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 int err;
63
64 atomic_inc(&s->s_active);
65 down_write(&s->s_umount);
66 err = reconfigure_single(s, flags, data);
67 if (err < 0) {
68 deactivate_locked_super(s);
69 return ERR_PTR(err);
70 }
71 return dget(s->s_root);
72 }
73
74 static struct file_system_type internal_fs_type = {
75 .name = "devtmpfs",
76 #ifdef CONFIG_TMPFS
77 .init_fs_context = shmem_init_fs_context,
78 .parameters = shmem_fs_parameters,
79 #else
80 .init_fs_context = ramfs_init_fs_context,
81 .parameters = ramfs_fs_parameters,
82 #endif
83 .kill_sb = kill_litter_super,
84 };
85
86 static struct file_system_type dev_fs_type = {
87 .name = "devtmpfs",
88 .mount = public_dev_mount,
89 };
90
91 #ifdef CONFIG_BLOCK
is_blockdev(struct device * dev)92 static inline int is_blockdev(struct device *dev)
93 {
94 return dev->class == &block_class;
95 }
96 #else
is_blockdev(struct device * dev)97 static inline int is_blockdev(struct device *dev) { return 0; }
98 #endif
99
devtmpfs_submit_req(struct req * req,const char * tmp)100 static int devtmpfs_submit_req(struct req *req, const char *tmp)
101 {
102 init_completion(&req->done);
103
104 spin_lock(&req_lock);
105 req->next = requests;
106 requests = req;
107 spin_unlock(&req_lock);
108
109 wake_up_process(thread);
110 wait_for_completion(&req->done);
111
112 kfree(tmp);
113
114 return req->err;
115 }
116
devtmpfs_create_node(struct device * dev)117 int devtmpfs_create_node(struct device *dev)
118 {
119 const char *tmp = NULL;
120 struct req req;
121
122 if (!thread)
123 return 0;
124
125 req.mode = 0;
126 req.uid = GLOBAL_ROOT_UID;
127 req.gid = GLOBAL_ROOT_GID;
128 req.name = device_get_devnode(dev, &req.mode, &req.uid, &req.gid, &tmp);
129 if (!req.name)
130 return -ENOMEM;
131
132 if (req.mode == 0)
133 req.mode = 0600;
134 if (is_blockdev(dev))
135 req.mode |= S_IFBLK;
136 else
137 req.mode |= S_IFCHR;
138
139 req.dev = dev;
140
141 return devtmpfs_submit_req(&req, tmp);
142 }
143
devtmpfs_delete_node(struct device * dev)144 int devtmpfs_delete_node(struct device *dev)
145 {
146 const char *tmp = NULL;
147 struct req req;
148
149 if (!thread)
150 return 0;
151
152 req.name = device_get_devnode(dev, NULL, NULL, NULL, &tmp);
153 if (!req.name)
154 return -ENOMEM;
155
156 req.mode = 0;
157 req.dev = dev;
158
159 return devtmpfs_submit_req(&req, tmp);
160 }
161
dev_mkdir(const char * name,umode_t mode)162 static int dev_mkdir(const char *name, umode_t mode)
163 {
164 struct dentry *dentry;
165 struct path path;
166 int err;
167
168 dentry = kern_path_create(AT_FDCWD, name, &path, LOOKUP_DIRECTORY);
169 if (IS_ERR(dentry))
170 return PTR_ERR(dentry);
171
172 err = vfs_mkdir(d_inode(path.dentry), dentry, mode);
173 if (!err)
174 /* mark as kernel-created inode */
175 d_inode(dentry)->i_private = &thread;
176 done_path_create(&path, dentry);
177 return err;
178 }
179
create_path(const char * nodepath)180 static int create_path(const char *nodepath)
181 {
182 char *path;
183 char *s;
184 int err = 0;
185
186 /* parent directories do not exist, create them */
187 path = kstrdup(nodepath, GFP_KERNEL);
188 if (!path)
189 return -ENOMEM;
190
191 s = path;
192 for (;;) {
193 s = strchr(s, '/');
194 if (!s)
195 break;
196 s[0] = '\0';
197 err = dev_mkdir(path, 0755);
198 if (err && err != -EEXIST)
199 break;
200 s[0] = '/';
201 s++;
202 }
203 kfree(path);
204 return err;
205 }
206
handle_create(const char * nodename,umode_t mode,kuid_t uid,kgid_t gid,struct device * dev)207 static int handle_create(const char *nodename, umode_t mode, kuid_t uid,
208 kgid_t gid, struct device *dev)
209 {
210 struct dentry *dentry;
211 struct path path;
212 int err;
213
