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 <linux/init_syscalls.h>
29 #include <uapi/linux/mount.h>
30 #include "base.h"
31
32 static struct task_struct *thread;
33
34 static int __initdata mount_dev = IS_ENABLED(CONFIG_DEVTMPFS_MOUNT);
35
36 static DEFINE_SPINLOCK(req_lock);
37
38 static struct req {
39 struct req *next;
40 struct completion done;
41 int err;
42 const char *name;
43 umode_t mode; /* 0 => delete */
44 kuid_t uid;
45 kgid_t gid;
46 struct device *dev;
47 } *requests;
48
mount_param(char * str)49 static int __init mount_param(char *str)
50 {
51 mount_dev = simple_strtoul(str, NULL, 0);
52 return 1;
53 }
54 __setup("devtmpfs.mount=", mount_param);
55
56 static struct vfsmount *mnt;
57
public_dev_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)58 static struct dentry *public_dev_mount(struct file_system_type *fs_type, int flags,
59 const char *dev_name, void *data)
60 {
61 struct super_block *s = mnt->mnt_sb;
62 atomic_inc(&s->s_active);
63 down_write(&s->s_umount);
64 return dget(s->s_root);
65 }
66
67 static struct file_system_type internal_fs_type = {
68 .name = "devtmpfs",
69 #ifdef CONFIG_TMPFS
70 .init_fs_context = shmem_init_fs_context,
71 .parameters = shmem_fs_parameters,
72 #else
73 .init_fs_context = ramfs_init_fs_context,
74 .parameters = ramfs_fs_parameters,
75 #endif
76 .kill_sb = kill_litter_super,
77 };
78
79 static struct file_system_type dev_fs_type = {
80 .name = "devtmpfs",
81 .mount = public_dev_mount,
82 };
83
84 #ifdef CONFIG_BLOCK
is_blockdev(struct device * dev)85 static inline int is_blockdev(struct device *dev)
86 {
87 return dev->class == &block_class;
88 }
89 #else
is_blockdev(struct device * dev)90 static inline int is_blockdev(struct device *dev) { return 0; }
91 #endif
92
devtmpfs_submit_req(struct req * req,const char * tmp)93 static int devtmpfs_submit_req(struct req *req, const char *tmp)
94 {
95 init_completion(&req->done);
96
97 spin_lock(&req_lock);
98 req->next = requests;
99 requests = req;
100 spin_unlock(&req_lock);
101
102 wake_up_process(thread);
103 wait_for_completion(&req->done);
104
105 kfree(tmp);
106
107 return req->err;
108 }
109
devtmpfs_create_node(struct device * dev)110 int devtmpfs_create_node(struct device *dev)
111 {
112 const char *tmp = NULL;
113 struct req req;
114
115 if (!thread)
116 return 0;
117
118 req.mode = 0;
119 req.uid = GLOBAL_ROOT_UID;
120 req.gid = GLOBAL_ROOT_GID;
121 req.name = device_get_devnode(dev, &req.mode, &req.uid, &req.gid, &tmp);
122 if (!req.name)
123 return -ENOMEM;
124
125 if (req.mode == 0)
126 req.mode = 0600;
127 if (is_blockdev(dev))
128 req.mode |= S_IFBLK;
129 else
130 req.mode |= S_IFCHR;
131
132 req.dev = dev;
133
134 return devtmpfs_submit_req(&req, tmp);
135 }
136
devtmpfs_delete_node(struct device * dev)137 int devtmpfs_delete_node(struct device *dev)
138 {
139 const char *tmp = NULL;
140 struct req req;
141
142 if (!thread)
143 return 0;
144
145 req.name = device_get_devnode(dev, NULL, NULL, NULL, &tmp);
146 if (!req.name)
147 return -ENOMEM;
148
149 req.mode = 0;
150 req.dev = dev;
151
152 return devtmpfs_submit_req(&req, tmp);
153 }
154
dev_mkdir(const char * name,umode_t mode)155 static int dev_mkdir(const char *name, umode_t mode)
156 {
157 struct dentry *dentry;
158 struct path path;
159 int err;
160
161 dentry = kern_path_create(AT_FDCWD, name, &path, LOOKUP_DIRECTORY);
162 if (IS_ERR(dentry))
163 return PTR_ERR(dentry);
164
165 err = vfs_mkdir(&init_user_ns, d_inode(path.dentry), dentry, mode);
166 if (!err)
167 /* mark as kernel-created inode */
168 d_inode(dentry)->i_private = &thread;
169 done_path_create(&path, dentry);
170 return err;
171 }
172
create_path(const char * nodepath)173 static int create_path(const char *nodepath)
174 {
175 char *path;
176 char *s;
177 int err = 0;
178
179 /* parent directories do not exist, create them */
180 path = kstrdup(nodepath, GFP_KERNEL);
181 if (!path)
182 return -ENOMEM;
183
184 s = path;
185 for (;;) {
186 s = strchr(s, '/');
187 if (!s)
188 break;
189 s[0] = '\0';
190 err = dev_mkdir(path, 0755);
191 if (err && err != -EEXIST)
192 break;
193 s[0] = '/';
194 s++;
