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 inode *inode, struct dentry *dentry, umode_t mode)
75 name = get_dname(dentry);
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.
85 ret = tracefs_ops.mkdir(name);
93 static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
98 name = get_dname(dentry);
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.
110 inode_unlock(dentry->d_inode);
112 ret = tracefs_ops.rmdir(name);
114 inode_lock_nested(inode, I_MUTEX_PARENT);
115 inode_lock(dentry->d_inode);
122 static const struct inode_operations tracefs_dir_inode_operations = {
123 .lookup = simple_lookup,
124 .mkdir = tracefs_syscall_mkdir,
125 .rmdir = tracefs_syscall_rmdir,
128 static struct inode *tracefs_get_inode(struct super_block *sb)
130 struct inode *inode = new_inode(sb);
132 inode->i_ino = get_next_ino();
133 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
138 struct tracefs_mount_opts {
151 static const match_table_t tokens = {
154 {Opt_mode, "mode=%o"},
158 struct tracefs_fs_info {
159 struct tracefs_mount_opts mount_opts;
162 static void change_gid(struct dentry *dentry, kgid_t gid)
164 if (!dentry->d_inode)
166 dentry->d_inode->i_gid = gid;
170 * Taken from d_walk, but without he need for handling renames.
171 * Nothing can be renamed while walking the list, as tracefs
172 * does not support renames. This is only called when mounting
173 * or remounting the file system, to set all the files to
176 static void set_gid(struct dentry *parent, kgid_t gid)
178 struct dentry *this_parent;
179 struct list_head *next;
181 this_parent = parent;
182 spin_lock(&this_parent->d_lock);
184 change_gid(this_parent, gid);
186 next = this_parent->d_subdirs.next;
188 while (next != &this_parent->d_subdirs) {
189 struct list_head *tmp = next;
190 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
193 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
195 change_gid(dentry, gid);
197 if (!list_empty(&dentry->d_subdirs)) {
198 spin_unlock(&this_parent->d_lock);
199 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
200 this_parent = dentry;
201 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
204 spin_unlock(&dentry->d_lock);
207 * All done at this level ... ascend and resume the search.
211 if (this_parent != parent) {
212 struct dentry *child = this_parent;
213 this_parent = child->d_parent;
215 spin_unlock(&child->d_lock);
216 spin_lock(&this_parent->d_lock);
218 /* go into the first sibling still alive */
220 next = child->d_child.next;
221 if (next == &this_parent->d_subdirs)
223 child = list_entry(next, struct dentry, d_child);
224 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
229 spin_unlock(&this_parent->d_lock);
233 static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
235 substring_t args[MAX_OPT_ARGS];
242 opts->mode = TRACEFS_DEFAULT_MODE;
244 while ((p = strsep(&data, ",")) != NULL) {
248 token = match_token(p, tokens, args);
251 if (match_int(&args[0], &option))
253 uid = make_kuid(current_user_ns(), option);
259 if (match_int(&args[0], &option))
261 gid = make_kgid(current_user_ns(), option);
267 if (match_octal(&args[0], &option))
269 opts->mode = option & S_IALLUGO;
272 * We might like to report bad mount options here;
273 * but traditionally tracefs has ignored all mount options
281 static int tracefs_apply_options(struct super_block *sb)
283 struct tracefs_fs_info *fsi = sb->s_fs_info;
284 struct inode *inode = sb->s_root->d_inode;
285 struct tracefs_mount_opts *opts = &fsi->mount_opts;
287 inode->i_mode &= ~S_IALLUGO;
288 inode->i_mode |= opts->mode;
290 inode->i_uid = opts->uid;
292 /* Set all the group ids to the mount option */
293 set_gid(sb->s_root, opts->gid);
298 static int tracefs_remount(struct super_block *sb, int *flags, char *data)
301 struct tracefs_fs_info *fsi = sb->s_fs_info;
304 err = tracefs_parse_options(data, &fsi->mount_opts);
308 tracefs_apply_options(sb);
314 static int tracefs_show_options(struct seq_file *m, struct dentry *root)
316 struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
317 struct tracefs_mount_opts *opts = &fsi->mount_opts;
319 if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
320 seq_printf(m, ",uid=%u",
321 from_kuid_munged(&init_user_ns, opts->uid));
322 if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
323 seq_printf(m, ",gid=%u",
324 from_kgid_munged(&init_user_ns, opts->gid));
325 if (opts->mode != TRACEFS_DEFAULT_MODE)
326 seq_printf(m, ",mode=%o", opts->mode);
331 static const struct super_operations tracefs_super_operations = {
332 .statfs = simple_statfs,
333 .remount_fs = tracefs_remount,
334 .show_options = tracefs_show_options,
337 static int trace_fill_super(struct super_block *sb, void *data, int silent)
339 static const struct tree_descr trace_files[] = {{""}};
340 struct tracefs_fs_info *fsi;
343 fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
350 err = tracefs_parse_options(data, &fsi->mount_opts);
354 err = simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
358 sb->s_op = &tracefs_super_operations;
360 tracefs_apply_options(sb);
366 sb->s_fs_info = NULL;
370 static struct dentry *trace_mount(struct file_system_type *fs_type,
371 int flags, const char *dev_name,
374 return mount_single(fs_type, flags, data, trace_fill_super);
377 static struct file_system_type trace_fs_type = {
378 .owner = THIS_MODULE,
380 .mount = trace_mount,
381 .kill_sb = kill_litter_super,
383 MODULE_ALIAS_FS("tracefs");
385 static struct dentry *start_creating(const char *name, struct dentry *parent)
387 struct dentry *dentry;
390 pr_debug("tracefs: creating file '%s'\n",name);
392 error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
393 &tracefs_mount_count);
395 return ERR_PTR(error);
397 /* If the parent is not specified, we create it in the root.
