1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 *  inode.c - part of tracefs, a pseudo file system for activating tracing
  4 *
  5 * Based on debugfs by: Greg Kroah-Hartman <greg@kroah.com>
  6 *
  7 *  Copyright (C) 2014 Red Hat Inc, author: Steven Rostedt <srostedt@redhat.com>
  8 *
  9 * tracefs is the file system that is used by the tracing infrastructure.
 10 */
 11
 12#include <linux/module.h>
 13#include <linux/fs.h>
 14#include <linux/mount.h>
 15#include <linux/kobject.h>
 16#include <linux/namei.h>
 17#include <linux/tracefs.h>
 18#include <linux/fsnotify.h>
 19#include <linux/security.h>
 20#include <linux/seq_file.h>
 21#include <linux/parser.h>
 22#include <linux/magic.h>
 23#include <linux/slab.h>
 24
 25#define TRACEFS_DEFAULT_MODE	0700
 26
 27static struct vfsmount *tracefs_mount;
 28static int tracefs_mount_count;
 29static bool tracefs_registered;
 30
 31static ssize_t default_read_file(struct file *file, char __user *buf,
 32				 size_t count, loff_t *ppos)
 33{
 34	return 0;
 35}
 36
 37static ssize_t default_write_file(struct file *file, const char __user *buf,
 38				   size_t count, loff_t *ppos)
 39{
 40	return count;
 41}
 42
 43static const struct file_operations tracefs_file_operations = {
 44	.read =		default_read_file,
 45	.write =	default_write_file,
 46	.open =		simple_open,
 47	.llseek =	noop_llseek,
 48};
 49
 50static struct tracefs_dir_ops {
 51	int (*mkdir)(const char *name);
 52	int (*rmdir)(const char *name);
 53} tracefs_ops __ro_after_init;
 54
 55static char *get_dname(struct dentry *dentry)
 56{
 57	const char *dname;
 58	char *name;
 59	int len = dentry->d_name.len;
 60
 61	dname = dentry->d_name.name;
 62	name = kmalloc(len + 1, GFP_KERNEL);
 63	if (!name)
 64		return NULL;
 65	memcpy(name, dname, len);
 66	name[len] = 0;
 67	return name;
 68}
 69
 70static int tracefs_syscall_mkdir(struct inode *inode, struct dentry *dentry, umode_t mode)
 71{
 72	char *name;
 73	int ret;
 74
 75	name = get_dname(dentry);
 76	if (!name)
 77		return -ENOMEM;
 78
 79	/*
 80	 * The mkdir call can call the generic functions that create
 81	 * the files within the tracefs system. It is up to the individual
 82	 * mkdir routine to handle races.
 83	 */
 84	inode_unlock(inode);
 85	ret = tracefs_ops.mkdir(name);
 86	inode_lock(inode);
 87
 88	kfree(name);
 89
 90	return ret;
 91}
 92
 93static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
 94{
 95	char *name;
 96	int ret;
 97
 98	name = get_dname(dentry);
 99	if (!name)
100		return -ENOMEM;
101
102	/*
103	 * The rmdir call can call the generic functions that create
104	 * the files within the tracefs system. It is up to the individual
105	 * rmdir routine to handle races.
106	 * This time we need to unlock not only the parent (inode) but
107	 * also the directory that is being deleted.
