	.statfs		= simple_statfs,
205 	.remount_fs	= debugfs_remount,
206 	.show_options	= debugfs_show_options,
207 	.free_inode	= debugfs_free_inode,
208 };
209 
210 static void debugfs_release_dentry(struct dentry *dentry)
211 {
212 	void *fsd = dentry->d_fsdata;
213 
214 	if (!((unsigned long)fsd & DEBUGFS_FSDATA_IS_REAL_FOPS_BIT))
215 		kfree(dentry->d_fsdata);
216 }
217 
218 static struct vfsmount *debugfs_automount(struct path *path)
219 {
220 	debugfs_automount_t f;
221 	f = (debugfs_automount_t)path->dentry->d_fsdata;
222 	return f(path->dentry, d_inode(path->dentry)->i_private);
223 }
224 
225 static const struct dentry_operations debugfs_dops = {
226 	.d_delete = always_delete_dentry,
227 	.d_release = debugfs_release_dentry,
228 	.d_automount = debugfs_automount,
229 };
230 
231 static int debug_fill_super(struct super_block *sb, void *data, int silent)
232 {
233 	static const struct tree_descr debug_files[] = {{""}};
234 	struct debugfs_fs_info *fsi;
235 	int err;
236 
237 	fsi = kzalloc(sizeof(struct debugfs_fs_info), GFP_KERNEL);
238 	sb->s_fs_info = fsi;
239 	if (!fsi) {
240 		err = -ENOMEM;
241 		goto fail;
242 	}
243 
244 	err = debugfs_parse_options(data, &fsi->mount_opts);
245 	if (err)
246 		goto fail;
247 
248 	err  =  simple_fill_super(sb, DEBUGFS_MAGIC, debug_files);
249 	if (err)
250 		goto fail;
251 
252 	sb->s_op = &debugfs_super_operations;
253 	sb->s_d_op = &debugfs_dops;
254 
255 	debugfs_apply_options(sb);
256 
257 	return 0;
258 
259 fail:
260 	kfree(fsi);
261 	sb->s_fs_info = NULL;
262 	return err;
263 }
264 
265 static struct dentry *debug_mount(struct file_system_type *fs_type,
266 			int flags, const char *dev_name,
267 			void *data)
268 {
269 	return mount_single(fs_type, flags, data, debug_fill_super);
270 }
271 
272 static struct file_system_type debug_fs_type = {
273 	.owner =	THIS_MODULE,
274 	.name =		"debugfs",
275 	.mount =	debug_mount,
276 	.kill_sb =	kill_litter_super,
277 };
278 MODULE_ALIAS_FS("debugfs");
279 
280 /**
281  * debugfs_lookup() - look up an existing debugfs file
282  * @name: a pointer to a string containing the name of the file to look up.
283  * @parent: a pointer to the parent dentry of the file.
284  *
285  * This function will return a pointer to a dentry if it succeeds.  If the file
286  * doesn't exist or an error occurs, %NULL will be returned.  The returned
287  * dentry must be passed to dput() when it is no longer needed.
288  *
289  * If debugfs is not enabled in the kernel, the value -%ENODEV will be
290  * returned.
291  */
292 struct dentry *debugfs_lookup(const char *name, struct dentry *parent)
293 {
294 	struct dentry *dentry;
295 
296 	if (IS_ERR(parent))
297 		return NULL;
298 
299 	if (!parent)
300 		parent = debugfs_mount->mnt_root;
301 
302 	dentry = lookup_positive_unlocked(name, parent, strlen(name));
303 	if (IS_ERR(dentry))
304 		return NULL;
305 	return dentry;
306 }
307 EXPORT_SYMBOL_GPL(debugfs_lookup);
308 
309 static struct dentry *start_creating(const char *name, struct dentry *parent)
310 {
311 	struct dentry *dentry;
312 	int error;
313 
314 	pr_debug("creating file '%s'\n", name);
315 
316 	if (IS_ERR(parent))
317 		return parent;
318 
319 	error = simple_pin_fs(&debug_fs_type, &debugfs_mount,
320 			      &debugfs_mount_count);
321 	if (error) {
322 		pr_err("Unable to pin filesystem for file '%s'\n", name);
323 		return ERR_PTR(error);
324 	}
325 
326 	/* If the parent is not specified, we create it in the root.
