1 // SPDX-License-Identifier: GPL-2.0-only
3 * inode.c - part of tracefs, a pseudo file system for activating tracing
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 void change_gid(struct dentry *dentry, kgid_t gid)
166 if (!dentry->d_inode)
168 dentry->d_inode->i_gid = gid;
172 * Taken from d_walk, but without he need for handling renames.
173 * Nothing can be renamed while walking the list, as tracefs
174 * does not support renames. This is only called when mounting
175 * or remounting the file system, to set all the files to
178 static void set_gid(struct dentry *parent, kgid_t gid)
180 struct dentry *this_parent;
181 struct list_head *next;
183 this_parent = parent;
184 spin_lock(&this_parent->d_lock);
186 change_gid(this_parent, gid);
188 next = this_parent->d_subdirs.next;
190 while (next != &this_parent->d_subdirs) {
191 struct list_head *tmp = next;
192 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
195 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
197 change_gid(dentry, gid);
199 if (!list_empty(&dentry->d_subdirs)) {
200 spin_unlock(&this_parent->d_lock);
201 spin_release(&dentry->d_lock.dep_map, _RET_IP_);
202 this_parent = dentry;
203 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
206 spin_unlock(&dentry->d_lock);
209 * All done at this level ... ascend and resume the search.
213 if (this_parent != parent) {
214 struct dentry *child = this_parent;
215 this_parent = child->d_parent;
217 spin_unlock(&child->d_lock);
218 spin_lock(&this_parent->d_lock);
220 /* go into the first sibling still alive */
222 next = child->d_child.next;
223 if (next == &this_parent->d_subdirs)
225 child = list_entry(next, struct dentry, d_child);
226 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
231 spin_unlock(&this_parent->d_lock);
235 static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
237 substring_t args[MAX_OPT_ARGS];
244 opts->mode = TRACEFS_DEFAULT_MODE;
246 while ((p = strsep(&data, ",")) != NULL) {
250 token = match_token(p, tokens, args);
253 if (match_int(&args[0], &option))
255 uid = make_kuid(current_user_ns(), option);
261 if (match_int(&args[0], &option))
263 gid = make_kgid(current_user_ns(), option);
267 set_gid(tracefs_mount->mnt_root, gid);
270 if (match_octal(&args[0], &option))
272 opts->mode = option & S_IALLUGO;
275 * We might like to report bad mount options here;
276 * but traditionally tracefs has ignored all mount options
284 static int tracefs_apply_options(struct super_block *sb)
286 struct tracefs_fs_info *fsi = sb->s_fs_info;
287 struct inode *inode = sb->s_root->d_inode;
288 struct tracefs_mount_opts *opts = &fsi->mount_opts;
290 inode->i_mode &= ~S_IALLUGO;
291 inode->i_mode |= opts->mode;
293 inode->i_uid = opts->uid;
294 inode->i_gid = opts->gid;
299 static int tracefs_remount(struct super_block *sb, int *flags, char *data)
302 struct tracefs_fs_info *fsi = sb->s_fs_info;
305 err = tracefs_parse_options(data, &fsi->mount_opts);
309 tracefs_apply_options(sb);
315 static int tracefs_show_options(struct seq_file *m, struct dentry *root)
317 struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
318 struct tracefs_mount_opts *opts = &fsi->mount_opts;
320 if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
321 seq_printf(m, ",uid=%u",
322 from_kuid_munged(&init_user_ns, opts->uid));
323 if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
324 seq_printf(m, ",gid=%u",
325 from_kgid_munged(&init_user_ns, opts->gid));
326 if (opts->mode != TRACEFS_DEFAULT_MODE)
327 seq_printf(m, ",mode=%o", opts->mode);
332 static const struct super_operations tracefs_super_operations = {
333 .statfs = simple_statfs,
334 .remount_fs = tracefs_remount,
335 .show_options = tracefs_show_options,
338 static int trace_fill_super(struct super_block *sb, void *data, int silent)
340 static const struct tree_descr trace_files[] = {{""}};
341 struct tracefs_fs_info *fsi;
344 fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
351 err = tracefs_parse_options(data, &fsi->mount_opts);
355 err = simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
359 sb->s_op = &tracefs_super_operations;
361 tracefs_apply_options(sb);
367 sb->s_fs_info = NULL;
371 static struct dentry *trace_mount(struct file_system_type *fs_type,
372 int flags, const char *dev_name,
375 return mount_single(fs_type, flags, data, trace_fill_super);
378 static struct file_system_type trace_fs_type = {
379 .owner = THIS_MODULE,
381 .mount = trace_mount,
382 .kill_sb = kill_litter_super,
384 MODULE_ALIAS_FS("tracefs");
386 static struct dentry *start_creating(const char *name, struct dentry *parent)
388 struct dentry *dentry;
391 pr_debug("tracefs: creating file '%s'\n",name);
393 error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
394 &tracefs_mount_count);
396 return ERR_PTR(error);
398 /* If the parent is not specified, we create it in the root.