195 }
196 kfree(path);
197 return err;
198 }
199
handle_create(const char * nodename,umode_t mode,kuid_t uid,kgid_t gid,struct device * dev)200 static int handle_create(const char *nodename, umode_t mode, kuid_t uid,
201 kgid_t gid, struct device *dev)
202 {
203 struct dentry *dentry;
204 struct path path;
205 int err;
206
207 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
208 if (dentry == ERR_PTR(-ENOENT)) {
209 create_path(nodename);
210 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
211 }
212 if (IS_ERR(dentry))
213 return PTR_ERR(dentry);
214
215 err = vfs_mknod(&init_user_ns, d_inode(path.dentry), dentry, mode,
216 dev->devt);
217 if (!err) {
218 struct iattr newattrs;
219
220 newattrs.ia_mode = mode;
221 newattrs.ia_uid = uid;
222 newattrs.ia_gid = gid;
223 newattrs.ia_valid = ATTR_MODE|ATTR_UID|ATTR_GID;
224 inode_lock(d_inode(dentry));
225 notify_change(&init_user_ns, dentry, &newattrs, NULL);
226 inode_unlock(d_inode(dentry));
227
228 /* mark as kernel-created inode */
229 d_inode(dentry)->i_private = &thread;
230 }
231 done_path_create(&path, dentry);
232 return err;
233 }
234
dev_rmdir(const char * name)235 static int dev_rmdir(const char *name)
236 {
237 struct path parent;
238 struct dentry *dentry;
239 int err;
240
241 dentry = kern_path_locked(name, &parent);
242 if (IS_ERR(dentry))
243 return PTR_ERR(dentry);
244 if (d_really_is_positive(dentry)) {
245 if (d_inode(dentry)->i_private == &thread)
246 err = vfs_rmdir(&init_user_ns, d_inode(parent.dentry),
247 dentry);
248 else
249 err = -EPERM;
250 } else {
251 err = -ENOENT;
252 }
253 dput(dentry);
254 inode_unlock(d_inode(parent.dentry));
255 path_put(&parent);
256 return err;
257 }
258
delete_path(const char * nodepath)259 static int delete_path(const char *nodepath)
260 {
261 char *path;
262 int err = 0;
263
264 path = kstrdup(nodepath, GFP_KERNEL);
265 if (!path)
266 return -ENOMEM;
267
268 for (;;) {
269 char *base;
270
271 base = strrchr(path, '/');
272 if (!base)
273 break;
274 base[0] = '\0';
275 err = dev_rmdir(path);
276 if (err)
277 break;
278 }
279
280 kfree(path);
281 return err;
282 }
283
dev_mynode(struct device * dev,struct inode * inode,struct kstat * stat)284 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
285 {
286 /* did we create it */
287 if (inode->i_private != &thread)
288 return 0;
289
290 /* does the dev_t match */
291 if (is_blockdev(dev)) {
292 if (!S_ISBLK(stat->mode))
293 return 0;
294 } else {
295 if (!S_ISCHR(stat->mode))
296 return 0;
297 }
298 if (stat->rdev != dev->devt)
299 return 0;
300
301 /* ours */
302 return 1;
303 }
304
handle_remove(const char * nodename,struct device * dev)305 static int handle_remove(const char *nodename, struct device *dev)
306 {
307 struct path parent;
308 struct dentry *dentry;
309 int deleted = 0;
310 int err;
311
312 dentry = kern_path_locked(nodename, &parent);
313 if (IS_ERR(dentry))
314 return PTR_ERR(dentry);
315
316 if (d_really_is_positive(dentry)) {
317 struct kstat stat;
318 struct path p = {.mnt = parent.mnt, .dentry = dentry};
319 err = vfs_getattr(&p, &stat, STATX_TYPE | STATX_MODE,
320 AT_STATX_SYNC_AS_STAT);
321 if (!err && dev_mynode(dev, d_inode(dentry), &stat)) {
322 struct iattr newattrs;
323 /*
324 * before unlinking this node, reset permissions
325 * of possible references like hardlinks
326 */
327 newattrs.ia_uid = GLOBAL_ROOT_UID;
328 newattrs.ia_gid = GLOBAL_ROOT_GID;
329 newattrs.ia_mode = stat.mode & ~0777;
330 newattrs.ia_valid =
331 ATTR_UID|ATTR_GID|ATTR_MODE;
332 inode_lock(d_inode(dentry));
333 notify_change(&init_user_ns, dentry, &newattrs, NULL);
334 inode_unlock(d_inode(dentry));
335 err = vfs_unlink(&init_user_ns, d_inode(parent.dentry),
336 dentry, NULL);
337 if (!err || err == -ENOENT)
338 deleted = 1;
339 }
340 } else {
341 err = -ENOENT;
342 }
343 dput(dentry);
344 inode_unlock(d_inode(parent.dentry));
345
346 path_put(&parent);
347 if (deleted && strchr(nodename, '/'))
348 delete_path(nodename);
349 return err;
350 }
351
352 /*
353 * If configured, or requested by the commandline, devtmpfs will be
354 * auto-mounted after the kernel mounted the root filesystem.
