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, 1);
 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         dput(dentry);
 160         mutex_unlock(&path.dentry->d_inode->i_mutex);
 161         path_put(&path);
 162         return err;
 163 }
 164
 165 static int create_path(const char *nodepath)
 166 {
 167         char *path;
 168         char *s;
 169         int err = 0;
 170
 171         /* parent directories do not exist, create them */
 172         path = kstrdup(nodepath, GFP_KERNEL);
 173         if (!path)
 174                 return -ENOMEM;
 175
 176         s = path;
 177         for (;;) {
 178                 s = strchr(s, '/');
 179                 if (!s)
 180                         break;
 181                 s[0] = '\0';
 182                 err = dev_mkdir(path, 0755);
 183                 if (err && err != -EEXIST)
 184                         break;
 185                 s[0] = '/';
 186                 s++;
 187         }
 188         kfree(path);
 189         return err;
 190 }
 191
 192 static int handle_create(const char *nodename, umode_t mode, struct device *dev)
 193 {
 194         struct dentry *dentry;
 195         struct path path;
 196         int err;
 197
 198         dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
 199         if (dentry == ERR_PTR(-ENOENT)) {
 200                 create_path(nodename);
 201                 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
 202         }
 203         if (IS_ERR(dentry))
 204                 return PTR_ERR(dentry);
 205
 206         err = vfs_mknod(path.dentry->d_inode,
 207                         dentry, mode, dev->devt);
 208         if (!err) {
 209                 struct iattr newattrs;
 210
 211                 /* fixup possibly umasked mode */
 212                 newattrs.ia_mode = mode;
 213                 newattrs.ia_valid = ATTR_MODE;
 214                 mutex_lock(&dentry->d_inode->i_mutex);
 215                 notify_change(dentry, &newattrs);
 216                 mutex_unlock(&dentry->d_inode->i_mutex);
 217
 218                 /* mark as kernel-created inode */
 219                 dentry->d_inode->i_private = &thread;
 220         }
 221         dput(dentry);
 222
 223         mutex_unlock(&path.dentry->d_inode->i_mutex);
 224         path_put(&path);
 225         return err;
 226 }
 227
 228 static int dev_rmdir(const char *name)
 229 {
 230         struct path parent;
 231         struct dentry *dentry;
 232         int err;
 233
 234         dentry = kern_path_locked(name, &parent);
 235         if (IS_ERR(dentry))
 236                 return PTR_ERR(dentry);
 237         if (dentry->d_inode) {
 238                 if (dentry->d_inode->i_private == &thread)
 239                         err = vfs_rmdir(parent.dentry->d_inode, dentry);
 240                 else
 241                         err = -EPERM;
 242         } else {
 243                 err = -ENOENT;
 244         }
 245         dput(dentry);
 246         mutex_unlock(&parent.dentry->d_inode->i_mutex);
 247         path_put(&parent);
 248         return err;
 249 }
 250
 251 static int delete_path(const char *nodepath)
 252 {
 253         const char *path;
 254         int err = 0;
 255
 256         path = kstrdup(nodepath, GFP_KERNEL);
 257         if (!path)
 258                 return -ENOMEM;
 259
 260         for (;;) {
 261                 char *base;
 262
 263                 base = strrchr(path, '/');
 264                 if (!base)
 265                         break;
 266                 base[0] = '\0';
 267                 err = dev_rmdir(path);
 268                 if (err)
 269                         break;
 270         }
 271
 272         kfree(path);
 273         return err;
 274 }
 275
 276 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
 277 {
 278         /* did we create it */
 279         if (inode->i_private != &thread)
 280                 return 0;
 281
 282         /* does the dev_t match */
 283         if (is_blockdev(dev)) {
 284                 if (!S_ISBLK(stat->mode))
 285                         return 0;
 286         } else {
 287                 if (!S_ISCHR(stat->mode))
 288                         return 0;
 289         }
 290         if (stat->rdev != dev->devt)
 291                 return 0;
 292
 293         /* ours */
 294         return 1;
 295 }
 296
 297 static int handle_remove(const char *nodename, struct device *dev)
 298 {
 299         struct path parent;
 300         struct dentry *dentry;
 301         int deleted = 1;
 302         int err;
 303
 304         dentry = kern_path_locked(nodename, &parent);
 305         if (IS_ERR(dentry))
 306                 return PTR_ERR(dentry);
 307
 308         if (dentry->d_inode) {
 309                 struct kstat stat;
