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 #include "base.h"
  28
  29 static struct task_struct *thread;
  30
  31 #if defined CONFIG_DEVTMPFS_MOUNT
  32 static int mount_dev = 1;
  33 #else
  34 static int mount_dev;
  35 #endif
  36
  37 static DEFINE_SPINLOCK(req_lock);
  38
  39 static struct req {
  40         struct req *next;
  41         struct completion done;
  42         int err;
  43         const char *name;
  44         umode_t mode;   /* 0 => delete */
  45         kuid_t uid;
  46         kgid_t gid;
  47         struct device *dev;
  48 } *requests;
  49
  50 static int __init mount_param(char *str)
  51 {
  52         mount_dev = simple_strtoul(str, NULL, 0);
  53         return 1;
  54 }
  55 __setup("devtmpfs.mount=", mount_param);
  56
  57 static struct dentry *dev_mount(struct file_system_type *fs_type, int flags,
  58                       const char *dev_name, void *data)
  59 {
  60 #ifdef CONFIG_TMPFS
  61         return mount_single(fs_type, flags, data, shmem_fill_super);
  62 #else
  63         return mount_single(fs_type, flags, data, ramfs_fill_super);
  64 #endif
  65 }
  66
  67 static struct file_system_type dev_fs_type = {
  68         .name = "devtmpfs",
  69         .mount = dev_mount,
  70         .kill_sb = kill_litter_super,
  71 };
  72
  73 #ifdef CONFIG_BLOCK
  74 static inline int is_blockdev(struct device *dev)
  75 {
  76         return dev->class == &block_class;
  77 }
  78 #else
  79 static inline int is_blockdev(struct device *dev) { return 0; }
  80 #endif
  81
  82 int devtmpfs_create_node(struct device *dev)
  83 {
  84         const char *tmp = NULL;
  85         struct req req;
  86
  87         if (!thread)
  88                 return 0;
  89
  90         req.mode = 0;
  91         req.uid = GLOBAL_ROOT_UID;
  92         req.gid = GLOBAL_ROOT_GID;
  93         req.name = device_get_devnode(dev, &req.mode, &req.uid, &req.gid, &tmp);
  94         if (!req.name)
  95                 return -ENOMEM;
  96
  97         if (req.mode == 0)
  98                 req.mode = 0600;
  99         if (is_blockdev(dev))
 100                 req.mode |= S_IFBLK;
 101         else
 102                 req.mode |= S_IFCHR;
 103
 104         req.dev = dev;
 105
 106         init_completion(&req.done);
 107
 108         spin_lock(&req_lock);
 109         req.next = requests;
 110         requests = &req;
 111         spin_unlock(&req_lock);
 112
 113         wake_up_process(thread);
 114         wait_for_completion(&req.done);
 115
 116         kfree(tmp);
 117
 118         return req.err;
 119 }
 120
 121 int devtmpfs_delete_node(struct device *dev)
 122 {
 123         const char *tmp = NULL;
 124         struct req req;
 125
 126         if (!thread)
 127                 return 0;
 128
 129         req.name = device_get_devnode(dev, NULL, NULL, NULL, &tmp);
 130         if (!req.name)
 131                 return -ENOMEM;
 132
 133         req.mode = 0;
 134         req.dev = dev;
 135
 136         init_completion(&req.done);
 137
 138         spin_lock(&req_lock);
 139         req.next = requests;
 140         requests = &req;
 141         spin_unlock(&req_lock);
 142
 143         wake_up_process(thread);
 144         wait_for_completion(&req.done);
 145
 146         kfree(tmp);
 147         return req.err;
 148 }
 149
 150 static int dev_mkdir(const char *name, umode_t mode)
 151 {
 152         struct dentry *dentry;
 153         struct path path;
 154         int err;
 155
 156         dentry = kern_path_create(AT_FDCWD, name, &path, LOOKUP_DIRECTORY);
 157         if (IS_ERR(dentry))
 158                 return PTR_ERR(dentry);
 159
 160         err = vfs_mkdir(d_inode(path.dentry), dentry, mode);
 161         if (!err)
 162                 /* mark as kernel-created inode */
 163                 d_inode(dentry)->i_private = &thread;
 164         done_path_create(&path, dentry);
 165         return err;
 166 }
 167
 168 static int create_path(const char *nodepath)
 169 {
 170         char *path;
 171         char *s;
 172         int err = 0;
 173
 174         /* parent directories do not exist, create them */
 175         path = kstrdup(nodepath, GFP_KERNEL);
 176         if (!path)
 177                 return -ENOMEM;
 178
 179         s = path;
 180         for (;;) {
 181                 s = strchr(s, '/');
 182                 if (!s)
 183                         break;
 184                 s[0] = '\0';
 185                 err = dev_mkdir(path, 0755);