327 	 * We need the root dentry to do this, which is in the super
328 	 * block. A pointer to that is in the struct vfsmount that we
329 	 * have around.
330 	 */
331 	if (!parent)
332 		parent = debugfs_mount->mnt_root;
333 
334 	inode_lock(d_inode(parent));
335 	dentry = lookup_one_len(name, parent, strlen(name));
336 	if (!IS_ERR(dentry) && d_really_is_positive(dentry)) {
337 		if (d_is_dir(dentry))
338 			pr_err("Directory '%s' with parent '%s' already present!\n",
339 			       name, parent->d_name.name);
340 		else
341 			pr_err("File '%s' in directory '%s' already present!\n",
342 			       name, parent->d_name.name);
343 		dput(dentry);
344 		dentry = ERR_PTR(-EEXIST);
345 	}
346 
347 	if (IS_ERR(dentry)) {
348 		inode_unlock(d_inode(parent));
349 		simple_release_fs(&debugfs_mount, &debugfs_mount_count);
350 	}
351 
352 	return dentry;
353 }
354 
355 static struct dentry *failed_creating(struct dentry *dentry)
356 {
357 	inode_unlock(d_inode(dentry->d_parent));
358 	dput(dentry);
359 	simple_release_fs(&debugfs_mount, &debugfs_mount_count);
360 	return ERR_PTR(-ENOMEM);
361 }
362 
363 static struct dentry *end_creating(struct dentry *dentry)
364 {
365 	inode_unlock(d_inode(dentry->d_parent));
366 	return dentry;
367 }
368 
369 static struct dentry *__debugfs_create_file(const char *name, umode_t mode,
370 				struct dentry *parent, void *data,
371 				const struct file_operations *proxy_fops,
372 				const struct file_operations *real_fops)
373 {
374 	struct dentry *dentry;
375 	struct inode *inode;
376 
377 	if (!(mode & S_IFMT))
378 		mode |= S_IFREG;
379 	BUG_ON(!S_ISREG(mode));
380 	dentry = start_creating(name, parent);
381 
382 	if (IS_ERR(dentry))
383 		return dentry;
384 
385 	inode = debugfs_get_inode(dentry->d_sb);
386 	if (unlikely(!inode)) {
387 		pr_err("out of free dentries, can not create file '%s'\n",
388 		       name);
389 		return failed_creating(dentry);
390 	}
391 
392 	inode->i_mode = mode;
393 	inode->i_private = data;
394 
395 	inode->i_op = &debugfs_file_inode_operations;
396 	inode->i_fop = proxy_fops;
397 	dentry->d_fsdata = (void *)((unsigned long)real_fops |
398 				DEBUGFS_FSDATA_IS_REAL_FOPS_BIT);
399 
400 	d_instantiate(dentry, inode);
401 	fsnotify_create(d_inode(dentry->d_parent), dentry);
402 	return end_creating(dentry);
403 }
404 
405 /**
406  * debugfs_create_file - create a file in the debugfs filesystem
407  * @name: a pointer to a string containing the name of the file to create.
408  * @mode: the permission that the file should have.
409  * @parent: a pointer to the parent dentry for this file.  This should be a
410  *          directory dentry if set.  If this parameter is NULL, then the
411  *          file will be created in the root of the debugfs filesystem.
412  * @data: a pointer to something that the caller will want to get to later
413  *        on.  The inode.i_private pointer will point to this value on
414  *        the open() call.
415  * @fops: a pointer to a struct file_operations that should be used for
416  *        this file.
