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>
26 #define TRACEFS_DEFAULT_MODE 0700
27 static struct kmem_cache *tracefs_inode_cachep __ro_after_init;
29 static struct vfsmount *tracefs_mount;
30 static int tracefs_mount_count;
31 static bool tracefs_registered;
33 static struct inode *tracefs_alloc_inode(struct super_block *sb)
35 struct tracefs_inode *ti;
37 ti = kmem_cache_alloc(tracefs_inode_cachep, GFP_KERNEL);
41 return &ti->vfs_inode;
44 static void tracefs_free_inode(struct inode *inode)
46 kmem_cache_free(tracefs_inode_cachep, get_tracefs(inode));
49 static ssize_t default_read_file(struct file *file, char __user *buf,
50 size_t count, loff_t *ppos)
55 static ssize_t default_write_file(struct file *file, const char __user *buf,
56 size_t count, loff_t *ppos)
61 static const struct file_operations tracefs_file_operations = {
62 .read = default_read_file,
63 .write = default_write_file,
65 .llseek = noop_llseek,
68 static struct tracefs_dir_ops {
69 int (*mkdir)(const char *name);
70 int (*rmdir)(const char *name);
71 } tracefs_ops __ro_after_init;
73 static char *get_dname(struct dentry *dentry)
77 int len = dentry->d_name.len;
79 dname = dentry->d_name.name;
80 name = kmalloc(len + 1, GFP_KERNEL);
83 memcpy(name, dname, len);
88 static int tracefs_syscall_mkdir(struct mnt_idmap *idmap,
89 struct inode *inode, struct dentry *dentry,
92 struct tracefs_inode *ti;
96 name = get_dname(dentry);
101 * This is a new directory that does not take the default of
102 * the rootfs. It becomes the default permissions for all the
103 * files and directories underneath it.
105 ti = get_tracefs(inode);
106 ti->flags |= TRACEFS_INSTANCE_INODE;
110 * The mkdir call can call the generic functions that create
111 * the files within the tracefs system. It is up to the individual
112 * mkdir routine to handle races.
115 ret = tracefs_ops.mkdir(name);
123 static int tracefs_syscall_rmdir(struct inode *inode, struct dentry *dentry)
128 name = get_dname(dentry);
133 * The rmdir call can call the generic functions that create
134 * the files within the tracefs system. It is up to the individual
135 * rmdir routine to handle races.
136 * This time we need to unlock not only the parent (inode) but
137 * also the directory that is being deleted.
140 inode_unlock(d_inode(dentry));
142 ret = tracefs_ops.rmdir(name);
144 inode_lock_nested(inode, I_MUTEX_PARENT);
145 inode_lock(d_inode(dentry));
152 static void set_tracefs_inode_owner(struct inode *inode)
154 struct tracefs_inode *ti = get_tracefs(inode);
155 struct inode *root_inode = ti->private;
158 * If this inode has never been referenced, then update
159 * the permissions to the superblock.