214 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
215 if (dentry == ERR_PTR(-ENOENT)) {
216 create_path(nodename);
217 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
218 }
219 if (IS_ERR(dentry))
220 return PTR_ERR(dentry);
221
222 err = vfs_mknod(d_inode(path.dentry), dentry, mode, dev->devt);
223 if (!err) {
224 struct iattr newattrs;
225
226 newattrs.ia_mode = mode;
227 newattrs.ia_uid = uid;
228 newattrs.ia_gid = gid;
229 newattrs.ia_valid = ATTR_MODE|ATTR_UID|ATTR_GID;
230 inode_lock(d_inode(dentry));
231 notify_change(dentry, &newattrs, NULL);
232 inode_unlock(d_inode(dentry));
233
234 /* mark as kernel-created inode */
235 d_inode(dentry)->i_private = &thread;
236 }
237 done_path_create(&path, dentry);
238 return err;
239 }
240
dev_rmdir(const char * name)241 static int dev_rmdir(const char *name)
242 {
243 struct path parent;
244 struct dentry *dentry;
245 int err;
246
247 dentry = kern_path_locked(name, &parent);
248 if (IS_ERR(dentry))
249 return PTR_ERR(dentry);
250 if (d_really_is_positive(dentry)) {
251 if (d_inode(dentry)->i_private == &thread)
252 err = vfs_rmdir(d_inode(parent.dentry), dentry);
253 else
254 err = -EPERM;
255 } else {
256 err = -ENOENT;
257 }
258 dput(dentry);
259 inode_unlock(d_inode(parent.dentry));
260 path_put(&parent);
261 return err;
262 }
263
delete_path(const char * nodepath)264 static int delete_path(const char *nodepath)
265 {
266 char *path;
267 int err = 0;
268
269 path = kstrdup(nodepath, GFP_KERNEL);
270 if (!path)
271 return -ENOMEM;
272
273 for (;;) {
274 char *base;
275
276 base = strrchr(path, '/');
277 if (!base)
278 break;
279 base[0] = '\0';
280 err = dev_rmdir(path);
281 if (err)
282 break;
283 }
284
285 kfree(path);
286 return err;
287 }
288
dev_mynode(struct device * dev,struct inode * inode,struct kstat * stat)289 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
290 {
291 /* did we create it */
292 if (inode->i_private != &thread)
293 return 0;
294
295 /* does the dev_t match */
296 if (is_blockdev(dev)) {
297 if (!S_ISBLK(stat->mode))
298 return 0;
299 } else {
300 if (!S_ISCHR(stat->mode))
301 return 0;
302 }
303 if (stat->rdev != dev->devt)
304 return 0;
305
306 /* ours */
307 return 1;
308 }
309
handle_remove(const char * nodename,struct device * dev)310 static int handle_remove(const char *nodename, struct device *dev)
311 {
312 struct path parent;
313 struct dentry *dentry;
314 int deleted = 0;
315 int err;
316
317 dentry = kern_path_locked(nodename, &parent);
318 if (IS_ERR(dentry))
319 return PTR_ERR(dentry);
320
321 if (d_really_is_positive(dentry)) {
322 struct kstat stat;
323 struct path p = {.mnt = parent.mnt, .dentry = dentry};
324 err = vfs_getattr(&p, &stat, STATX_TYPE | STATX_MODE,
325 AT_STATX_SYNC_AS_STAT);
326 if (!err && dev_mynode(dev, d_inode(dentry), &stat)) {
327 struct iattr newattrs;
328 /*
329 * before unlinking this node, reset permissions
330 * of possible references like hardlinks
331 */
332 newattrs.ia_uid = GLOBAL_ROOT_UID;
333 newattrs.ia_gid = GLOBAL_ROOT_GID;
334 newattrs.ia_mode = stat.mode & ~0777;
335 newattrs.ia_valid =
336 ATTR_UID|ATTR_GID|ATTR_MODE;
337 inode_lock(d_inode(dentry));
338 notify_change(dentry, &newattrs, NULL);
339 inode_unlock(d_inode(dentry));
340 err = vfs_unlink(d_inode(parent.dentry), dentry, NULL);
341 if (!err || err == -ENOENT)
342 deleted = 1;
343 }
344 } else {
345 err = -ENOENT;
346 }
347 dput(dentry);
348 inode_unlock(d_inode(parent.dentry));
349
350 path_put(&parent);
351 if (deleted && strchr(nodename, '/'))
352 delete_path(nodename);
353 return err;
354 }
355
356 /*
357 * If configured, or requested by the commandline, devtmpfs will be
358 * auto-mounted after the kernel mounted the root filesystem.
