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 <linux/init_syscalls.h>
  29 #include <uapi/linux/mount.h>
  30 #include "base.h"
  31
  32 static struct task_struct *thread;
  33
  34 static int __initdata mount_dev = IS_ENABLED(CONFIG_DEVTMPFS_MOUNT);
  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         kuid_t uid;
  45         kgid_t gid;
  46         struct device *dev;
  47 } *requests;
  48
  49 static int __init mount_param(char *str)
  50 {
  51         mount_dev = simple_strtoul(str, NULL, 0);
  52         return 1;
  53 }
  54 __setup("devtmpfs.mount=", mount_param);
  55
  56 static struct vfsmount *mnt;
  57
  58 static struct dentry *public_dev_mount(struct file_system_type *fs_type, int flags,
  59                       const char *dev_name, void *data)
  60 {
  61         struct super_block *s = mnt->mnt_sb;
  62         atomic_inc(&s->s_active);
  63         down_write(&s->s_umount);
  64         return dget(s->s_root);
  65 }
  66
  67 static struct file_system_type internal_fs_type = {
  68         .name = "devtmpfs",
  69 #ifdef CONFIG_TMPFS
  70         .init_fs_context = shmem_init_fs_context,
  71         .parameters     = shmem_fs_parameters,
  72 #else
  73         .init_fs_context = ramfs_init_fs_context,
  74         .parameters     = ramfs_fs_parameters,
  75 #endif
  76         .kill_sb = kill_litter_super,
  77 };
  78
  79 static struct file_system_type dev_fs_type = {
  80         .name = "devtmpfs",
  81         .mount = public_dev_mount,
  82 };
  83
  84 #ifdef CONFIG_BLOCK
  85 static inline int is_blockdev(struct device *dev)
  86 {
  87         return dev->class == &block_class;
  88 }
  89 #else
  90 static inline int is_blockdev(struct device *dev) { return 0; }
  91 #endif
  92
  93 static int devtmpfs_submit_req(struct req *req, const char *tmp)
  94 {
  95         init_completion(&req->done);
  96
  97         spin_lock(&req_lock);
  98         req->next = requests;
  99         requests = req;
 100         spin_unlock(&req_lock);
 101
 102         wake_up_process(thread);
 103         wait_for_completion(&req->done);
 104
 105         kfree(tmp);
 106
 107         return req->err;
 108 }
 109
 110 int devtmpfs_create_node(struct device *dev)
 111 {
 112         const char *tmp = NULL;
 113         struct req req;
 114
 115         if (!thread)
 116                 return 0;
 117
 118         req.mode = 0;
 119         req.uid = GLOBAL_ROOT_UID;
 120         req.gid = GLOBAL_ROOT_GID;
 121         req.name = device_get_devnode(dev, &req.mode, &req.uid, &req.gid, &tmp);
 122         if (!req.name)
 123                 return -ENOMEM;
 124
 125         if (req.mode == 0)
 126                 req.mode = 0600;
 127         if (is_blockdev(dev))
 128                 req.mode |= S_IFBLK;
 129         else
 130                 req.mode |= S_IFCHR;
 131
 132         req.dev = dev;
 133
 134         return devtmpfs_submit_req(&req, tmp);
 135 }
 136
 137 int devtmpfs_delete_node(struct device *dev)
 138 {
 139         const char *tmp = NULL;
 140         struct req req;
 141
 142         if (!thread)
 143                 return 0;
 144
 145         req.name = device_get_devnode(dev, NULL, NULL, NULL, &tmp);
 146         if (!req.name)
 147                 return -ENOMEM;
 148
 149         req.mode = 0;
 150         req.dev = dev;
 151
 152         return devtmpfs_submit_req(&req, tmp);
 153 }
 154
 155 static int dev_mkdir(const char *name, umode_t mode)
 156 {
 157         struct dentry *dentry;
 158         struct path path;
 159         int err;
 160
 161         dentry = kern_path_create(AT_FDCWD, name, &path, LOOKUP_DIRECTORY);
 162         if (IS_ERR(dentry))
 163                 return PTR_ERR(dentry);
 164
 165         err = vfs_mkdir(d_inode(path.dentry), dentry, mode);
 166         if (!err)
 167                 /* mark as kernel-created inode */
 168                 d_inode(dentry)->i_private = &thread;
 169         done_path_create(&path, dentry);
 170         return err;
 171 }
 172
 173 static int create_path(const char *nodepath)
 174 {
 175         char *path;
 176         char *s;
 177         int err = 0;
 178
 179         /* parent directories do not exist, create them */
 180         path = kstrdup(nodepath, GFP_KERNEL);
 181         if (!path)
 182                 return -ENOMEM;
 183
 184         s = path;
 185         for (;;) {
 186                 s = strchr(s, '/');
 187                 if (!s)
 188                         break;
 189                 s[0] = '\0';
 190                 err = dev_mkdir(path, 0755);
 191                 if (err && err != -EEXIST)
 192                         break;
 193                 s[0] = '/';
 194                 s++;
 195         }
 196         kfree(path);
 197         return err;
 198 }
 199
 200 static int handle_create(const char *nodename, umode_t mode, kuid_t uid,
 201                          kgid_t gid, struct device *dev)
 202 {
 203         struct dentry *dentry;
 204         struct path path;
 205         int err;
 206
 207         dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
 208         if (dentry == ERR_PTR(-ENOENT)) {
