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(d_inode(dentry));
114 ret = tracefs_ops.rmdir(name);
116 inode_lock_nested(inode, I_MUTEX_PARENT);
117 inode_lock(d_inode(dentry));
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 {
144 /* Opt_* bitfield. */
155 static const match_table_t tokens = {
158 {Opt_mode, "mode=%o"},
162 struct tracefs_fs_info {
163 struct tracefs_mount_opts mount_opts;
166 static void change_gid(struct dentry *dentry, kgid_t gid)
168 if (!dentry->d_inode)
170 dentry->d_inode->i_gid = gid;
174 * Taken from d_walk, but without he need for handling renames.
175 * Nothing can be renamed while walking the list, as tracefs
176 * does not support renames. This is only called when mounting
177 * or remounting the file system, to set all the files to
180 static void set_gid(struct dentry *parent, kgid_t gid)
182 struct dentry *this_parent;
183 struct list_head *next;
185 this_parent = parent;
186 spin_lock(&this_parent->d_lock);
188 change_gid(this_parent, gid);
190 next = this_parent->d_subdirs.next;
192 while (next != &this_parent->d_subdirs) {
193 struct list_head *tmp = next;
194 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
197 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
199 change_gid(dentry, gid);
201 if (!list_empty(&dentry->d_subdirs)) {
202 spin_unlock(&this_parent->d_lock);
203 spin_release(&dentry->d_lock.dep_map, _RET_IP_);
204 this_parent = dentry;
205 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
208 spin_unlock(&dentry->d_lock);
211 * All done at this level ... ascend and resume the search.
215 if (this_parent != parent) {
216 struct dentry *child = this_parent;
217 this_parent = child->d_parent;
219 spin_unlock(&child->d_lock);
220 spin_lock(&this_parent->d_lock);
222 /* go into the first sibling still alive */
224 next = child->d_child.next;
225 if (next == &this_parent->d_subdirs)
227 child = list_entry(next, struct dentry, d_child);
228 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
233 spin_unlock(&this_parent->d_lock);
237 static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
239 substring_t args[MAX_OPT_ARGS];
247 opts->mode = TRACEFS_DEFAULT_MODE;
249 while ((p = strsep(&data, ",")) != NULL) {
253 token = match_token(p, tokens, args);
256 if (match_int(&args[0], &option))
258 uid = make_kuid(current_user_ns(), option);
264 if (match_int(&args[0], &option))
266 gid = make_kgid(current_user_ns(), option);
272 if (match_octal(&args[0], &option))
274 opts->mode = option & S_IALLUGO;
277 * We might like to report bad mount options here;
278 * but traditionally tracefs has ignored all mount options
282 opts->opts |= BIT(token);
288 static int tracefs_apply_options(struct super_block *sb, bool remount)
290 struct tracefs_fs_info *fsi = sb->s_fs_info;
291 struct inode *inode = d_inode(sb->s_root);
292 struct tracefs_mount_opts *opts = &fsi->mount_opts;
295 * On remount, only reset mode/uid/gid if they were provided as mount
299 if (!remount || opts->opts & BIT(Opt_mode)) {
300 inode->i_mode &= ~S_IALLUGO;
301 inode->i_mode |= opts->mode;
304 if (!remount || opts->opts & BIT(Opt_uid))
305 inode->i_uid = opts->uid;
307 if (!remount || opts->opts & BIT(Opt_gid)) {
308 /* Set all the group ids to the mount option */
309 set_gid(sb->s_root, opts->gid);
315 static int tracefs_remount(struct super_block *sb, int *flags, char *data)
318 struct tracefs_fs_info *fsi = sb->s_fs_info;
321 err = tracefs_parse_options(data, &fsi->mount_opts);
325 tracefs_apply_options(sb, true);
331 static int tracefs_show_options(struct seq_file *m, struct dentry *root)
333 struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
334 struct tracefs_mount_opts *opts = &fsi->mount_opts;
