1/*
2 * devtmpfs - kernel-maintained tmpfs-based /dev
3 *
4 * Copyright (C) 2009, Kay Sievers <kay.sievers@vrfy.org>
5 *
6 * During bootup, before any driver core device is registered,
7 * devtmpfs, a tmpfs-based filesystem is created. Every driver-core
8 * device which requests a device node, will add a node in this
9 * filesystem.
10 * By default, all devices are named after the the name of the
11 * device, owned by root and have a default mode of 0600. Subsystems
12 * can overwrite the default setting if needed.
13 */
14
15#include <linux/kernel.h>
16#include <linux/syscalls.h>
17#include <linux/mount.h>
18#include <linux/device.h>
19#include <linux/genhd.h>
20#include <linux/namei.h>
21#include <linux/fs.h>
22#include <linux/shmem_fs.h>
23#include <linux/ramfs.h>
24#include <linux/sched.h>
25#include <linux/slab.h>
26#include <linux/kthread.h>
27
28static struct task_struct *thread;
29
30#if defined CONFIG_DEVTMPFS_MOUNT
31static int mount_dev = 1;
32#else
33static int mount_dev;
34#endif
35
36static DEFINE_SPINLOCK(req_lock);
37
38static struct req {
39	struct req *next;
40	struct completion done;
41	int err;
42	const char *name;
43	umode_t mode;	/* 0 => delete */
44	struct device *dev;
45} *requests;
46
47static int __init mount_param(char *str)
48{
49	mount_dev = simple_strtoul(str, NULL, 0);
50	return 1;
51}
52__setup("devtmpfs.mount=", mount_param);
53
54static struct dentry *dev_mount(struct file_system_type *fs_type, int flags,
55		      const char *dev_name, void *data)
56{
57#ifdef CONFIG_TMPFS
58	return mount_single(fs_type, flags, data, shmem_fill_super);
59#else
60	return mount_single(fs_type, flags, data, ramfs_fill_super);
61#endif
62}
63
64static struct file_system_type dev_fs_type = {
65	.name = "devtmpfs",
66	.mount = dev_mount,
67	.kill_sb = kill_litter_super,
68};
69
70#ifdef CONFIG_BLOCK
71static inline int is_blockdev(struct device *dev)
72{
73	return dev->class == &block_class;
74}
75#else
76static inline int is_blockdev(struct device *dev) { return 0; }
77#endif
78
79int devtmpfs_create_node(struct device *dev)
80{
81	const char *tmp = NULL;
82	struct req req;
83
84	if (!thread)
85		return 0;
86
87	req.mode = 0;
88	req.name = device_get_devnode(dev, &req.mode, &tmp);
89	if (!req.name)
90		return -ENOMEM;
91
92	if (req.mode == 0)
93		req.mode = 0600;
94	if (is_blockdev(dev))
95		req.mode |= S_IFBLK;
96	else
97		req.mode |= S_IFCHR;
98
99	req.dev = dev;
100
101	init_completion(&req.done);
102
103	spin_lock(&req_lock);
104	req.next = requests;
105	requests = &req;
106	spin_unlock(&req_lock);
107
108	wake_up_process(thread);
109	wait_for_completion(&req.done);
110
111	kfree(tmp);
112
113	return req.err;
114}
115
116int devtmpfs_delete_node(struct device *dev)
117{
118	const char *tmp = NULL;
119	struct req req;
120
121	if (!thread)
122		return 0;
123
124	req.name = device_get_devnode(dev, NULL, &tmp);
125	if (!req.name)
126		return -ENOMEM;
127
128	req.mode = 0;
129	req.dev = dev;
130
131	init_completion(&req.done);
132
133	spin_lock(&req_lock);
134	req.next = requests;
135	requests = &req;
136	spin_unlock(&req_lock);
137
138	wake_up_process(thread);
139	wait_for_completion(&req.done);
140
141	kfree(tmp);
142	return req.err;
143}
144
145static int dev_mkdir(const char *name, umode_t mode)
146{
147	struct dentry *dentry;
148	struct path path;
149	int err;
150
151	dentry = kern_path_create(AT_FDCWD, name, &path, 1);
152	if (IS_ERR(dentry))
153		return PTR_ERR(dentry);
154
155	err = vfs_mkdir(path.dentry->d_inode, dentry, mode);
156	if (!err)
