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 name of the device,
  11  * owned by root and have a default mode of 0600. Subsystems can
  12  * 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
  28 static struct task_struct *thread;
  29
  30 #if defined CONFIG_DEVTMPFS_MOUNT
  31 static int mount_dev = 1;
  32 #else
  33 static int mount_dev;
  34 #endif
  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         struct device *dev;
  45 } *requests;
  46
  47 static 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
  54 static 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
  64 static 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
  71 static inline int is_blockdev(struct device *dev)
  72 {
  73         return dev->class == &block_class;
  74 }
  75 #else
  76 static inline int is_blockdev(struct device *dev) { return 0; }
  77 #endif
  78
  79 int 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
 116 int 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
 145 static 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, LOOKUP_DIRECTORY);
 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         done_path_create(&path, dentry);
 160         return err;
 161 }
 162
 163 static int create_path(const char *nodepath)
 164 {
 165         char *path;
 166         char *s;
 167         int err = 0;
 168
 169         /* parent directories do not exist, create them */
 170         path = kstrdup(nodepath, GFP_KERNEL);
 171         if (!path)
 172                 return -ENOMEM;
 173
 174         s = path;
 175         for (;;) {
 176                 s = strchr(s, '/');
 177                 if (!s)
 178                         break;
 179                 s[0] = '\0';
 180                 err = dev_mkdir(path, 0755);
 181                 if (err && err != -EEXIST)
 182                         break;
 183                 s[0] = '/';
 184                 s++;
 185         }
 186         kfree(path);
 187         return err;
 188 }
 189
 190 static int handle_create(const char *nodename, umode_t mode, struct device *dev)
 191 {
 192         struct dentry *dentry;
 193         struct path path;
 194         int err;
 195
 196         dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
 197         if (dentry == ERR_PTR(-ENOENT)) {
 198                 create_path(nodename);
 199                 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
 200         }
 201         if (IS_ERR(dentry))
 202                 return PTR_ERR(dentry);
 203
 204         err = vfs_mknod(path.dentry->d_inode,
 205                         dentry, mode, dev->devt);
 206         if (!err) {
 207                 struct iattr newattrs;
 208
 209                 /* fixup possibly umasked mode */
 210                 newattrs.ia_mode = mode;
 211                 newattrs.ia_valid = ATTR_MODE;
 212                 mutex_lock(&dentry->d_inode->i_mutex);
 213                 notify_change(dentry, &newattrs);
 214                 mutex_unlock(&dentry->d_inode->i_mutex);
 215
 216                 /* mark as kernel-created inode */
 217                 dentry->d_inode->i_private = &thread;
 218         }
 219         done_path_create(&path, dentry);
 220         return err;
 221 }
 222
 223 static int dev_rmdir(const char *name)
 224 {
 225         struct path parent;
 226         struct dentry *dentry;
 227         int err;
 228
 229         dentry = kern_path_locked(name, &parent);
 230         if (IS_ERR(dentry))
 231                 return PTR_ERR(dentry);
 232         if (dentry->d_inode) {
 233                 if (dentry->d_inode->i_private == &thread)
 234                         err = vfs_rmdir(parent.dentry->d_inode, dentry);
 235                 else
 236                         err = -EPERM;
 237         } else {
 238                 err = -ENOENT;
 239         }
 240         dput(dentry);
 241         mutex_unlock(&parent.dentry->d_inode->i_mutex);
 242         path_put(&parent);
 243         return err;
 244 }
 245
 246 static int delete_path(const char *nodepath)
 247 {
 248         const char *path;
 249         int err = 0;
 250
 251         path = kstrdup(nodepath, GFP_KERNEL);
 252         if (!path)
 253                 return -ENOMEM;
 254
 255         for (;;) {
 256                 char *base;
 257
 258                 base = strrchr(path, '/');
 259                 if (!base)
 260                         break;
 261                 base[0] = '\0';
 262                 err = dev_rmdir(path);
 263                 if (err)
 264                         break;
 265         }
 266
 267         kfree(path);
 268         return err;
 269 }
 270
 271 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
 272 {
 273         /* did we create it */
 274         if (inode->i_private != &thread)
 275                 return 0;
 276
 277         /* does the dev_t match */
 278         if (is_blockdev(dev)) {
 279                 if (!S_ISBLK(stat->mode))
 280                         return 0;
 281         } else {
 282                 if (!S_ISCHR(stat->mode))
 283                         return 0;
 284         }
 285         if (stat->rdev != dev->devt)
 286                 return 0;
 287
 288         /* ours */
 289         return 1;
 290 }
 291
