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 nameidata nd;
 148         struct dentry *dentry;
 149         int err;
 150
 151         err = kern_path_parent(name, &nd);
 152         if (err)
 153                 return err;
 154
 155         dentry = lookup_create(&nd, 1);
 156         if (!IS_ERR(dentry)) {
 157                 err = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
 158                 if (!err)
 159                         /* mark as kernel-created inode */
 160                         dentry->d_inode->i_private = &thread;
 161                 dput(dentry);
 162         } else {
 163                 err = PTR_ERR(dentry);
 164         }
 165
 166         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 167         path_put(&nd.path);
 168         return err;
 169 }
 170
 171 static int create_path(const char *nodepath)
 172 {
 173         int err;
 174
 175         err = dev_mkdir(nodepath, 0755);
 176         if (err == -ENOENT) {
 177                 char *path;
 178                 char *s;
 179
 180                 /* parent directories do not exist, create them */
 181                 path = kstrdup(nodepath, GFP_KERNEL);
 182                 if (!path) {
 183                         err = -ENOMEM;
 184                         goto out;
 185                 }
 186                 s = path;
 187                 for (;;) {
 188                         s = strchr(s, '/');
 189                         if (!s)
 190                                 break;
 191                         s[0] = '\0';
 192                         err = dev_mkdir(path, 0755);
 193                         if (err && err != -EEXIST)
 194                                 break;
 195                         s[0] = '/';
 196                         s++;
 197                 }
 198                 kfree(path);
 199         }
 200 out:
 201         return err;
 202 }
 203
 204 static int handle_create(const char *nodename, mode_t mode, struct device *dev)
 205 {
 206         struct nameidata nd;
 207         struct dentry *dentry;
 208         int err;
 209
 210         err = kern_path_parent(nodename, &nd);
 211         if (err == -ENOENT) {
 212                 create_path(nodename);
 213                 err = kern_path_parent(nodename, &nd);
 214         }
 215         if (err)
 216                 return err;
 217
 218         dentry = lookup_create(&nd, 0);
 219         if (!IS_ERR(dentry)) {
 220                 err = vfs_mknod(nd.path.dentry->d_inode,
 221                                 dentry, mode, dev->devt);
 222                 if (!err) {
 223                         struct iattr newattrs;
 224
 225                         /* fixup possibly umasked mode */
 226                         newattrs.ia_mode = mode;
 227                         newattrs.ia_valid = ATTR_MODE;
 228                         mutex_lock(&dentry->d_inode->i_mutex);
 229                         notify_change(dentry, &newattrs);
 230                         mutex_unlock(&dentry->d_inode->i_mutex);
 231
 232                         /* mark as kernel-created inode */
 233                         dentry->d_inode->i_private = &thread;
 234                 }
 235                 dput(dentry);
 236         } else {
 237                 err = PTR_ERR(dentry);
 238         }
 239
 240         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 241         path_put(&nd.path);
 242         return err;
 243 }
 244
 245 static int dev_rmdir(const char *name)
 246 {
 247         struct nameidata nd;
 248         struct dentry *dentry;
 249         int err;
 250
 251         err = kern_path_parent(name, &nd);
 252         if (err)
 253                 return err;
 254
 255         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 256         dentry = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len);
 257         if (!IS_ERR(dentry)) {
 258                 if (dentry->d_inode) {
 259                         if (dentry->d_inode->i_private == &thread)
 260                                 err = vfs_rmdir(nd.path.dentry->d_inode,
 261                                                 dentry);
 262                         else
 263                                 err = -EPERM;
 264                 } else {
 265                         err = -ENOENT;
 266                 }
 267                 dput(dentry);
 268         } else {
 269                 err = PTR_ERR(dentry);
 270         }
 271
 272         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 273         path_put(&nd.path);
 274         return err;
 275 }
 276
 277 static int delete_path(const char *nodepath)
 278 {
 279         const char *path;
 280         int err = 0;
 281
 282         path = kstrdup(nodepath, GFP_KERNEL);
 283         if (!path)
 284                 return -ENOMEM;
 285
 286         for (;;) {
 287                 char *base;
 288
 289                 base = strrchr(path, '/');
 290                 if (!base)
 291                         break;
 292                 base[0] = '\0';
 293                 err = dev_rmdir(path);
 294                 if (err)
 295                         break;
 296         }
 297
 298         kfree(path);
 299         return err;
 300 }
 301
 302 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
 303 {
 304         /* did we create it */
 305         if (inode->i_private != &thread)
 306                 return 0;
 307
 308         /* does the dev_t match */
 309         if (is_blockdev(dev)) {
 310                 if (!S_ISBLK(stat->mode))
 311                         return 0;
 312         } else {
 313                 if (!S_ISCHR(stat->mode))
 314                         return 0;
 315         }
 316         if (stat->rdev != dev->devt)
 317                 return 0;
 318