 186                 if (err && err != -EEXIST)
 187                         break;
 188                 s[0] = '/';
 189                 s++;
 190         }
 191         kfree(path);
 192         return err;
 193 }
 194
 195 static int handle_create(const char *nodename, umode_t mode, kuid_t uid,
 196                          kgid_t gid, struct device *dev)
 197 {
 198         struct dentry *dentry;
 199         struct path path;
 200         int err;
 201
 202         dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
 203         if (dentry == ERR_PTR(-ENOENT)) {
 204                 create_path(nodename);
 205                 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
 206         }
 207         if (IS_ERR(dentry))
 208                 return PTR_ERR(dentry);
 209
 210         err = vfs_mknod(d_inode(path.dentry), dentry, mode, dev->devt);
 211         if (!err) {
 212                 struct iattr newattrs;
 213
 214                 newattrs.ia_mode = mode;
 215                 newattrs.ia_uid = uid;
 216                 newattrs.ia_gid = gid;
 217                 newattrs.ia_valid = ATTR_MODE|ATTR_UID|ATTR_GID;
 218                 mutex_lock(&d_inode(dentry)->i_mutex);
 219                 notify_change(dentry, &newattrs, NULL);
 220                 mutex_unlock(&d_inode(dentry)->i_mutex);
 221
 222                 /* mark as kernel-created inode */
 223                 d_inode(dentry)->i_private = &thread;
 224         }
 225         done_path_create(&path, dentry);
 226         return err;
 227 }
 228
 229 static int dev_rmdir(const char *name)
 230 {
 231         struct path parent;
 232         struct dentry *dentry;
 233         int err;
 234
 235         dentry = kern_path_locked(name, &parent);
 236         if (IS_ERR(dentry))
 237                 return PTR_ERR(dentry);
 238         if (d_really_is_positive(dentry)) {
 239                 if (d_inode(dentry)->i_private == &thread)
 240                         err = vfs_rmdir(d_inode(parent.dentry), dentry);
 241                 else
 242                         err = -EPERM;
 243         } else {
 244                 err = -ENOENT;
 245         }
 246         dput(dentry);
 247         mutex_unlock(&d_inode(parent.dentry)->i_mutex);
 248         path_put(&parent);
 249         return err;
 250 }
 251
 252 static int delete_path(const char *nodepath)
 253 {
 254         const char *path;
 255         int err = 0;
 256
 257         path = kstrdup(nodepath, GFP_KERNEL);
 258         if (!path)
 259                 return -ENOMEM;
 260
 261         for (;;) {
 262                 char *base;
 263
 264                 base = strrchr(path, '/');
 265                 if (!base)
 266                         break;
 267                 base[0] = '\0';
 268                 err = dev_rmdir(path);
 269                 if (err)
 270                         break;
 271         }
 272
 273         kfree(path);
 274         return err;
 275 }
 276
 277 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
 278 {
 279         /* did we create it */
 280         if (inode->i_private != &thread)
 281                 return 0;
 282
 283         /* does the dev_t match */
 284         if (is_blockdev(dev)) {
 285                 if (!S_ISBLK(stat->mode))
 286                         return 0;
 287         } else {
 288                 if (!S_ISCHR(stat->mode))
 289                         return 0;
 290         }
 291         if (stat->rdev != dev->devt)
 292                 return 0;
 293
 294         /* ours */
 295         return 1;
 296 }
 297
 298 static int handle_remove(const char *nodename, struct device *dev)
 299 {
 300         struct path parent;
 301         struct dentry *dentry;
 302         int deleted = 0;
 303         int err;
 304
 305         dentry = kern_path_locked(nodename, &parent);
 306         if (IS_ERR(dentry))
 307                 return PTR_ERR(dentry);
 308
 309         if (d_really_is_positive(dentry)) {
 310                 struct kstat stat;
 311                 struct path p = {.mnt = parent.mnt, .dentry = dentry};
 312                 err = vfs_getattr(&p, &stat);
 313                 if (!err && dev_mynode(dev, d_inode(dentry), &stat)) {
 314                         struct iattr newattrs;
 315                         /*
 316                          * before unlinking this node, reset permissions
 317                          * of possible references like hardlinks
 318                          */
 319                         newattrs.ia_uid = GLOBAL_ROOT_UID;
