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#include "internal.h"
25
26#define TRACEFS_DEFAULT_MODE	0700
27static struct kmem_cache *tracefs_inode_cachep __ro_after_init;
28
29static struct vfsmount *tracefs_mount;
30static int tracefs_mount_count;
31static bool tracefs_registered;
32
33static struct inode *tracefs_alloc_inode(struct super_block *sb)
34{
35	struct tracefs_inode *ti;
36
37	ti = kmem_cache_alloc(tracefs_inode_cachep, GFP_KERNEL);
38	if (!ti)
39		return NULL;
40
41	return &ti->vfs_inode;
42}
43
44static void tracefs_free_inode(struct inode *inode)
45{
46	kmem_cache_free(tracefs_inode_cachep, get_tracefs(inode));
47}
48
49static ssize_t default_read_file(struct file *file, char __user *buf,
50				 size_t count, loff_t *ppos)
51{
52	return 0;
53}
54
55static ssize_t default_write_file(struct file *file, const char __user *buf,
56				   size_t count, loff_t *ppos)
57{
58	return count;
59}
60
61static const struct file_operations tracefs_file_operations = {
62	.read =		default_read_file,
63	.write =	default_write_file,
64	.open =		simple_open,
65	.llseek =	noop_llseek,
66};
67
68static struct tracefs_dir_ops {
69	int (*mkdir)(const char *name);
70	int (*rmdir)(const char *name);
71} tracefs_ops __ro_after_init;
72
73static char *get_dname(struct dentry *dentry)
74{
75	const char *dname;
76	char *name;
77	int len = dentry->d_name.len;
78
79	dname = dentry->d_name.name;
80	name = kmalloc(len + 1, GFP_KERNEL);
81	if (!name)
82		return NULL;
83	memcpy(name, dname, len);
84	name[len] = 0;
85	return name;
86}
87
88static int tracefs_syscall_mkdir(struct mnt_idmap *idmap,
89				 struct inode *inode, struct dentry *dentry,
90				 umode_t mode)
91{
92	struct tracefs_inode *ti;
93	char *name;
94	int ret;
95
96	name = get_dname(dentry);
97	if (!name)
98		return -ENOMEM;
99
100	/*
101	 * This is a new directory that does not take the default of
102	 * the rootfs. It becomes the default permissions for all the
103	 * files and directories underneath it.
104	 */
105	ti = get_tracefs(inode);
106	ti->flags |= TRACEFS_INSTANCE_INODE;
107	ti->private = inode;
108
109	/*
110	 * The mkdir call can call the generic functions that create
111	 * the files within the tracefs system. It is up to the individual
112	 * mkdir routine to handle races.
113	 */
114	inode_unlock(inode);
115	ret = tracefs_ops.mkdir(name);
116	inode_lock(inode);
117
118	kfree(name);
119
120	return ret;
121}
122
123static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
124{
125	char *name;
126	int ret;
127
128	name = get_dname(dentry);
129	if (!name)
130		return -ENOMEM;
131
132	/*
133	 * The rmdir call can call the generic functions that create
134	 * the files within the tracefs system. It is up to the individual
135	 * rmdir routine to handle races.
136	 * This time we need to unlock not only the parent (inode) but
137	 * also the directory that is being deleted.
138	 */
139	inode_unlock(inode);
140	inode_unlock(d_inode(dentry));
141
142	ret = tracefs_ops.rmdir(name);
143
144	inode_lock_nested(inode, I_MUTEX_PARENT);
145	inode_lock(d_inode(dentry));
146
147	kfree(name);
148
149	return ret;
150}
151
152static void set_tracefs_inode_owner(struct inode *inode)
153{
154	struct tracefs_inode *ti = get_tracefs(inode);
155	struct inode *root_inode = ti->private;
156
157	/*
158	 * If this inode has never been referenced, then update
159	 * the permissions to the superblock.
