2 * inode.c - part of tracefs, a pseudo file system for activating tracing
4 * Based on debugfs by: Greg Kroah-Hartman <greg@kroah.com>
6 * Copyright (C) 2014 Red Hat Inc, author: Steven Rostedt <srostedt@redhat.com>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License version
10 * 2 as published by the Free Software Foundation.
12 * tracefs is the file system that is used by the tracing infrastructure.
16 #include <linux/module.h>
18 #include <linux/mount.h>
19 #include <linux/kobject.h>
20 #include <linux/namei.h>
21 #include <linux/tracefs.h>
22 #include <linux/fsnotify.h>
23 #include <linux/seq_file.h>
24 #include <linux/parser.h>
25 #include <linux/magic.h>
26 #include <linux/slab.h>
28 #define TRACEFS_DEFAULT_MODE 0700
30 static struct vfsmount *tracefs_mount;
31 static int tracefs_mount_count;
32 static bool tracefs_registered;
34 static ssize_t default_read_file(struct file *file, char __user *buf,
35 size_t count, loff_t *ppos)
40 static ssize_t default_write_file(struct file *file, const char __user *buf,
41 size_t count, loff_t *ppos)
46 static const struct file_operations tracefs_file_operations = {
47 .read = default_read_file,
48 .write = default_write_file,
50 .llseek = noop_llseek,
53 static struct tracefs_dir_ops {
54 int (*mkdir)(const char *name);
55 int (*rmdir)(const char *name);
56 } tracefs_ops __ro_after_init;
58 static char *get_dname(struct dentry *dentry)
62 int len = dentry->d_name.len;
64 dname = dentry->d_name.name;
65 name = kmalloc(len + 1, GFP_KERNEL);
68 memcpy(name, dname, len);
73 static int tracefs_syscall_mkdir(struct inode *inode, struct dentry *dentry, umode_t mode)
78 name = get_dname(dentry);
83 * The mkdir call can call the generic functions that create
84 * the files within the tracefs system. It is up to the individual
85 * mkdir routine to handle races.
88 ret = tracefs_ops.mkdir(name);
96 static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
101 name = get_dname(dentry);
106 * The rmdir call can call the generic functions that create
107 * the files within the tracefs system. It is up to the individual
108 * rmdir routine to handle races.
109 * This time we need to unlock not only the parent (inode) but
110 * also the directory that is being deleted.
113 inode_unlock(dentry->d_inode);
115 ret = tracefs_ops.rmdir(name);
117 inode_lock_nested(inode, I_MUTEX_PARENT);
118 inode_lock(dentry->d_inode);
125 static const struct inode_operations tracefs_dir_inode_operations = {
126 .lookup = simple_lookup,
127 .mkdir = tracefs_syscall_mkdir,
128 .rmdir = tracefs_syscall_rmdir,
131 static struct inode *tracefs_get_inode(struct super_block *sb)
133 struct inode *inode = new_inode(sb);
135 inode->i_ino = get_next_ino();
136 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
141 struct tracefs_mount_opts {
145 /* Opt_* bitfield. */
156 static const match_table_t tokens = {
159 {Opt_mode, "mode=%o"},
163 struct tracefs_fs_info {
164 struct tracefs_mount_opts mount_opts;
167 static void change_gid(struct dentry *dentry, kgid_t gid)
169 if (!dentry->d_inode)
171 dentry->d_inode->i_gid = gid;
175 * Taken from d_walk, but without he need for handling renames.
176 * Nothing can be renamed while walking the list, as tracefs
177 * does not support renames. This is only called when mounting
178 * or remounting the file system, to set all the files to
181 static void set_gid(struct dentry *parent, kgid_t gid)
183 struct dentry *this_parent;
184 struct list_head *next;
186 this_parent = parent;
187 spin_lock(&this_parent->d_lock);
189 change_gid(this_parent, gid);
191 next = this_parent->d_subdirs.next;
193 while (next != &this_parent->d_subdirs) {
194 struct list_head *tmp = next;
195 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
198 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
200 change_gid(dentry, gid);
202 if (!list_empty(&dentry->d_subdirs)) {
203 spin_unlock(&this_parent->d_lock);
204 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
205 this_parent = dentry;
206 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
209 spin_unlock(&dentry->d_lock);
212 * All done at this level ... ascend and resume the search.
