blob: 5e9b00711357b0a93e2e77f5a08afac44e2c1005 [file] [log] [blame]
b.liue9582032025-04-17 19:18:16 +08001// 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/fs_context.h>
29#include <uapi/linux/mount.h>
30#include "base.h"
31
32static struct task_struct *thread;
33
34#if defined CONFIG_DEVTMPFS_MOUNT
35static int mount_dev = 1;
36#else
37static int mount_dev;
38#endif
39
40static DEFINE_SPINLOCK(req_lock);
41
42static struct req {
43 struct req *next;
44 struct completion done;
45 int err;
46 const char *name;
47 umode_t mode; /* 0 => delete */
48 kuid_t uid;
49 kgid_t gid;
50 struct device *dev;
51} *requests;
52
53static int __init mount_param(char *str)
54{
55 mount_dev = simple_strtoul(str, NULL, 0);
56 return 1;
57}
58__setup("devtmpfs.mount=", mount_param);
59
60static struct vfsmount *mnt;
61
62static struct dentry *public_dev_mount(struct file_system_type *fs_type, int flags,
63 const char *dev_name, void *data)
64{
65 struct super_block *s = mnt->mnt_sb;
66 int err;
67
68 atomic_inc(&s->s_active);
69 down_write(&s->s_umount);
70 err = reconfigure_single(s, flags, data);
71 if (err < 0) {
72 deactivate_locked_super(s);
73 return ERR_PTR(err);
74 }
75 return dget(s->s_root);
76}
77
78static struct file_system_type internal_fs_type = {
79 .name = "devtmpfs",
80#ifdef CONFIG_TMPFS
81 .init_fs_context = shmem_init_fs_context,
82 .parameters = &shmem_fs_parameters,
83#else
84 .init_fs_context = ramfs_init_fs_context,
85 .parameters = &ramfs_fs_parameters,
86#endif
87 .kill_sb = kill_litter_super,
88};
89
90static struct file_system_type dev_fs_type = {
91 .name = "devtmpfs",
92 .mount = public_dev_mount,
93};
94
95#ifdef CONFIG_BLOCK
96static inline int is_blockdev(struct device *dev)
97{
98 return dev->class == &block_class;
99}
100#else
101static inline int is_blockdev(struct device *dev) { return 0; }
102#endif
103
104int devtmpfs_create_node(struct device *dev)
105{
106 const char *tmp = NULL;
107 struct req req;
108
109 if (!thread)
110 return 0;
111
112 req.mode = 0;
113 req.uid = GLOBAL_ROOT_UID;
114 req.gid = GLOBAL_ROOT_GID;
115 req.name = device_get_devnode(dev, &req.mode, &req.uid, &req.gid, &tmp);
116 if (!req.name)
117 return -ENOMEM;
118
119 if (req.mode == 0)
120 req.mode = 0600;
121 if (is_blockdev(dev))
122 req.mode |= S_IFBLK;
123 else
124 req.mode |= S_IFCHR;
125
126 req.dev = dev;
127
128 init_completion(&req.done);
129
130 spin_lock(&req_lock);
131 req.next = requests;
132 requests = &req;
133 spin_unlock(&req_lock);
134
135 wake_up_process(thread);
136 wait_for_completion(&req.done);
137
138 kfree(tmp);
139
140 return req.err;
141}
142
143int devtmpfs_delete_node(struct device *dev)
144{
145 const char *tmp = NULL;
146 struct req req;
147
148 if (!thread)
149 return 0;
150
151 req.name = device_get_devnode(dev, NULL, NULL, NULL, &tmp);
152 if (!req.name)
153 return -ENOMEM;
154
155 req.mode = 0;
156 req.dev = dev;
157
158 init_completion(&req.done);
159
160 spin_lock(&req_lock);
161 req.next = requests;
162 requests = &req;
163 spin_unlock(&req_lock);
164
165 wake_up_process(thread);
166 wait_for_completion(&req.done);
167
168 kfree(tmp);
169 return req.err;
170}
171
172static int dev_mkdir(const char *name, umode_t mode)
173{
174 struct dentry *dentry;
175 struct path path;
176 int err;
177
178 dentry = kern_path_create(AT_FDCWD, name, &path, LOOKUP_DIRECTORY);
179 if (IS_ERR(dentry))
180 return PTR_ERR(dentry);
181