108	 */
109	inode_unlock(inode);
110	inode_unlock(dentry->d_inode);
111
112	ret = tracefs_ops.rmdir(name);
113
114	inode_lock_nested(inode, I_MUTEX_PARENT);
115	inode_lock(dentry->d_inode);
116
117	kfree(name);
118
119	return ret;
120}
121
122static const struct inode_operations tracefs_dir_inode_operations = {
123	.lookup		= simple_lookup,
124	.mkdir		= tracefs_syscall_mkdir,
125	.rmdir		= tracefs_syscall_rmdir,
126};
127
128static struct inode *tracefs_get_inode(struct super_block *sb)
129{
130	struct inode *inode = new_inode(sb);
131	if (inode) {
132		inode->i_ino = get_next_ino();
133		inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
134	}
135	return inode;
136}
137
138struct tracefs_mount_opts {
139	kuid_t uid;
140	kgid_t gid;
141	umode_t mode;
142};
143
144enum {
145	Opt_uid,
146	Opt_gid,
147	Opt_mode,
148	Opt_err
149};
150
151static const match_table_t tokens = {
152	{Opt_uid, "uid=%u"},
153	{Opt_gid, "gid=%u"},
154	{Opt_mode, "mode=%o"},
155	{Opt_err, NULL}
156};
157
158struct tracefs_fs_info {
159	struct tracefs_mount_opts mount_opts;
160};
161
162static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
163{
164	substring_t args[MAX_OPT_ARGS];
165	int option;
166	int token;
167	kuid_t uid;
168	kgid_t gid;
169	char *p;
170
171	opts->mode = TRACEFS_DEFAULT_MODE;
172
173	while ((p = strsep(&data, ",")) != NULL) {
174		if (!*p)
175			continue;
176
177		token = match_token(p, tokens, args);
178		switch (token) {
179		case Opt_uid:
180			if (match_int(&args[0], &option))
181				return -EINVAL;
182			uid = make_kuid(current_user_ns(), option);
183			if (!uid_valid(uid))
184				return -EINVAL;
185			opts->uid = uid;
186			break;
187		case Opt_gid:
188			if (match_int(&args[0], &option))
189				return -EINVAL;
190			gid = make_kgid(current_user_ns(), option);
191			if (!gid_valid(gid))
192				return -EINVAL;
193			opts->gid = gid;
194			break;
195		case Opt_mode:
196			if (match_octal(&args[0], &option))
197				return -EINVAL;
198			opts->mode = option & S_IALLUGO;
199			break;
200		/*
201		 * We might like to report bad mount options here;
202		 * but traditionally tracefs has ignored all mount options
203		 */
204		}
205	}
206
207	return 0;
208}
209
210static int tracefs_apply_options(struct super_block *sb)
211{
212	struct tracefs_fs_info *fsi = sb->s_fs_info;
213	struct inode *inode = sb->s_root->d_inode;
214	struct tracefs_mount_opts *opts = &fsi->mount_opts;
215
216	inode->i_mode &= ~S_IALLUGO;
217	inode->i_mode |= opts->mode;
218
219	inode->i_uid = opts->uid;
220	inode->i_gid = opts->gid;
221
222	return 0;
223}
224
225static int tracefs_remount(struct super_block *sb, int *flags, char *data)
226{
227	int err;
228	struct tracefs_fs_info *fsi = sb->s_fs_info;
229
230	sync_filesystem(sb);
231	err = tracefs_parse_options(data, &fsi->mount_opts);
232	if (err)
233		goto fail;
234
235	tracefs_apply_options(sb);
236
237fail:
238	return err;
239}
240
241static int tracefs_show_options(struct seq_file *m, struct dentry *root)
242{
243	struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
244	struct tracefs_mount_opts *opts = &fsi->mount_opts;
245
246	if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
247		seq_printf(m, ",uid=%u",
248			   from_kuid_munged(&init_user_ns, opts->uid));
249	if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
250		seq_printf(m, ",gid=%u",
251			   from_kgid_munged(&init_user_ns, opts->gid));
252	if (opts->mode != TRACEFS_DEFAULT_MODE)
253		seq_printf(m, ",mode=%o", opts->mode);
254
255	return 0;
256}
257
258static const struct super_operations tracefs_super_operations = {
259	.statfs		= simple_statfs,
260	.remount_fs	= tracefs_remount,
261	.show_options	= tracefs_show_options,
262};
263
264static int trace_fill_super(struct super_block *sb, void *data, int silent)
265{
266	static const struct tree_descr trace_files[] = {{""}};
267	struct tracefs_fs_info *fsi;
268	int err;
269
270	fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
271	sb->s_fs_info = fsi;
272	if (!fsi) {
273		err = -ENOMEM;
274		goto fail;
275	}
276
277	err = tracefs_parse_options(data, &fsi->mount_opts);
278	if (err)
279		goto fail;
280
281	err  =  simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
282	if (err)
283		goto fail;
284
285	sb->s_op = &tracefs_super_operations;
286
287	tracefs_apply_options(sb);
288
289	return 0;
290
291fail:
292	kfree(fsi);
293	sb->s_fs_info = NULL;
294	return err;
295}
296
297static struct dentry *trace_mount(struct file_system_type *fs_type,
298			int flags, const char *dev_name,
299			void *data)
300{
301	return mount_single(fs_type, flags, data, trace_fill_super);
302}
303
304static struct file_system_type trace_fs_type = {
305	.owner =	THIS_MODULE,
306	.name =		"tracefs",
307	.mount =	trace_mount,
308	.kill_sb =	kill_litter_super,
309};
310MODULE_ALIAS_FS("tracefs");
311
312static struct dentry *start_creating(const char *name, struct dentry *parent)
313{
314	struct dentry *dentry;
315	int error;
316
317	pr_debug("tracefs: creating file '%s'\n",name);
318
319	error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
320			      &tracefs_mount_count);
321	if (error)
322		return ERR_PTR(error);
323
324	/* If the parent is not specified, we create it in the root.