216 if (this_parent != parent) {
217 struct dentry *child = this_parent;
218 this_parent = child->d_parent;
220 spin_unlock(&child->d_lock);
221 spin_lock(&this_parent->d_lock);
223 /* go into the first sibling still alive */
225 next = child->d_child.next;
226 if (next == &this_parent->d_subdirs)
228 child = list_entry(next, struct dentry, d_child);
229 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
234 spin_unlock(&this_parent->d_lock);
238 static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
240 substring_t args[MAX_OPT_ARGS];
248 opts->mode = TRACEFS_DEFAULT_MODE;
250 while ((p = strsep(&data, ",")) != NULL) {
254 token = match_token(p, tokens, args);
257 if (match_int(&args[0], &option))
259 uid = make_kuid(current_user_ns(), option);
265 if (match_int(&args[0], &option))
267 gid = make_kgid(current_user_ns(), option);
273 if (match_octal(&args[0], &option))
275 opts->mode = option & S_IALLUGO;
278 * We might like to report bad mount options here;
279 * but traditionally tracefs has ignored all mount options
283 opts->opts |= BIT(token);
289 static int tracefs_apply_options(struct super_block *sb, bool remount)
291 struct tracefs_fs_info *fsi = sb->s_fs_info;
292 struct inode *inode = sb->s_root->d_inode;
293 struct tracefs_mount_opts *opts = &fsi->mount_opts;
296 * On remount, only reset mode/uid/gid if they were provided as mount
300 if (!remount || opts->opts & BIT(Opt_mode)) {
301 inode->i_mode &= ~S_IALLUGO;
302 inode->i_mode |= opts->mode;
305 if (!remount || opts->opts & BIT(Opt_uid))
306 inode->i_uid = opts->uid;
308 if (!remount || opts->opts & BIT(Opt_gid)) {
309 /* Set all the group ids to the mount option */
310 set_gid(sb->s_root, opts->gid);
316 static int tracefs_remount(struct super_block *sb, int *flags, char *data)
319 struct tracefs_fs_info *fsi = sb->s_fs_info;
322 err = tracefs_parse_options(data, &fsi->mount_opts);
326 tracefs_apply_options(sb, true);
332 static int tracefs_show_options(struct seq_file *m, struct dentry *root)
334 struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
335 struct tracefs_mount_opts *opts = &fsi->mount_opts;
337 if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
338 seq_printf(m, ",uid=%u",
339 from_kuid_munged(&init_user_ns, opts->uid));
340 if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
341 seq_printf(m, ",gid=%u",
342 from_kgid_munged(&init_user_ns, opts->gid));
343 if (opts->mode != TRACEFS_DEFAULT_MODE)
344 seq_printf(m, ",mode=%o", opts->mode);
349 static const struct super_operations tracefs_super_operations = {
350 .statfs = simple_statfs,
351 .remount_fs = tracefs_remount,
352 .show_options = tracefs_show_options,
355 static int trace_fill_super(struct super_block *sb, void *data, int silent)
357 static const struct tree_descr trace_files[] = {{""}};
358 struct tracefs_fs_info *fsi;
361 fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
368 err = tracefs_parse_options(data, &fsi->mount_opts);
372 err = simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
376 sb->s_op = &tracefs_super_operations;
378 tracefs_apply_options(sb, false);
384 sb->s_fs_info = NULL;
388 static struct dentry *trace_mount(struct file_system_type *fs_type,
389 int flags, const char *dev_name,
392 return mount_single(fs_type, flags, data, trace_fill_super);
395 static struct file_system_type trace_fs_type = {
396 .owner = THIS_MODULE,
398 .mount = trace_mount,
399 .kill_sb = kill_litter_super,
401 MODULE_ALIAS_FS("tracefs");
403 static struct dentry *start_creating(const char *name, struct dentry *parent)
405 struct dentry *dentry;
408 pr_debug("tracefs: creating file '%s'\n",name);
410 error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
411 &tracefs_mount_count);
413 return ERR_PTR(error);
415 /* If the parent is not specified, we create it in the root.