160	 */
161	if (!(ti->flags & TRACEFS_UID_PERM_SET))
162		inode->i_uid = root_inode->i_uid;
163
164	if (!(ti->flags & TRACEFS_GID_PERM_SET))
165		inode->i_gid = root_inode->i_gid;
166}
167
168static int tracefs_permission(struct mnt_idmap *idmap,
169			      struct inode *inode, int mask)
170{
171	set_tracefs_inode_owner(inode);
172	return generic_permission(idmap, inode, mask);
173}
174
175static int tracefs_getattr(struct mnt_idmap *idmap,
176			   const struct path *path, struct kstat *stat,
177			   u32 request_mask, unsigned int flags)
178{
179	struct inode *inode = d_backing_inode(path->dentry);
180
181	set_tracefs_inode_owner(inode);
182	generic_fillattr(idmap, request_mask, inode, stat);
183	return 0;
184}
185
186static int tracefs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
187			   struct iattr *attr)
188{
189	unsigned int ia_valid = attr->ia_valid;
190	struct inode *inode = d_inode(dentry);
191	struct tracefs_inode *ti = get_tracefs(inode);
192
193	if (ia_valid & ATTR_UID)
194		ti->flags |= TRACEFS_UID_PERM_SET;
195
196	if (ia_valid & ATTR_GID)
197		ti->flags |= TRACEFS_GID_PERM_SET;
198
199	return simple_setattr(idmap, dentry, attr);
200}
201
202static const struct inode_operations tracefs_instance_dir_inode_operations = {
203	.lookup		= simple_lookup,
204	.mkdir		= tracefs_syscall_mkdir,
205	.rmdir		= tracefs_syscall_rmdir,
206	.permission	= tracefs_permission,
207	.getattr	= tracefs_getattr,
208	.setattr	= tracefs_setattr,
209};
210
211static const struct inode_operations tracefs_dir_inode_operations = {
212	.lookup		= simple_lookup,
213	.permission	= tracefs_permission,
214	.getattr	= tracefs_getattr,
215	.setattr	= tracefs_setattr,
216};
217
218static const struct inode_operations tracefs_file_inode_operations = {
219	.permission	= tracefs_permission,
220	.getattr	= tracefs_getattr,
221	.setattr	= tracefs_setattr,
222};
223
224struct inode *tracefs_get_inode(struct super_block *sb)
225{
226	struct inode *inode = new_inode(sb);
227	if (inode) {
228		inode->i_ino = get_next_ino();
229		simple_inode_init_ts(inode);
230	}
231	return inode;
232}
233
234struct tracefs_mount_opts {
235	kuid_t uid;
236	kgid_t gid;
237	umode_t mode;
238	/* Opt_* bitfield. */
239	unsigned int opts;
240};
241
242enum {
243	Opt_uid,
244	Opt_gid,
245	Opt_mode,
246	Opt_err
247};
248
249static const match_table_t tokens = {
250	{Opt_uid, "uid=%u"},
251	{Opt_gid, "gid=%u"},
252	{Opt_mode, "mode=%o"},
253	{Opt_err, NULL}
254};
255
256struct tracefs_fs_info {
257	struct tracefs_mount_opts mount_opts;
258};
259
260static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
261{
262	substring_t args[MAX_OPT_ARGS];
263	int option;
264	int token;
265	kuid_t uid;
266	kgid_t gid;
267	char *p;
268
269	opts->opts = 0;
270	opts->mode = TRACEFS_DEFAULT_MODE;
271
272	while ((p = strsep(&data, ",")) != NULL) {
273		if (!*p)
274			continue;
275
276		token = match_token(p, tokens, args);
277		switch (token) {
278		case Opt_uid:
279			if (match_int(&args[0], &option))
280				return -EINVAL;
281			uid = make_kuid(current_user_ns(), option);
282			if (!uid_valid(uid))
283				return -EINVAL;
284			opts->uid = uid;
285			break;
286		case Opt_gid:
287			if (match_int(&args[0], &option))
288				return -EINVAL;
289			gid = make_kgid(current_user_ns(), option);
290			if (!gid_valid(gid))
291				return -EINVAL;
292			opts->gid = gid;
293			break;
294		case Opt_mode:
295			if (match_octal(&args[0], &option))
296				return -EINVAL;
297			opts->mode = option & S_IALLUGO;
298			break;
299		/*
300		 * We might like to report bad mount options here;
301		 * but traditionally tracefs has ignored all mount options
302		 */
303		}
304
305		opts->opts |= BIT(token);
306	}
307
308	return 0;
309}
310
311static int tracefs_apply_options(struct super_block *sb, bool remount)
312{
313	struct tracefs_fs_info *fsi = sb->s_fs_info;
314	struct inode *inode = d_inode(sb->s_root);
315	struct tracefs_mount_opts *opts = &fsi->mount_opts;
316	umode_t tmp_mode;
317
318	/*
319	 * On remount, only reset mode/uid/gid if they were provided as mount
320	 * options.