417  *
418  * This is the basic "create a file" function for debugfs.  It allows for a
419  * wide range of flexibility in creating a file, or a directory (if you want
420  * to create a directory, the debugfs_create_dir() function is
421  * recommended to be used instead.)
422  *
423  * This function will return a pointer to a dentry if it succeeds.  This
424  * pointer must be passed to the debugfs_remove() function when the file is
425  * to be removed (no automatic cleanup happens if your module is unloaded,
426  * you are responsible here.)  If an error occurs, %ERR_PTR(-ERROR) will be
427  * returned.
428  *
429  * If debugfs is not enabled in the kernel, the value -%ENODEV will be
430  * returned.
431  */
432 struct dentry *debugfs_create_file(const char *name, umode_t mode,
433 				   struct dentry *parent, void *data,
434 				   const struct file_operations *fops)
435 {
436 
437 	return __debugfs_create_file(name, mode, parent, data,
438 				fops ? &debugfs_full_proxy_file_operations :
439 					&debugfs_noop_file_operations,
440 				fops);
441 }
442 EXPORT_SYMBOL_GPL(debugfs_create_file);
443 
444 /**
445  * debugfs_create_file_unsafe - create a file in the debugfs filesystem
446  * @name: a pointer to a string containing the name of the file to create.
447  * @mode: the permission that the file should have.
448  * @parent: a pointer to the parent dentry for this file.  This should be a
449  *          directory dentry if set.  If this parameter is NULL, then the
450  *          file will be created in the root of the debugfs filesystem.
451  * @data: a pointer to something that the caller will want to get to later
452  *        on.  The inode.i_private pointer will point to this value on
453  *        the open() call.
454  * @fops: a pointer to a struct file_operations that should be used for
455  *        this file.
456  *
457  * debugfs_create_file_unsafe() is completely analogous to
458  * debugfs_create_file(), the only difference being that the fops
459  * handed it will not get protected against file removals by the
460  * debugfs core.
461  *
462  * It is your responsibility to protect your struct file_operation
463  * methods against file removals by means of debugfs_file_get()
464  * and debugfs_file_put(). ->open() is still protected by
465  * debugfs though.
466  *
467  * Any struct file_operations defined by means of
468  * DEFINE_DEBUGFS_ATTRIBUTE() is protected against file removals and
469  * thus, may be used here.
470  */
471 struct dentry *debugfs_create_file_unsafe(const char *name, umode_t mode,
472 				   struct dentry *parent, void *data,
473 				   const struct file_operations *fops)
474 {
475 
476 	return __debugfs_create_file(name, mode, parent, data,
477 				fops ? &debugfs_open_proxy_file_operations :
478 					&debugfs_noop_file_operations,
479 				fops);
480 }
481 EXPORT_SYMBOL_GPL(debugfs_create_file_unsafe);
482 
483 /**
484  * debugfs_create_file_size - create a file in the debugfs filesystem
485  * @name: a pointer to a string containing the name of the file to create.
486  * @mode: the permission that the file should have.
487  * @parent: a pointer to the parent dentry for this file.  This should be a
488  *          directory dentry if set.  If this parameter is NULL, then the
489  *          file will be created in the root of the debugfs filesystem.
490  * @data: a pointer to something that the caller will want to get to later
491  *        on.  The inode.i_private pointer will point to this value on
492  *        the open() call.
493  * @fops: a pointer to a struct file_operations that should be used for
494  *        this file.
495  * @file_size: initial file size
496  *
497  * This is the basic "create a file" function for debugfs.  It allows for a
498  * wide range of flexibility in creating a file, or a directory (if you want
499  * to create a directory, the debugfs_create_dir() function is
500  * recommended to be used instead.)
501  *
502  * This function will return a pointer to a dentry if it succeeds.  This
503  * pointer must be passed to the debugfs_remove() function when the file is
504  * to be removed (no automatic cleanup happens if your module is unloaded,
505  * you are responsible here.)  If an error occurs, %ERR_PTR(-ERROR) will be
506  * returned.