398 * We need the root dentry to do this, which is in the super
399 * block. A pointer to that is in the struct vfsmount that we
403 parent = tracefs_mount->mnt_root;
405 inode_lock(parent->d_inode);
406 dentry = lookup_one_len(name, parent, strlen(name));
407 if (!IS_ERR(dentry) && dentry->d_inode) {
409 dentry = ERR_PTR(-EEXIST);
412 if (IS_ERR(dentry)) {
413 inode_unlock(parent->d_inode);
414 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
420 static struct dentry *failed_creating(struct dentry *dentry)
422 inode_unlock(dentry->d_parent->d_inode);
424 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
428 static struct dentry *end_creating(struct dentry *dentry)
430 inode_unlock(dentry->d_parent->d_inode);
435 * tracefs_create_file - create a file in the tracefs filesystem
436 * @name: a pointer to a string containing the name of the file to create.
437 * @mode: the permission that the file should have.
438 * @parent: a pointer to the parent dentry for this file. This should be a
439 * directory dentry if set. If this parameter is NULL, then the
440 * file will be created in the root of the tracefs filesystem.
441 * @data: a pointer to something that the caller will want to get to later
442 * on. The inode.i_private pointer will point to this value on
444 * @fops: a pointer to a struct file_operations that should be used for
447 * This is the basic "create a file" function for tracefs. It allows for a
448 * wide range of flexibility in creating a file, or a directory (if you want
449 * to create a directory, the tracefs_create_dir() function is
450 * recommended to be used instead.)
452 * This function will return a pointer to a dentry if it succeeds. This
453 * pointer must be passed to the tracefs_remove() function when the file is
454 * to be removed (no automatic cleanup happens if your module is unloaded,
455 * you are responsible here.) If an error occurs, %NULL will be returned.
457 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
460 struct dentry *tracefs_create_file(const char *name, umode_t mode,
461 struct dentry *parent, void *data,
462 const struct file_operations *fops)
464 struct dentry *dentry;
467 if (security_locked_down(LOCKDOWN_TRACEFS))
470 if (!(mode & S_IFMT))
472 BUG_ON(!S_ISREG(mode));
473 dentry = start_creating(name, parent);
478 inode = tracefs_get_inode(dentry->d_sb);
479 if (unlikely(!inode))
480 return failed_creating(dentry);
482 inode->i_mode = mode;
483 inode->i_fop = fops ? fops : &tracefs_file_operations;
484 inode->i_private = data;
485 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
486 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
487 d_instantiate(dentry, inode);
488 fsnotify_create(dentry->d_parent->d_inode, dentry);
489 return end_creating(dentry);
492 static struct dentry *__create_dir(const char *name, struct dentry *parent,
493 const struct inode_operations *ops)
495 struct dentry *dentry = start_creating(name, parent);
501 inode = tracefs_get_inode(dentry->d_sb);
502 if (unlikely(!inode))
503 return failed_creating(dentry);
505 /* Do not set bits for OTH */
506 inode->i_mode = S_IFDIR | S_IRWXU | S_IRUSR| S_IRGRP | S_IXUSR | S_IXGRP;
508 inode->i_fop = &simple_dir_operations;
509 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
510 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
512 /* directory inodes start off with i_nlink == 2 (for "." entry) */
514 d_instantiate(dentry, inode);
515 inc_nlink(dentry->d_parent->d_inode);
516 fsnotify_mkdir(dentry->d_parent->d_inode, dentry);
517 return end_creating(dentry);
521 * tracefs_create_dir - create a directory in the tracefs filesystem
522 * @name: a pointer to a string containing the name of the directory to
524 * @parent: a pointer to the parent dentry for this file. This should be a
525 * directory dentry if set. If this parameter is NULL, then the
526 * directory will be created in the root of the tracefs filesystem.