336 if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
337 seq_printf(m, ",uid=%u",
338 from_kuid_munged(&init_user_ns, opts->uid));
339 if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
340 seq_printf(m, ",gid=%u",
341 from_kgid_munged(&init_user_ns, opts->gid));
342 if (opts->mode != TRACEFS_DEFAULT_MODE)
343 seq_printf(m, ",mode=%o", opts->mode);
348 static const struct super_operations tracefs_super_operations = {
349 .statfs = simple_statfs,
350 .remount_fs = tracefs_remount,
351 .show_options = tracefs_show_options,
354 static int trace_fill_super(struct super_block *sb, void *data, int silent)
356 static const struct tree_descr trace_files[] = {{""}};
357 struct tracefs_fs_info *fsi;
360 fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
367 err = tracefs_parse_options(data, &fsi->mount_opts);
371 err = simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
375 sb->s_op = &tracefs_super_operations;
377 tracefs_apply_options(sb, false);
383 sb->s_fs_info = NULL;
387 static struct dentry *trace_mount(struct file_system_type *fs_type,
388 int flags, const char *dev_name,
391 return mount_single(fs_type, flags, data, trace_fill_super);
394 static struct file_system_type trace_fs_type = {
395 .owner = THIS_MODULE,
397 .mount = trace_mount,
398 .kill_sb = kill_litter_super,
400 MODULE_ALIAS_FS("tracefs");
402 static struct dentry *start_creating(const char *name, struct dentry *parent)
404 struct dentry *dentry;
407 pr_debug("tracefs: creating file '%s'\n",name);
409 error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
410 &tracefs_mount_count);
412 return ERR_PTR(error);
414 /* If the parent is not specified, we create it in the root.
415 * We need the root dentry to do this, which is in the super
416 * block. A pointer to that is in the struct vfsmount that we
420 parent = tracefs_mount->mnt_root;
422 inode_lock(d_inode(parent));
423 if (unlikely(IS_DEADDIR(d_inode(parent))))
424 dentry = ERR_PTR(-ENOENT);
426 dentry = lookup_one_len(name, parent, strlen(name));
427 if (!IS_ERR(dentry) && d_inode(dentry)) {
429 dentry = ERR_PTR(-EEXIST);
432 if (IS_ERR(dentry)) {
433 inode_unlock(d_inode(parent));
434 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
440 static struct dentry *failed_creating(struct dentry *dentry)
442 inode_unlock(d_inode(dentry->d_parent));
444 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
448 static struct dentry *end_creating(struct dentry *dentry)
450 inode_unlock(d_inode(dentry->d_parent));
455 * tracefs_create_file - create a file in the tracefs filesystem
456 * @name: a pointer to a string containing the name of the file to create.
457 * @mode: the permission that the file should have.
458 * @parent: a pointer to the parent dentry for this file. This should be a
459 * directory dentry if set. If this parameter is NULL, then the
460 * file will be created in the root of the tracefs filesystem.
461 * @data: a pointer to something that the caller will want to get to later
462 * on. The inode.i_private pointer will point to this value on
464 * @fops: a pointer to a struct file_operations that should be used for
467 * This is the basic "create a file" function for tracefs. It allows for a
468 * wide range of flexibility in creating a file, or a directory (if you want
469 * to create a directory, the tracefs_create_dir() function is
470 * recommended to be used instead.)
472 * This function will return a pointer to a dentry if it succeeds. This
473 * pointer must be passed to the tracefs_remove() function when the file is
474 * to be removed (no automatic cleanup happens if your module is unloaded,
475 * you are responsible here.) If an error occurs, %NULL will be returned.