161 if (!(ti->flags & TRACEFS_UID_PERM_SET))
162 inode->i_uid = root_inode->i_uid;
164 if (!(ti->flags & TRACEFS_GID_PERM_SET))
165 inode->i_gid = root_inode->i_gid;
168 static int tracefs_permission(struct mnt_idmap *idmap,
169 struct inode *inode, int mask)
171 set_tracefs_inode_owner(inode);
172 return generic_permission(idmap, inode, mask);
175 static int tracefs_getattr(struct mnt_idmap *idmap,
176 const struct path *path, struct kstat *stat,
177 u32 request_mask, unsigned int flags)
179 struct inode *inode = d_backing_inode(path->dentry);
181 set_tracefs_inode_owner(inode);
182 generic_fillattr(idmap, request_mask, inode, stat);
186 static int tracefs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
189 unsigned int ia_valid = attr->ia_valid;
190 struct inode *inode = d_inode(dentry);
191 struct tracefs_inode *ti = get_tracefs(inode);
193 if (ia_valid & ATTR_UID)
194 ti->flags |= TRACEFS_UID_PERM_SET;
196 if (ia_valid & ATTR_GID)
197 ti->flags |= TRACEFS_GID_PERM_SET;
199 return simple_setattr(idmap, dentry, attr);
202 static const struct inode_operations tracefs_instance_dir_inode_operations = {
203 .lookup = simple_lookup,
204 .mkdir = tracefs_syscall_mkdir,
205 .rmdir = tracefs_syscall_rmdir,
206 .permission = tracefs_permission,
207 .getattr = tracefs_getattr,
208 .setattr = tracefs_setattr,
211 static const struct inode_operations tracefs_dir_inode_operations = {
212 .lookup = simple_lookup,
213 .permission = tracefs_permission,
214 .getattr = tracefs_getattr,
215 .setattr = tracefs_setattr,
218 static const struct inode_operations tracefs_file_inode_operations = {
219 .permission = tracefs_permission,
220 .getattr = tracefs_getattr,
221 .setattr = tracefs_setattr,
224 struct inode *tracefs_get_inode(struct super_block *sb)
226 struct inode *inode = new_inode(sb);
228 inode->i_ino = get_next_ino();
229 simple_inode_init_ts(inode);
234 struct tracefs_mount_opts {
238 /* Opt_* bitfield. */
249 static const match_table_t tokens = {
252 {Opt_mode, "mode=%o"},
256 struct tracefs_fs_info {
257 struct tracefs_mount_opts mount_opts;
260 static int tracefs_parse_options(char *data, struct tracefs_mount_opts *opts)
262 substring_t args[MAX_OPT_ARGS];
270 opts->mode = TRACEFS_DEFAULT_MODE;
272 while ((p = strsep(&data, ",")) != NULL) {
276 token = match_token(p, tokens, args);
279 if (match_int(&args[0], &option))
281 uid = make_kuid(current_user_ns(), option);
287 if (match_int(&args[0], &option))
289 gid = make_kgid(current_user_ns(), option);
295 if (match_octal(&args[0], &option))
297 opts->mode = option & S_IALLUGO;
300 * We might like to report bad mount options here;
301 * but traditionally tracefs has ignored all mount options
305 opts->opts |= BIT(token);
311 static int tracefs_apply_options(struct super_block *sb, bool remount)
313 struct tracefs_fs_info *fsi = sb->s_fs_info;
314 struct inode *inode = d_inode(sb->s_root);
315 struct tracefs_mount_opts *opts = &fsi->mount_opts;
319 * On remount, only reset mode/uid/gid if they were provided as mount
323 if (!remount || opts->opts & BIT(Opt_mode)) {
324 tmp_mode = READ_ONCE(inode->i_mode) & ~S_IALLUGO;
325 tmp_mode |= opts->mode;
326 WRITE_ONCE(inode->i_mode, tmp_mode);
329 if (!remount || opts->opts & BIT(Opt_uid))
330 inode->i_uid = opts->uid;
332 if (!remount || opts->opts & BIT(Opt_gid))
333 inode->i_gid = opts->gid;
338 static int tracefs_remount(struct super_block *sb, int *flags, char *data)
341 struct tracefs_fs_info *fsi = sb->s_fs_info;
344 err = tracefs_parse_options(data, &fsi->mount_opts);
348 tracefs_apply_options(sb, true);
354 static int tracefs_show_options(struct seq_file *m, struct dentry *root)
356 struct tracefs_fs_info *fsi = root->d_sb->s_fs_info;
357 struct tracefs_mount_opts *opts = &fsi->mount_opts;
359 if (!uid_eq(opts->uid, GLOBAL_ROOT_UID))
360 seq_printf(m, ",uid=%u",
361 from_kuid_munged(&init_user_ns, opts->uid));
362 if (!gid_eq(opts->gid, GLOBAL_ROOT_GID))
363 seq_printf(m, ",gid=%u",
364 from_kgid_munged(&init_user_ns, opts->gid));
365 if (opts->mode != TRACEFS_DEFAULT_MODE)