507  *
508  * If debugfs is not enabled in the kernel, the value -%ENODEV will be
509  * returned.
510  */
511 struct dentry *debugfs_create_file_size(const char *name, umode_t mode,
512 					struct dentry *parent, void *data,
513 					const struct file_operations *fops,
514 					loff_t file_size)
515 {
516 	struct dentry *de = debugfs_create_file(name, mode, parent, data, fops);
517 
518 	if (de)
519 		d_inode(de)->i_size = file_size;
520 	return de;
521 }
522 EXPORT_SYMBOL_GPL(debugfs_create_file_size);
523 
524 /**
525  * debugfs_create_dir - create a directory in the debugfs filesystem
526  * @name: a pointer to a string containing the name of the directory to
527  *        create.
528  * @parent: a pointer to the parent dentry for this file.  This should be a
529  *          directory dentry if set.  If this parameter is NULL, then the
530  *          directory will be created in the root of the debugfs filesystem.
531  *
532  * This function creates a directory in debugfs with the given name.
533  *
534  * This function will return a pointer to a dentry if it succeeds.  This
535  * pointer must be passed to the debugfs_remove() function when the file is
536  * to be removed (no automatic cleanup happens if your module is unloaded,
537  * you are responsible here.)  If an error occurs, %ERR_PTR(-ERROR) will be
538  * returned.
539  *
540  * If debugfs is not enabled in the kernel, the value -%ENODEV will be
541  * returned.
542  */
543 struct dentry *debugfs_create_dir(const char *name, struct dentry *parent)
544 {
545 	struct dentry *dentry = start_creating(name, parent);
546 	struct inode *inode;
547 
548 	if (IS_ERR(dentry))
549 		return dentry;
550 
551 	inode = debugfs_get_inode(dentry->d_sb);
552 	if (unlikely(!inode)) {
553 		pr_err("out of free dentries, can not create directory '%s'\n",
554 		       name);
555 		return failed_creating(dentry);
556 	}
557 
558 	inode->i_mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
559 	inode->i_op = &debugfs_dir_inode_operations;
560 	inode->i_fop = &simple_dir_operations;
561 
562 	/* directory inodes start off with i_nlink == 2 (for "." entry) */
563 	inc_nlink(inode);
564 	d_instantiate(dentry, inode);
565 	inc_nlink(d_inode(dentry->d_parent));
566 	fsnotify_mkdir(d_inode(dentry->d_parent), dentry);
567 	return end_creating(dentry);
568 }
569 EXPORT_SYMBOL_GPL(debugfs_create_dir);
570 
571 /**
572  * debugfs_create_automount - create automount point in the debugfs filesystem
573  * @name: a pointer to a string containing the name of the file to create.
574  * @parent: a pointer to the parent dentry for this file.  This should be a
575  *          directory dentry if set.  If this parameter is NULL, then the
576  *          file will be created in the root of the debugfs filesystem.
577  * @f: function to be called when pathname resolution steps on that one.
578  * @data: opaque argument to pass to f().
579  *
580  * @f should return what ->d_automount() would.
581  */
582 struct dentry *debugfs_create_automount(const char *name,
583 					struct dentry *parent,
584 					debugfs_automount_t f,
585 					void *data)
586 {
587 	struct dentry *dentry = start_creating(name, parent);
588 	struct inode *inode;
589 
590 	if (IS_ERR(dentry))
591 		return dentry;
592 
593 	inode = debugfs_get_inode(dentry->d_sb);
594 	if (unlikely(!inode)) {
595 		pr_err("out of free dentries, can not create automount '%s'\n",
596 		       name);
597 		return failed_creating(dentry);
598 	}
599 
600 	make_empty_dir_inode(inode);
601 	inode->i_flags |= S_AUTOMOUNT;
602 	inode->i_private = data;
603 	dentry->d_fsdata = (void *)f;
604 	/* directory inodes start off with i_nlink == 2 (for "." entry) */
605 	inc_nlink(inode);
606 	d_instantiate(dentry, inode);
607 	inc_nlink(d_inode(dentry->d_parent));
608 	fsnotify_mkdir(d_inode(dentry->d_parent), dentry);
609 	return end_creating(dentry);
610 }
611 EXPORT_SYMBOL(debugfs_create_automount);
612 
613 /**
614  * debugfs_create_symlink- create a symbolic link in the debugfs filesystem
615  * @name: a pointer to a string containing the name of the symbolic link to
616  *        create.