477 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
480 struct dentry *tracefs_create_file(const char *name, umode_t mode,
481 struct dentry *parent, void *data,
482 const struct file_operations *fops)
484 struct dentry *dentry;
487 if (security_locked_down(LOCKDOWN_TRACEFS))
490 if (!(mode & S_IFMT))
492 BUG_ON(!S_ISREG(mode));
493 dentry = start_creating(name, parent);
498 inode = tracefs_get_inode(dentry->d_sb);
499 if (unlikely(!inode))
500 return failed_creating(dentry);
502 inode->i_mode = mode;
503 inode->i_fop = fops ? fops : &tracefs_file_operations;
504 inode->i_private = data;
505 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
506 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
507 d_instantiate(dentry, inode);
508 fsnotify_create(d_inode(dentry->d_parent), dentry);
509 return end_creating(dentry);
512 static struct dentry *__create_dir(const char *name, struct dentry *parent,
513 const struct inode_operations *ops)
515 struct dentry *dentry = start_creating(name, parent);
521 inode = tracefs_get_inode(dentry->d_sb);
522 if (unlikely(!inode))
523 return failed_creating(dentry);
525 /* Do not set bits for OTH */
526 inode->i_mode = S_IFDIR | S_IRWXU | S_IRUSR| S_IRGRP | S_IXUSR | S_IXGRP;
528 inode->i_fop = &simple_dir_operations;
529 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
530 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
532 /* directory inodes start off with i_nlink == 2 (for "." entry) */
534 d_instantiate(dentry, inode);
535 inc_nlink(d_inode(dentry->d_parent));
536 fsnotify_mkdir(d_inode(dentry->d_parent), dentry);
537 return end_creating(dentry);
541 * tracefs_create_dir - create a directory in the tracefs filesystem
542 * @name: a pointer to a string containing the name of the directory to
544 * @parent: a pointer to the parent dentry for this file. This should be a
545 * directory dentry if set. If this parameter is NULL, then the
546 * directory will be created in the root of the tracefs filesystem.
548 * This function creates a directory in tracefs with the given name.
550 * This function will return a pointer to a dentry if it succeeds. This
551 * pointer must be passed to the tracefs_remove() function when the file is
552 * to be removed. If an error occurs, %NULL will be returned.
554 * If tracing is not enabled in the kernel, the value -%ENODEV will be
557 struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
559 return __create_dir(name, parent, &simple_dir_inode_operations);
563 * tracefs_create_instance_dir - create the tracing instances directory
564 * @name: The name of the instances directory to create
565 * @parent: The parent directory that the instances directory will exist
566 * @mkdir: The function to call when a mkdir is performed.
567 * @rmdir: The function to call when a rmdir is performed.
569 * Only one instances directory is allowed.
571 * The instances directory is special as it allows for mkdir and rmdir
572 * to be done by userspace. When a mkdir or rmdir is performed, the inode
573 * locks are released and the methods passed in (@mkdir and @rmdir) are
574 * called without locks and with the name of the directory being created
575 * within the instances directory.
577 * Returns the dentry of the instances directory.
579 __init struct dentry *tracefs_create_instance_dir(const char *name,
580 struct dentry *parent,
581 int (*mkdir)(const char *name),
582 int (*rmdir)(const char *name))
584 struct dentry *dentry;
586 /* Only allow one instance of the instances directory. */
587 if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
590 dentry = __create_dir(name, parent, &tracefs_dir_inode_operations);
594 tracefs_ops.mkdir = mkdir;
595 tracefs_ops.rmdir = rmdir;
600 static void remove_one(struct dentry *victim)
602 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
606 * tracefs_remove - recursively removes a directory
607 * @dentry: a pointer to a the dentry of the directory to be removed.
609 * This function recursively removes a directory tree in tracefs that
610 * was previously created with a call to another tracefs function
611 * (like tracefs_create_file() or variants thereof.)
613 void tracefs_remove(struct dentry *dentry)
615 if (IS_ERR_OR_NULL(dentry))
618 simple_pin_fs(&trace_fs_type, &tracefs_mount, &tracefs_mount_count);
619 simple_recursive_removal(dentry, remove_one);
620 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
624 * tracefs_initialized - Tells whether tracefs has been registered
626 bool tracefs_initialized(void)
628 return tracefs_registered;
631 static int __init tracefs_init(void)
635 retval = sysfs_create_mount_point(kernel_kobj, "tracing");
639 retval = register_filesystem(&trace_fs_type);
641 tracefs_registered = true;
645 core_initcall(tracefs_init);