 292 static int handle_remove(const char *nodename, struct device *dev)
 293 {
 294         struct path parent;
 295         struct dentry *dentry;
 296         int deleted = 1;
 297         int err;
 298
 299         dentry = kern_path_locked(nodename, &parent);
 300         if (IS_ERR(dentry))
 301                 return PTR_ERR(dentry);
 302
 303         if (dentry->d_inode) {
 304                 struct kstat stat;
 305                 err = vfs_getattr(parent.mnt, dentry, &stat);
 306                 if (!err && dev_mynode(dev, dentry->d_inode, &stat)) {
 307                         struct iattr newattrs;
 308                         /*
 309                          * before unlinking this node, reset permissions
 310                          * of possible references like hardlinks
 311                          */
 312                         newattrs.ia_uid = GLOBAL_ROOT_UID;
 313                         newattrs.ia_gid = GLOBAL_ROOT_GID;
 314                         newattrs.ia_mode = stat.mode & ~0777;
 315                         newattrs.ia_valid =
 316                                 ATTR_UID|ATTR_GID|ATTR_MODE;
 317                         mutex_lock(&dentry->d_inode->i_mutex);
 318                         notify_change(dentry, &newattrs);
 319                         mutex_unlock(&dentry->d_inode->i_mutex);
 320                         err = vfs_unlink(parent.dentry->d_inode, dentry);
 321                         if (!err || err == -ENOENT)
 322                                 deleted = 1;
 323                 }
 324         } else {
 325                 err = -ENOENT;
 326         }
 327         dput(dentry);
 328         mutex_unlock(&parent.dentry->d_inode->i_mutex);
 329
 330         path_put(&parent);
 331         if (deleted && strchr(nodename, '/'))
 332                 delete_path(nodename);
 333         return err;
 334 }
 335
 336 /*
 337  * If configured, or requested by the commandline, devtmpfs will be
 338  * auto-mounted after the kernel mounted the root filesystem.
 339  */
 340 int devtmpfs_mount(const char *mntdir)
 341 {
 342         int err;
 343
 344         if (!mount_dev)
 345                 return 0;
 346
 347         if (!thread)
 348                 return 0;
 349
 350         err = sys_mount("devtmpfs", (char *)mntdir, "devtmpfs", MS_SILENT, NULL);
 351         if (err)
 352                 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
 353         else
 354                 printk(KERN_INFO "devtmpfs: mounted\n");
 355         return err;
 356 }
 357
 358 static DECLARE_COMPLETION(setup_done);
 359
 360 static int handle(const char *name, umode_t mode, struct device *dev)
 361 {
 362         if (mode)
 363                 return handle_create(name, mode, dev);
 364         else
 365                 return handle_remove(name, dev);
 366 }
 367
 368 static int devtmpfsd(void *p)
 369 {
 370         char options[] = "mode=0755";
 371         int *err = p;
 372         *err = sys_unshare(CLONE_NEWNS);
 373         if (*err)
 374                 goto out;
 375         *err = sys_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, options);
 376         if (*err)
 377                 goto out;
 378         sys_chdir("/.."); /* will traverse into overmounted root */
 379         sys_chroot(".");
 380         complete(&setup_done);
 381         while (1) {
 382                 spin_lock(&req_lock);
 383                 while (requests) {
 384                         struct req *req = requests;
 385                         requests = NULL;
 386                         spin_unlock(&req_lock);
 387                         while (req) {
 388                                 struct req *next = req->next;
 389                                 req->err = handle(req->name, req->mode, req->dev);
 390                                 complete(&req->done);
 391                                 req = next;
 392                         }
 393                         spin_lock(&req_lock);
 394                 }
 395                 __set_current_state(TASK_INTERRUPTIBLE);
 396                 spin_unlock(&req_lock);
 397                 schedule();
 398         }
 399         return 0;
 400 out:
 401         complete(&setup_done);
 402         return *err;
 403 }
 404
 405 /*
 406  * Create devtmpfs instance, driver-core devices will add their device
 407  * nodes here.
 408  */
 409 int __init devtmpfs_init(void)
 410 {
 411         int err = register_filesystem(&dev_fs_type);
 412         if (err) {
 413                 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
 414                        "type %i\n", err);
 415                 return err;
 416         }
 417
 418         thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
 419         if (!IS_ERR(thread)) {
 420                 wait_for_completion(&setup_done);
 421         } else {
 422                 err = PTR_ERR(thread);
 423                 thread = NULL;
 424         }
 425
 426         if (err) {
 427                 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
 428                 unregister_filesystem(&dev_fs_type);
 429                 return err;
 430         }
 431
 432         printk(KERN_INFO "devtmpfs: initialized\n");
 433         return 0;
 434 }