Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
[linux-2.6.git] / drivers / base / devtmpfs.c
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/cred.h>
24 #include <linux/sched.h>
25 #include <linux/init_task.h>
26
27 static struct vfsmount *dev_mnt;
28
29 #if defined CONFIG_DEVTMPFS_MOUNT
30 static int dev_mount = 1;
31 #else
32 static int dev_mount;
33 #endif
34
35 static DEFINE_MUTEX(dirlock);
36
37 static int __init mount_param(char *str)
38 {
39         dev_mount = simple_strtoul(str, NULL, 0);
40         return 1;
41 }
42 __setup("devtmpfs.mount=", mount_param);
43
44 static int dev_get_sb(struct file_system_type *fs_type, int flags,
45                       const char *dev_name, void *data, struct vfsmount *mnt)
46 {
47         return get_sb_single(fs_type, flags, data, shmem_fill_super, mnt);
48 }
49
50 static struct file_system_type dev_fs_type = {
51         .name = "devtmpfs",
52         .get_sb = dev_get_sb,
53         .kill_sb = kill_litter_super,
54 };
55
56 #ifdef CONFIG_BLOCK
57 static inline int is_blockdev(struct device *dev)
58 {
59         return dev->class == &block_class;
60 }
61 #else
62 static inline int is_blockdev(struct device *dev) { return 0; }
63 #endif
64
65 static int dev_mkdir(const char *name, mode_t mode)
66 {
67         struct nameidata nd;
68         struct dentry *dentry;
69         int err;
70
71         err = vfs_path_lookup(dev_mnt->mnt_root, dev_mnt,
72                               name, LOOKUP_PARENT, &nd);
73         if (err)
74                 return err;
75
76         dentry = lookup_create(&nd, 1);
77         if (!IS_ERR(dentry)) {
78                 err = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
79                 if (!err)
80                         /* mark as kernel-created inode */
81                         dentry->d_inode->i_private = &dev_mnt;
82                 dput(dentry);
83         } else {
84                 err = PTR_ERR(dentry);
85         }
86
87         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
88         path_put(&nd.path);
89         return err;
90 }
91
92 static int create_path(const char *nodepath)
93 {
94         int err;
95
96         mutex_lock(&dirlock);
97         err = dev_mkdir(nodepath, 0755);
98         if (err == -ENOENT) {
99                 char *path;
100                 char *s;
101
102                 /* parent directories do not exist, create them */
103                 path = kstrdup(nodepath, GFP_KERNEL);
104                 if (!path) {
105                         err = -ENOMEM;
106                         goto out;
107                 }
108                 s = path;
109                 for (;;) {
110                         s = strchr(s, '/');
111                         if (!s)
112                                 break;
113                         s[0] = '\0';
114                         err = dev_mkdir(path, 0755);
115                         if (err && err != -EEXIST)
116                                 break;
117                         s[0] = '/';
118                         s++;
119                 }
120                 kfree(path);
121         }
122 out:
123         mutex_unlock(&dirlock);
124         return err;
125 }
126
127 int devtmpfs_create_node(struct device *dev)
128 {
129         const char *tmp = NULL;
130         const char *nodename;
131         const struct cred *curr_cred;
132         mode_t mode = 0;
133         struct nameidata nd;
134         struct dentry *dentry;
135         int err;
136
137         if (!dev_mnt)
138                 return 0;
139
140         nodename = device_get_devnode(dev, &mode, &tmp);
141         if (!nodename)
142                 return -ENOMEM;
143
144         if (mode == 0)
145                 mode = 0600;
146         if (is_blockdev(dev))
147                 mode |= S_IFBLK;
148         else
149                 mode |= S_IFCHR;
150
151         curr_cred = override_creds(&init_cred);
152
153         err = vfs_path_lookup(dev_mnt->mnt_root, dev_mnt,
154                               nodename, LOOKUP_PARENT, &nd);
155         if (err == -ENOENT) {
156                 create_path(nodename);
157                 err = vfs_path_lookup(dev_mnt->mnt_root, dev_mnt,
158                                       nodename, LOOKUP_PARENT, &nd);
159         }
160         if (err)
161                 goto out;
162
163         dentry = lookup_create(&nd, 0);
164         if (!IS_ERR(dentry)) {
165                 err = vfs_mknod(nd.path.dentry->d_inode,
166                                 dentry, mode, dev->devt);
167                 if (!err) {
168                         struct iattr newattrs;
169
170                         /* fixup possibly umasked mode */
171                         newattrs.ia_mode = mode;
172                         newattrs.ia_valid = ATTR_MODE;
173                         mutex_lock(&dentry->d_inode->i_mutex);
174                         notify_change(dentry, &newattrs);
175                         mutex_unlock(&dentry->d_inode->i_mutex);
176
177                         /* mark as kernel-created inode */
178                         dentry->d_inode->i_private = &dev_mnt;
179                 }
180                 dput(dentry);
181         } else {
182                 err = PTR_ERR(dentry);
183         }
184
185         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
186         path_put(&nd.path);
187 out:
188         kfree(tmp);
189         revert_creds(curr_cred);
190         return err;
191 }
192
193 static int dev_rmdir(const char *name)
194 {
195         struct nameidata nd;
196         struct dentry *dentry;
197         int err;
198
199         err = vfs_path_lookup(dev_mnt->mnt_root, dev_mnt,
200                               name, LOOKUP_PARENT, &nd);
201         if (err)
202                 return err;
203
204         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
205         dentry = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len);
206         if (!IS_ERR(dentry)) {
207                 if (dentry->d_inode) {
208                         if (dentry->d_inode->i_private == &dev_mnt)
