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/blkdev.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 #ifdef CONFIG_DEVTMPFS_SAFE 33 #define DEVTMPFS_MFLAGS (MS_SILENT | MS_NOEXEC | MS_NOSUID) 34 #else 35 #define DEVTMPFS_MFLAGS (MS_SILENT) 36 #endif 37 38 static struct task_struct *thread; 39 40 static int __initdata mount_dev = IS_ENABLED(CONFIG_DEVTMPFS_MOUNT); 41 42 static DEFINE_SPINLOCK(req_lock); 43 44 static struct req { 45 struct req *next; 46 struct completion done; 47 int err; 48 const char *name; 49 umode_t mode; /* 0 => delete */ 50 kuid_t uid; 51 kgid_t gid; 52 struct device *dev; 53 } *requests; 54 55 static int __init mount_param(char *str) 56 { 57 mount_dev = simple_strtoul(str, NULL, 0); 58 return 1; 59 } 60 __setup("devtmpfs.mount=", mount_param); 61 62 static struct vfsmount *mnt; 63 64 static struct dentry *public_dev_mount(struct file_system_type *fs_type, int flags, 65 const char *dev_name, void *data) 66 { 67 struct super_block *s = mnt->mnt_sb; 68 int err; 69 70 atomic_inc(&s->s_active); 71 down_write(&s->s_umount); 72 err = reconfigure_single(s, flags, data); 73 if (err < 0) { 74 deactivate_locked_super(s); 75 return ERR_PTR(err); 76 } 77 return dget(s->s_root); 78 } 79 80 static struct file_system_type internal_fs_type = { 81 .name = "devtmpfs", 82 #ifdef CONFIG_TMPFS 83 .init_fs_context = shmem_init_fs_context, 84 #else 85 .init_fs_context = ramfs_init_fs_context, 86 #endif 87 .kill_sb = kill_litter_super, 88 }; 89 90 static struct file_system_type dev_fs_type = { 91 .name = "devtmpfs", 92 .mount = public_dev_mount, 93 }; 94 95 #ifdef CONFIG_BLOCK 96 static inline int is_blockdev(struct device *dev) 97 { 98 return dev->class == &block_class; 99 } 100 #else 101 static inline int is_blockdev(struct device *dev) { return 0; } 102 #endif 103 104 static int devtmpfs_submit_req(struct req *req, const char *tmp) 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_create_node(struct device *dev) 122 { 123 const char *tmp = NULL; 124 struct req req; 125 126 if (!thread) 127 return 0; 128 129 req.mode = 0; 130 req.uid = GLOBAL_ROOT_UID; 131 req.gid = GLOBAL_ROOT_GID; 132 req.name = device_get_devnode(dev, &req.mode, &req.uid, &req.gid, &tmp); 133 if (!req.name) 134 return -ENOMEM; 135 136 if (req.mode == 0) 137 req.mode = 0600; 138 if (is_blockdev(dev)) 139 req.mode |= S_IFBLK; 140 else 141 req.mode |= S_IFCHR; 142 143 req.dev = dev; 144 145 return devtmpfs_submit_req(&req, tmp); 146 } 147 148 int devtmpfs_delete_node(struct device *dev) 149 { 150 const char *tmp = NULL; 151 struct req req; 152 153 if (!thread) 154 return 0; 155 156 req.name = device_get_devnode(dev, NULL, NULL, NULL, &tmp); 157 if (!req.name) 158 return -ENOMEM; 159 160 req.mode = 0; 161 req.dev = dev; 162 163 return devtmpfs_submit_req(&req, tmp); 164 } 165 166 static int dev_mkdir(const char *name, umode_t mode) 167 { 168 struct dentry *dentry; 169 struct path path; 170 int err; 171 172 dentry = kern_path_create(AT_FDCWD, name, &path, LOOKUP_DIRECTORY); 173 if (IS_ERR(dentry)) 174 return PTR_ERR(dentry); 175 176 err = vfs_mkdir(&init_user_ns, d_inode(path.dentry), dentry, mode); 177 if (!err) 178 /* mark as kernel-created inode */ 179 d_inode(dentry)->i_private = &thread; 180 done_path_create(&path, dentry); 181 return err; 182 } 183 184 static int create_path(const char *nodepath) 185 { 186 char *path; 187 char *s; 188 int err = 0; 189 190 /* parent directories do not exist, create them */ 191 path = kstrdup(nodepath, GFP_KERNEL); 192 if (!path) 193 return -ENOMEM; 194 195 s = path; 196 for (;;) { 197 s = strchr(s, '/'); 198 if (!s) 199 break; 200 s[0] = '\0'; 201 err = dev_mkdir(path, 0755); 202 if (err && err != -EEXIST) 203 break; 204 s[0] = '/'; 205 s++; 206 } 207 kfree(path); 208 return err; 209 } 210 211 static int handle_create(const char *nodename, umode_t mode, kuid_t uid, 212 kgid_t gid, struct device *dev) 213 { 214 struct dentry *dentry; 215 struct path path; 216 int err; 217 218 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0); 219 if (dentry == ERR_PTR(-ENOENT)) { 220 create_path(nodename); 221 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0); 222 } 223 if (IS_ERR(dentry)) 224 return PTR_ERR(dentry); 225 226 err = vfs_mknod(&init_user_ns, d_inode(path.dentry), dentry, mode, 227 dev->devt); 228 if (!err) { 229 struct iattr newattrs; 230 231 newattrs.ia_mode = mode; 232 newattrs.ia_uid = uid; 233 newattrs.ia_gid = gid; 234 newattrs.ia_valid = ATTR_MODE|ATTR_UID|ATTR_GID; 235 inode_lock(d_inode(dentry)); 236 notify_change(&init_user_ns, dentry, &newattrs, NULL); 237 inode_unlock(d_inode(dentry)); 238 239 /* mark as kernel-created inode */ 240 d_inode(dentry)->i_private = &thread; 241 } 242 done_path_create(&path, dentry); 243 return err; 244 } 245 246 static int dev_rmdir(const char *name) 247 { 248 struct path parent; 249 struct dentry *dentry; 250 int err; 251 252 dentry = kern_path_locked(name, &parent); 253 if (IS_ERR(dentry)) 254 return PTR_ERR(dentry); 255 if (d_really_is_positive(dentry)) { 256 if (d_inode(dentry)->i_private == &thread) 257 err = vfs_rmdir(&init_user_ns, d_inode(parent.dentry), 258 dentry); 259 else 260 err = -EPERM; 261 } else { 262 err = -ENOENT; 263 } 264 dput(dentry); 265 inode_unlock(d_inode(parent.dentry)); 266 path_put(&parent); 267 return err; 268 } 269 270 static int delete_path(const char *nodepath) 271 { 272 char *path; 273 int err = 0; 274 275 path = kstrdup(nodepath, GFP_KERNEL); 276 if (!path) 277 return -ENOMEM; 278 279 for (;;) { 280 char *base; 281 282 base = strrchr(path, '/'); 283 if (!base) 284 break; 285 base[0] = '\0'; 286 err = dev_rmdir(path); 287 if (err) 288 break; 289 } 290 291 kfree(path); 292 return err; 293 } 294 295 static int dev_mynode(struct device *dev, struct inode *inode, struct kstat *stat) 296 { 297 /* did we create it */ 298 if (inode->i_private != &thread) 299 return 0; 300 301 /* does the dev_t match */ 302 if (is_blockdev(dev)) { 303 if (!S_ISBLK(stat->mode)) 304 return 0; 305 } else { 306 if (!S_ISCHR(stat->mode)) 307 return 0; 308 } 309 if (stat->rdev != dev->devt) 310 return 0; 311 312 /* ours */ 313 return 1; 314 } 315 316 static int handle_remove(const char *nodename, struct device *dev) 317 { 318 struct path parent; 319 struct dentry *dentry; 320 int deleted = 0; 321 int err; 322 323 dentry = kern_path_locked(nodename, &parent); 324 if (IS_ERR(dentry)) 325 return PTR_ERR(dentry); 326 327 if (d_really_is_positive(dentry)) { 328 struct kstat stat; 329 struct path p = {.mnt = parent.mnt, .dentry = dentry}; 330 err = vfs_getattr(&p, &stat, STATX_TYPE | STATX_MODE, 331 AT_STATX_SYNC_AS_STAT); 332 if (!err && dev_mynode(dev, d_inode(dentry), &stat)) { 333 struct iattr newattrs; 334 /* 335 * before unlinking this node, reset permissions 336 * of possible references like hardlinks 337 */ 338 newattrs.ia_uid = GLOBAL_ROOT_UID; 339 newattrs.ia_gid = GLOBAL_ROOT_GID; 340 newattrs.ia_mode = stat.mode & ~0777; 341 newattrs.ia_valid = 342 ATTR_UID|ATTR_GID|ATTR_MODE; 343 inode_lock(d_inode(dentry)); 344 notify_change(&init_user_ns, dentry, &newattrs, NULL); 