1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * devtmpfs - kernel-maintained tmpfs-based /dev
4  *
5  * Copyright (C) 2009, Kay Sievers <kay.sievers@vrfy.org>
6  *
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
10  * filesystem.
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.
14  */
15 
16 #include <linux/kernel.h>
17 #include <linux/syscalls.h>
18 #include <linux/mount.h>
19 #include <linux/device.h>
20 #include <linux/genhd.h>
21 #include <linux/namei.h>
22 #include <linux/fs.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 "base.h"
29 
30 static struct task_struct *thread;
31 
32 #if defined CONFIG_DEVTMPFS_MOUNT
33 static int mount_dev = 1;
34 #else
35 static int mount_dev;
36 #endif
37 
38 static DEFINE_SPINLOCK(req_lock);
39 
40 static struct req {
41 	struct req *next;
42 	struct completion done;
43 	int err;
44 	const char *name;
45 	umode_t mode;	/* 0 => delete */
46 	kuid_t uid;
47 	kgid_t gid;
48 	struct device *dev;
49 } *requests;
50 
mount_param(char * str)51 static int __init mount_param(char *str)
52 {
53 	mount_dev = simple_strtoul(str, NULL, 0);
54 	return 1;
55 }
56 __setup("devtmpfs.mount=", mount_param);
57 
dev_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)58 static struct dentry *dev_mount(struct file_system_type *fs_type, int flags,
59 		      const char *dev_name, void *data)
60 {
61 #ifdef CONFIG_TMPFS
62 	return mount_single(fs_type, flags, data, shmem_fill_super);
63 #else
64 	return mount_single(fs_type, flags, data, ramfs_fill_super);
65 #endif
66 }
67 
68 static struct file_system_type dev_fs_type = {
69 	.name = "devtmpfs",
70 	.mount = dev_mount,
71 	.kill_sb = kill_litter_super,
72 };
73 
74 #ifdef CONFIG_BLOCK
is_blockdev(struct device * dev)75 static inline int is_blockdev(struct device *dev)
76 {
77 	return dev->class == &block_class;
78 }
79 #else
is_blockdev(struct device * dev)80 static inline int is_blockdev(struct device *dev) { return 0; }
81 #endif
82 
devtmpfs_create_node(struct device * dev)83 int devtmpfs_create_node(struct device *dev)
84 {
85 	const char *tmp = NULL;
86 	struct req req;
87 
88 	if (!thread)
89 		return 0;
90 
91 	req.mode = 0;
92 	req.uid = GLOBAL_ROOT_UID;
93 	req.gid = GLOBAL_ROOT_GID;
94 	req.name = device_get_devnode(dev, &req.mode, &req.uid, &req.gid, &tmp);
95 	if (!req.name)
96 		return -ENOMEM;
97 
98 	if (req.mode == 0)
99 		req.mode = 0600;
100 	if (is_blockdev(dev))
101 		req.mode |= S_IFBLK;
102 	else
103 		req.mode |= S_IFCHR;
104 
105 	req.dev = dev;
106 
107 	init_completion(&req.done);
108 
109 	spin_lock(&req_lock);
110 	req.next = requests;
111 	requests = &req;
112 	spin_unlock(&req_lock);
113 
114 	wake_up_process(thread);
115 	wait_for_completion(&req.done);
116 
117 	kfree(tmp);
118 
119 	return req.err;
120 }
121 
devtmpfs_delete_node(struct device * dev)122 int devtmpfs_delete_node(struct device *dev)
123 {
124 	const char *tmp = NULL;
125 	struct req req;
126 
127 	if (!thread)
128 		return 0;
129 
130 	req.name = device_get_devnode(dev, NULL, NULL, NULL, &tmp);
131 	if (!req.name)
132 		return -ENOMEM;
133 
134 	req.mode = 0;
135 	req.dev = dev;
136 
137 	init_completion(&req.done);
138 
139 	spin_lock(&req_lock);
140 	req.next = requests;
141 	requests = &req;
142 	spin_unlock(&req_lock);
143 
144 	wake_up_process(thread);
145 	wait_for_completion(&req.done);
146 
147 	kfree(tmp);
148 	return req.err;
149 }
150 
dev_mkdir(const char * name,umode_t mode)151 static int dev_mkdir(const char *name, umode_t mode)
152 {
153 	struct dentry *dentry;
154 	struct path path;
155 	int err;
156 
157 	dentry = kern_path_create(AT_FDCWD, name, &path, LOOKUP_DIRECTORY);
158 	if (IS_ERR(dentry))
159 		return PTR_ERR(dentry);
160 
161 	err = vfs_mkdir(d_inode(path.dentry), dentry, mode);
162 	if (!err)