182 err = vfs_mkdir(d_inode(path.dentry), dentry, mode);
183 if (!err)
184 /* mark as kernel-created inode */
185 d_inode(dentry)->i_private = &thread;
186 done_path_create(&path, dentry);
187 return err;
188}
189
190static int create_path(const char *nodepath)
191{
192 char *path;
193 char *s;
194 int err = 0;
195
196 /* parent directories do not exist, create them */
197 path = kstrdup(nodepath, GFP_KERNEL);
198 if (!path)
199 return -ENOMEM;
200
201 s = path;
202 for (;;) {
203 s = strchr(s, '/');
204 if (!s)
205 break;
206 s[0] = '\0';
207 err = dev_mkdir(path, 0755);
208 if (err && err != -EEXIST)
209 break;
210 s[0] = '/';
211 s++;
212 }
213 kfree(path);
214 return err;
215}
216
217static int handle_create(const char *nodename, umode_t mode, kuid_t uid,
218 kgid_t gid, struct device *dev)
219{
220 struct dentry *dentry;
221 struct path path;
222 int err;
223
224 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
225 if (dentry == ERR_PTR(-ENOENT)) {
226 create_path(nodename);
227 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
228 }
229 if (IS_ERR(dentry))
230 return PTR_ERR(dentry);
231
232 err = vfs_mknod(d_inode(path.dentry), dentry, mode, dev->devt);
233 if (!err) {
234 struct iattr newattrs;
235
236 newattrs.ia_mode = mode;
237 newattrs.ia_uid = uid;
238 newattrs.ia_gid = gid;
239 newattrs.ia_valid = ATTR_MODE|ATTR_UID|ATTR_GID;
240 inode_lock(d_inode(dentry));
241 notify_change(dentry, &newattrs, NULL);
242 inode_unlock(d_inode(dentry));
243
244 /* mark as kernel-created inode */
245 d_inode(dentry)->i_private = &thread;
246 }
247 done_path_create(&path, dentry);
248 return err;
249}
250
251static int dev_rmdir(const char *name)
252{
253 struct path parent;
254 struct dentry *dentry;
255 int err;
256
257 dentry = kern_path_locked(name, &parent);
258 if (IS_ERR(dentry))
259 return PTR_ERR(dentry);
260 if (d_really_is_positive(dentry)) {
261 if (d_inode(dentry)->i_private == &thread)
262 err = vfs_rmdir(d_inode(parent.dentry), dentry);
263 else
264 err = -EPERM;
265 } else {
266 err = -ENOENT;
267 }
268 dput(dentry);
269 inode_unlock(d_inode(parent.dentry));
270 path_put(&parent);
271 return err;
272}
273
274static int delete_path(const char *nodepath)
275{
276 char *path;
277 int err = 0;
278
279 path = kstrdup(nodepath, GFP_KERNEL);
280 if (!path)
281 return -ENOMEM;
282
283 for (;;) {
284 char *base;
285
286 base = strrchr(path, '/');
287 if (!base)
288 break;
289 base[0] = '\0';
290 err = dev_rmdir(path);
291 if (err)
292 break;
293 }
294
295 kfree(path);
296 return err;
297}
298
299static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
300{
301 /* did we create it */
302 if (inode->i_private != &thread)
303 return 0;
304
305 /* does the dev_t match */
306 if (is_blockdev(dev)) {
307 if (!S_ISBLK(stat->mode))
308 return 0;
309 } else {
310 if (!S_ISCHR(stat->mode))
311 return 0;
312 }
313 if (stat->rdev != dev->devt)
314 return 0;
315
316 /* ours */
317 return 1;
318}
319
320static int handle_remove(const char *nodename, struct device *dev)
321{
322 struct path parent;
323 struct dentry *dentry;
324 int deleted = 0;
325 int err;
326
327 dentry = kern_path_locked(nodename, &parent);
328 if (IS_ERR(dentry))
329 return PTR_ERR(dentry);
330
331 if (d_really_is_positive(dentry)) {
332 struct kstat stat;
333 struct path p = {.mnt = parent.mnt, .dentry = dentry};
334 err = vfs_getattr(&p, &stat, STATX_TYPE | STATX_MODE,
335 AT_STATX_SYNC_AS_STAT);