416 * We need the root dentry to do this, which is in the super
417 * block. A pointer to that is in the struct vfsmount that we
421 parent = tracefs_mount->mnt_root;
423 inode_lock(parent->d_inode);
424 dentry = lookup_one_len(name, parent, strlen(name));
425 if (!IS_ERR(dentry) && dentry->d_inode) {
427 dentry = ERR_PTR(-EEXIST);
430 if (IS_ERR(dentry)) {
431 inode_unlock(parent->d_inode);
432 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
438 static struct dentry *failed_creating(struct dentry *dentry)
440 inode_unlock(dentry->d_parent->d_inode);
442 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
446 static struct dentry *end_creating(struct dentry *dentry)
448 inode_unlock(dentry->d_parent->d_inode);
453 * tracefs_create_file - create a file in the tracefs filesystem
454 * @name: a pointer to a string containing the name of the file to create.
455 * @mode: the permission that the file should have.
456 * @parent: a pointer to the parent dentry for this file. This should be a
457 * directory dentry if set. If this parameter is NULL, then the
458 * file will be created in the root of the tracefs filesystem.
459 * @data: a pointer to something that the caller will want to get to later
460 * on. The inode.i_private pointer will point to this value on
462 * @fops: a pointer to a struct file_operations that should be used for
465 * This is the basic "create a file" function for tracefs. It allows for a
466 * wide range of flexibility in creating a file, or a directory (if you want
467 * to create a directory, the tracefs_create_dir() function is
468 * recommended to be used instead.)
470 * This function will return a pointer to a dentry if it succeeds. This
471 * pointer must be passed to the tracefs_remove() function when the file is
472 * to be removed (no automatic cleanup happens if your module is unloaded,
473 * you are responsible here.) If an error occurs, %NULL will be returned.
475 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
478 struct dentry *tracefs_create_file(const char *name, umode_t mode,
479 struct dentry *parent, void *data,
480 const struct file_operations *fops)
482 struct dentry *dentry;
485 if (!(mode & S_IFMT))
487 BUG_ON(!S_ISREG(mode));
488 dentry = start_creating(name, parent);
493 inode = tracefs_get_inode(dentry->d_sb);
494 if (unlikely(!inode))
495 return failed_creating(dentry);
497 inode->i_mode = mode;
498 inode->i_fop = fops ? fops : &tracefs_file_operations;
499 inode->i_private = data;
500 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
501 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
502 d_instantiate(dentry, inode);
503 fsnotify_create(dentry->d_parent->d_inode, dentry);
504 return end_creating(dentry);
507 static struct dentry *__create_dir(const char *name, struct dentry *parent,
508 const struct inode_operations *ops)
510 struct dentry *dentry = start_creating(name, parent);
516 inode = tracefs_get_inode(dentry->d_sb);
517 if (unlikely(!inode))
518 return failed_creating(dentry);
520 /* Do not set bits for OTH */
521 inode->i_mode = S_IFDIR | S_IRWXU | S_IRUSR| S_IRGRP | S_IXUSR | S_IXGRP;
523 inode->i_fop = &simple_dir_operations;
524 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
525 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
527 /* directory inodes start off with i_nlink == 2 (for "." entry) */
529 d_instantiate(dentry, inode);
530 inc_nlink(dentry->d_parent->d_inode);
531 fsnotify_mkdir(dentry->d_parent->d_inode, dentry);
532 return end_creating(dentry);
536 * tracefs_create_dir - create a directory in the tracefs filesystem
537 * @name: a pointer to a string containing the name of the directory to
539 * @parent: a pointer to the parent dentry for this file. This should be a
540 * directory dentry if set. If this parameter is NULL, then the
541 * directory will be created in the root of the tracefs filesystem.
543 * This function creates a directory in tracefs with the given name.
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. If an error occurs, %NULL will be returned.
549 * If tracing is not enabled in the kernel, the value -%ENODEV will be
552 struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
554 return __create_dir(name, parent, &simple_dir_inode_operations);
558 * tracefs_create_instance_dir - create the tracing instances directory
559 * @name: The name of the instances directory to create
560 * @parent: The parent directory that the instances directory will exist
561 * @mkdir: The function to call when a mkdir is performed.
