1 // SPDX-License-Identifier: GPL-2.0
3 * devtmpfs - kernel-maintained tmpfs-based /dev
5 * Copyright (C) 2009, Kay Sievers <kay.sievers@vrfy.org>
7 * During bootup, before any driver core device is registered,
8 * devtmpfs, a tmpfs-based filesystem is created. Every driver-core
9 * device which requests a device node, will add a node in this
11 * By default, all devices are named after the name of the device,
12 * owned by root and have a default mode of 0600. Subsystems can
13 * overwrite the default setting if needed.
16 #include <linux/kernel.h>
17 #include <linux/syscalls.h>
18 #include <linux/mount.h>
19 #include <linux/device.h>
20 #include <linux/blkdev.h>
21 #include <linux/namei.h>
23 #include <linux/shmem_fs.h>
24 #include <linux/ramfs.h>
25 #include <linux/sched.h>
26 #include <linux/slab.h>
27 #include <linux/kthread.h>
28 #include <linux/init_syscalls.h>
29 #include <uapi/linux/mount.h>
32 #ifdef CONFIG_DEVTMPFS_SAFE
33 #define DEVTMPFS_MFLAGS (MS_SILENT | MS_NOEXEC | MS_NOSUID)
35 #define DEVTMPFS_MFLAGS (MS_SILENT)
38 static struct task_struct *thread;
40 static int __initdata mount_dev = IS_ENABLED(CONFIG_DEVTMPFS_MOUNT);
42 static DEFINE_SPINLOCK(req_lock);
46 struct completion done;
49 umode_t mode; /* 0 => delete */
55 static int __init mount_param(char *str)
57 mount_dev = simple_strtoul(str, NULL, 0);
60 __setup("devtmpfs.mount=", mount_param);
62 static struct vfsmount *mnt;
64 static struct dentry *public_dev_mount(struct file_system_type *fs_type, int flags,
65 const char *dev_name, void *data)
67 struct super_block *s = mnt->mnt_sb;
70 atomic_inc(&s->s_active);
71 down_write(&s->s_umount);
72 err = reconfigure_single(s, flags, data);
74 deactivate_locked_super(s);
77 return dget(s->s_root);
80 static struct file_system_type internal_fs_type = {
83 .init_fs_context = shmem_init_fs_context,
85 .init_fs_context = ramfs_init_fs_context,
87 .kill_sb = kill_litter_super,
90 static struct file_system_type dev_fs_type = {
92 .mount = public_dev_mount,
96 static inline int is_blockdev(struct device *dev)
98 return dev->class == &block_class;
101 static inline int is_blockdev(struct device *dev) { return 0; }
104 static int devtmpfs_submit_req(struct req *req, const char *tmp)
106 init_completion(&req->done);
108 spin_lock(&req_lock);
109 req->next = requests;
111 spin_unlock(&req_lock);
113 wake_up_process(thread);
114 wait_for_completion(&req->done);
121 int devtmpfs_create_node(struct device *dev)
123 const char *tmp = NULL;
130 req.uid = GLOBAL_ROOT_UID;
131 req.gid = GLOBAL_ROOT_GID;
132 req.name = device_get_devnode(dev, &req.mode, &req.uid, &req.gid, &tmp);
138 if (is_blockdev(dev))
145 return devtmpfs_submit_req(&req, tmp);
148 int devtmpfs_delete_node(struct device *dev)
150 const char *tmp = NULL;
156 req.name = device_get_devnode(dev, NULL, NULL, NULL, &tmp);
163 return devtmpfs_submit_req(&req, tmp);
166 static int dev_mkdir(const char *name, umode_t mode)
168 struct dentry *dentry;
172 dentry = kern_path_create(AT_FDCWD, name, &path, LOOKUP_DIRECTORY);
174 return PTR_ERR(dentry);
176 err = vfs_mkdir(&init_user_ns, d_inode(path.dentry), dentry, mode);
178 /* mark as kernel-created inode */
179 d_inode(dentry)->i_private = &thread;
180 done_path_create(&path, dentry);
184 static int create_path(const char *nodepath)
190 /* parent directories do not exist, create them */
191 path = kstrdup(nodepath, GFP_KERNEL);
201 err = dev_mkdir(path, 0755);
202 if (err && err != -EEXIST)
211 static int handle_create(const char *nodename, umode_t mode, kuid_t uid,
212 kgid_t gid, struct device *dev)
214 struct dentry *dentry;
218 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
219 if (dentry == ERR_PTR(-ENOENT)) {
220 create_path(nodename);
221 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
224 return PTR_ERR(dentry);
226 err = vfs_mknod(&init_user_ns, d_inode(path.dentry), dentry, mode,
229 struct iattr newattrs;
231 newattrs.ia_mode = mode;
232 newattrs.ia_uid = uid;
233 newattrs.ia_gid = gid;
234 newattrs.ia_valid = ATTR_MODE|ATTR_UID|ATTR_GID;
235 inode_lock(d_inode(dentry));
236 notify_change(&init_user_ns, dentry, &newattrs, NULL);