163 		/* mark as kernel-created inode */
164 		d_inode(dentry)->i_private = &thread;
165 	done_path_create(&path, dentry);
166 	return err;
167 }
168 
create_path(const char * nodepath)169 static int create_path(const char *nodepath)
170 {
171 	char *path;
172 	char *s;
173 	int err = 0;
174 
175 	/* parent directories do not exist, create them */
176 	path = kstrdup(nodepath, GFP_KERNEL);
177 	if (!path)
178 		return -ENOMEM;
179 
180 	s = path;
181 	for (;;) {
182 		s = strchr(s, '/');
183 		if (!s)
184 			break;
185 		s[0] = '\0';
186 		err = dev_mkdir(path, 0755);
187 		if (err && err != -EEXIST)
188 			break;
189 		s[0] = '/';
190 		s++;
191 	}
192 	kfree(path);
193 	return err;
194 }
195 
handle_create(const char * nodename,umode_t mode,kuid_t uid,kgid_t gid,struct device * dev)196 static int handle_create(const char *nodename, umode_t mode, kuid_t uid,
197 			 kgid_t gid, struct device *dev)
198 {
199 	struct dentry *dentry;
200 	struct path path;
201 	int err;
202 
203 	dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
204 	if (dentry == ERR_PTR(-ENOENT)) {
205 		create_path(nodename);
206 		dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
207 	}
208 	if (IS_ERR(dentry))
209 		return PTR_ERR(dentry);
210 
211 	err = vfs_mknod(d_inode(path.dentry), dentry, mode, dev->devt);
212 	if (!err) {
213 		struct iattr newattrs;
214 
215 		newattrs.ia_mode = mode;
216 		newattrs.ia_uid = uid;
217 		newattrs.ia_gid = gid;
218 		newattrs.ia_valid = ATTR_MODE|ATTR_UID|ATTR_GID;
219 		inode_lock(d_inode(dentry));
220 		notify_change(dentry, &newattrs, NULL);
221 		inode_unlock(d_inode(dentry));
222 
223 		/* mark as kernel-created inode */
224 		d_inode(dentry)->i_private = &thread;
225 	}
226 	done_path_create(&path, dentry);
227 	return err;
228 }
229 
dev_rmdir(const char * name)230 static int dev_rmdir(const char *name)
231 {
232 	struct path parent;
233 	struct dentry *dentry;
234 	int err;
235 
236 	dentry = kern_path_locked(name, &parent);
237 	if (IS_ERR(dentry))
238 		return PTR_ERR(dentry);
239 	if (d_really_is_positive(dentry)) {
240 		if (d_inode(dentry)->i_private == &thread)
241 			err = vfs_rmdir(d_inode(parent.dentry), dentry);
242 		else
243 			err = -EPERM;
244 	} else {
245 		err = -ENOENT;
246 	}
247 	dput(dentry);
248 	inode_unlock(d_inode(parent.dentry));
249 	path_put(&parent);
250 	return err;
251 }
252 
delete_path(const char * nodepath)253 static int delete_path(const char *nodepath)
254 {
255 	const char *path;
256 	int err = 0;
257 
258 	path = kstrdup(nodepath, GFP_KERNEL);
259 	if (!path)
260 		return -ENOMEM;
261 
262 	for (;;) {
263 		char *base;
264 
265 		base = strrchr(path, '/');
266 		if (!base)
267 			break;
268 		base[0] = '\0';
269 		err = dev_rmdir(path);
270 		if (err)
271 			break;
272 	}
273 
274 	kfree(path);
275 	return err;
276 }
277 
dev_mynode(struct device * dev,struct inode * inode,struct kstat * stat)278 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
279 {
280 	/* did we create it */
281 	if (inode->i_private != &thread)
282 		return 0;
283 
284 	/* does the dev_t match */
285 	if (is_blockdev(dev)) {
286 		if (!S_ISBLK(stat->mode))
287 			return 0;
288 	} else {
289 		if (!S_ISCHR(stat->mode))
290 			return 0;
291 	}
292 	if (stat->rdev != dev->devt)
293 		return 0;
294 
295 	/* ours */
296 	return 1;
297 }
298 
handle_remove(const char * nodename,struct device * dev)299 static int handle_remove(const char *nodename, struct device *dev)
300 {
301 	struct path parent;
302 	struct dentry *dentry;
303 	int deleted = 0;
304 	int err;
305 
306 	dentry = kern_path_locked(nodename, &parent);
307 	if (IS_ERR(dentry))
308 		return PTR_ERR(dentry);
309 
310 	if (d_really_is_positive(dentry)) {
311 		struct kstat stat;
