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);
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 {
154 static const match_table_t tokens = {
157 {Opt_mode, "mode=%o"},
161 struct tracefs_fs_info {
162 struct tracefs_mount_opts mount_opts;
165 static void change_gid(struct dentry *dentry, kgid_t gid)
167 if (!dentry->d_inode)
169 dentry->d_inode->i_gid = gid;
173 * Taken from d_walk, but without he need for handling renames.
174 * Nothing can be renamed while walking the list, as tracefs
175 * does not support renames. This is only called when mounting
176 * or remounting the file system, to set all the files to
179 static void set_gid(struct dentry *parent, kgid_t gid)
181 struct dentry *this_parent;
182 struct list_head *next;
184 this_parent = parent;
185 spin_lock(&this_parent->d_lock);
187 change_gid(this_parent, gid);
189 next = this_parent->d_subdirs.next;
191 while (next != &this_parent->d_subdirs) {
192 struct list_head *tmp = next;
193 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
196 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
198 change_gid(dentry, gid);
200 if (!list_empty(&dentry->d_subdirs)) {
201 spin_unlock(&this_parent->d_lock);
202 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
203 this_parent = dentry;
204 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
207 spin_unlock(&dentry->d_lock);
210 * All done at this level ... ascend and resume the search.
214 if (this_parent != parent) {
215 struct dentry *child = this_parent;
216 this_parent = child->d_parent;
218 spin_unlock(&child->d_lock);
219 spin_lock(&this_parent->d_lock);
221 /* go into the first sibling still alive */
223 next = child->d_child.next;
224 if (next == &this_parent->d_subdirs)
226 child = list_entry(next, struct dentry, d_child);
227 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
232 spin_unlock(&this_parent->d_lock);
236 static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
238 substring_t args[MAX_OPT_ARGS];
245 opts->mode = TRACEFS_DEFAULT_MODE;
247 while ((p = strsep(&data, ",")) != NULL) {
251 token = match_token(p, tokens, args);
254 if (match_int(&args[0], &option))
256 uid = make_kuid(current_user_ns(), option);
262 if (match_int(&args[0], &option))
264 gid = make_kgid(current_user_ns(), option);
268 set_gid(tracefs_mount->mnt_root, gid);
271 if (match_octal(&args[0], &option))
273 opts->mode = option & S_IALLUGO;
276 * We might like to report bad mount options here;
277 * but traditionally tracefs has ignored all mount options
285 static int tracefs_apply_options(struct super_block *sb)
287 struct tracefs_fs_info *fsi = sb->s_fs_info;
288 struct inode *inode = sb->s_root->d_inode;
289 struct tracefs_mount_opts *opts = &fsi->mount_opts;
291 inode->i_mode &= ~S_IALLUGO;
292 inode->i_mode |= opts->mode;
294 inode->i_uid = opts->uid;
295 inode->i_gid = opts->gid;
300 static int tracefs_remount(struct super_block *sb, int *flags, char *data)
303 struct tracefs_fs_info *fsi = sb->s_fs_info;
306 err = tracefs_parse_options(data, &fsi->mount_opts);
310 tracefs_apply_options(sb);
316 static int tracefs_show_options(struct seq_file *m, struct dentry *root)
318 struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
319 struct tracefs_mount_opts *opts = &fsi->mount_opts;
321 if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
322 seq_printf(m, ",uid=%u",
323 from_kuid_munged(&init_user_ns, opts->uid));
324 if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
325 seq_printf(m, ",gid=%u",
326 from_kgid_munged(&init_user_ns, opts->gid));
327 if (opts->mode != TRACEFS_DEFAULT_MODE)
328 seq_printf(m, ",mode=%o", opts->mode);
333 static const struct super_operations tracefs_super_operations = {
334 .statfs = simple_statfs,
335 .remount_fs = tracefs_remount,
336 .show_options = tracefs_show_options,
339 static int trace_fill_super(struct super_block *sb, void *data, int silent)
341 static struct tree_descr trace_files[] = {{""}};
342 struct tracefs_fs_info *fsi;
345 save_mount_options(sb, data);
347 fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
354 err = tracefs_parse_options(data, &fsi->mount_opts);
358 err = simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
362 sb->s_op = &tracefs_super_operations;
364 tracefs_apply_options(sb);
370 sb->s_fs_info = NULL;
374 static struct dentry *trace_mount(struct file_system_type *fs_type,
375 int flags, const char *dev_name,
378 return mount_single(fs_type, flags, data, trace_fill_super);
381 static struct file_system_type trace_fs_type = {
382 .owner = THIS_MODULE,
384 .mount = trace_mount,
385 .kill_sb = kill_litter_super,
387 MODULE_ALIAS_FS("tracefs");
389 static struct dentry *start_creating(const char *name, struct dentry *parent)
391 struct dentry *dentry;
394 pr_debug("tracefs: creating file '%s'\n",name);
396 error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
397 &tracefs_mount_count);
399 return ERR_PTR(error);
401 /* If the parent is not specified, we create it in the root.
