drivers/base/devtmpfs.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * devtmpfs - kernel-maintained tmpfs-based /dev
   4  *
   5  * Copyright (C) 2009, Kay Sievers <kay.sievers@vrfy.org>
   6  *
   7  * During bootup, before any driver core device is registered,
   8  * devtmpfs, a tmpfs-based filesystem is created. Every driver-core
   9  * device which requests a device node, will add a node in this
  10  * filesystem.
  11  * By default, all devices are named after the name of the device,
  12  * owned by root and have a default mode of 0600. Subsystems can
  13  * overwrite the default setting if needed.
  14  */
  15
  16 #include <linux/kernel.h>
  17 #include <linux/syscalls.h>
  18 #include <linux/mount.h>
  19 #include <linux/device.h>
  20 #include <linux/genhd.h>
  21 #include <linux/namei.h>
  22 #include <linux/fs.h>
  23 #include <linux/shmem_fs.h>
  24 #include <linux/ramfs.h>
  25 #include <linux/sched.h>
  26 #include <linux/slab.h>
  27 #include <linux/kthread.h>
  28 #include <uapi/linux/mount.h>
  29 #include "base.h"
  30
  31 static struct task_struct *thread;
  32
  33 #if defined CONFIG_DEVTMPFS_MOUNT
  34 static int mount_dev = 1;
  35 #else
  36 static int mount_dev;
  37 #endif
  38
  39 static DEFINE_SPINLOCK(req_lock);
  40
  41 static struct req {
  42         struct req *next;
  43         struct completion done;
  44         int err;
  45         const char *name;
  46         umode_t mode;   /* 0 => delete */
  47         kuid_t uid;
  48         kgid_t gid;
  49         struct device *dev;
  50 } *requests;
  51
  52 static int __init mount_param(char *str)
  53 {
  54         mount_dev = simple_strtoul(str, NULL, 0);
  55         return 1;
  56 }
  57 __setup("devtmpfs.mount=", mount_param);
  58
  59 static struct dentry *dev_mount(struct file_system_type *fs_type, int flags,
  60                       const char *dev_name, void *data)
  61 {
  62 #ifdef CONFIG_TMPFS
  63         return mount_single(fs_type, flags, data, shmem_fill_super);
  64 #else
  65         return mount_single(fs_type, flags, data, ramfs_fill_super);
  66 #endif
  67 }
  68
  69 static struct file_system_type dev_fs_type = {
  70         .name = "devtmpfs",
  71         .mount = dev_mount,
  72         .kill_sb = kill_litter_super,
  73 };
  74
  75 #ifdef CONFIG_BLOCK
  76 static inline int is_blockdev(struct device *dev)
  77 {
  78         return dev->class == &block_class;
  79 }
  80 #else
  81 static inline int is_blockdev(struct device *dev) { return 0; }
  82 #endif
  83
  84 int devtmpfs_create_node(struct device *dev)
  85 {
  86         const char *tmp = NULL;
  87         struct req req;
  88
  89         if (!thread)
  90                 return 0;
  91
  92         req.mode = 0;
  93         req.uid = GLOBAL_ROOT_UID;
  94         req.gid = GLOBAL_ROOT_GID;
  95         req.name = device_get_devnode(dev, &req.mode, &req.uid, &req.gid, &tmp);
  96         if (!req.name)
  97                 return -ENOMEM;
  98
  99         if (req.mode == 0)
 100                 req.mode = 0600;
 101         if (is_blockdev(dev))
 102                 req.mode |= S_IFBLK;
 103         else
 104                 req.mode |= S_IFCHR;
 105
 106         req.dev = dev;
 107
 108         init_completion(&req.done);
 109
 110         spin_lock(&req_lock);
 111         req.next = requests;
 112         requests = &req;
 113         spin_unlock(&req_lock);
 114
 115         wake_up_process(thread);
 116         wait_for_completion(&req.done);
 117
 118         kfree(tmp);
 119
 120         return req.err;
 121 }
 122
 123 int devtmpfs_delete_node(struct device *dev)
 124 {
 125         const char *tmp = NULL;
 126         struct req req;
 127
 128         if (!thread)
 129                 return 0;
 130
 131         req.name = device_get_devnode(dev, NULL, NULL, NULL, &tmp);
 132         if (!req.name)
 133                 return -ENOMEM;
 134
 135         req.mode = 0;
 136         req.dev = dev;
 137
 138         init_completion(&req.done);
 139
 140         spin_lock(&req_lock);
 141         req.next = requests;
 142         requests = &req;
 143         spin_unlock(&req_lock);
 144
 145         wake_up_process(thread);
 146         wait_for_completion(&req.done);
 147
 148         kfree(tmp);
 149         return req.err;
 150 }
 151
 152 static int dev_mkdir(const char *name, umode_t mode)
 153 {
 154         struct dentry *dentry;
 155         struct path path;
 156         int err;
 157
 158         dentry = kern_path_create(AT_FDCWD, name, &path, LOOKUP_DIRECTORY);
 159         if (IS_ERR(dentry))
 160                 return PTR_ERR(dentry);
 161
 162         err = vfs_mkdir(d_inode(path.dentry), dentry, mode);
 163         if (!err)