157		/* mark as kernel-created inode */
158		dentry->d_inode->i_private = &thread;
159	dput(dentry);
160	mutex_unlock(&path.dentry->d_inode->i_mutex);
161	path_put(&path);
162	return err;
163}
164
165static int create_path(const char *nodepath)
166{
167	char *path;
168	char *s;
169	int err = 0;
170
171	/* parent directories do not exist, create them */
172	path = kstrdup(nodepath, GFP_KERNEL);
173	if (!path)
174		return -ENOMEM;
175
176	s = path;
177	for (;;) {
178		s = strchr(s, '/');
179		if (!s)
180			break;
181		s[0] = '\0';
182		err = dev_mkdir(path, 0755);
183		if (err && err != -EEXIST)
184			break;
185		s[0] = '/';
186		s++;
187	}
188	kfree(path);
189	return err;
190}
191
192static int handle_create(const char *nodename, umode_t mode, struct device *dev)
193{
194	struct dentry *dentry;
195	struct path path;
196	int err;
197
198	dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
199	if (dentry == ERR_PTR(-ENOENT)) {
200		create_path(nodename);
201		dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
202	}
203	if (IS_ERR(dentry))
204		return PTR_ERR(dentry);
205
206	err = vfs_mknod(path.dentry->d_inode,
207			dentry, mode, dev->devt);
208	if (!err) {
209		struct iattr newattrs;
210
211		/* fixup possibly umasked mode */
212		newattrs.ia_mode = mode;
213		newattrs.ia_valid = ATTR_MODE;
214		mutex_lock(&dentry->d_inode->i_mutex);
215		notify_change(dentry, &newattrs);
216		mutex_unlock(&dentry->d_inode->i_mutex);
217
218		/* mark as kernel-created inode */
219		dentry->d_inode->i_private = &thread;
220	}
221	dput(dentry);
222
223	mutex_unlock(&path.dentry->d_inode->i_mutex);
224	path_put(&path);
225	return err;
226}
227
228static int dev_rmdir(const char *name)
229{
230	struct nameidata nd;
231	struct dentry *dentry;
232	int err;
233
234	err = kern_path_parent(name, &nd);
235	if (err)
236		return err;
237
238	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
239	dentry = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len);
240	if (!IS_ERR(dentry)) {
241		if (dentry->d_inode) {
242			if (dentry->d_inode->i_private == &thread)
243				err = vfs_rmdir(nd.path.dentry->d_inode,
244						dentry);
245			else
246				err = -EPERM;
247		} else {
248			err = -ENOENT;
249		}
250		dput(dentry);
251	} else {
252		err = PTR_ERR(dentry);
253	}
254
255	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
256	path_put(&nd.path);
257	return err;
258}
259
260static int delete_path(const char *nodepath)
261{
262	const char *path;
263	int err = 0;
264
265	path = kstrdup(nodepath, GFP_KERNEL);
266	if (!path)
267		return -ENOMEM;
268
269	for (;;) {
270		char *base;
271
272		base = strrchr(path, '/');
273		if (!base)
274			break;
275		base[0] = '\0';
276		err = dev_rmdir(path);
277		if (err)
278			break;
279	}
280
281	kfree(path);
282	return err;
283}
284
285static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
286{
287	/* did we create it */
288	if (inode->i_private != &thread)
289		return 0;
290
291	/* does the dev_t match */
292	if (is_blockdev(dev)) {
293		if (!S_ISBLK(stat->mode))
294			return 0;
295	} else {
296		if (!S_ISCHR(stat->mode))
297			return 0;
298	}
299	if (stat->rdev != dev->devt)
300		return 0;
301
302	/* ours */
303	return 1;
304}
305
306static int handle_remove(const char *nodename, struct device *dev)
307{
308	struct nameidata nd;
309	struct dentry *dentry;
310	struct kstat stat;
311	int deleted = 1;
312	int err;
313
314	err = kern_path_parent(nodename, &nd);
315	if (err)
316		return err;
317
318	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