366 seq_printf(m, ",mode=%o", opts->mode);
371 static const struct super_operations tracefs_super_operations = {
372 .alloc_inode = tracefs_alloc_inode,
373 .free_inode = tracefs_free_inode,
374 .drop_inode = generic_delete_inode,
375 .statfs = simple_statfs,
376 .remount_fs = tracefs_remount,
377 .show_options = tracefs_show_options,
380 static void tracefs_dentry_iput(struct dentry *dentry, struct inode *inode)
382 struct tracefs_inode *ti;
384 if (!dentry || !inode)
387 ti = get_tracefs(inode);
388 if (ti && ti->flags & TRACEFS_EVENT_INODE)
389 eventfs_set_ei_status_free(ti, dentry);
393 static const struct dentry_operations tracefs_dentry_operations = {
394 .d_iput = tracefs_dentry_iput,
397 static int trace_fill_super(struct super_block *sb, void *data, int silent)
399 static const struct tree_descr trace_files[] = {{""}};
400 struct tracefs_fs_info *fsi;
403 fsi = kzalloc(sizeof(struct tracefs_fs_info), GFP_KERNEL);
410 err = tracefs_parse_options(data, &fsi->mount_opts);
414 err = simple_fill_super(sb, TRACEFS_MAGIC, trace_files);
418 sb->s_op = &tracefs_super_operations;
419 sb->s_d_op = &tracefs_dentry_operations;
421 tracefs_apply_options(sb, false);
427 sb->s_fs_info = NULL;
431 static struct dentry *trace_mount(struct file_system_type *fs_type,
432 int flags, const char *dev_name,
435 return mount_single(fs_type, flags, data, trace_fill_super);
438 static struct file_system_type trace_fs_type = {
439 .owner = THIS_MODULE,
441 .mount = trace_mount,
442 .kill_sb = kill_litter_super,
444 MODULE_ALIAS_FS("tracefs");
446 struct dentry *tracefs_start_creating(const char *name, struct dentry *parent)
448 struct dentry *dentry;
451 pr_debug("tracefs: creating file '%s'\n",name);
453 error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
454 &tracefs_mount_count);
456 return ERR_PTR(error);
458 /* If the parent is not specified, we create it in the root.
459 * We need the root dentry to do this, which is in the super
460 * block. A pointer to that is in the struct vfsmount that we
464 parent = tracefs_mount->mnt_root;
466 inode_lock(d_inode(parent));
467 if (unlikely(IS_DEADDIR(d_inode(parent))))
468 dentry = ERR_PTR(-ENOENT);
470 dentry = lookup_one_len(name, parent, strlen(name));
471 if (!IS_ERR(dentry) && d_inode(dentry)) {
473 dentry = ERR_PTR(-EEXIST);
476 if (IS_ERR(dentry)) {
477 inode_unlock(d_inode(parent));
478 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
484 struct dentry *tracefs_failed_creating(struct dentry *dentry)
486 inode_unlock(d_inode(dentry->d_parent));
488 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
492 struct dentry *tracefs_end_creating(struct dentry *dentry)
494 inode_unlock(d_inode(dentry->d_parent));
499 * eventfs_start_creating - start the process of creating a dentry
500 * @name: Name of the file created for the dentry
501 * @parent: The parent dentry where this dentry will be created
503 * This is a simple helper function for the dynamically created eventfs
504 * files. When the directory of the eventfs files are accessed, their
505 * dentries are created on the fly. This function is used to start that
508 struct dentry *eventfs_start_creating(const char *name, struct dentry *parent)
510 struct dentry *dentry;
513 /* Must always have a parent. */
514 if (WARN_ON_ONCE(!parent))
515 return ERR_PTR(-EINVAL);
517 error = simple_pin_fs(&trace_fs_type, &tracefs_mount,
518 &tracefs_mount_count);
520 return ERR_PTR(error);
522 if (unlikely(IS_DEADDIR(parent->d_inode)))
523 dentry = ERR_PTR(-ENOENT);
525 dentry = lookup_one_len(name, parent, strlen(name));
527 if (!IS_ERR(dentry) && dentry->d_inode) {
529 dentry = ERR_PTR(-EEXIST);
533 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
539 * eventfs_failed_creating - clean up a failed eventfs dentry creation
540 * @dentry: The dentry to clean up
542 * If after calling eventfs_start_creating(), a failure is detected, the
543 * resources created by eventfs_start_creating() needs to be cleaned up. In
544 * that case, this function should be called to perform that clean up.