359 */
devtmpfs_mount(void)360 int __init devtmpfs_mount(void)
361 {
362 int err;
363
364 if (!mount_dev)
365 return 0;
366
367 if (!thread)
368 return 0;
369
370 err = init_mount("devtmpfs", "dev", "devtmpfs", MS_SILENT, NULL);
371 if (err)
372 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
373 else
374 printk(KERN_INFO "devtmpfs: mounted\n");
375 return err;
376 }
377
378 static DECLARE_COMPLETION(setup_done);
379
handle(const char * name,umode_t mode,kuid_t uid,kgid_t gid,struct device * dev)380 static int handle(const char *name, umode_t mode, kuid_t uid, kgid_t gid,
381 struct device *dev)
382 {
383 if (mode)
384 return handle_create(name, mode, uid, gid, dev);
385 else
386 return handle_remove(name, dev);
387 }
388
devtmpfs_work_loop(void)389 static void __noreturn devtmpfs_work_loop(void)
390 {
391 while (1) {
392 spin_lock(&req_lock);
393 while (requests) {
394 struct req *req = requests;
395 requests = NULL;
396 spin_unlock(&req_lock);
397 while (req) {
398 struct req *next = req->next;
399 req->err = handle(req->name, req->mode,
400 req->uid, req->gid, req->dev);
401 complete(&req->done);
402 req = next;
403 }
404 spin_lock(&req_lock);
405 }
406 __set_current_state(TASK_INTERRUPTIBLE);
407 spin_unlock(&req_lock);
408 schedule();
409 }
410 }
411
devtmpfs_setup(void * p)412 static int __init devtmpfs_setup(void *p)
413 {
414 int err;
415
416 err = ksys_unshare(CLONE_NEWNS);
417 if (err)
418 goto out;
419 err = init_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, NULL);
420 if (err)
421 goto out;
422 init_chdir("/.."); /* will traverse into overmounted root */
423 init_chroot(".");
424 out:
425 *(int *)p = err;
426 return err;
427 }
428
429 /*
430 * The __ref is because devtmpfs_setup needs to be __init for the routines it
431 * calls. That call is done while devtmpfs_init, which is marked __init,
432 * synchronously waits for it to complete.
433 */
devtmpfsd(void * p)434 static int __ref devtmpfsd(void *p)
435 {
436 int err = devtmpfs_setup(p);
437
438 complete(&setup_done);
439 if (err)
440 return err;
441 devtmpfs_work_loop();
442 return 0;
443 }
444
445 /*
446 * Create devtmpfs instance, driver-core devices will add their device
447 * nodes here.
448 */
devtmpfs_init(void)449 int __init devtmpfs_init(void)
450 {
451 char opts[] = "mode=0755";
452 int err;
453
454 mnt = vfs_kern_mount(&internal_fs_type, 0, "devtmpfs", opts);
455 if (IS_ERR(mnt)) {
456 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %ld\n",
457 PTR_ERR(mnt));
458 return PTR_ERR(mnt);
459 }
460 err = register_filesystem(&dev_fs_type);
461 if (err) {
462 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
463 "type %i\n", err);
464 return err;
465 }
466
467 thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
468 if (!IS_ERR(thread)) {
469 wait_for_completion(&setup_done);
470 } else {
471 err = PTR_ERR(thread);
472 thread = NULL;
473 }
474
475 if (err) {
476 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
477 unregister_filesystem(&dev_fs_type);
478 return err;
479 }
480
481 printk(KERN_INFO "devtmpfs: initialized\n");
482 return 0;
483 }
484