319	dentry = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len);
320	if (!IS_ERR(dentry)) {
321		if (dentry->d_inode) {
322			err = vfs_getattr(nd.path.mnt, dentry, &stat);
323			if (!err && dev_mynode(dev, dentry->d_inode, &stat)) {
324				struct iattr newattrs;
325				/*
326				 * before unlinking this node, reset permissions
327				 * of possible references like hardlinks
328				 */
329				newattrs.ia_uid = 0;
330				newattrs.ia_gid = 0;
331				newattrs.ia_mode = stat.mode & ~0777;
332				newattrs.ia_valid =
333					ATTR_UID|ATTR_GID|ATTR_MODE;
334				mutex_lock(&dentry->d_inode->i_mutex);
335				notify_change(dentry, &newattrs);
336				mutex_unlock(&dentry->d_inode->i_mutex);
337				err = vfs_unlink(nd.path.dentry->d_inode,
338						 dentry);
339				if (!err || err == -ENOENT)
340					deleted = 1;
341			}
342		} else {
343			err = -ENOENT;
344		}
345		dput(dentry);
346	} else {
347		err = PTR_ERR(dentry);
348	}
349	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
350
351	path_put(&nd.path);
352	if (deleted && strchr(nodename, '/'))
353		delete_path(nodename);
354	return err;
355}
356
357/*
358 * If configured, or requested by the commandline, devtmpfs will be
359 * auto-mounted after the kernel mounted the root filesystem.
360 */
361int devtmpfs_mount(const char *mntdir)
362{
363	int err;
364
365	if (!mount_dev)
366		return 0;
367
368	if (!thread)
369		return 0;
370
371	err = sys_mount("devtmpfs", (char *)mntdir, "devtmpfs", MS_SILENT, NULL);
372	if (err)
373		printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
374	else
375		printk(KERN_INFO "devtmpfs: mounted\n");
376	return err;
377}
378
379static DECLARE_COMPLETION(setup_done);
380
381static int handle(const char *name, umode_t mode, struct device *dev)
382{
383	if (mode)
384		return handle_create(name, mode, dev);
385	else
386		return handle_remove(name, dev);
387}
388
389static int devtmpfsd(void *p)
390{
391	char options[] = "mode=0755";
392	int *err = p;
393	*err = sys_unshare(CLONE_NEWNS);
394	if (*err)
395		goto out;
396	*err = sys_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, options);
397	if (*err)
398		goto out;
399	sys_chdir("/.."); /* will traverse into overmounted root */
400	sys_chroot(".");
401	complete(&setup_done);
402	while (1) {
403		spin_lock(&req_lock);
404		while (requests) {
405			struct req *req = requests;
406			requests = NULL;
407			spin_unlock(&req_lock);
408			while (req) {
409				struct req *next = req->next;
410				req->err = handle(req->name, req->mode, req->dev);
411				complete(&req->done);
412				req = next;
413			}
414			spin_lock(&req_lock);
415		}
416		__set_current_state(TASK_INTERRUPTIBLE);
417		spin_unlock(&req_lock);
418		schedule();
419	}
420	return 0;
421out:
422	complete(&setup_done);
423	return *err;
424}
425
426/*
427 * Create devtmpfs instance, driver-core devices will add their device
428 * nodes here.
429 */
430int __init devtmpfs_init(void)
431{
432	int err = register_filesystem(&dev_fs_type);
433	if (err) {
434		printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
435		       "type %i\n", err);
436		return err;
437	}
438
439	thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
440	if (!IS_ERR(thread)) {
441		wait_for_completion(&setup_done);
442	} else {
443		err = PTR_ERR(thread);
444		thread = NULL;
445	}
446
447	if (err) {
448		printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
449		unregister_filesystem(&dev_fs_type);
450		return err;
451	}
452
453	printk(KERN_INFO "devtmpfs: initialized\n");
454	return 0;
455}
456