321	 */
322
323	if (!remount || opts->opts & BIT(Opt_mode)) {
324		tmp_mode = READ_ONCE(inode->i_mode) & ~S_IALLUGO;
325		tmp_mode |= opts->mode;
326		WRITE_ONCE(inode->i_mode, tmp_mode);
327	}
328
329	if (!remount || opts->opts & BIT(Opt_uid))
330		inode->i_uid = opts->uid;
331
332	if (!remount || opts->opts & BIT(Opt_gid))
333		inode->i_gid = opts->gid;
334
335	return 0;
336}
337
338static int tracefs_remount(struct super_block *sb, int *flags, char *data)
339{
340	int err;
341	struct tracefs_fs_info *fsi = sb->s_fs_info;
342
343	sync_filesystem(sb);
344	err = tracefs_parse_options(data, &fsi->mount_opts);
345	if (err)
346		goto fail;
347
348	tracefs_apply_options(sb, true);
349
350fail:
351	return err;
352}
353
354static int tracefs_show_options(struct seq_file *m, struct dentry *root)
355{
356	struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
357	struct tracefs_mount_opts *opts = &fsi->mount_opts;
358
359	if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
360		seq_printf(m, ",uid=%u",
361			   from_kuid_munged(&init_user_ns, opts->uid));
362	if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
363		seq_printf(m, ",gid=%u",
364			   from_kgid_munged(&init_user_ns, opts->gid));
365	if (opts->mode != TRACEFS_DEFAULT_MODE)
366		seq_printf(m, ",mode=%o", opts->mode);
367
368	return 0;
369}
370
371static const struct super_operations tracefs_super_operations = {
372	.alloc_inode    = tracefs_alloc_inode,
373	.free_inode     = tracefs_free_inode,
374	.drop_inode     = generic_delete_inode,
375	.statfs		= simple_statfs,
376	.remount_fs	= tracefs_remount,
377	.show_options	= tracefs_show_options,
378};
379
380/*
381 * It would be cleaner if eventfs had its own dentry ops.
382 *
383 * Note that d_revalidate is called potentially under RCU,
384 * so it can't take the eventfs mutex etc. It's fine - if
385 * we open a file just as it's marked dead, things will
386 * still work just fine, and just see the old stale case.
