1 // SPDX-License-Identifier: GPL-2.0-only
3 * inode.c - part of tracefs, a pseudo file system for activating tracing
5 * Based on debugfs by: Greg Kroah-Hartman <greg@kroah.com>
7 * Copyright (C) 2014 Red Hat Inc, author: Steven Rostedt <srostedt@redhat.com>
9 * tracefs is the file system that is used by the tracing infrastructure.
12 #include <linux/module.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>
25 #define TRACEFS_DEFAULT_MODE 0700
27 static struct vfsmount *tracefs_mount;
28 static int tracefs_mount_count;
29 static bool tracefs_registered;
31 static ssize_t default_read_file(struct file *file, char __user *buf,
32 size_t count, loff_t *ppos)
37 static ssize_t default_write_file(struct file *file, const char __user *buf,
38 size_t count, loff_t *ppos)
43 static const struct file_operations tracefs_file_operations = {
44 .read = default_read_file,
45 .write = default_write_file,
47 .llseek = noop_llseek,
50 static struct tracefs_dir_ops {
51 int (*mkdir)(const char *name);
52 int (*rmdir)(const char *name);
53 } tracefs_ops __ro_after_init;
55 static char *get_dname(struct dentry *dentry)
59 int len = dentry->d_name.len;
61 dname = dentry->d_name.name;
62 name = kmalloc(len + 1, GFP_KERNEL);
65 memcpy(name, dname, len);
70 static int tracefs_syscall_mkdir(struct user_namespace *mnt_userns,
71 struct inode *inode, struct dentry *dentry,
77 name = get_dname(dentry);
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.
87 ret = tracefs_ops.mkdir(name);
95 static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
100 name = get_dname(dentry);
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.
112 inode_unlock(dentry->d_inode);
114 ret = tracefs_ops.rmdir(name);
116 inode_lock_nested(inode, I_MUTEX_PARENT);
117 inode_lock(dentry->d_inode);
124 static const struct inode_operations tracefs_dir_inode_operations = {
125 .lookup = simple_lookup,
126 .mkdir = tracefs_syscall_mkdir,
127 .rmdir = tracefs_syscall_rmdir,
130 static struct inode *tracefs_get_inode(struct super_block *sb)
132 struct inode *inode = new_inode(sb);
134 inode->i_ino = get_next_ino();
135 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
140 struct tracefs_mount_opts {
153 static const match_table_t tokens = {
156 {Opt_mode, "mode=%o"},
160 struct tracefs_fs_info {
161 struct tracefs_mount_opts mount_opts;
164 static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
166 substring_t args[MAX_OPT_ARGS];
173 opts->mode = TRACEFS_DEFAULT_MODE;
175 while ((p = strsep(&data, ",")) != NULL) {
179 token = match_token(p, tokens, args);
182 if (match_int(&args[0], &option))
184 uid = make_kuid(current_user_ns(), option);
190 if (match_int(&args[0], &option))
192 gid = make_kgid(current_user_ns(), option);
198 if (match_octal(&args[0], &option))
200 opts->mode = option & S_IALLUGO;
203 * We might like to report bad mount options here;
204 * but traditionally tracefs has ignored all mount options
212 static int tracefs_apply_options(struct super_block *sb)
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;
218 inode->i_mode &= ~S_IALLUGO;
219 inode->i_mode |= opts->mode;
221 inode->i_uid = opts->uid;
222 inode->i_gid = opts->gid;
227 static int tracefs_remount(struct super_block *sb, int *flags, char *data)
230 struct tracefs_fs_info *fsi = sb->s_fs_info;
233 err = tracefs_parse_options(data, &fsi->mount_opts);
237 tracefs_apply_options(sb);
243 static int tracefs_show_options(struct seq_file *m, struct dentry *root)
245 struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
246 struct tracefs_mount_opts *opts = &fsi->mount_opts;
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);
260 static const struct super_operations tracefs_super_operations = {
261 .statfs = simple_statfs,
262 .remount_fs = tracefs_remount,
263 .show_options = tracefs_show_options,
266 static int trace_fill_super(struct super_block *sb, void *data, int silent)
268 static const struct tree_descr trace_files[] = {{""}};
269 struct tracefs_fs_info *fsi;
272 fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
279 err = tracefs_parse_options(data, &fsi->mount_opts);
283 err = simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
287 sb->s_op = &tracefs_super_operations;
289 tracefs_apply_options(sb);
295 sb->s_fs_info = NULL;
299 static struct dentry *trace_mount(struct file_system_type *fs_type,