336 if (!err && dev_mynode(dev, d_inode(dentry), &stat)) {
337 struct iattr newattrs;
338 /*
339 * before unlinking this node, reset permissions
340 * of possible references like hardlinks
341 */
342 newattrs.ia_uid = GLOBAL_ROOT_UID;
343 newattrs.ia_gid = GLOBAL_ROOT_GID;
344 newattrs.ia_mode = stat.mode & ~0777;
345 newattrs.ia_valid =
346 ATTR_UID|ATTR_GID|ATTR_MODE;
347 inode_lock(d_inode(dentry));
348 notify_change(dentry, &newattrs, NULL);
349 inode_unlock(d_inode(dentry));
350 err = vfs_unlink(d_inode(parent.dentry), dentry, NULL);
351 if (!err || err == -ENOENT)
352 deleted = 1;
353 }
354 } else {
355 err = -ENOENT;
356 }
357 dput(dentry);
358 inode_unlock(d_inode(parent.dentry));
359
360 path_put(&parent);
361 if (deleted && strchr(nodename, '/'))
362 delete_path(nodename);
363 return err;
364}
365
366/*
367 * If configured, or requested by the commandline, devtmpfs will be
368 * auto-mounted after the kernel mounted the root filesystem.
369 */
370int devtmpfs_mount(const char *mntdir)
371{
372 int err;
373
374 if (!mount_dev)
375 return 0;
376
377 if (!thread)
378 return 0;
379
380 err = ksys_mount("devtmpfs", mntdir, "devtmpfs", MS_SILENT, NULL);
381 if (err)
382 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
383 else
384 printk(KERN_INFO "devtmpfs: mounted\n");
385 return err;
386}
387
388static DECLARE_COMPLETION(setup_done);
389
390static int handle(const char *name, umode_t mode, kuid_t uid, kgid_t gid,
391 struct device *dev)
392{
393 if (mode)
394 return handle_create(name, mode, uid, gid, dev);
395 else
396 return handle_remove(name, dev);
397}
398
399static int devtmpfsd(void *p)
400{
401 int *err = p;
402 *err = ksys_unshare(CLONE_NEWNS);
403 if (*err)
404 goto out;
405 *err = ksys_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, NULL);
406 if (*err)
407 goto out;
408 ksys_chdir("/.."); /* will traverse into overmounted root */
409 ksys_chroot(".");
410 complete(&setup_done);
411 while (1) {
412 spin_lock(&req_lock);
413 while (requests) {
414 struct req *req = requests;
415 requests = NULL;
416 spin_unlock(&req_lock);
417 while (req) {
418 struct req *next = req->next;
419 req->err = handle(req->name, req->mode,
420 req->uid, req->gid, req->dev);
421 complete(&req->done);
422 req = next;
423 }
424 spin_lock(&req_lock);
425 }
426 __set_current_state(TASK_INTERRUPTIBLE);
427 spin_unlock(&req_lock);
428 schedule();
429 }
430 return 0;
431out:
432 complete(&setup_done);
433 return *err;
434}
435
436/*
437 * Create devtmpfs instance, driver-core devices will add their device
438 * nodes here.
439 */
440int __init devtmpfs_init(void)
441{
442 char opts[] = "mode=0755";
443 int err;
444
445 mnt = vfs_kern_mount(&internal_fs_type, 0, "devtmpfs", opts);
446 if (IS_ERR(mnt)) {
447 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %ld\n",
448 PTR_ERR(mnt));
449 return PTR_ERR(mnt);
450 }
451 err = register_filesystem(&dev_fs_type);
452 if (err) {
453 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
454 "type %i\n", err);
455 return err;
456 }
457
458 thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
459 if (!IS_ERR(thread)) {
460 wait_for_completion(&setup_done);
461 } else {
462 err = PTR_ERR(thread);
463 thread = NULL;
464 }
465
466 if (err) {
467 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
468 unregister_filesystem(&dev_fs_type);
469 return err;
470 }
471
472 printk(KERN_INFO "devtmpfs: initialized\n");
473 return 0;
474}