387 */
388static void tracefs_d_release(struct dentry *dentry)
389{
390	if (dentry->d_fsdata)
391		eventfs_d_release(dentry);
392}
393
394static int tracefs_d_revalidate(struct dentry *dentry, unsigned int flags)
395{
396	struct eventfs_inode *ei = dentry->d_fsdata;
397
398	return !(ei && ei->is_freed);
399}
400
401static const struct dentry_operations tracefs_dentry_operations = {
402	.d_revalidate = tracefs_d_revalidate,
403	.d_release = tracefs_d_release,
404};
405
406static int trace_fill_super(struct super_block *sb, void *data, int silent)
407{
408	static const struct tree_descr trace_files[] = {{""}};
409	struct tracefs_fs_info *fsi;
410	int err;
411
412	fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
413	sb->s_fs_info = fsi;
414	if (!fsi) {
415		err = -ENOMEM;
416		goto fail;
417	}
418
419	err = tracefs_parse_options(data, &fsi->mount_opts);
420	if (err)
421		goto fail;
422
423	err  =  simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
424	if (err)
425		goto fail;
426
427	sb->s_op = &tracefs_super_operations;
428	sb->s_d_op = &tracefs_dentry_operations;
429
430	tracefs_apply_options(sb, false);
431
432	return 0;
433
434fail:
435	kfree(fsi);
436	sb->s_fs_info = NULL;
437	return err;
438}
439
440static struct dentry *trace_mount(struct file_system_type *fs_type,
441			int flags, const char *dev_name,
442			void *data)
443{
444	return mount_single(fs_type, flags, data, trace_fill_super);
445}
446
447static struct file_system_type trace_fs_type = {
448	.owner =	THIS_MODULE,
449	.name =		"tracefs",
450	.mount =	trace_mount,
451	.kill_sb =	kill_litter_super,
452};
453MODULE_ALIAS_FS("tracefs");
454
455struct dentry *tracefs_start_creating(const char *name, struct dentry *parent)
456{
457	struct dentry *dentry;
458	int error;
459
460	pr_debug("tracefs: creating file '%s'\n",name);
461
462	error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
463			      &tracefs_mount_count);
464	if (error)
465		return ERR_PTR(error);
466
467	/* If the parent is not specified, we create it in the root.
468	 * We need the root dentry to do this, which is in the super
469	 * block. A pointer to that is in the struct vfsmount that we
470	 * have around.
471	 */
472	if (!parent)
473		parent = tracefs_mount->mnt_root;
474
475	inode_lock(d_inode(parent));
476	if (unlikely(IS_DEADDIR(d_inode(parent))))
477		dentry = ERR_PTR(-ENOENT);
478	else
479		dentry = lookup_one_len(name, parent, strlen(name));
480	if (!IS_ERR(dentry) && d_inode(dentry)) {
481		dput(dentry);
482		dentry = ERR_PTR(-EEXIST);
483	}
484
485	if (IS_ERR(dentry)) {
486		inode_unlock(d_inode(parent));
487		simple_release_fs(&tracefs_mount, &tracefs_mount_count);
488	}
489
490	return dentry;
491}
492
493struct dentry *tracefs_failed_creating(struct dentry *dentry)
494{
495	inode_unlock(d_inode(dentry->d_parent));
496	dput(dentry);
497	simple_release_fs(&tracefs_mount, &tracefs_mount_count);
498	return NULL;
499}
500
501struct dentry *tracefs_end_creating(struct dentry *dentry)
502{
503	inode_unlock(d_inode(dentry->d_parent));
504	return dentry;
505}
506
507/* Find the inode that this will use for default */
508static struct inode *instance_inode(struct dentry *parent, struct inode *inode)
509{
510	struct tracefs_inode *ti;
511
512	/* If parent is NULL then use root inode */
513	if (!parent)
514		return d_inode(inode->i_sb->s_root);
515
516	/* Find the inode that is flagged as an instance or the root inode */
517	while (!IS_ROOT(parent)) {
518		ti = get_tracefs(d_inode(parent));
519		if (ti->flags & TRACEFS_INSTANCE_INODE)
520			break;
521		parent = parent->d_parent;
522	}
523
524	return d_inode(parent);
525}
526
527/**
528 * tracefs_create_file - create a file in the tracefs filesystem
529 * @name: a pointer to a string containing the name of the file to create.
530 * @mode: the permission that the file should have.
531 * @parent: a pointer to the parent dentry for this file.  This should be a
532 *          directory dentry if set.  If this parameter is NULL, then the
533 *          file will be created in the root of the tracefs filesystem.