300 int flags, const char *dev_name,
303 return mount_single(fs_type, flags, data, trace_fill_super);
306 static struct file_system_type trace_fs_type = {
307 .owner = THIS_MODULE,
309 .mount = trace_mount,
310 .kill_sb = kill_litter_super,
312 MODULE_ALIAS_FS("tracefs");
314 static struct dentry *start_creating(const char *name, struct dentry *parent)
316 struct dentry *dentry;
319 pr_debug("tracefs: creating file '%s'\n",name);
321 error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
322 &tracefs_mount_count);
324 return ERR_PTR(error);
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
332 parent = tracefs_mount->mnt_root;
334 inode_lock(parent->d_inode);
335 if (unlikely(IS_DEADDIR(parent->d_inode)))
336 dentry = ERR_PTR(-ENOENT);
338 dentry = lookup_one_len(name, parent, strlen(name));
339 if (!IS_ERR(dentry) && dentry->d_inode) {
341 dentry = ERR_PTR(-EEXIST);
344 if (IS_ERR(dentry)) {
345 inode_unlock(parent->d_inode);
346 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
352 static struct dentry *failed_creating(struct dentry *dentry)
354 inode_unlock(dentry->d_parent->d_inode);
356 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
360 static struct dentry *end_creating(struct dentry *dentry)
362 inode_unlock(dentry->d_parent->d_inode);
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
376 * @fops: a pointer to a struct file_operations that should be used for
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.)
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.
389 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
392 struct dentry *tracefs_create_file(const char *name, umode_t mode,
393 struct dentry *parent, void *data,
394 const struct file_operations *fops)
396 struct dentry *dentry;
399 if (security_locked_down(LOCKDOWN_TRACEFS))
402 if (!(mode & S_IFMT))
404 BUG_ON(!S_ISREG(mode));
405 dentry = start_creating(name, parent);
410 inode = tracefs_get_inode(dentry->d_sb);
411 if (unlikely(!inode))
412 return failed_creating(dentry);
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);
422 static struct dentry *__create_dir(const char *name, struct dentry *parent,
423 const struct inode_operations *ops)
425 struct dentry *dentry = start_creating(name, parent);
431 inode = tracefs_get_inode(dentry->d_sb);
432 if (unlikely(!inode))
433 return failed_creating(dentry);
435 inode->i_mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
437 inode->i_fop = &simple_dir_operations;
439 /* directory inodes start off with i_nlink == 2 (for "." entry) */
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);
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
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.
455 * This function creates a directory in tracefs with the given name.
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.
461 * If tracing is not enabled in the kernel, the value -%ENODEV will be
464 struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
466 return __create_dir(name, parent, &simple_dir_inode_operations);
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.
476 * Only one instances directory is allowed.
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.
484 * Returns the dentry of the instances directory.
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))
491 struct dentry *dentry;
493 /* Only allow one instance of the instances directory. */
494 if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
497 dentry = __create_dir(name, parent, &tracefs_dir_inode_operations);
501 tracefs_ops.mkdir = mkdir;
502 tracefs_ops.rmdir = rmdir;
507 static void remove_one(struct dentry *victim)
509 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
513 * tracefs_remove - recursively removes a directory
514 * @dentry: a pointer to a the dentry of the directory to be removed.
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.)
520 void tracefs_remove(struct dentry *dentry)
522 if (IS_ERR_OR_NULL(dentry))
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);
531 * tracefs_initialized - Tells whether tracefs has been registered
533 bool tracefs_initialized(void)
535 return tracefs_registered;
538 static int __init tracefs_init(void)
542 retval = sysfs_create_mount_point(kernel_kobj, "tracing");
546 retval = register_filesystem(&trace_fs_type);
548 tracefs_registered = true;
552 core_initcall(tracefs_init);