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 user_namespace *mnt_userns,
 71				 struct inode *inode, struct dentry *dentry,
 72				 umode_t mode)
 73{
 74	char *name;
 75	int ret;
 76
 77	name = get_dname(dentry);
 78	if (!name)
 79		return -ENOMEM;
 80
 81	/*
 82	 * The mkdir call can call the generic functions that create
 83	 * the files within the tracefs system. It is up to the individual
 84	 * mkdir routine to handle races.
 85	 */
 86	inode_unlock(inode);
 87	ret = tracefs_ops.mkdir(name);
 88	inode_lock(inode);
 89
 90	kfree(name);
 91
 92	return ret;
 93}
 94
 95static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
 96{
 97	char *name;
 98	int ret;
 99
100	name = get_dname(dentry);
101	if (!name)
102		return -ENOMEM;
103
104	/*
105	 * The rmdir call can call the generic functions that create
106	 * the files within the tracefs system. It is up to the individual
107	 * rmdir routine to handle races.
108	 * This time we need to unlock not only the parent (inode) but
109	 * also the directory that is being deleted.
110	 */
111	inode_unlock(inode);
112	inode_unlock(dentry->d_inode);
113
114	ret = tracefs_ops.rmdir(name);
115
116	inode_lock_nested(inode, I_MUTEX_PARENT);
117	inode_lock(dentry->d_inode);
118
119	kfree(name);
120
121	return ret;
122}
123
124static const struct inode_operations tracefs_dir_inode_operations = {
125	.lookup		= simple_lookup,
126	.mkdir		= tracefs_syscall_mkdir,
127	.rmdir		= tracefs_syscall_rmdir,
128};
129
130static struct inode *tracefs_get_inode(struct super_block *sb)
131{
132	struct inode *inode = new_inode(sb);
133	if (inode) {
134		inode->i_ino = get_next_ino();
135		inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
136	}
137	return inode;
138}
139
140struct tracefs_mount_opts {
141	kuid_t uid;
142	kgid_t gid;
143	umode_t mode;
144};
145
146enum {
147	Opt_uid,
148	Opt_gid,
149	Opt_mode,
150	Opt_err
151};
152
153static const match_table_t tokens = {
154	{Opt_uid, "uid=%u"},
155	{Opt_gid, "gid=%u"},
156	{Opt_mode, "mode=%o"},
157	{Opt_err, NULL}
158};
159
160struct tracefs_fs_info {
161	struct tracefs_mount_opts mount_opts;
162};
163
164static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
165{
166	substring_t args[MAX_OPT_ARGS];
167	int option;
168	int token;
169	kuid_t uid;
170	kgid_t gid;
171	char *p;
172
173	opts->mode = TRACEFS_DEFAULT_MODE;
174
175	while ((p = strsep(&data, ",")) != NULL) {
176		if (!*p)
177			continue;
178
179		token = match_token(p, tokens, args);
180		switch (token) {
181		case Opt_uid:
182			if (match_int(&args[0], &option))
183				return -EINVAL;
184			uid = make_kuid(current_user_ns(), option);
185			if (!uid_valid(uid))
186				return -EINVAL;
187			opts->uid = uid;
188			break;
189		case Opt_gid:
190			if (match_int(&args[0], &option))
191				return -EINVAL;
192			gid = make_kgid(current_user_ns(), option);
193			if (!gid_valid(gid))
194				return -EINVAL;
195			opts->gid = gid;
196			break;
197		case Opt_mode:
198			if (match_octal(&args[0], &option))
199				return -EINVAL;
200			opts->mode = option & S_IALLUGO;
201			break;
202		/*
203		 * We might like to report bad mount options here;
204		 * but traditionally tracefs has ignored all mount options
205		 */
206		}
207	}
208
209	return 0;
210}
211
212static int tracefs_apply_options(struct super_block *sb)
213{
214	struct tracefs_fs_info *fsi = sb->s_fs_info;
215	struct inode *inode = sb->s_root->d_inode;
216	struct tracefs_mount_opts *opts = &fsi->mount_opts;
217
218	inode->i_mode &= ~S_IALLUGO;
219	inode->i_mode |= opts->mode;
220
221	inode->i_uid = opts->uid;
222	inode->i_gid = opts->gid;
223
224	return 0;
225}
226
227static int tracefs_remount(struct super_block *sb, int *flags, char *data)
228{
229	int err;
230	struct tracefs_fs_info *fsi = sb->s_fs_info;
231
232	sync_filesystem(sb);
233	err = tracefs_parse_options(data, &fsi->mount_opts);
234	if (err)
235		goto fail;
236
237	tracefs_apply_options(sb);
238
239fail:
240	return err;
241}
242
243static int tracefs_show_options(struct seq_file *m, struct dentry *root)
244{
245	struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
246	struct tracefs_mount_opts *opts = &fsi->mount_opts;
247
248	if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