402 * We need the root dentry to do this, which is in the super
403 * block. A pointer to that is in the struct vfsmount that we
407 parent = tracefs_mount->mnt_root;
409 inode_lock(parent->d_inode);
410 dentry = lookup_one_len(name, parent, strlen(name));
411 if (!IS_ERR(dentry) && dentry->d_inode) {
413 dentry = ERR_PTR(-EEXIST);
416 if (IS_ERR(dentry)) {
417 inode_unlock(parent->d_inode);
418 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
424 static struct dentry *failed_creating(struct dentry *dentry)
426 inode_unlock(dentry->d_parent->d_inode);
428 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
432 static struct dentry *end_creating(struct dentry *dentry)
434 inode_unlock(dentry->d_parent->d_inode);
439 * tracefs_create_file - create a file in the tracefs filesystem
440 * @name: a pointer to a string containing the name of the file to create.
441 * @mode: the permission that the file should have.
442 * @parent: a pointer to the parent dentry for this file. This should be a
443 * directory dentry if set. If this parameter is NULL, then the
444 * file will be created in the root of the tracefs filesystem.
445 * @data: a pointer to something that the caller will want to get to later
446 * on. The inode.i_private pointer will point to this value on
448 * @fops: a pointer to a struct file_operations that should be used for
451 * This is the basic "create a file" function for tracefs. It allows for a
452 * wide range of flexibility in creating a file, or a directory (if you want
453 * to create a directory, the tracefs_create_dir() function is
454 * recommended to be used instead.)
456 * This function will return a pointer to a dentry if it succeeds. This
457 * pointer must be passed to the tracefs_remove() function when the file is
458 * to be removed (no automatic cleanup happens if your module is unloaded,
459 * you are responsible here.) If an error occurs, %NULL will be returned.
461 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
464 struct dentry *tracefs_create_file(const char *name, umode_t mode,
465 struct dentry *parent, void *data,
466 const struct file_operations *fops)
468 struct dentry *dentry;
471 if (!(mode & S_IFMT))
473 BUG_ON(!S_ISREG(mode));
474 dentry = start_creating(name, parent);
479 inode = tracefs_get_inode(dentry->d_sb);
480 if (unlikely(!inode))
481 return failed_creating(dentry);
483 inode->i_mode = mode;
484 inode->i_fop = fops ? fops : &tracefs_file_operations;
485 inode->i_private = data;
486 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
487 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
488 d_instantiate(dentry, inode);
489 fsnotify_create(dentry->d_parent->d_inode, dentry);
490 return end_creating(dentry);
493 static struct dentry *__create_dir(const char *name, struct dentry *parent,
494 const struct inode_operations *ops)
496 struct dentry *dentry = start_creating(name, parent);
502 inode = tracefs_get_inode(dentry->d_sb);
503 if (unlikely(!inode))
504 return failed_creating(dentry);
506 /* Do not set bits for OTH */
507 inode->i_mode = S_IFDIR | S_IRWXU | S_IRUSR| S_IRGRP | S_IXUSR | S_IXGRP;
509 inode->i_fop = &simple_dir_operations;
510 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
511 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
513 /* directory inodes start off with i_nlink == 2 (for "." entry) */
515 d_instantiate(dentry, inode);
516 inc_nlink(dentry->d_parent->d_inode);
517 fsnotify_mkdir(dentry->d_parent->d_inode, dentry);
518 return end_creating(dentry);
522 * tracefs_create_dir - create a directory in the tracefs filesystem
523 * @name: a pointer to a string containing the name of the directory to
525 * @parent: a pointer to the parent dentry for this file. This should be a
526 * directory dentry if set. If this parameter is NULL, then the
527 * directory will be created in the root of the tracefs filesystem.