617  * @parent: a pointer to the parent dentry for this symbolic link.  This
618  *          should be a directory dentry if set.  If this parameter is NULL,
619  *          then the symbolic link will be created in the root of the debugfs
620  *          filesystem.
621  * @target: a pointer to a string containing the path to the target of the
622  *          symbolic link.
623  *
624  * This function creates a symbolic link with the given name in debugfs that
625  * links to the given target path.
626  *
627  * This function will return a pointer to a dentry if it succeeds.  This
628  * pointer must be passed to the debugfs_remove() function when the symbolic
629  * link is to be removed (no automatic cleanup happens if your module is
630  * unloaded, you are responsible here.)  If an error occurs, %ERR_PTR(-ERROR)
631  * will be returned.
632  *
633  * If debugfs is not enabled in the kernel, the value -%ENODEV will be
634  * returned.
635  */
636 struct dentry *debugfs_create_symlink(const char *name, struct dentry *parent,
637 				      const char *target)
638 {
639 	struct dentry *dentry;
640 	struct inode *inode;
641 	char *link = kstrdup(target, GFP_KERNEL);
642 	if (!link)
643 		return ERR_PTR(-ENOMEM);
644 
645 	dentry = start_creating(name, parent);
646 	if (IS_ERR(dentry)) {
647 		kfree(link);
648 		return dentry;
649 	}
650 
651 	inode = debugfs_get_inode(dentry->d_sb);
652 	if (unlikely(!inode)) {
653 		pr_err("out of free dentries, can not create symlink '%s'\n",
654 		       name);
655 		kfree(link);
656 		return failed_creating(dentry);
657 	}
658 	inode->i_mode = S_IFLNK | S_IRWXUGO;
659 	inode->i_op = &debugfs_symlink_inode_operations;
660 	inode->i_link = link;
661 	d_instantiate(dentry, inode);
662 	return end_creating(dentry);
663 }
664 EXPORT_SYMBOL_GPL(debugfs_create_symlink);
665 
666 static void __debugfs_file_removed(struct dentry *dentry)
667 {
668 	struct debugfs_fsdata *fsd;
669 
670 	/*
671 	 * Paired with the closing smp_mb() implied by a successful
672 	 * cmpxchg() in debugfs_file_get(): either
673 	 * debugfs_file_get() must see a dead dentry or we must see a
674 	 * debugfs_fsdata instance at ->d_fsdata here (or both).
675 	 */
676 	smp_mb();
677 	fsd = READ_ONCE(dentry->d_fsdata);
678 	if ((unsigned long)fsd & DEBUGFS_FSDATA_IS_REAL_FOPS_BIT)
679 		return;
680 	if (!refcount_dec_and_test(&fsd->active_users))
681 		wait_for_completion(&fsd->active_users_drained);
682 }
683 
684 static int __debugfs_remove(struct dentry *dentry, struct dentry *parent)
685 {
686 	int ret = 0;
687 
688 	if (simple_positive(dentry)) {
689 		dget(dentry);
690 		if (d_is_dir(dentry)) {
691 			ret = simple_rmdir(d_inode(parent), dentry);
692 			if (!ret)
693 				fsnotify_rmdir(d_inode(parent), dentry);
694 		} else {
695 			simple_unlink(d_inode(parent), dentry);
696 			fsnotify_unlink(d_inode(parent), dentry);
697 		}
698 		if (!ret)
699 			d_delete(dentry);
700 		if (d_is_reg(dentry))
701 			__debugfs_file_removed(dentry);
702 		dput(dentry);
703 	}
704 	return ret;
705 }
706 
707 /**
708  * debugfs_remove - removes a file or directory from the debugfs filesystem
709  * @dentry: a pointer to a the dentry of the file or directory to be
710  *          removed.  If this parameter is NULL or an error value, nothing
711  *          will be done.