325	 * We need the root dentry to do this, which is in the super
326	 * block. A pointer to that is in the struct vfsmount that we
327	 * have around.
328	 */
329	if (!parent)
330		parent = tracefs_mount->mnt_root;
331
332	inode_lock(parent->d_inode);
333	if (unlikely(IS_DEADDIR(parent->d_inode)))
334		dentry = ERR_PTR(-ENOENT);
335	else
336		dentry = lookup_one_len(name, parent, strlen(name));
337	if (!IS_ERR(dentry) && dentry->d_inode) {
338		dput(dentry);
339		dentry = ERR_PTR(-EEXIST);
340	}
341
342	if (IS_ERR(dentry)) {
343		inode_unlock(parent->d_inode);
344		simple_release_fs(&tracefs_mount, &tracefs_mount_count);
345	}
346
347	return dentry;
348}
349
350static struct dentry *failed_creating(struct dentry *dentry)
351{
352	inode_unlock(dentry->d_parent->d_inode);
353	dput(dentry);
354	simple_release_fs(&tracefs_mount, &tracefs_mount_count);
355	return NULL;
356}
357
358static struct dentry *end_creating(struct dentry *dentry)
359{
360	inode_unlock(dentry->d_parent->d_inode);
361	return dentry;
362}
363
364/**
365 * tracefs_create_file - create a file in the tracefs filesystem
366 * @name: a pointer to a string containing the name of the file to create.
367 * @mode: the permission that the file should have.
368 * @parent: a pointer to the parent dentry for this file.  This should be a
369 *          directory dentry if set.  If this parameter is NULL, then the
370 *          file will be created in the root of the tracefs filesystem.
371 * @data: a pointer to something that the caller will want to get to later
372 *        on.  The inode.i_private pointer will point to this value on
373 *        the open() call.
374 * @fops: a pointer to a struct file_operations that should be used for
375 *        this file.
376 *
377 * This is the basic "create a file" function for tracefs.  It allows for a
378 * wide range of flexibility in creating a file, or a directory (if you want
379 * to create a directory, the tracefs_create_dir() function is
380 * recommended to be used instead.)
381 *
382 * This function will return a pointer to a dentry if it succeeds.  This
383 * pointer must be passed to the tracefs_remove() function when the file is
384 * to be removed (no automatic cleanup happens if your module is unloaded,
385 * you are responsible here.)  If an error occurs, %NULL will be returned.
386 *
387 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
388 * returned.
389 */
390struct dentry *tracefs_create_file(const char *name, umode_t mode,
391				   struct dentry *parent, void *data,
392				   const struct file_operations *fops)
393{
394	struct dentry *dentry;
395	struct inode *inode;
396
397	if (security_locked_down(LOCKDOWN_TRACEFS))
398		return NULL;
399
400	if (!(mode & S_IFMT))
401		mode |= S_IFREG;
402	BUG_ON(!S_ISREG(mode));
403	dentry = start_creating(name, parent);
404
405	if (IS_ERR(dentry))
406		return NULL;
407
408	inode = tracefs_get_inode(dentry->d_sb);
409	if (unlikely(!inode))
410		return failed_creating(dentry);
411
412	inode->i_mode = mode;
413	inode->i_fop = fops ? fops : &tracefs_file_operations;
414	inode->i_private = data;
415	d_instantiate(dentry, inode);
416	fsnotify_create(dentry->d_parent->d_inode, dentry);
417	return end_creating(dentry);
418}
419
420static struct dentry *__create_dir(const char *name, struct dentry *parent,
421				   const struct inode_operations *ops)
422{
423	struct dentry *dentry = start_creating(name, parent);
424	struct inode *inode;
425
426	if (IS_ERR(dentry))
427		return NULL;
428
429	inode = tracefs_get_inode(dentry->d_sb);
430	if (unlikely(!inode))
431		return failed_creating(dentry);
432
433	inode->i_mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
434	inode->i_op = ops;
435	inode->i_fop = &simple_dir_operations;
436
437	/* directory inodes start off with i_nlink == 2 (for "." entry) */
438	inc_nlink(inode);
439	d_instantiate(dentry, inode);
440	inc_nlink(dentry->d_parent->d_inode);
441	fsnotify_mkdir(dentry->d_parent->d_inode, dentry);
442	return end_creating(dentry);
443}
444
445/**
446 * tracefs_create_dir - create a directory in the tracefs filesystem
447 * @name: a pointer to a string containing the name of the directory to
448 *        create.