529 * This function creates a directory in tracefs with the given name.
531 * This function will return a pointer to a dentry if it succeeds. This
532 * pointer must be passed to the tracefs_remove() function when the file is
533 * to be removed. If an error occurs, %NULL will be returned.
535 * If tracing is not enabled in the kernel, the value -%ENODEV will be
538 struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
540 return __create_dir(name, parent, &simple_dir_inode_operations);
544 * tracefs_create_instance_dir - create the tracing instances directory
545 * @name: The name of the instances directory to create
546 * @parent: The parent directory that the instances directory will exist
547 * @mkdir: The function to call when a mkdir is performed.
548 * @rmdir: The function to call when a rmdir is performed.
550 * Only one instances directory is allowed.
552 * The instances directory is special as it allows for mkdir and rmdir to
553 * to be done by userspace. When a mkdir or rmdir is performed, the inode
554 * locks are released and the methhods passed in (@mkdir and @rmdir) are
555 * called without locks and with the name of the directory being created
556 * within the instances directory.
558 * Returns the dentry of the instances directory.
560 struct dentry *tracefs_create_instance_dir(const char *name, 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);
592 simple_unlink(parent->d_inode, dentry);
604 * tracefs_remove - removes a file or directory from the tracefs filesystem
605 * @dentry: a pointer to a the dentry of the file or directory to be
608 * This function removes a file or directory in tracefs that was previously
609 * created with a call to another tracefs function (like
610 * tracefs_create_file() or variants thereof.)
612 void tracefs_remove(struct dentry *dentry)
614 struct dentry *parent;
617 if (IS_ERR_OR_NULL(dentry))
620 parent = dentry->d_parent;
621 inode_lock(parent->d_inode);
622 ret = __tracefs_remove(dentry, parent);
623 inode_unlock(parent->d_inode);
625 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
629 * tracefs_remove_recursive - recursively removes a directory
630 * @dentry: a pointer to a the dentry of the directory to be removed.
632 * This function recursively removes a directory tree in tracefs that
633 * was previously created with a call to another tracefs function
634 * (like tracefs_create_file() or variants thereof.)
636 void tracefs_remove_recursive(struct dentry *dentry)
638 struct dentry *child, *parent;
640 if (IS_ERR_OR_NULL(dentry))
645 inode_lock(parent->d_inode);
648 * The parent->d_subdirs is protected by the d_lock. Outside that
649 * lock, the child can be unlinked and set to be freed which can
650 * use the d_u.d_child as the rcu head and corrupt this list.
652 spin_lock(&parent->d_lock);
653 list_for_each_entry(child, &parent->d_subdirs, d_child) {
654 if (!simple_positive(child))
657 /* perhaps simple_empty(child) makes more sense */
658 if (!list_empty(&child->d_subdirs)) {
659 spin_unlock(&parent->d_lock);
660 inode_unlock(parent->d_inode);
665 spin_unlock(&parent->d_lock);
667 if (!__tracefs_remove(child, parent))
668 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
671 * The parent->d_lock protects agaist child from unlinking
672 * from d_subdirs. When releasing the parent->d_lock we can
673 * no longer trust that the next pointer is valid.
674 * Restart the loop. We'll skip this one with the
675 * simple_positive() check.
679 spin_unlock(&parent->d_lock);
681 inode_unlock(parent->d_inode);
683 parent = parent->d_parent;
684 inode_lock(parent->d_inode);
690 if (!__tracefs_remove(child, parent))
691 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
692 inode_unlock(parent->d_inode);
696 * tracefs_initialized - Tells whether tracefs has been registered
698 bool tracefs_initialized(void)
700 return tracefs_registered;
703 static int __init tracefs_init(void)
707 retval = sysfs_create_mount_point(kernel_kobj, "tracing");
711 retval = register_filesystem(&trace_fs_type);
713 tracefs_registered = true;
717 core_initcall(tracefs_init);