 164                 /* mark as kernel-created inode */
 165                 d_inode(dentry)->i_private = &thread;
 166         done_path_create(&path, dentry);
 167         return err;
 168 }
 169
 170 static int create_path(const char *nodepath)
 171 {
 172         char *path;
 173         char *s;
 174         int err = 0;
 175
 176         /* parent directories do not exist, create them */
 177         path = kstrdup(nodepath, GFP_KERNEL);
 178         if (!path)
 179                 return -ENOMEM;
 180
 181         s = path;
 182         for (;;) {
 183                 s = strchr(s, '/');
 184                 if (!s)
 185                         break;
 186                 s[0] = '\0';
 187                 err = dev_mkdir(path, 0755);
 188                 if (err && err != -EEXIST)
 189                         break;
 190                 s[0] = '/';
 191                 s++;
 192         }
 193         kfree(path);
 194         return err;
 195 }
 196
 197 static int handle_create(const char *nodename, umode_t mode, kuid_t uid,
 198                          kgid_t gid, 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(d_inode(path.dentry), dentry, mode, dev->devt);
 213         if (!err) {
 214                 struct iattr newattrs;
 215
 216                 newattrs.ia_mode = mode;
 217                 newattrs.ia_uid = uid;
 218                 newattrs.ia_gid = gid;
 219                 newattrs.ia_valid = ATTR_MODE|ATTR_UID|ATTR_GID;
 220                 inode_lock(d_inode(dentry));
 221                 notify_change(dentry, &newattrs, NULL);
 222                 inode_unlock(d_inode(dentry));
 223
 224                 /* mark as kernel-created inode */
 225                 d_inode(dentry)->i_private = &thread;
 226         }
 227         done_path_create(&path, dentry);
 228         return err;
 229 }
 230
 231 static int dev_rmdir(const char *name)
 232 {
 233         struct path parent;
 234         struct dentry *dentry;
 235         int err;
 236
 237         dentry = kern_path_locked(name, &parent);
 238         if (IS_ERR(dentry))
 239                 return PTR_ERR(dentry);
 240         if (d_really_is_positive(dentry)) {
 241                 if (d_inode(dentry)->i_private == &thread)
 242                         err = vfs_rmdir(d_inode(parent.dentry), dentry);
 243                 else
 244                         err = -EPERM;
 245         } else {
 246                 err = -ENOENT;
 247         }
 248         dput(dentry);
 249         inode_unlock(d_inode(parent.dentry));
 250         path_put(&parent);
 251         return err;
 252 }
 253
 254 static int delete_path(const char *nodepath)
 255 {
 256         char *path;
 257         int err = 0;
 258
 259         path = kstrdup(nodepath, GFP_KERNEL);
 260         if (!path)
 261                 return -ENOMEM;
 262
 263         for (;;) {
 264                 char *base;
 265
 266                 base = strrchr(path, '/');
 267                 if (!base)
 268                         break;
 269                 base[0] = '\0';
 270                 err = dev_rmdir(path);
 271                 if (err)
 272                         break;
 273         }
 274
 275         kfree(path);
 276         return err;
 277 }
 278
 279 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
 280 {
 281         /* did we create it */
 282         if (inode->i_private != &thread)
 283                 return 0;
 284
 285         /* does the dev_t match */
 286         if (is_blockdev(dev)) {
 287                 if (!S_ISBLK(stat->mode))
 288                         return 0;
 289         } else {
 290                 if (!S_ISCHR(stat->mode))
 291                         return 0;
 292         }
 293         if (stat->rdev != dev->devt)
 294                 return 0;
 295
 296         /* ours */
 297         return 1;
 298 }
 299
 300 static int handle_remove(const char *nodename, struct device *dev)
 301 {
 302         struct path parent;
 303         struct dentry *dentry;
 304         int deleted = 0;
 305         int err;
 306
 307         dentry = kern_path_locked(nodename, &parent);
 308         if (IS_ERR(dentry))
 309                 return PTR_ERR(dentry);
 310
 311         if (d_really_is_positive(dentry)) {
 312                 struct kstat stat;
 313                 struct path p = {.mnt = parent.mnt, .dentry = dentry};
 314                 err = vfs_getattr(&p, &stat, STATX_TYPE | STATX_MODE,
 315                                   AT_STATX_SYNC_AS_STAT);
 316                 if (!err && dev_mynode(dev, d_inode(dentry), &stat)) {
 317                         struct iattr newattrs;
 318                         /*
 319                          * before unlinking this node, reset permissions