249		seq_printf(m, ",uid=%u",
250			   from_kuid_munged(&init_user_ns, opts->uid));
251	if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
252		seq_printf(m, ",gid=%u",
253			   from_kgid_munged(&init_user_ns, opts->gid));
254	if (opts->mode != TRACEFS_DEFAULT_MODE)
255		seq_printf(m, ",mode=%o", opts->mode);
256
257	return 0;
258}
259
260static const struct super_operations tracefs_super_operations = {
261	.statfs		= simple_statfs,
262	.remount_fs	= tracefs_remount,
263	.show_options	= tracefs_show_options,
264};
265
266static int trace_fill_super(struct super_block *sb, void *data, int silent)
267{
268	static const struct tree_descr trace_files[] = {{""}};
269	struct tracefs_fs_info *fsi;
270	int err;
271
272	fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
273	sb->s_fs_info = fsi;
274	if (!fsi) {
275		err = -ENOMEM;
276		goto fail;
277	}
278
279	err = tracefs_parse_options(data, &fsi->mount_opts);
280	if (err)
281		goto fail;
282
283	err  =  simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
284	if (err)
285		goto fail;
286
287	sb->s_op = &tracefs_super_operations;
288
289	tracefs_apply_options(sb);
290
291	return 0;
292
293fail:
294	kfree(fsi);
295	sb->s_fs_info = NULL;
296	return err;
297}
298
299static struct dentry *trace_mount(struct file_system_type *fs_type,
300			int flags, const char *dev_name,
301			void *data)
302{
303	return mount_single(fs_type, flags, data, trace_fill_super);
304}
305
306static struct file_system_type trace_fs_type = {
307	.owner =	THIS_MODULE,
308	.name =		"tracefs",
309	.mount =	trace_mount,
310	.kill_sb =	kill_litter_super,
311};
312MODULE_ALIAS_FS("tracefs");
313
314static struct dentry *start_creating(const char *name, struct dentry *parent)
315{
316	struct dentry *dentry;
317	int error;
318
319	pr_debug("tracefs: creating file '%s'\n",name);
320
321	error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
322			      &tracefs_mount_count);
323	if (error)
324		return ERR_PTR(error);
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 = tracefs_mount->mnt_root;
333
334	inode_lock(parent->d_inode);
335	if (unlikely(IS_DEADDIR(parent->d_inode)))
336		dentry = ERR_PTR(-ENOENT);
337	else
338		dentry = lookup_one_len(name, parent, strlen(name));
339	if (!IS_ERR(dentry) && dentry->d_inode) {
340		dput(dentry);
341		dentry = ERR_PTR(-EEXIST);
342	}
343
344	if (IS_ERR(dentry)) {
345		inode_unlock(parent->d_inode);
346		simple_release_fs(&tracefs_mount, &tracefs_mount_count);
347	}
348
349	return dentry;
350}
351
352static struct dentry *failed_creating(struct dentry *dentry)
353{
354	inode_unlock(dentry->d_parent->d_inode);
355	dput(dentry);
356	simple_release_fs(&tracefs_mount, &tracefs_mount_count);
357	return NULL;
358}
359
360static struct dentry *end_creating(struct dentry *dentry)
361{
362	inode_unlock(dentry->d_parent->d_inode);
363	return dentry;
364}
365
366/**
367 * tracefs_create_file - create a file in the tracefs filesystem
368 * @name: a pointer to a string containing the name of the file to create.
369 * @mode: the permission that the file should have.
370 * @parent: a pointer to the parent dentry for this file.  This should be a
371 *          directory dentry if set.  If this parameter is NULL, then the
372 *          file will be created in the root of the tracefs filesystem.
373 * @data: a pointer to something that the caller will want to get to later
374 *        on.  The inode.i_private pointer will point to this value on
375 *        the open() call.
376 * @fops: a pointer to a struct file_operations that should be used for
377 *        this file.
378 *
379 * This is the basic "create a file" function for tracefs.  It allows for a
380 * wide range of flexibility in creating a file, or a directory (if you want
381 * to create a directory, the tracefs_create_dir() function is
382 * recommended to be used instead.)
383 *
384 * This function will return a pointer to a dentry if it succeeds.  This
385 * pointer must be passed to the tracefs_remove() function when the file is
386 * to be removed (no automatic cleanup happens if your module is unloaded,
387 * you are responsible here.)  If an error occurs, %NULL will be returned.
388 *
389 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
390 * returned.