209                                 err = vfs_rmdir(nd.path.dentry->d_inode,
210                                                 dentry);
211                         else
212                                 err = -EPERM;
213                 } else {
214                         err = -ENOENT;
215                 }
216                 dput(dentry);
217         } else {
218                 err = PTR_ERR(dentry);
219         }
220
221         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
222         path_put(&nd.path);
223         return err;
224 }
225
226 static int delete_path(const char *nodepath)
227 {
228         const char *path;
229         int err = 0;
230
231         path = kstrdup(nodepath, GFP_KERNEL);
232         if (!path)
233                 return -ENOMEM;
234
235         mutex_lock(&dirlock);
236         for (;;) {
237                 char *base;
238
239                 base = strrchr(path, '/');
240                 if (!base)
241                         break;
242                 base[0] = '\0';
243                 err = dev_rmdir(path);
244                 if (err)
245                         break;
246         }
247         mutex_unlock(&dirlock);
248
249         kfree(path);
250         return err;
251 }
252
253 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
254 {
255         /* did we create it */
256         if (inode->i_private != &dev_mnt)
257                 return 0;
258
259         /* does the dev_t match */
260         if (is_blockdev(dev)) {
261                 if (!S_ISBLK(stat->mode))
262                         return 0;
263         } else {
264                 if (!S_ISCHR(stat->mode))
265                         return 0;
266         }
267         if (stat->rdev != dev->devt)
268                 return 0;
269
270         /* ours */
271         return 1;
272 }
273
274 int devtmpfs_delete_node(struct device *dev)
275 {
276         const char *tmp = NULL;
277         const char *nodename;
278         const struct cred *curr_cred;
279         struct nameidata nd;
280         struct dentry *dentry;
281         struct kstat stat;
282         int deleted = 1;
283         int err;
284
285         if (!dev_mnt)
286                 return 0;
287
288         nodename = device_get_devnode(dev, NULL, &tmp);
289         if (!nodename)
290                 return -ENOMEM;
291
292         curr_cred = override_creds(&init_cred);
293         err = vfs_path_lookup(dev_mnt->mnt_root, dev_mnt,
294                               nodename, LOOKUP_PARENT, &nd);
295         if (err)
296                 goto out;
297
298         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
299         dentry = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len);
300         if (!IS_ERR(dentry)) {
301                 if (dentry->d_inode) {
302                         err = vfs_getattr(nd.path.mnt, dentry, &stat);
303                         if (!err && dev_mynode(dev, dentry->d_inode, &stat)) {
304                                 struct iattr newattrs;
305                                 /*
306                                  * before unlinking this node, reset permissions
307                                  * of possible references like hardlinks
308                                  */
309                                 newattrs.ia_uid = 0;
310                                 newattrs.ia_gid = 0;
311                                 newattrs.ia_mode = stat.mode & ~0777;
312                                 newattrs.ia_valid =
313                                         ATTR_UID|ATTR_GID|ATTR_MODE;
314                                 mutex_lock(&dentry->d_inode->i_mutex);
315                                 notify_change(dentry, &newattrs);
316                                 mutex_unlock(&dentry->d_inode->i_mutex);
317                                 err = vfs_unlink(nd.path.dentry->d_inode,
318                                                  dentry);
319                                 if (!err || err == -ENOENT)
320                                         deleted = 1;
321                         }
322                 } else {
323                         err = -ENOENT;
324                 }
325                 dput(dentry);
326         } else {
327                 err = PTR_ERR(dentry);
328         }
329         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
330
331         path_put(&nd.path);
332         if (deleted && strchr(nodename, '/'))
333                 delete_path(nodename);
334 out:
335         kfree(tmp);
336         revert_creds(curr_cred);
337         return err;
338 }
339
340 /*
341  * If configured, or requested by the commandline, devtmpfs will be
342  * auto-mounted after the kernel mounted the root filesystem.
343  */
344 int devtmpfs_mount(const char *mntdir)
345 {
346         int err;
347
348         if (!dev_mount)
349                 return 0;
350
351         if (!dev_mnt)
352                 return 0;
353
354         err = sys_mount("devtmpfs", (char *)mntdir, "devtmpfs", MS_SILENT, NULL);
355         if (err)
356                 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
357         else
358                 printk(KERN_INFO "devtmpfs: mounted\n");
359         return err;
360 }
361
362 /*
363  * Create devtmpfs instance, driver-core devices will add their device
364  * nodes here.
365  */
366 int __init devtmpfs_init(void)
367 {
368         int err;
369         struct vfsmount *mnt;
370         char options[] = "mode=0755";
371
372         err = register_filesystem(&dev_fs_type);
373         if (err) {
374                 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
375                        "type %i\n", err);
376                 return err;
377         }
378
379         mnt = kern_mount_data(&dev_fs_type, options);
380         if (IS_ERR(mnt)) {
381                 err = PTR_ERR(mnt);
382                 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
383                 unregister_filesystem(&dev_fs_type);
384                 return err;
385         }
386         dev_mnt = mnt;
387
388         printk(KERN_INFO "devtmpfs: initialized\n");
389         return 0;
390 }