712  *
713  * This function removes a file or directory in debugfs that was previously
714  * created with a call to another debugfs function (like
715  * debugfs_create_file() or variants thereof.)
716  *
717  * This function is required to be called in order for the file to be
718  * removed, no automatic cleanup of files will happen when a module is
719  * removed, you are responsible here.
720  */
721 void debugfs_remove(struct dentry *dentry)
722 {
723 	struct dentry *parent;
724 	int ret;
725 
726 	if (IS_ERR_OR_NULL(dentry))
727 		return;
728 
729 	parent = dentry->d_parent;
730 	inode_lock(d_inode(parent));
731 	ret = __debugfs_remove(dentry, parent);
732 	inode_unlock(d_inode(parent));
733 	if (!ret)
734 		simple_release_fs(&debugfs_mount, &debugfs_mount_count);
735 }
736 EXPORT_SYMBOL_GPL(debugfs_remove);
737 
738 /**
739  * debugfs_remove_recursive - recursively removes a directory
740  * @dentry: a pointer to a the dentry of the directory to be removed.  If this
741  *          parameter is NULL or an error value, nothing will be done.
742  *
743  * This function recursively removes a directory tree in debugfs that
744  * was previously created with a call to another debugfs function
745  * (like debugfs_create_file() or variants thereof.)
746  *
747  * This function is required to be called in order for the file to be
748  * removed, no automatic cleanup of files will happen when a module is
749  * removed, you are responsible here.
750  */
751 void debugfs_remove_recursive(struct dentry *dentry)
752 {
753 	struct dentry *child, *parent;
754 
755 	if (IS_ERR_OR_NULL(dentry))
756 		return;
757 
758 	parent = dentry;
759  down:
760 	inode_lock(d_inode(parent));
761  loop:
762 	/*
763 	 * The parent->d_subdirs is protected by the d_lock. Outside that
764 	 * lock, the child can be unlinked and set to be freed which can
765 	 * use the d_u.d_child as the rcu head and corrupt this list.
766 	 */
767 	spin_lock(&parent->d_lock);
768 	list_for_each_entry(child, &parent->d_subdirs, d_child) {
769 		if (!simple_positive(child))
770 			continue;
771 
772 		/* perhaps simple_empty(child) makes more sense */
773 		if (!list_empty(&child->d_subdirs)) {
774 			spin_unlock(&parent->d_lock);
775 			inode_unlock(d_inode(parent));
776 			parent = child;
777 			goto down;
778 		}
779 
780 		spin_unlock(&parent->d_lock);
781 
782 		if (!__debugfs_remove(child, parent))
783 			simple_release_fs(&debugfs_mount, &debugfs_mount_count);
784 
785 		/*
786 		 * The parent->d_lock protects agaist child from unlinking
787 		 * from d_subdirs. When releasing the parent->d_lock we can
788 		 * no longer trust that the next pointer is valid.
789 		 * Restart the loop. We'll skip this one with the
790 		 * simple_positive() check.
791 		 */
792 		goto loop;
793 	}
794 	spin_unlock(&parent->d_lock);
795 
796 	inode_unlock(d_inode(parent));
797 	child = parent;
798 	parent = parent->d_parent;
799 	inode_lock(d_inode(parent));
800 
801 	if (child != dentry)
802 		/* go up */
803 		goto loop;
804 
805 	if (!__debugfs_remove(child, parent))
806 		simple_release_fs(&debugfs_mount, &debugfs_mount_count);
807 	inode_unlock(d_inode(parent));
808 }
809 EXPORT_SYMBOL_GPL(debugfs_remove_recursive);
810 
811 /**
812  * debugfs_rename - rename a file/directory in the debugfs filesystem
813  * @old_dir: a pointer to the parent dentry for the renamed object. This
814  *          should be a directory dentry.