 209                 create_path(nodename);
 210                 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
 211         }
 212         if (IS_ERR(dentry))
 213                 return PTR_ERR(dentry);
 214
 215         err = vfs_mknod(d_inode(path.dentry), dentry, mode, dev->devt);
 216         if (!err) {
 217                 struct iattr newattrs;
 218
 219                 newattrs.ia_mode = mode;
 220                 newattrs.ia_uid = uid;
 221                 newattrs.ia_gid = gid;
 222                 newattrs.ia_valid = ATTR_MODE|ATTR_UID|ATTR_GID;
 223                 inode_lock(d_inode(dentry));
 224                 notify_change(dentry, &newattrs, NULL);
 225                 inode_unlock(d_inode(dentry));
 226
 227                 /* mark as kernel-created inode */
 228                 d_inode(dentry)->i_private = &thread;
 229         }
 230         done_path_create(&path, dentry);
 231         return err;
 232 }
 233
 234 static int dev_rmdir(const char *name)
 235 {
 236         struct path parent;
 237         struct dentry *dentry;
 238         int err;
 239
 240         dentry = kern_path_locked(name, &parent);
 241         if (IS_ERR(dentry))
 242                 return PTR_ERR(dentry);
 243         if (d_really_is_positive(dentry)) {
 244                 if (d_inode(dentry)->i_private == &thread)
 245                         err = vfs_rmdir(d_inode(parent.dentry), dentry);
 246                 else
 247                         err = -EPERM;
 248         } else {
 249                 err = -ENOENT;
 250         }
 251         dput(dentry);
 252         inode_unlock(d_inode(parent.dentry));
 253         path_put(&parent);
 254         return err;
 255 }
 256
 257 static int delete_path(const char *nodepath)
 258 {
 259         char *path;
 260         int err = 0;
 261
 262         path = kstrdup(nodepath, GFP_KERNEL);
 263         if (!path)
 264                 return -ENOMEM;
 265
 266         for (;;) {
 267                 char *base;
 268
 269                 base = strrchr(path, '/');
 270                 if (!base)
 271                         break;
 272                 base[0] = '\0';
 273                 err = dev_rmdir(path);
 274                 if (err)
 275                         break;
 276         }
 277
 278         kfree(path);
 279         return err;
 280 }
 281
 282 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat)
 283 {
 284         /* did we create it */
 285         if (inode->i_private != &thread)
 286                 return 0;
 287
 288         /* does the dev_t match */
 289         if (is_blockdev(dev)) {
 290                 if (!S_ISBLK(stat->mode))
 291                         return 0;
 292         } else {
 293                 if (!S_ISCHR(stat->mode))
 294                         return 0;
 295         }
 296         if (stat->rdev != dev->devt)
 297                 return 0;
 298
 299         /* ours */
 300         return 1;
 301 }
 302
 303 static int handle_remove(const char *nodename, struct device *dev)
 304 {
 305         struct path parent;
 306         struct dentry *dentry;
 307         int deleted = 0;
 308         int err;
 309
 310         dentry = kern_path_locked(nodename, &parent);
 311         if (IS_ERR(dentry))
 312                 return PTR_ERR(dentry);
 313
 314         if (d_really_is_positive(dentry)) {
 315                 struct kstat stat;
 316                 struct path p = {.mnt = parent.mnt, .dentry = dentry};
 317                 err = vfs_getattr(&p, &stat, STATX_TYPE | STATX_MODE,
 318                                   AT_STATX_SYNC_AS_STAT);
 319                 if (!err && dev_mynode(dev, d_inode(dentry), &stat)) {
 320                         struct iattr newattrs;
 321                         /*
 322                          * before unlinking this node, reset permissions
 323                          * of possible references like hardlinks
 324                          */
 325                         newattrs.ia_uid = GLOBAL_ROOT_UID;
 326                         newattrs.ia_gid = GLOBAL_ROOT_GID;
 327                         newattrs.ia_mode = stat.mode & ~0777;
 328                         newattrs.ia_valid =
 329                                 ATTR_UID|ATTR_GID|ATTR_MODE;
 330                         inode_lock(d_inode(dentry));
 331                         notify_change(dentry, &newattrs, NULL);
 332                         inode_unlock(d_inode(dentry));
 333                         err = vfs_unlink(d_inode(parent.dentry), dentry, NULL);
 334                         if (!err || err == -ENOENT)
 335                                 deleted = 1;
 336                 }
 337         } else {
 338                 err = -ENOENT;
 339         }
 340         dput(dentry);
 341         inode_unlock(d_inode(parent.dentry));
 342
 343         path_put(&parent);
 344         if (deleted && strchr(nodename, '/'))
 345                 delete_path(nodename);
 346         return err;
 347 }
 348
 349 /*
 350  * If configured, or requested by the commandline, devtmpfs will be
 351  * auto-mounted after the kernel mounted the root filesystem.