 320                          * of possible references like hardlinks
 321                          */
 322                         newattrs.ia_uid = GLOBAL_ROOT_UID;
 323                         newattrs.ia_gid = GLOBAL_ROOT_GID;
 324                         newattrs.ia_mode = stat.mode & ~0777;
 325                         newattrs.ia_valid =
 326                                 ATTR_UID|ATTR_GID|ATTR_MODE;
 327                         inode_lock(d_inode(dentry));
 328                         notify_change(dentry, &newattrs, NULL);
 329                         inode_unlock(d_inode(dentry));
 330                         err = vfs_unlink(d_inode(parent.dentry), dentry, NULL);
 331                         if (!err || err == -ENOENT)
 332                                 deleted = 1;
 333                 }
 334         } else {
 335                 err = -ENOENT;
 336         }
 337         dput(dentry);
 338         inode_unlock(d_inode(parent.dentry));
 339
 340         path_put(&parent);
 341         if (deleted && strchr(nodename, '/'))
 342                 delete_path(nodename);
 343         return err;
 344 }
 345
 346 /*
 347  * If configured, or requested by the commandline, devtmpfs will be
 348  * auto-mounted after the kernel mounted the root filesystem.
 349  */
 350 int devtmpfs_mount(const char *mntdir)
 351 {
 352         int err;
 353
 354         if (!mount_dev)
 355                 return 0;
 356
 357         if (!thread)
 358                 return 0;
 359
 360         err = ksys_mount("devtmpfs", (char *)mntdir, "devtmpfs", MS_SILENT,
 361                          NULL);
 362         if (err)
 363                 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
 364         else
 365                 printk(KERN_INFO "devtmpfs: mounted\n");
 366         return err;
 367 }
 368
 369 static DECLARE_COMPLETION(setup_done);
 370
 371 static int handle(const char *name, umode_t mode, kuid_t uid, kgid_t gid,
 372                   struct device *dev)
 373 {
 374         if (mode)
 375                 return handle_create(name, mode, uid, gid, dev);
 376         else
 377                 return handle_remove(name, dev);
 378 }
 379
 380 static int devtmpfsd(void *p)
 381 {
 382         char options[] = "mode=0755";
 383         int *err = p;
 384         *err = ksys_unshare(CLONE_NEWNS);
 385         if (*err)
 386                 goto out;
 387         *err = ksys_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, options);
 388         if (*err)
 389                 goto out;
 390         ksys_chdir("/.."); /* will traverse into overmounted root */
 391         ksys_chroot(".");
 392         complete(&setup_done);
 393         while (1) {
 394                 spin_lock(&req_lock);
 395                 while (requests) {
 396                         struct req *req = requests;
 397                         requests = NULL;
 398                         spin_unlock(&req_lock);
 399                         while (req) {
 400                                 struct req *next = req->next;
 401                                 req->err = handle(req->name, req->mode,
 402                                                   req->uid, req->gid, req->dev);
 403                                 complete(&req->done);
 404                                 req = next;
 405                         }
 406                         spin_lock(&req_lock);
 407                 }
 408                 __set_current_state(TASK_INTERRUPTIBLE);
 409                 spin_unlock(&req_lock);
 410                 schedule();
 411         }
 412         return 0;
 413 out:
 414         complete(&setup_done);
 415         return *err;
 416 }
 417
 418 /*
 419  * Create devtmpfs instance, driver-core devices will add their device
 420  * nodes here.
 421  */
 422 int __init devtmpfs_init(void)
 423 {
 424         int err = register_filesystem(&dev_fs_type);
 425         if (err) {
 426                 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
 427                        "type %i\n", err);
 428                 return err;
 429         }
 430
 431         thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
 432         if (!IS_ERR(thread)) {
 433                 wait_for_completion(&setup_done);
 434         } else {
 435                 err = PTR_ERR(thread);
 436                 thread = NULL;
 437         }
 438
 439         if (err) {
 440                 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
 441                 unregister_filesystem(&dev_fs_type);
 442                 return err;
 443         }
 444
 445         printk(KERN_INFO "devtmpfs: initialized\n");
 446         return 0;
 447 }