237 inode_unlock(d_inode(dentry));
239 /* mark as kernel-created inode */
240 d_inode(dentry)->i_private = &thread;
242 done_path_create(&path, dentry);
246 static int dev_rmdir(const char *name)
249 struct dentry *dentry;
252 dentry = kern_path_locked(name, &parent);
254 return PTR_ERR(dentry);
255 if (d_really_is_positive(dentry)) {
256 if (d_inode(dentry)->i_private == &thread)
257 err = vfs_rmdir(&init_user_ns, d_inode(parent.dentry),
265 inode_unlock(d_inode(parent.dentry));
270 static int delete_path(const char *nodepath)
275 path = kstrdup(nodepath, GFP_KERNEL);
282 base = strrchr(path, '/');
286 err = dev_rmdir(path);
295 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
297 /* did we create it */
298 if (inode->i_private != &thread)
301 /* does the dev_t match */
302 if (is_blockdev(dev)) {
303 if (!S_ISBLK(stat->mode))
306 if (!S_ISCHR(stat->mode))
309 if (stat->rdev != dev->devt)
316 static int handle_remove(const char *nodename, struct device *dev)
319 struct dentry *dentry;
323 dentry = kern_path_locked(nodename, &parent);
325 return PTR_ERR(dentry);
327 if (d_really_is_positive(dentry)) {
329 struct path p = {.mnt = parent.mnt, .dentry = dentry};
330 err = vfs_getattr(&p, &stat, STATX_TYPE | STATX_MODE,
331 AT_STATX_SYNC_AS_STAT);
332 if (!err && dev_mynode(dev, d_inode(dentry), &stat)) {
333 struct iattr newattrs;
335 * before unlinking this node, reset permissions
336 * of possible references like hardlinks
338 newattrs.ia_uid = GLOBAL_ROOT_UID;
339 newattrs.ia_gid = GLOBAL_ROOT_GID;
340 newattrs.ia_mode = stat.mode & ~0777;
342 ATTR_UID|ATTR_GID|ATTR_MODE;
343 inode_lock(d_inode(dentry));
344 notify_change(&init_user_ns, dentry, &newattrs, NULL);
345 inode_unlock(d_inode(dentry));
346 err = vfs_unlink(&init_user_ns, d_inode(parent.dentry),
348 if (!err || err == -ENOENT)
355 inode_unlock(d_inode(parent.dentry));
358 if (deleted && strchr(nodename, '/'))
359 delete_path(nodename);
364 * If configured, or requested by the commandline, devtmpfs will be
365 * auto-mounted after the kernel mounted the root filesystem.
367 int __init devtmpfs_mount(void)
377 err = init_mount("devtmpfs", "dev", "devtmpfs", DEVTMPFS_MFLAGS, NULL);
379 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
381 printk(KERN_INFO "devtmpfs: mounted\n");
385 static __initdata DECLARE_COMPLETION(setup_done);
387 static int handle(const char *name, umode_t mode, kuid_t uid, kgid_t gid,
391 return handle_create(name, mode, uid, gid, dev);
393 return handle_remove(name, dev);
396 static void __noreturn devtmpfs_work_loop(void)
399 spin_lock(&req_lock);
401 struct req *req = requests;
403 spin_unlock(&req_lock);
405 struct req *next = req->next;
406 req->err = handle(req->name, req->mode,
407 req->uid, req->gid, req->dev);
408 complete(&req->done);
411 spin_lock(&req_lock);
413 __set_current_state(TASK_INTERRUPTIBLE);
414 spin_unlock(&req_lock);
419 static noinline int __init devtmpfs_setup(void *p)
423 err = ksys_unshare(CLONE_NEWNS);
426 err = init_mount("devtmpfs", "/", "devtmpfs", DEVTMPFS_MFLAGS, NULL);
429 init_chdir("/.."); /* will traverse into overmounted root */
437 * The __ref is because devtmpfs_setup needs to be __init for the routines it
438 * calls. That call is done while devtmpfs_init, which is marked __init,
439 * synchronously waits for it to complete.
441 static int __ref devtmpfsd(void *p)
443 int err = devtmpfs_setup(p);
445 complete(&setup_done);
448 devtmpfs_work_loop();
453 * Create devtmpfs instance, driver-core devices will add their device
456 int __init devtmpfs_init(void)
458 char opts[] = "mode=0755";
461 mnt = vfs_kern_mount(&internal_fs_type, 0, "devtmpfs", opts);
463 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %ld\n",
467 err = register_filesystem(&dev_fs_type);
469 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
474 thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
475 if (!IS_ERR(thread)) {
476 wait_for_completion(&setup_done);
478 err = PTR_ERR(thread);
483 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
484 unregister_filesystem(&dev_fs_type);
489 printk(KERN_INFO "devtmpfs: initialized\n");