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/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         mode_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, mode_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         int err;
 168
 169         err = dev_mkdir(nodepath, 0755);
 170         if (err == -ENOENT) {
 171                 char *path;
 172                 char *s;
 173
 174                 /* parent directories do not exist, create them */
 175                 path = kstrdup(nodepath, GFP_KERNEL);
 176                 if (!path) {
 177                         err = -ENOMEM;
 178                         goto out;
 179                 }
 180                 s = path;
 181                 for (;;) {
 182                         s = strchr(s, '/');
 183                         if (!s)
 184                                 break;
 185                         s[0] = '\0';
 186                         err = dev_mkdir(path, 0755);
 187                         if (err && err != -EEXIST)
 188                                 break;
 189                         s[0] = '/';
 190                         s++;
 191                 }
 192                 kfree(path);
 193         }
 194 out:
 195         return err;
 196 }
 197
 198 static int handle_create(const char *nodename, mode_t mode, struct device *dev)
 199 {
 200         struct dentry *dentry;
 201         struct path path;
 202         int err;
 203
 204         dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
 205         if (dentry == ERR_PTR(-ENOENT)) {
 206                 create_path(nodename);
 207                 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
 208         }
 209         if (IS_ERR(dentry))
 210                 return PTR_ERR(dentry);
 211
 212         err = vfs_mknod(path.dentry->d_inode,
 213                         dentry, mode, dev->devt);
 214         if (!err) {
 215                 struct iattr newattrs;
 216
 217                 /* fixup possibly umasked mode */
 218                 newattrs.ia_mode = mode;
 219                 newattrs.ia_valid = ATTR_MODE;
 220                 mutex_lock(&dentry->d_inode->i_mutex);
 221                 notify_change(dentry, &newattrs);
 222                 mutex_unlock(&dentry->d_inode->i_mutex);
 223
 224                 /* mark as kernel-created inode */
 225                 dentry->d_inode->i_private = &thread;
 226         }
 227         dput(dentry);
 228
 229         mutex_unlock(&path.dentry->d_inode->i_mutex);
 230         path_put(&path);
 231         return err;
 232 }
 233
 234 static int dev_rmdir(const char *name)
 235 {
 236         struct nameidata nd;
 237         struct dentry *dentry;
 238         int err;
 239
 240         err = kern_path_parent(name, &nd);
 241         if (err)
 242                 return err;
 243
 244         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 245         dentry = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len);
 246         if (!IS_ERR(dentry)) {
 247                 if (dentry->d_inode) {
 248                         if (dentry->d_inode->i_private == &thread)
 249                                 err = vfs_rmdir(nd.path.dentry->d_inode,
 250                                                 dentry);
 251                         else
 252                                 err = -EPERM;
 253                 } else {
 254                         err = -ENOENT;
 255                 }
 256                 dput(dentry);
 257         } else {
 258                 err = PTR_ERR(dentry);
 259         }
 260
 261         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 262         path_put(&nd.path);
 263         return err;
 264 }
 265
 266 static int delete_path(const char *nodepath)
 267 {
 268         const char *path;
 269         int err = 0;
 270
 271         path = kstrdup(nodepath, GFP_KERNEL);
 272         if (!path)
 273                 return -ENOMEM;
 274
 275         for (;;) {
 276                 char *base;
 277
 278                 base = strrchr(path, '/');
 279                 if (!base)
 280                         break;
 281                 base[0] = '\0';
 282                 err = dev_rmdir(path);
 283                 if (err)
 284                         break;
 285         }
 286
 287         kfree(path);
 288         return err;
 289 }
 290
 291 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
 292 {
 293         /* did we create it */
 294         if (inode->i_private != &thread)
 295                 return 0;
 296
 297         /* does the dev_t match */
 298         if (is_blockdev(dev)) {
 299                 if (!S_ISBLK(stat->mode))
 300                         return 0;
 301         } else {
 302                 if (!S_ISCHR(stat->mode))
 303                         return 0;
 304         }
 305         if (stat->rdev != dev->devt)
 306                 return 0;
 307
 308         /* ours */
 309         return 1;
 310 }
 311
 312 static int handle_remove(const char *nodename, struct device *dev)
 313 {
 314         struct nameidata nd;
 315         struct dentry *dentry;
 316         struct kstat stat;
 317         int deleted = 1;
 318         int err;
 319
 320         err = kern_path_parent(nodename, &nd);
 321         if (err)
 322                 return err;
 323