 320                         newattrs.ia_gid = GLOBAL_ROOT_GID;
 321                         newattrs.ia_mode = stat.mode & ~0777;
 322                         newattrs.ia_valid =
 323                                 ATTR_UID|ATTR_GID|ATTR_MODE;
 324                         mutex_lock(&d_inode(dentry)->i_mutex);
 325                         notify_change(dentry, &newattrs, NULL);
 326                         mutex_unlock(&d_inode(dentry)->i_mutex);
 327                         err = vfs_unlink(d_inode(parent.dentry), dentry, NULL);
 328                         if (!err || err == -ENOENT)
 329                                 deleted = 1;
 330                 }
 331         } else {
 332                 err = -ENOENT;
 333         }
 334         dput(dentry);
 335         mutex_unlock(&d_inode(parent.dentry)->i_mutex);
 336
 337         path_put(&parent);
 338         if (deleted && strchr(nodename, '/'))
 339                 delete_path(nodename);
 340         return err;
 341 }
 342
 343 /*
 344  * If configured, or requested by the commandline, devtmpfs will be
 345  * auto-mounted after the kernel mounted the root filesystem.
 346  */
 347 int devtmpfs_mount(const char *mntdir)
 348 {
 349         int err;
 350
 351         if (!mount_dev)
 352                 return 0;
 353
 354         if (!thread)
 355                 return 0;
 356
 357         err = sys_mount("devtmpfs", (char *)mntdir, "devtmpfs", MS_SILENT, NULL);
 358         if (err)
 359                 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
 360         else
 361                 printk(KERN_INFO "devtmpfs: mounted\n");
 362         return err;
 363 }
 364
 365 static DECLARE_COMPLETION(setup_done);
 366
 367 static int handle(const char *name, umode_t mode, kuid_t uid, kgid_t gid,
 368                   struct device *dev)
 369 {
 370         if (mode)
 371                 return handle_create(name, mode, uid, gid, dev);
 372         else
 373                 return handle_remove(name, dev);
 374 }
 375
 376 static int devtmpfsd(void *p)
 377 {
 378         char options[] = "mode=0755";
 379         int *err = p;
 380         *err = sys_unshare(CLONE_NEWNS);
 381         if (*err)
 382                 goto out;
 383         *err = sys_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, options);
 384         if (*err)
 385                 goto out;
 386         sys_chdir("/.."); /* will traverse into overmounted root */
 387         sys_chroot(".");
 388         complete(&setup_done);
 389         while (1) {
 390                 spin_lock(&req_lock);
 391                 while (requests) {
 392                         struct req *req = requests;
 393                         requests = NULL;
 394                         spin_unlock(&req_lock);
 395                         while (req) {
 396                                 struct req *next = req->next;
 397                                 req->err = handle(req->name, req->mode,
 398                                                   req->uid, req->gid, req->dev);
 399                                 complete(&req->done);
 400                                 req = next;
 401                         }
 402                         spin_lock(&req_lock);
 403                 }
 404                 __set_current_state(TASK_INTERRUPTIBLE);
 405                 spin_unlock(&req_lock);
 406                 schedule();
 407         }
 408         return 0;
 409 out:
 410         complete(&setup_done);
 411         return *err;
 412 }
 413
 414 /*
 415  * Create devtmpfs instance, driver-core devices will add their device
 416  * nodes here.
 417  */
 418 int __init devtmpfs_init(void)
 419 {
 420         int err = register_filesystem(&dev_fs_type);
 421         if (err) {
 422                 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
 423                        "type %i\n", err);
 424                 return err;
 425         }
 426
 427         thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
 428         if (!IS_ERR(thread)) {
 429                 wait_for_completion(&setup_done);
 430         } else {
 431                 err = PTR_ERR(thread);
 432                 thread = NULL;
 433         }
 434
 435         if (err) {
 436                 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
 437                 unregister_filesystem(&dev_fs_type);
 438                 return err;
 439         }
 440
 441         printk(KERN_INFO "devtmpfs: initialized\n");
 442         return 0;
 443 }