312 		struct path p = {.mnt = parent.mnt, .dentry = dentry};
313 		err = vfs_getattr(&p, &stat, STATX_TYPE | STATX_MODE,
314 				  AT_STATX_SYNC_AS_STAT);
315 		if (!err && dev_mynode(dev, d_inode(dentry), &stat)) {
316 			struct iattr newattrs;
317 			/*
318 			 * before unlinking this node, reset permissions
319 			 * of possible references like hardlinks
320 			 */
321 			newattrs.ia_uid = GLOBAL_ROOT_UID;
322 			newattrs.ia_gid = GLOBAL_ROOT_GID;
323 			newattrs.ia_mode = stat.mode & ~0777;
324 			newattrs.ia_valid =
325 				ATTR_UID|ATTR_GID|ATTR_MODE;
326 			inode_lock(d_inode(dentry));
327 			notify_change(dentry, &newattrs, NULL);
328 			inode_unlock(d_inode(dentry));
329 			err = vfs_unlink(d_inode(parent.dentry), dentry, NULL);
330 			if (!err || err == -ENOENT)
331 				deleted = 1;
332 		}
333 	} else {
334 		err = -ENOENT;
335 	}
336 	dput(dentry);
337 	inode_unlock(d_inode(parent.dentry));
338 
339 	path_put(&parent);
340 	if (deleted && strchr(nodename, '/'))
341 		delete_path(nodename);
342 	return err;
343 }
344 
345 /*
346  * If configured, or requested by the commandline, devtmpfs will be
347  * auto-mounted after the kernel mounted the root filesystem.
348  */
devtmpfs_mount(const char * mntdir)349 int devtmpfs_mount(const char *mntdir)
350 {
351 	int err;
352 
353 	if (!mount_dev)
354 		return 0;
355 
356 	if (!thread)
357 		return 0;
358 
359 	err = ksys_mount("devtmpfs", (char *)mntdir, "devtmpfs", MS_SILENT,
360 			 NULL);
361 	if (err)
362 		printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
363 	else
364 		printk(KERN_INFO "devtmpfs: mounted\n");
365 	return err;
366 }
367 
368 static DECLARE_COMPLETION(setup_done);
369 
handle(const char * name,umode_t mode,kuid_t uid,kgid_t gid,struct device * dev)370 static int handle(const char *name, umode_t mode, kuid_t uid, kgid_t gid,
371 		  struct device *dev)
372 {
373 	if (mode)
374 		return handle_create(name, mode, uid, gid, dev);
375 	else
376 		return handle_remove(name, dev);
377 }
378 
devtmpfsd(void * p)379 static int devtmpfsd(void *p)
380 {
381 	char options[] = "mode=0755";
382 	int *err = p;
383 	*err = ksys_unshare(CLONE_NEWNS);
384 	if (*err)
385 		goto out;
386 	*err = ksys_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, options);
387 	if (*err)
388 		goto out;
389 	ksys_chdir("/.."); /* will traverse into overmounted root */
390 	ksys_chroot(".");
391 	complete(&setup_done);
392 	while (1) {
393 		spin_lock(&req_lock);
394 		while (requests) {
395 			struct req *req = requests;
396 			requests = NULL;
397 			spin_unlock(&req_lock);
398 			while (req) {
399 				struct req *next = req->next;
400 				req->err = handle(req->name, req->mode,
401 						  req->uid, req->gid, req->dev);
402 				complete(&req->done);
403 				req = next;
404 			}
405 			spin_lock(&req_lock);
406 		}
407 		__set_current_state(TASK_INTERRUPTIBLE);
408 		spin_unlock(&req_lock);
409 		schedule();
410 	}
411 	return 0;
412 out:
413 	complete(&setup_done);
414 	return *err;
415 }
416 
417 /*
418  * Create devtmpfs instance, driver-core devices will add their device
419  * nodes here.
420  */
devtmpfs_init(void)421 int __init devtmpfs_init(void)
422 {
423 	int err = register_filesystem(&dev_fs_type);
424 	if (err) {
425 		printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
426 		       "type %i\n", err);
427 		return err;
428 	}
429 
430 	thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
431 	if (!IS_ERR(thread)) {
432 		wait_for_completion(&setup_done);
433 	} else {
434 		err = PTR_ERR(thread);
435 		thread = NULL;
436 	}
437 
438 	if (err) {
439 		printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
440 		unregister_filesystem(&dev_fs_type);
441 		return err;
442 	}
443 
444 	printk(KERN_INFO "devtmpfs: initialized\n");
445 	return 0;
446 }
447