562 * @rmdir: The function to call when a rmdir is performed.
564 * Only one instances directory is allowed.
566 * The instances directory is special as it allows for mkdir and rmdir to
567 * to be done by userspace. When a mkdir or rmdir is performed, the inode
568 * locks are released and the methhods passed in (@mkdir and @rmdir) are
569 * called without locks and with the name of the directory being created
570 * within the instances directory.
572 * Returns the dentry of the instances directory.
574 __init struct dentry *tracefs_create_instance_dir(const char *name,
575 struct dentry *parent,
576 int (*mkdir)(const char *name),
577 int (*rmdir)(const char *name))
579 struct dentry *dentry;
581 /* Only allow one instance of the instances directory. */
582 if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
585 dentry = __create_dir(name, parent, &tracefs_dir_inode_operations);
589 tracefs_ops.mkdir = mkdir;
590 tracefs_ops.rmdir = rmdir;
595 static int __tracefs_remove(struct dentry *dentry, struct dentry *parent)
599 if (simple_positive(dentry)) {
600 if (dentry->d_inode) {
602 switch (dentry->d_inode->i_mode & S_IFMT) {
604 ret = simple_rmdir(parent->d_inode, dentry);
607 simple_unlink(parent->d_inode, dentry);
619 * tracefs_remove - removes a file or directory from the tracefs filesystem
620 * @dentry: a pointer to a the dentry of the file or directory to be
623 * This function removes a file or directory in tracefs that was previously
624 * created with a call to another tracefs function (like
625 * tracefs_create_file() or variants thereof.)
627 void tracefs_remove(struct dentry *dentry)
629 struct dentry *parent;
632 if (IS_ERR_OR_NULL(dentry))
635 parent = dentry->d_parent;
636 inode_lock(parent->d_inode);
637 ret = __tracefs_remove(dentry, parent);
638 inode_unlock(parent->d_inode);
640 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
644 * tracefs_remove_recursive - recursively removes a directory
645 * @dentry: a pointer to a the dentry of the directory to be removed.
647 * This function recursively removes a directory tree in tracefs that
648 * was previously created with a call to another tracefs function
649 * (like tracefs_create_file() or variants thereof.)
651 void tracefs_remove_recursive(struct dentry *dentry)
653 struct dentry *child, *parent;
655 if (IS_ERR_OR_NULL(dentry))
660 inode_lock(parent->d_inode);
663 * The parent->d_subdirs is protected by the d_lock. Outside that
664 * lock, the child can be unlinked and set to be freed which can
665 * use the d_u.d_child as the rcu head and corrupt this list.
667 spin_lock(&parent->d_lock);
668 list_for_each_entry(child, &parent->d_subdirs, d_child) {
669 if (!simple_positive(child))
672 /* perhaps simple_empty(child) makes more sense */
673 if (!list_empty(&child->d_subdirs)) {
674 spin_unlock(&parent->d_lock);
675 inode_unlock(parent->d_inode);
680 spin_unlock(&parent->d_lock);
682 if (!__tracefs_remove(child, parent))
683 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
686 * The parent->d_lock protects agaist child from unlinking
687 * from d_subdirs. When releasing the parent->d_lock we can
688 * no longer trust that the next pointer is valid.
689 * Restart the loop. We'll skip this one with the
690 * simple_positive() check.
694 spin_unlock(&parent->d_lock);
696 inode_unlock(parent->d_inode);
698 parent = parent->d_parent;
699 inode_lock(parent->d_inode);
705 if (!__tracefs_remove(child, parent))
706 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
707 inode_unlock(parent->d_inode);
711 * tracefs_initialized - Tells whether tracefs has been registered
713 bool tracefs_initialized(void)
715 return tracefs_registered;
718 static int __init tracefs_init(void)
722 retval = sysfs_create_mount_point(kernel_kobj, "tracing");
726 retval = register_filesystem(&trace_fs_type);
728 tracefs_registered = true;
732 core_initcall(tracefs_init);