 310                 err = vfs_getattr(parent.mnt, dentry, &stat);
 311                 if (!err && dev_mynode(dev, dentry->d_inode, &stat)) {
 312                         struct iattr newattrs;
 313                         /*
 314                          * before unlinking this node, reset permissions
 315                          * of possible references like hardlinks
 316                          */
 317                         newattrs.ia_uid = 0;
 318                         newattrs.ia_gid = 0;
 319                         newattrs.ia_mode = stat.mode & ~0777;
 320                         newattrs.ia_valid =
 321                                 ATTR_UID|ATTR_GID|ATTR_MODE;
 322                         mutex_lock(&dentry->d_inode->i_mutex);
 323                         notify_change(dentry, &newattrs);
 324                         mutex_unlock(&dentry->d_inode->i_mutex);
 325                         err = vfs_unlink(parent.dentry->d_inode, dentry);
 326                         if (!err || err == -ENOENT)
 327                                 deleted = 1;
 328                 }
 329         } else {
 330                 err = -ENOENT;
 331         }
 332         dput(dentry);
 333         mutex_unlock(&parent.dentry->d_inode->i_mutex);
 334
 335         path_put(&parent);
 336         if (deleted && strchr(nodename, '/'))
 337                 delete_path(nodename);
 338         return err;
 339 }
 340
 341 /*
 342  * If configured, or requested by the commandline, devtmpfs will be
 343  * auto-mounted after the kernel mounted the root filesystem.
 344  */
 345 int devtmpfs_mount(const char *mntdir)
 346 {
 347         int err;
 348
 349         if (!mount_dev)
 350                 return 0;
 351
 352         if (!thread)
 353                 return 0;
 354
 355         err = sys_mount("devtmpfs", (char *)mntdir, "devtmpfs", MS_SILENT, NULL);
 356         if (err)
 357                 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
 358         else
 359                 printk(KERN_INFO "devtmpfs: mounted\n");
 360         return err;
 361 }
 362
 363 static DECLARE_COMPLETION(setup_done);
 364
 365 static int handle(const char *name, umode_t mode, struct device *dev)
 366 {
 367         if (mode)
 368                 return handle_create(name, mode, dev);
 369         else
 370                 return handle_remove(name, dev);
 371 }
 372
 373 static int devtmpfsd(void *p)
 374 {
 375         char options[] = "mode=0755";
 376         int *err = p;
 377         *err = sys_unshare(CLONE_NEWNS);
 378         if (*err)
 379                 goto out;
 380         *err = sys_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, options);
 381         if (*err)
 382                 goto out;
 383         sys_chdir("/.."); /* will traverse into overmounted root */
 384         sys_chroot(".");
 385         complete(&setup_done);
 386         while (1) {
 387                 spin_lock(&req_lock);
 388                 while (requests) {
 389                         struct req *req = requests;
 390                         requests = NULL;
 391                         spin_unlock(&req_lock);
 392                         while (req) {
 393                                 struct req *next = req->next;
 394                                 req->err = handle(req->name, req->mode, req->dev);
 395                                 complete(&req->done);
 396                                 req = next;
 397                         }
 398                         spin_lock(&req_lock);
 399                 }
 400                 __set_current_state(TASK_INTERRUPTIBLE);
 401                 spin_unlock(&req_lock);
 402                 schedule();
 403         }
 404         return 0;
 405 out:
 406         complete(&setup_done);
 407         return *err;
 408 }
 409
 410 /*
 411  * Create devtmpfs instance, driver-core devices will add their device
 412  * nodes here.
 413  */
 414 int __init devtmpfs_init(void)
 415 {
 416         int err = register_filesystem(&dev_fs_type);
 417         if (err) {
 418                 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
 419                        "type %i\n", err);
 420                 return err;
 421         }
 422
 423         thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
 424         if (!IS_ERR(thread)) {
 425                 wait_for_completion(&setup_done);
 426         } else {
 427                 err = PTR_ERR(thread);
 428                 thread = NULL;
 429         }
 430
 431         if (err) {
 432                 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
 433                 unregister_filesystem(&dev_fs_type);
 434                 return err;
 435         }
 436
 437         printk(KERN_INFO "devtmpfs: initialized\n");
 438         return 0;
 439 }