534 * @data: a pointer to something that the caller will want to get to later
535 *        on.  The inode.i_private pointer will point to this value on
536 *        the open() call.
537 * @fops: a pointer to a struct file_operations that should be used for
538 *        this file.
539 *
540 * This is the basic "create a file" function for tracefs.  It allows for a
541 * wide range of flexibility in creating a file, or a directory (if you want
542 * to create a directory, the tracefs_create_dir() function is
543 * recommended to be used instead.)
544 *
545 * This function will return a pointer to a dentry if it succeeds.  This
546 * pointer must be passed to the tracefs_remove() function when the file is
547 * to be removed (no automatic cleanup happens if your module is unloaded,
548 * you are responsible here.)  If an error occurs, %NULL will be returned.
549 *
550 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
551 * returned.
552 */
553struct dentry *tracefs_create_file(const char *name, umode_t mode,
554				   struct dentry *parent, void *data,
555				   const struct file_operations *fops)
556{
557	struct tracefs_inode *ti;
558	struct dentry *dentry;
559	struct inode *inode;
560
561	if (security_locked_down(LOCKDOWN_TRACEFS))
562		return NULL;
563
564	if (!(mode & S_IFMT))
565		mode |= S_IFREG;
566	BUG_ON(!S_ISREG(mode));
567	dentry = tracefs_start_creating(name, parent);
568
569	if (IS_ERR(dentry))
570		return NULL;
571
572	inode = tracefs_get_inode(dentry->d_sb);
573	if (unlikely(!inode))
574		return tracefs_failed_creating(dentry);
575
576	ti = get_tracefs(inode);
577	ti->private = instance_inode(parent, inode);
578
579	inode->i_mode = mode;
580	inode->i_op = &tracefs_file_inode_operations;
581	inode->i_fop = fops ? fops : &tracefs_file_operations;
582	inode->i_private = data;
583	inode->i_uid = d_inode(dentry->d_parent)->i_uid;
584	inode->i_gid = d_inode(dentry->d_parent)->i_gid;
585	d_instantiate(dentry, inode);
586	fsnotify_create(d_inode(dentry->d_parent), dentry);
587	return tracefs_end_creating(dentry);
588}
589
590static struct dentry *__create_dir(const char *name, struct dentry *parent,
591				   const struct inode_operations *ops)
592{
593	struct tracefs_inode *ti;
594	struct dentry *dentry = tracefs_start_creating(name, parent);
595	struct inode *inode;
596
597	if (IS_ERR(dentry))
598		return NULL;
599
600	inode = tracefs_get_inode(dentry->d_sb);
601	if (unlikely(!inode))
602		return tracefs_failed_creating(dentry);
603
604	/* Do not set bits for OTH */
605	inode->i_mode = S_IFDIR | S_IRWXU | S_IRUSR| S_IRGRP | S_IXUSR | S_IXGRP;
606	inode->i_op = ops;
607	inode->i_fop = &simple_dir_operations;
608	inode->i_uid = d_inode(dentry->d_parent)->i_uid;
609	inode->i_gid = d_inode(dentry->d_parent)->i_gid;
610
611	ti = get_tracefs(inode);
612	ti->private = instance_inode(parent, inode);
613
614	/* directory inodes start off with i_nlink == 2 (for "." entry) */
615	inc_nlink(inode);
616	d_instantiate(dentry, inode);
617	inc_nlink(d_inode(dentry->d_parent));
618	fsnotify_mkdir(d_inode(dentry->d_parent), dentry);
619	return tracefs_end_creating(dentry);
620}
621
622/**
623 * tracefs_create_dir - create a directory in the tracefs filesystem
624 * @name: a pointer to a string containing the name of the directory to
625 *        create.
626 * @parent: a pointer to the parent dentry for this file.  This should be a
627 *          directory dentry if set.  If this parameter is NULL, then the
628 *          directory will be created in the root of the tracefs filesystem.
629 *
630 * This function creates a directory in tracefs with the given name.