546 struct dentry *eventfs_failed_creating(struct dentry *dentry)
549 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
554 * eventfs_end_creating - Finish the process of creating a eventfs dentry
555 * @dentry: The dentry that has successfully been created.
557 * This function is currently just a place holder to match
558 * eventfs_start_creating(). In case any synchronization needs to be added,
559 * this function will be used to implement that without having to modify
560 * the callers of eventfs_start_creating().
562 struct dentry *eventfs_end_creating(struct dentry *dentry)
567 /* Find the inode that this will use for default */
568 static struct inode *instance_inode(struct dentry *parent, struct inode *inode)
570 struct tracefs_inode *ti;
572 /* If parent is NULL then use root inode */
574 return d_inode(inode->i_sb->s_root);
576 /* Find the inode that is flagged as an instance or the root inode */
577 while (!IS_ROOT(parent)) {
578 ti = get_tracefs(d_inode(parent));
579 if (ti->flags & TRACEFS_INSTANCE_INODE)
581 parent = parent->d_parent;
584 return d_inode(parent);
588 * tracefs_create_file - create a file in the tracefs filesystem
589 * @name: a pointer to a string containing the name of the file to create.
590 * @mode: the permission that the file should have.
591 * @parent: a pointer to the parent dentry for this file. This should be a
592 * directory dentry if set. If this parameter is NULL, then the
593 * file will be created in the root of the tracefs filesystem.
594 * @data: a pointer to something that the caller will want to get to later
595 * on. The inode.i_private pointer will point to this value on
597 * @fops: a pointer to a struct file_operations that should be used for
600 * This is the basic "create a file" function for tracefs. It allows for a
601 * wide range of flexibility in creating a file, or a directory (if you want
602 * to create a directory, the tracefs_create_dir() function is
603 * recommended to be used instead.)
605 * This function will return a pointer to a dentry if it succeeds. This
606 * pointer must be passed to the tracefs_remove() function when the file is
607 * to be removed (no automatic cleanup happens if your module is unloaded,
608 * you are responsible here.) If an error occurs, %NULL will be returned.
610 * If tracefs is not enabled in the kernel, the value -%ENODEV will be
613 struct dentry *tracefs_create_file(const char *name, umode_t mode,
614 struct dentry *parent, void *data,
615 const struct file_operations *fops)
617 struct tracefs_inode *ti;
618 struct dentry *dentry;
621 if (security_locked_down(LOCKDOWN_TRACEFS))
624 if (!(mode & S_IFMT))
626 BUG_ON(!S_ISREG(mode));
627 dentry = tracefs_start_creating(name, parent);
632 inode = tracefs_get_inode(dentry->d_sb);
633 if (unlikely(!inode))
634 return tracefs_failed_creating(dentry);
636 ti = get_tracefs(inode);
637 ti->private = instance_inode(parent, inode);
639 inode->i_mode = mode;
640 inode->i_op = &tracefs_file_inode_operations;
641 inode->i_fop = fops ? fops : &tracefs_file_operations;
642 inode->i_private = data;
643 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
644 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
645 d_instantiate(dentry, inode);
646 fsnotify_create(d_inode(dentry->d_parent), dentry);
647 return tracefs_end_creating(dentry);
650 static struct dentry *__create_dir(const char *name, struct dentry *parent,
651 const struct inode_operations *ops)
653 struct tracefs_inode *ti;
654 struct dentry *dentry = tracefs_start_creating(name, parent);
660 inode = tracefs_get_inode(dentry->d_sb);
661 if (unlikely(!inode))
662 return tracefs_failed_creating(dentry);
664 /* Do not set bits for OTH */
665 inode->i_mode = S_IFDIR | S_IRWXU | S_IRUSR| S_IRGRP | S_IXUSR | S_IXGRP;
667 inode->i_fop = &simple_dir_operations;
668 inode->i_uid = d_inode(dentry->d_parent)->i_uid;
669 inode->i_gid = d_inode(dentry->d_parent)->i_gid;
671 ti = get_tracefs(inode);
672 ti->private = instance_inode(parent, inode);
674 /* directory inodes start off with i_nlink == 2 (for "." entry) */
676 d_instantiate(dentry, inode);
677 inc_nlink(d_inode(dentry->d_parent));
678 fsnotify_mkdir(d_inode(dentry->d_parent), dentry);
679 return tracefs_end_creating(dentry);
683 * tracefs_create_dir - create a directory in the tracefs filesystem
684 * @name: a pointer to a string containing the name of the directory to
686 * @parent: a pointer to the parent dentry for this file. This should be a
687 * directory dentry if set. If this parameter is NULL, then the
688 * directory will be created in the root of the tracefs filesystem.