345 inode_unlock(d_inode(dentry)); 346 err = vfs_unlink(&init_user_ns, d_inode(parent.dentry), 347 dentry, NULL); 348 if (!err || err == -ENOENT) 349 deleted = 1; 350 } 351 } else { 352 err = -ENOENT; 353 } 354 dput(dentry); 355 inode_unlock(d_inode(parent.dentry)); 356 357 path_put(&parent); 358 if (deleted && strchr(nodename, '/')) 359 delete_path(nodename); 360 return err; 361 } 362 363 /* 364 * If configured, or requested by the commandline, devtmpfs will be 365 * auto-mounted after the kernel mounted the root filesystem. 366 */ 367 int __init devtmpfs_mount(void) 368 { 369 int err; 370 371 if (!mount_dev) 372 return 0; 373 374 if (!thread) 375 return 0; 376 377 err = init_mount("devtmpfs", "dev", "devtmpfs", DEVTMPFS_MFLAGS, NULL); 378 if (err) 379 printk(KERN_INFO "devtmpfs: error mounting %i\n", err); 380 else 381 printk(KERN_INFO "devtmpfs: mounted\n"); 382 return err; 383 } 384 385 static __initdata DECLARE_COMPLETION(setup_done); 386 387 static int handle(const char *name, umode_t mode, kuid_t uid, kgid_t gid, 388 struct device *dev) 389 { 390 if (mode) 391 return handle_create(name, mode, uid, gid, dev); 392 else 393 return handle_remove(name, dev); 394 } 395 396 static void __noreturn devtmpfs_work_loop(void) 397 { 398 while (1) { 399 spin_lock(&req_lock); 400 while (requests) { 401 struct req *req = requests; 402 requests = NULL; 403 spin_unlock(&req_lock); 404 while (req) { 405 struct req *next = req->next; 406 req->err = handle(req->name, req->mode, 407 req->uid, req->gid, req->dev); 408 complete(&req->done); 409 req = next; 410 } 411 spin_lock(&req_lock); 412 } 413 __set_current_state(TASK_INTERRUPTIBLE); 414 spin_unlock(&req_lock); 415 schedule(); 416 } 417 } 418 419 static noinline int __init devtmpfs_setup(void *p) 420 { 421 int err; 422 423 err = ksys_unshare(CLONE_NEWNS); 424 if (err) 425 goto out; 426 err = init_mount("devtmpfs", "/", "devtmpfs", DEVTMPFS_MFLAGS, NULL); 427 if (err) 428 goto out; 429 init_chdir("/.."); /* will traverse into overmounted root */ 430 init_chroot("."); 431 out: 432 *(int *)p = err; 433 return err; 434 } 435 436 /* 437 * The __ref is because devtmpfs_setup needs to be __init for the routines it 438 * calls. That call is done while devtmpfs_init, which is marked __init, 439 * synchronously waits for it to complete. 440 */ 441 static int __ref devtmpfsd(void *p) 442 { 443 int err = devtmpfs_setup(p); 444 445 complete(&setup_done); 446 if (err) 447 return err; 448 devtmpfs_work_loop(); 449 return 0; 450 } 451 452 /* 453 * Create devtmpfs instance, driver-core devices will add their device 454 * nodes here. 455 */ 456 int __init devtmpfs_init(void) 457 { 458 char opts[] = "mode=0755"; 459 int err; 460 461 mnt = vfs_kern_mount(&internal_fs_type, 0, "devtmpfs", opts); 462 if (IS_ERR(mnt)) { 463 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %ld\n", 464 PTR_ERR(mnt)); 465 return PTR_ERR(mnt); 466 } 467 err = register_filesystem(&dev_fs_type); 468 if (err) { 469 printk(KERN_ERR "devtmpfs: unable to register devtmpfs " 470 "type %i\n", err); 471 return err; 472 } 473 474 thread = kthread_run(devtmpfsd, &err, "kdevtmpfs"); 475 if (!IS_ERR(thread)) { 476 wait_for_completion(&setup_done); 477 } else { 478 err = PTR_ERR(thread); 479 thread = NULL; 480 } 481 482 if (err) { 483 printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err); 484 unregister_filesystem(&dev_fs_type); 485 thread = NULL; 486 return err; 487 } 488 489 printk(KERN_INFO "devtmpfs: initialized\n"); 490 return 0; 491 } 492