815  * @old_dentry: dentry of an object to be renamed.
816  * @new_dir: a pointer to the parent dentry where the object should be
817  *          moved. This should be a directory dentry.
818  * @new_name: a pointer to a string containing the target name.
819  *
820  * This function renames a file/directory in debugfs.  The target must not
821  * exist for rename to succeed.
822  *
823  * This function will return a pointer to old_dentry (which is updated to
824  * reflect renaming) if it succeeds. If an error occurs, %NULL will be
825  * returned.
826  *
827  * If debugfs is not enabled in the kernel, the value -%ENODEV will be
828  * returned.
829  */
830 struct dentry *debugfs_rename(struct dentry *old_dir, struct dentry *old_dentry,
831 		struct dentry *new_dir, const char *new_name)
832 {
833 	int error;
834 	struct dentry *dentry = NULL, *trap;
835 	struct name_snapshot old_name;
836 
837 	if (IS_ERR(old_dir))
838 		return old_dir;
839 	if (IS_ERR(new_dir))
840 		return new_dir;
841 	if (IS_ERR_OR_NULL(old_dentry))
842 		return old_dentry;
843 
844 	trap = lock_rename(new_dir, old_dir);
845 	/* Source or destination directories don't exist? */
846 	if (d_really_is_negative(old_dir) || d_really_is_negative(new_dir))
847 		goto exit;
848 	/* Source does not exist, cyclic rename, or mountpoint? */
849 	if (d_really_is_negative(old_dentry) || old_dentry == trap ||
850 	    d_mountpoint(old_dentry))
851 		goto exit;
852 	dentry = lookup_one_len(new_name, new_dir, strlen(new_name));
853 	/* Lookup failed, cyclic rename or target exists? */
854 	if (IS_ERR(dentry) || dentry == trap || d_really_is_positive(dentry))
855 		goto exit;
856 
857 	take_dentry_name_snapshot(&old_name, old_dentry);
858 
859 	error = simple_rename(d_inode(old_dir), old_dentry, d_inode(new_dir),
860 			      dentry, 0);
861 	if (error) {
862 		release_dentry_name_snapshot(&old_name);
863 		goto exit;
864 	}
865 	d_move(old_dentry, dentry);
866 	fsnotify_move(d_inode(old_dir), d_inode(new_dir), &old_name.name,
867 		d_is_dir(old_dentry),
868 		NULL, old_dentry);
869 	release_dentry_name_snapshot(&old_name);
870 	unlock_rename(new_dir, old_dir);
871 	dput(dentry);
872 	return old_dentry;
873 exit:
874 	if (dentry && !IS_ERR(dentry))
875 		dput(dentry);
876 	unlock_rename(new_dir, old_dir);
877 	if (IS_ERR(dentry))
878 		return dentry;
879 	return ERR_PTR(-EINVAL);
880 }
881 EXPORT_SYMBOL_GPL(debugfs_rename);
882 
883 /**
884  * debugfs_initialized - Tells whether debugfs has been registered
885  */
886 bool debugfs_initialized(void)
887 {
888 	return debugfs_registered;
889 }
890 EXPORT_SYMBOL_GPL(debugfs_initialized);
891 
892 static int __init debugfs_init(void)
893 {
894 	int retval;
895 
896 	retval = sysfs_create_mount_point(kernel_kobj, "debug");
897 	if (retval)
898 		return retval;
899 
900 	retval = register_filesystem(&debug_fs_type);
901 	if (retval)
902 		sysfs_remove_mount_point(kernel_kobj, "debug");
903 	else
904 		debugfs_registered = true;
905 
906 	return retval;
907 }
908 core_initcall(debugfs_init);
909 
910