631 *
632 * This function will return a pointer to a dentry if it succeeds.  This
633 * pointer must be passed to the tracefs_remove() function when the file is
634 * to be removed. If an error occurs, %NULL will be returned.
635 *
636 * If tracing is not enabled in the kernel, the value -%ENODEV will be
637 * returned.
638 */
639struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
640{
641	if (security_locked_down(LOCKDOWN_TRACEFS))
642		return NULL;
643
644	return __create_dir(name, parent, &tracefs_dir_inode_operations);
645}
646
647/**
648 * tracefs_create_instance_dir - create the tracing instances directory
649 * @name: The name of the instances directory to create
650 * @parent: The parent directory that the instances directory will exist
651 * @mkdir: The function to call when a mkdir is performed.
652 * @rmdir: The function to call when a rmdir is performed.
653 *
654 * Only one instances directory is allowed.
655 *
656 * The instances directory is special as it allows for mkdir and rmdir
657 * to be done by userspace. When a mkdir or rmdir is performed, the inode
658 * locks are released and the methods passed in (@mkdir and @rmdir) are
659 * called without locks and with the name of the directory being created
660 * within the instances directory.
661 *
662 * Returns the dentry of the instances directory.
663 */
664__init struct dentry *tracefs_create_instance_dir(const char *name,
665					  struct dentry *parent,
666					  int (*mkdir)(const char *name),
667					  int (*rmdir)(const char *name))
668{
669	struct dentry *dentry;
670
671	/* Only allow one instance of the instances directory. */
672	if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
673		return NULL;
674
675	dentry = __create_dir(name, parent, &tracefs_instance_dir_inode_operations);
676	if (!dentry)
677		return NULL;
678
679	tracefs_ops.mkdir = mkdir;
680	tracefs_ops.rmdir = rmdir;
681
682	return dentry;
683}
684
685static void remove_one(struct dentry *victim)
686{
687	simple_release_fs(&tracefs_mount, &tracefs_mount_count);
688}
689
690/**
691 * tracefs_remove - recursively removes a directory
692 * @dentry: a pointer to a the dentry of the directory to be removed.
693 *
694 * This function recursively removes a directory tree in tracefs that
695 * was previously created with a call to another tracefs function
696 * (like tracefs_create_file() or variants thereof.)
697 */
698void tracefs_remove(struct dentry *dentry)
699{
700	if (IS_ERR_OR_NULL(dentry))
701		return;
702
703	simple_pin_fs(&trace_fs_type, &tracefs_mount, &tracefs_mount_count);
704	simple_recursive_removal(dentry, remove_one);
705	simple_release_fs(&tracefs_mount, &tracefs_mount_count);
706}
707
708/**
709 * tracefs_initialized - Tells whether tracefs has been registered
710 */
711bool tracefs_initialized(void)
712{
713	return tracefs_registered;
714}
715
716static void init_once(void *foo)
717{
718	struct tracefs_inode *ti = (struct tracefs_inode *) foo;
719
720	/* inode_init_once() calls memset() on the vfs_inode portion */
721	inode_init_once(&ti->vfs_inode);
722
723	/* Zero out the rest */
724	memset_after(ti, 0, vfs_inode);
725}
726
727static int __init tracefs_init(void)
728{
729	int retval;
730
731	tracefs_inode_cachep = kmem_cache_create("tracefs_inode_cache",
732						 sizeof(struct tracefs_inode),
733						 0, (SLAB_RECLAIM_ACCOUNT|
734						     SLAB_ACCOUNT),
735						 init_once);
736	if (!tracefs_inode_cachep)
737		return -ENOMEM;
738
739	retval = sysfs_create_mount_point(kernel_kobj, "tracing");
740	if (retval)
741		return -EINVAL;
742
743	retval = register_filesystem(&trace_fs_type);
744	if (!retval)
745		tracefs_registered = true;
746
747	return retval;
748}
749core_initcall(tracefs_init);
750