399 * We need the root dentry to do this, which is in the super
400 * block. A pointer to that is in the struct vfsmount that we
404 parent = tracefs_mount->mnt_root;
406 inode_lock(parent->d_inode);
407 if (unlikely(IS_DEADDIR(parent->d_inode)))
408 dentry = ERR_PTR(-ENOENT);
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 (security_locked_down(LOCKDOWN_TRACEFS))
474 if (!(mode & S_IFMT))
476 BUG_ON(!S_ISREG(mode));
477 dentry = start_creating(name, parent);
482 inode = tracefs_get_inode(dentry->d_sb);
483 if (unlikely(!inode))
484 return failed_creating(dentry);
486 inode->i_mode = mode;
487 inode->i_fop = fops ? fops : &tracefs_file_operations;
488 inode->i_private = data;
489 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
490 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
491 d_instantiate(dentry, inode);
492 fsnotify_create(dentry->d_parent->d_inode, dentry);
493 return end_creating(dentry);
496 static struct dentry *__create_dir(const char *name, struct dentry *parent,
497 const struct inode_operations *ops)
499 struct dentry *dentry = start_creating(name, parent);
505 inode = tracefs_get_inode(dentry->d_sb);
506 if (unlikely(!inode))
507 return failed_creating(dentry);
509 /* Do not set bits for OTH */
510 inode->i_mode = S_IFDIR | S_IRWXU | S_IRUSR| S_IRGRP | S_IXUSR | S_IXGRP;
512 inode->i_fop = &simple_dir_operations;
513 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
514 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
516 /* directory inodes start off with i_nlink == 2 (for "." entry) */
518 d_instantiate(dentry, inode);
519 inc_nlink(dentry->d_parent->d_inode);
520 fsnotify_mkdir(dentry->d_parent->d_inode, dentry);
521 return end_creating(dentry);
525 * tracefs_create_dir - create a directory in the tracefs filesystem
526 * @name: a pointer to a string containing the name of the directory to
528 * @parent: a pointer to the parent dentry for this file. This should be a
529 * directory dentry if set. If this parameter is NULL, then the
530 * directory will be created in the root of the tracefs filesystem.
532 * This function creates a directory in tracefs with the given name.
534 * This function will return a pointer to a dentry if it succeeds. This
535 * pointer must be passed to the tracefs_remove() function when the file is
536 * to be removed. If an error occurs, %NULL will be returned.
538 * If tracing is not enabled in the kernel, the value -%ENODEV will be
541 struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
543 return __create_dir(name, parent, &simple_dir_inode_operations);
547 * tracefs_create_instance_dir - create the tracing instances directory
548 * @name: The name of the instances directory to create
549 * @parent: The parent directory that the instances directory will exist
550 * @mkdir: The function to call when a mkdir is performed.
551 * @rmdir: The function to call when a rmdir is performed.
553 * Only one instances directory is allowed.
555 * The instances directory is special as it allows for mkdir and rmdir to
556 * to be done by userspace. When a mkdir or rmdir is performed, the inode
557 * locks are released and the methods passed in (@mkdir and @rmdir) are
558 * called without locks and with the name of the directory being created
559 * within the instances directory.
561 * Returns the dentry of the instances directory.
563 __init struct dentry *tracefs_create_instance_dir(const char *name,
564 struct dentry *parent,
565 int (*mkdir)(const char *name),
566 int (*rmdir)(const char *name))
568 struct dentry *dentry;
570 /* Only allow one instance of the instances directory. */
571 if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
574 dentry = __create_dir(name, parent, &tracefs_dir_inode_operations);
578 tracefs_ops.mkdir = mkdir;
579 tracefs_ops.rmdir = rmdir;
584 static void remove_one(struct dentry *victim)
586 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
590 * tracefs_remove - recursively removes a directory
591 * @dentry: a pointer to a the dentry of the directory to be removed.
593 * This function recursively removes a directory tree in tracefs that
594 * was previously created with a call to another tracefs function
595 * (like tracefs_create_file() or variants thereof.)
597 void tracefs_remove(struct dentry *dentry)
599 if (IS_ERR_OR_NULL(dentry))
602 simple_pin_fs(&trace_fs_type, &tracefs_mount, &tracefs_mount_count);
603 simple_recursive_removal(dentry, remove_one);
604 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
608 * tracefs_initialized - Tells whether tracefs has been registered
610 bool tracefs_initialized(void)
612 return tracefs_registered;
615 static int __init tracefs_init(void)
619 retval = sysfs_create_mount_point(kernel_kobj, "tracing");
623 retval = register_filesystem(&trace_fs_type);
625 tracefs_registered = true;
629 core_initcall(tracefs_init);