690 * This function creates a directory in tracefs with the given name.
692 * This function will return a pointer to a dentry if it succeeds. This
693 * pointer must be passed to the tracefs_remove() function when the file is
694 * to be removed. If an error occurs, %NULL will be returned.
696 * If tracing is not enabled in the kernel, the value -%ENODEV will be
699 struct dentry *tracefs_create_dir(const char *name, struct dentry *parent)
701 if (security_locked_down(LOCKDOWN_TRACEFS))
704 return __create_dir(name, parent, &tracefs_dir_inode_operations);
708 * tracefs_create_instance_dir - create the tracing instances directory
709 * @name: The name of the instances directory to create
710 * @parent: The parent directory that the instances directory will exist
711 * @mkdir: The function to call when a mkdir is performed.
712 * @rmdir: The function to call when a rmdir is performed.
714 * Only one instances directory is allowed.
716 * The instances directory is special as it allows for mkdir and rmdir
717 * to be done by userspace. When a mkdir or rmdir is performed, the inode
718 * locks are released and the methods passed in (@mkdir and @rmdir) are
719 * called without locks and with the name of the directory being created
720 * within the instances directory.
722 * Returns the dentry of the instances directory.
724 __init struct dentry *tracefs_create_instance_dir(const char *name,
725 struct dentry *parent,
726 int (*mkdir)(const char *name),
727 int (*rmdir)(const char *name))
729 struct dentry *dentry;
731 /* Only allow one instance of the instances directory. */
732 if (WARN_ON(tracefs_ops.mkdir || tracefs_ops.rmdir))
735 dentry = __create_dir(name, parent, &tracefs_instance_dir_inode_operations);
739 tracefs_ops.mkdir = mkdir;
740 tracefs_ops.rmdir = rmdir;
745 static void remove_one(struct dentry *victim)
747 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
751 * tracefs_remove - recursively removes a directory
752 * @dentry: a pointer to a the dentry of the directory to be removed.
754 * This function recursively removes a directory tree in tracefs that
755 * was previously created with a call to another tracefs function
756 * (like tracefs_create_file() or variants thereof.)
758 void tracefs_remove(struct dentry *dentry)
760 if (IS_ERR_OR_NULL(dentry))
763 simple_pin_fs(&trace_fs_type, &tracefs_mount, &tracefs_mount_count);
764 simple_recursive_removal(dentry, remove_one);
765 simple_release_fs(&tracefs_mount, &tracefs_mount_count);
769 * tracefs_initialized - Tells whether tracefs has been registered
771 bool tracefs_initialized(void)
773 return tracefs_registered;
776 static void init_once(void *foo)
778 struct tracefs_inode *ti = (struct tracefs_inode *) foo;
780 /* inode_init_once() calls memset() on the vfs_inode portion */
781 inode_init_once(&ti->vfs_inode);
783 /* Zero out the rest */
784 memset_after(ti, 0, vfs_inode);
787 static int __init tracefs_init(void)
791 tracefs_inode_cachep = kmem_cache_create("tracefs_inode_cache",
792 sizeof(struct tracefs_inode),
793 0, (SLAB_RECLAIM_ACCOUNT|
797 if (!tracefs_inode_cachep)
800 retval = sysfs_create_mount_point(kernel_kobj, "tracing");
804 retval = register_filesystem(&trace_fs_type);
806 tracefs_registered = true;
810 core_initcall(tracefs_init);