 352  */
 353 int __init devtmpfs_mount(void)
 354 {
 355         int err;
 356
 357         if (!mount_dev)
 358                 return 0;
 359
 360         if (!thread)
 361                 return 0;
 362
 363         err = init_mount("devtmpfs", "dev", "devtmpfs", MS_SILENT, NULL);
 364         if (err)
 365                 printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
 366         else
 367                 printk(KERN_INFO "devtmpfs: mounted\n");
 368         return err;
 369 }
 370
 371 static DECLARE_COMPLETION(setup_done);
 372
 373 static int handle(const char *name, umode_t mode, kuid_t uid, kgid_t gid,
 374                   struct device *dev)
 375 {
 376         if (mode)
 377                 return handle_create(name, mode, uid, gid, dev);
 378         else
 379                 return handle_remove(name, dev);
 380 }
 381
 382 static void __noreturn devtmpfs_work_loop(void)
 383 {
 384         while (1) {
 385                 spin_lock(&req_lock);
 386                 while (requests) {
 387                         struct req *req = requests;
 388                         requests = NULL;
 389                         spin_unlock(&req_lock);
 390                         while (req) {
 391                                 struct req *next = req->next;
 392                                 req->err = handle(req->name, req->mode,
 393                                                   req->uid, req->gid, req->dev);
 394                                 complete(&req->done);
 395                                 req = next;
 396                         }
 397                         spin_lock(&req_lock);
 398                 }
 399                 __set_current_state(TASK_INTERRUPTIBLE);
 400                 spin_unlock(&req_lock);
 401                 schedule();
 402         }
 403 }
 404
 405 static int __init devtmpfs_setup(void *p)
 406 {
 407         int err;
 408
 409         err = ksys_unshare(CLONE_NEWNS);
 410         if (err)
 411                 goto out;
 412         err = init_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, NULL);
 413         if (err)
 414                 goto out;
 415         init_chdir("/.."); /* will traverse into overmounted root */
 416         init_chroot(".");
 417 out:
 418         *(int *)p = err;
 419         return err;
 420 }
 421
 422 /*
 423  * The __ref is because devtmpfs_setup needs to be __init for the routines it
 424  * calls.  That call is done while devtmpfs_init, which is marked __init,
 425  * synchronously waits for it to complete.
 426  */
 427 static int __ref devtmpfsd(void *p)
 428 {
 429         int err = devtmpfs_setup(p);
 430
 431         complete(&setup_done);
 432         if (err)
 433                 return err;
 434         devtmpfs_work_loop();
 435         return 0;
 436 }
 437
 438 /*
 439  * Create devtmpfs instance, driver-core devices will add their device
 440  * nodes here.
 441  */
 442 int __init devtmpfs_init(void)
 443 {
 444         char opts[] = "mode=0755";
 445         int err;
 446
 447         mnt = vfs_kern_mount(&internal_fs_type, 0, "devtmpfs", opts);
 448         if (IS_ERR(mnt)) {
 449                 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %ld\n",
 450                                 PTR_ERR(mnt));
 451                 return PTR_ERR(mnt);
 452         }
 453         err = register_filesystem(&dev_fs_type);
 454         if (err) {
 455                 printk(KERN_ERR "devtmpfs: unable to register devtmpfs "
 456                        "type %i\n", err);
 457                 return err;
 458         }
 459
 460         thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
 461         if (!IS_ERR(thread)) {
 462                 wait_for_completion(&setup_done);
 463         } else {
 464                 err = PTR_ERR(thread);
 465                 thread = NULL;
 466         }
 467
 468         if (err) {
 469                 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
 470                 unregister_filesystem(&dev_fs_type);
 471                 return err;
 472         }
 473
 474         printk(KERN_INFO "devtmpfs: initialized\n");
 475         return 0;
 476 }