391 */
392struct dentry *tracefs_create_file(const char *name, umode_t mode,
393				   struct dentry *parent, void *data,
394				   const struct file_operations *fops)
395{
396	struct dentry *dentry;
397	struct inode *inode;
398
399	if (security_locked_down(LOCKDOWN_TRACEFS))
400		return NULL;
401
402	if (!(mode & S_IFMT))
403		mode |= S_IFREG;
404	BUG_ON(!S_ISREG(mode));
405	dentry = start_creating(name, parent);
406
407	if (IS_ERR(dentry))
408		return NULL;
409
410	inode = tracefs_get_inode(dentry->d_sb);
411	if (unlikely(!inode))
412		return failed_creating(dentry);
413
414	inode->i_mode = mode;
415	inode->i_fop = fops ? fops : &tracefs_file_operations;
416	inode->i_private = data;
417	d_instantiate(dentry, inode);
418	fsnotify_create(dentry->d_parent->d_inode, dentry);
419	return end_creating(dentry);
420}
421
422static struct dentry *__create_dir(const char *name, struct dentry *parent,
423				   const struct inode_operations *ops)
424{
425	struct dentry *dentry = start_creating(name, parent);
426	struct inode *inode;
427
428	if (IS_ERR(dentry))
429		return NULL;
430
431	inode = tracefs_get_inode(dentry->d_sb);
432	if (unlikely(!inode))
433		return failed_creating(dentry);
434
435	inode->i_mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
436	inode->i_op = ops;
437	inode->i_fop = &simple_dir_operations;
438
439	/* directory inodes start off with i_nlink == 2 (for "." entry) */
440	inc_nlink(inode);
441	d_instantiate(dentry, inode);
442	inc_nlink(dentry->d_parent->d_inode);
443	fsnotify_mkdir(dentry->d_parent->d_inode, dentry);
444	return end_creating(dentry);
445}
446
447/**
448 * tracefs_create_dir - create a directory in the tracefs filesystem
449 * @name: a pointer to a string containing the name of the directory to
450 *        create.
451 * @parent: a pointer to the parent dentry for this file.  This should be a
452 *          directory dentry if set.  If this parameter is NULL, then the
453 *          directory will be created in the root of the tracefs filesystem.
454 *
455 * This function creates a directory in tracefs with the given name.
456 *
457 * This function will return a pointer to a dentry if it succeeds.  This
458 * pointer must be passed to the tracefs_remove() function when the file is
459 * to be removed. If an error occurs, %NULL will be returned.
460 *
461 * If tracing is not enabled in the kernel, the value -%ENODEV will be
462 * returned.
463 */
464struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
465{
466	return __create_dir(name, parent, &simple_dir_inode_operations);
467}
468
469/**
470 * tracefs_create_instance_dir - create the tracing instances directory
471 * @name: The name of the instances directory to create
472 * @parent: The parent directory that the instances directory will exist
473 * @mkdir: The function to call when a mkdir is performed.
474 * @rmdir: The function to call when a rmdir is performed.
475 *
476 * Only one instances directory is allowed.
477 *
478 * The instances directory is special as it allows for mkdir and rmdir to
479 * to be done by userspace. When a mkdir or rmdir is performed, the inode
480 * locks are released and the methods passed in (@mkdir and @rmdir) are
481 * called without locks and with the name of the directory being created
482 * within the instances directory.
483 *
484 * Returns the dentry of the instances directory.
485 */
486__init struct dentry *tracefs_create_instance_dir(const char *name,
487					  struct dentry *parent,
488					  int (*mkdir)(const char *name),
489					  int (*rmdir)(const char *name))
490{
491	struct dentry *dentry;
492
493	/* Only allow one instance of the instances directory. */
494	if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
495		return NULL;
496
497	dentry = __create_dir(name, parent, &tracefs_dir_inode_operations);
498	if (!dentry)
499		return NULL;
500
501	tracefs_ops.mkdir = mkdir;
502	tracefs_ops.rmdir = rmdir;
503
504	return dentry;
505}
506
507static void remove_one(struct dentry *victim)
508{
509	simple_release_fs(&tracefs_mount, &tracefs_mount_count);
510}
511
512/**
513 * tracefs_remove - recursively removes a directory
514 * @dentry: a pointer to a the dentry of the directory to be removed.
515 *
516 * This function recursively removes a directory tree in tracefs that
517 * was previously created with a call to another tracefs function
518 * (like tracefs_create_file() or variants thereof.)
519 */
520void tracefs_remove(struct dentry *dentry)
521{
522	if (IS_ERR_OR_NULL(dentry))
523		return;
524
525	simple_pin_fs(&trace_fs_type, &tracefs_mount, &tracefs_mount_count);
526	simple_recursive_removal(dentry, remove_one);
527	simple_release_fs(&tracefs_mount, &tracefs_mount_count);
528}
529
530/**
531 * tracefs_initialized - Tells whether tracefs has been registered
532 */
533bool tracefs_initialized(void)
534{
535	return tracefs_registered;
536}
537
538static int __init tracefs_init(void)
539{
540	int retval;
541
542	retval = sysfs_create_mount_point(kernel_kobj, "tracing");
543	if (retval)
544		return -EINVAL;
545
546	retval = register_filesystem(&trace_fs_type);
547	if (!retval)
548		tracefs_registered = true;
549
550	return retval;
551}
552core_initcall(tracefs_init);