 319         /* ours */
 320         return 1;
 321 }
 322
 323 static int handle_remove(const char *nodename, struct device *dev)
 324 {
 325         struct nameidata nd;
 326         struct dentry *dentry;
 327         struct kstat stat;
 328         int deleted = 1;
 329         int err;
 330
 331         err = kern_path_parent(nodename, &nd);
 332         if (err)
 333                 return err;
 334
 335         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
 336         dentry = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len);
 337         if (!IS_ERR(dentry)) {
 338                 if (dentry->d_inode) {
 339                         err = vfs_getattr(nd.path.mnt, dentry, &stat);
 340                         if (!err && dev_mynode(dev, dentry->d_inode, &stat)) {
 341                                 struct iattr newattrs;
 342                                 /*
 343                                  * before unlinking this node, reset permissions
 344                                  * of possible references like hardlinks
 345                                  */
 346                                 newattrs.ia_uid = 0;
 347                                 newattrs.ia_gid = 0;
 348                                 newattrs.ia_mode = stat.mode & ~0777;
 349                                 newattrs.ia_valid =
 350                                         ATTR_UID|ATTR_GID|ATTR_MODE;
 351                                 mutex_lock(&dentry->d_inode->i_mutex);
 352                                 notify_change(dentry, &newattrs);
 353                                 mutex_unlock(&dentry->d_inode->i_mutex);
 354                                 err = vfs_unlink(nd.path.dentry->d_inode,
 355                                                  dentry);
 356                                 if (!err || err == -ENOENT)
 357                                         deleted = 1;
 358                         }
 359                 } else {
 360                         err = -ENOENT;
 361                 }
 362                 dput(dentry);
 363         } else {
 364                 err = PTR_ERR(dentry);
 365         }
 366         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 367
 368         path_put(&nd.path);
 369         if (deleted && strchr(nodename, '/'))
 370                 delete_path(nodename);
 371         return err;
 372 }
 373
 374 /*
 375  * If configured, or requested by the commandline, devtmpfs will be
 376  * auto-mounted after the kernel mounted the root filesystem.
 377  */
 378 int devtmpfs_mount(const char *mntdir)
 379 {
 380         int err;
 381
 382         if (!mount_dev)
 383                 return 0;
 384
 385         if (!thread)
 386                 return 0;
 387
 388         err = sys_mount("devtmpfs", (char *)mntdir, "devtmpfs", MS_SILENT, NULL);
 389         if (err)
 390                 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
 391         else
 392                 printk(KERN_INFO "devtmpfs: mounted\n");
 393         return err;
 394 }
 395
 396 static __initdata DECLARE_COMPLETION(setup_done);
 397
 398 static int handle(const char *name, mode_t mode, struct device *dev)
 399 {
 400         if (mode)
 401                 return handle_create(name, mode, dev);
 402         else
 403                 return handle_remove(name, dev);
 404 }
 405
 406 static int devtmpfsd(void *p)
 407 {
 408         char options[] = "mode=0755";
 409         int *err = p;
 410         *err = sys_unshare(CLONE_NEWNS);
 411         if (*err)
 412                 goto out;
 413         *err = sys_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, options);
 414         if (*err)
 415                 goto out;
 416         sys_chdir("/.."); /* will traverse into overmounted root */
 417         sys_chroot(".");
 418         complete(&setup_done);
 419         while (1) {
 420                 spin_lock(&req_lock);
 421                 while (requests) {
 422                         struct req *req = requests;
 423                         requests = NULL;
 424                         spin_unlock(&req_lock);
 425                         while (req) {
 426                                 req->err = handle(req->name, req->mode, req->dev);
 427                                 complete(&req->done);
 428                                 req = req->next;
 429                         }
 430                         spin_lock(&req_lock);
 431                 }
 432                 set_current_state(TASK_INTERRUPTIBLE);
 433                 spin_unlock(&req_lock);
 434                 schedule();
 435                 __set_current_state(TASK_RUNNING);
 436         }
 437         return 0;
 438 out:
 439         complete(&setup_done);
 440         return *err;
 441 }
 442
 443 /*
 444  * Create devtmpfs instance, driver-core devices will add their device
 445  * nodes here.
 446  */
 447 int __init devtmpfs_init(void)
 448 {
 449         int err = register_filesystem(&dev_fs_type);
 450         if (err) {
 451                 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
 452                        "type %i\n", err);
 453                 return err;
 454         }
 455
 456         thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
 457         if (!IS_ERR(thread)) {
 458                 wait_for_completion(&setup_done);
 459         } else {
 460                 err = PTR_ERR(thread);
 461                 thread = NULL;
 462         }
 463
 464         if (err) {
 465                 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
 466                 unregister_filesystem(&dev_fs_type);
 467                 return err;
 468         }
 469
 470         printk(KERN_INFO "devtmpfs: initialized\n");
 471         return 0;
 472 }