355 */
devtmpfs_mount(void)356 int __init devtmpfs_mount(void)
357 {
358 int err;
359
360 if (!mount_dev)
361 return 0;
362
363 if (!thread)
364 return 0;
365
366 err = init_mount("devtmpfs", "dev", "devtmpfs", MS_SILENT, NULL);
367 if (err)
368 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
369 else
370 printk(KERN_INFO "devtmpfs: mounted\n");
371 return err;
372 }
373
374 static __initdata DECLARE_COMPLETION(setup_done);
375
handle(const char * name,umode_t mode,kuid_t uid,kgid_t gid,struct device * dev)376 static int handle(const char *name, umode_t mode, kuid_t uid, kgid_t gid,
377 struct device *dev)
378 {
379 if (mode)
380 return handle_create(name, mode, uid, gid, dev);
381 else
382 return handle_remove(name, dev);
383 }
384
devtmpfs_work_loop(void)385 static void __noreturn devtmpfs_work_loop(void)
386 {
387 while (1) {
388 spin_lock(&req_lock);
389 while (requests) {
390 struct req *req = requests;
391 requests = NULL;
392 spin_unlock(&req_lock);
393 while (req) {
394 struct req *next = req->next;
395 req->err = handle(req->name, req->mode,
396 req->uid, req->gid, req->dev);
397 complete(&req->done);
398 req = next;
399 }
400 spin_lock(&req_lock);
401 }
402 __set_current_state(TASK_INTERRUPTIBLE);
403 spin_unlock(&req_lock);
404 schedule();
405 }
406 }
407
devtmpfs_setup(void * p)408 static noinline int __init devtmpfs_setup(void *p)
409 {
410 int err;
411
412 err = ksys_unshare(CLONE_NEWNS);
413 if (err)
414 goto out;
415 err = init_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, NULL);
416 if (err)
417 goto out;
418 init_chdir("/.."); /* will traverse into overmounted root */
419 init_chroot(".");
420 out:
421 *(int *)p = err;
422 return err;
423 }
424
425 /*
426 * The __ref is because devtmpfs_setup needs to be __init for the routines it
427 * calls. That call is done while devtmpfs_init, which is marked __init,
428 * synchronously waits for it to complete.
429 */
devtmpfsd(void * p)430 static int __ref devtmpfsd(void *p)
431 {
432 int err = devtmpfs_setup(p);
433
434 complete(&setup_done);
435 if (err)
436 return err;
437 devtmpfs_work_loop();
438 return 0;
439 }
440
441 /*
442 * Create devtmpfs instance, driver-core devices will add their device
443 * nodes here.
444 */
devtmpfs_init(void)445 int __init devtmpfs_init(void)
446 {
447 char opts[] = "mode=0755";
448 int err;
449
450 mnt = vfs_kern_mount(&internal_fs_type, 0, "devtmpfs", opts);
451 if (IS_ERR(mnt)) {
452 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %ld\n",
453 PTR_ERR(mnt));
454 return PTR_ERR(mnt);
455 }
456 err = register_filesystem(&dev_fs_type);
457 if (err) {
458 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
459 "type %i\n", err);
460 return err;
461 }
462
463 thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
464 if (!IS_ERR(thread)) {
465 wait_for_completion(&setup_done);
466 } else {
467 err = PTR_ERR(thread);
468 thread = NULL;
469 }
470
471 if (err) {
472 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
473 unregister_filesystem(&dev_fs_type);
474 return err;
475 }
476
477 printk(KERN_INFO "devtmpfs: initialized\n");
478 return 0;
479 }
480