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