449 * @parent: a pointer to the parent dentry for this file.  This should be a
450 *          directory dentry if set.  If this parameter is NULL, then the
451 *          directory will be created in the root of the tracefs filesystem.
452 *
453 * This function creates a directory in tracefs with the given name.
454 *
455 * This function will return a pointer to a dentry if it succeeds.  This
456 * pointer must be passed to the tracefs_remove() function when the file is
457 * to be removed. If an error occurs, %NULL will be returned.
458 *
459 * If tracing is not enabled in the kernel, the value -%ENODEV will be
460 * returned.
461 */
462struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
463{
464	return __create_dir(name, parent, &simple_dir_inode_operations);
465}
466
467/**
468 * tracefs_create_instance_dir - create the tracing instances directory
469 * @name: The name of the instances directory to create
470 * @parent: The parent directory that the instances directory will exist
471 * @mkdir: The function to call when a mkdir is performed.
472 * @rmdir: The function to call when a rmdir is performed.
473 *
474 * Only one instances directory is allowed.
475 *
476 * The instances directory is special as it allows for mkdir and rmdir to
477 * to be done by userspace. When a mkdir or rmdir is performed, the inode
478 * locks are released and the methhods passed in (@mkdir and @rmdir) are
479 * called without locks and with the name of the directory being created
480 * within the instances directory.
481 *
482 * Returns the dentry of the instances directory.
483 */
484__init struct dentry *tracefs_create_instance_dir(const char *name,
485					  struct dentry *parent,
486					  int (*mkdir)(const char *name),
487					  int (*rmdir)(const char *name))
488{
489	struct dentry *dentry;
490
491	/* Only allow one instance of the instances directory. */
492	if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
493		return NULL;
494
495	dentry = __create_dir(name, parent, &tracefs_dir_inode_operations);
496	if (!dentry)
497		return NULL;
498
499	tracefs_ops.mkdir = mkdir;
500	tracefs_ops.rmdir = rmdir;
501
502	return dentry;
503}
504
505static void remove_one(struct dentry *victim)
506{
507	simple_release_fs(&tracefs_mount, &tracefs_mount_count);
508}
509
510/**
511 * tracefs_remove - recursively removes a directory
512 * @dentry: a pointer to a the dentry of the directory to be removed.
513 *
514 * This function recursively removes a directory tree in tracefs that
515 * was previously created with a call to another tracefs function
516 * (like tracefs_create_file() or variants thereof.)
517 */
518void tracefs_remove(struct dentry *dentry)
519{
520	if (IS_ERR_OR_NULL(dentry))
521		return;
522
523	simple_pin_fs(&trace_fs_type, &tracefs_mount, &tracefs_mount_count);
524	simple_recursive_removal(dentry, remove_one);
525	simple_release_fs(&tracefs_mount, &tracefs_mount_count);
526}
527
528/**
529 * tracefs_initialized - Tells whether tracefs has been registered
530 */
531bool tracefs_initialized(void)
532{
533	return tracefs_registered;
534}
535
536static int __init tracefs_init(void)
537{
538	int retval;
539
540	retval = sysfs_create_mount_point(kernel_kobj, "tracing");
541	if (retval)
542		return -EINVAL;
543
544	retval = register_filesystem(&trace_fs_type);
545	if (!retval)
546		tracefs_registered = true;
547
548	return retval;
549}
550core_initcall(tracefs_init);