 324         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 325         dentry = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len);
 326         if (!IS_ERR(dentry)) {
 327                 if (dentry->d_inode) {
 328                         err = vfs_getattr(nd.path.mnt, dentry, &stat);
 329                         if (!err && dev_mynode(dev, dentry->d_inode, &stat)) {
 330                                 struct iattr newattrs;
 331                                 /*
 332                                  * before unlinking this node, reset permissions
 333                                  * of possible references like hardlinks
 334                                  */
 335                                 newattrs.ia_uid = 0;
 336                                 newattrs.ia_gid = 0;
 337                                 newattrs.ia_mode = stat.mode & ~0777;
 338                                 newattrs.ia_valid =
 339                                         ATTR_UID|ATTR_GID|ATTR_MODE;
 340                                 mutex_lock(&dentry->d_inode->i_mutex);
 341                                 notify_change(dentry, &newattrs);
 342                                 mutex_unlock(&dentry->d_inode->i_mutex);
 343                                 err = vfs_unlink(nd.path.dentry->d_inode,
 344                                                  dentry);
 345                                 if (!err || err == -ENOENT)
 346                                         deleted = 1;
 347                         }
 348                 } else {
 349                         err = -ENOENT;
 350                 }
 351                 dput(dentry);
 352         } else {
 353                 err = PTR_ERR(dentry);
 354         }
 355         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 356
 357         path_put(&nd.path);
 358         if (deleted && strchr(nodename, '/'))
 359                 delete_path(nodename);
 360         return err;
 361 }
 362
 363 /*
 364  * If configured, or requested by the commandline, devtmpfs will be
 365  * auto-mounted after the kernel mounted the root filesystem.
 366  */
 367 int devtmpfs_mount(const char *mntdir)
 368 {
 369         int err;
 370
 371         if (!mount_dev)
 372                 return 0;
 373
 374         if (!thread)
 375                 return 0;
 376
 377         err = sys_mount("devtmpfs", (char *)mntdir, "devtmpfs", MS_SILENT, NULL);
 378         if (err)
 379                 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
 380         else
 381                 printk(KERN_INFO "devtmpfs: mounted\n");
 382         return err;
 383 }
 384
 385 static __initdata DECLARE_COMPLETION(setup_done);
 386
 387 static int handle(const char *name, mode_t mode, struct device *dev)
 388 {
 389         if (mode)
 390                 return handle_create(name, mode, dev);
 391         else
 392                 return handle_remove(name, dev);
 393 }
 394
 395 static int devtmpfsd(void *p)
 396 {
 397         char options[] = "mode=0755";
 398         int *err = p;
 399         *err = sys_unshare(CLONE_NEWNS);
 400         if (*err)
 401                 goto out;
 402         *err = sys_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, options);
 403         if (*err)
 404                 goto out;
 405         sys_chdir("/.."); /* will traverse into overmounted root */
 406         sys_chroot(".");
 407         complete(&setup_done);
 408         while (1) {
 409                 spin_lock(&req_lock);
 410                 while (requests) {
 411                         struct req *req = requests;
 412                         requests = NULL;
 413                         spin_unlock(&req_lock);
 414                         while (req) {
 415                                 req->err = handle(req->name, req->mode, req->dev);
 416                                 complete(&req->done);
 417                                 req = req->next;
 418                         }
 419                         spin_lock(&req_lock);
 420                 }
 421                 set_current_state(TASK_INTERRUPTIBLE);
 422                 spin_unlock(&req_lock);
 423                 schedule();
 424                 __set_current_state(TASK_RUNNING);
 425         }
 426         return 0;
 427 out:
 428         complete(&setup_done);
 429         return *err;
 430 }
 431
 432 /*
 433  * Create devtmpfs instance, driver-core devices will add their device
 434  * nodes here.
 435  */
 436 int __init devtmpfs_init(void)
 437 {
 438         int err = register_filesystem(&dev_fs_type);
 439         if (err) {
 440                 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
 441                        "type %i\n", err);
 442                 return err;
 443         }
 444
 445         thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
 446         if (!IS_ERR(thread)) {
 447                 wait_for_completion(&setup_done);
 448         } else {
 449                 err = PTR_ERR(thread);
 450                 thread = NULL;
 451         }
 452
 453         if (err) {
 454                 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
 455                 unregister_filesystem(&dev_fs_type);
 456                 return err;
 457         }
 458
 459         printk(KERN_INFO "devtmpfs: initialized\n");
 460         return 0;
 461 }