528 * This function creates a directory in tracefs with the given name.
530 * This function will return a pointer to a dentry if it succeeds. This
531 * pointer must be passed to the tracefs_remove() function when the file is
532 * to be removed. If an error occurs, %NULL will be returned.
534 * If tracing is not enabled in the kernel, the value -%ENODEV will be
537 struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
539 return __create_dir(name, parent, &simple_dir_inode_operations);
543 * tracefs_create_instance_dir - create the tracing instances directory
544 * @name: The name of the instances directory to create
545 * @parent: The parent directory that the instances directory will exist
546 * @mkdir: The function to call when a mkdir is performed.
547 * @rmdir: The function to call when a rmdir is performed.
549 * Only one instances directory is allowed.
551 * The instances directory is special as it allows for mkdir and rmdir to
552 * to be done by userspace. When a mkdir or rmdir is performed, the inode
553 * locks are released and the methhods passed in (@mkdir and @rmdir) are
554 * called without locks and with the name of the directory being created
555 * within the instances directory.
557 * Returns the dentry of the instances directory.
559 __init struct dentry *tracefs_create_instance_dir(const char *name,
560 struct dentry *parent,
561 int (*mkdir)(const char *name),
562 int (*rmdir)(const char *name))
564 struct dentry *dentry;
566 /* Only allow one instance of the instances directory. */
567 if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
570 dentry = __create_dir(name, parent, &tracefs_dir_inode_operations);
574 tracefs_ops.mkdir = mkdir;
575 tracefs_ops.rmdir = rmdir;
580 static int __tracefs_remove(struct dentry *dentry, struct dentry *parent)
584 if (simple_positive(dentry)) {
585 if (dentry->d_inode) {
587 switch (dentry->d_inode->i_mode & S_IFMT) {
589 ret = simple_rmdir(parent->d_inode, dentry);
591 fsnotify_rmdir(parent->d_inode, dentry);
594 simple_unlink(parent->d_inode, dentry);
595 fsnotify_unlink(parent->d_inode, dentry);
607 * tracefs_remove - removes a file or directory from the tracefs filesystem
608 * @dentry: a pointer to a the dentry of the file or directory to be
611 * This function removes a file or directory in tracefs that was previously
612 * created with a call to another tracefs function (like
613 * tracefs_create_file() or variants thereof.)
615 void tracefs_remove(struct dentry *dentry)
617 struct dentry *parent;
620 if (IS_ERR_OR_NULL(dentry))
623 parent = dentry->d_parent;
624 inode_lock(parent->d_inode);
625 ret = __tracefs_remove(dentry, parent);
626 inode_unlock(parent->d_inode);
628 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
632 * tracefs_remove_recursive - recursively removes a directory
633 * @dentry: a pointer to a the dentry of the directory to be removed.
635 * This function recursively removes a directory tree in tracefs that
636 * was previously created with a call to another tracefs function
637 * (like tracefs_create_file() or variants thereof.)
639 void tracefs_remove_recursive(struct dentry *dentry)
641 struct dentry *child, *parent;
643 if (IS_ERR_OR_NULL(dentry))
648 inode_lock(parent->d_inode);
651 * The parent->d_subdirs is protected by the d_lock. Outside that
652 * lock, the child can be unlinked and set to be freed which can
653 * use the d_u.d_child as the rcu head and corrupt this list.
655 spin_lock(&parent->d_lock);
656 list_for_each_entry(child, &parent->d_subdirs, d_child) {
657 if (!simple_positive(child))
660 /* perhaps simple_empty(child) makes more sense */
661 if (!list_empty(&child->d_subdirs)) {
662 spin_unlock(&parent->d_lock);
663 inode_unlock(parent->d_inode);
668 spin_unlock(&parent->d_lock);
670 if (!__tracefs_remove(child, parent))
671 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
674 * The parent->d_lock protects agaist child from unlinking
675 * from d_subdirs. When releasing the parent->d_lock we can
676 * no longer trust that the next pointer is valid.
677 * Restart the loop. We'll skip this one with the
678 * simple_positive() check.
682 spin_unlock(&parent->d_lock);
684 inode_unlock(parent->d_inode);
686 parent = parent->d_parent;
687 inode_lock(parent->d_inode);
693 if (!__tracefs_remove(child, parent))
694 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
695 inode_unlock(parent->d_inode);
699 * tracefs_initialized - Tells whether tracefs has been registered
701 bool tracefs_initialized(void)
703 return tracefs_registered;
706 static int __init tracefs_init(void)
710 retval = sysfs_create_mount_point(kernel_kobj, "tracing");
714 retval = register_filesystem(&trace_fs_type);
716 tracefs_registered = true;
720 core_initcall(tracefs_init);