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 #define pr_fmt(fmt) "devtmpfs: " fmt 17 18 #include <linux/kernel.h> 19 #include <linux/syscalls.h> 20 #include <linux/mount.h> 21 #include <linux/device.h> 22 #include <linux/blkdev.h> 23 #include <linux/namei.h> 24 #include <linux/fs.h> 25 #include <linux/shmem_fs.h> 26 #include <linux/ramfs.h> 27 #include <linux/sched.h> 28 #include <linux/slab.h> 29 #include <linux/kthread.h> 30 #include <linux/init_syscalls.h> 31 #include <uapi/linux/mount.h> 32 #include "base.h" 33 34 #ifdef CONFIG_DEVTMPFS_SAFE 35 #define DEVTMPFS_MFLAGS (MS_SILENT | MS_NOEXEC | MS_NOSUID) 36 #else 37 #define DEVTMPFS_MFLAGS (MS_SILENT) 38 #endif 39 40 static struct task_struct *thread; 41 42 static int __initdata mount_dev = IS_ENABLED(CONFIG_DEVTMPFS_MOUNT); 43 44 static DEFINE_SPINLOCK(req_lock); 45 46 static struct req { 47 struct req *next; 48 struct completion done; 49 int err; 50 const char *name; 51 umode_t mode; /* 0 => delete */ 52 kuid_t uid; 53 kgid_t gid; 54 struct device *dev; 55 } *requests; 56 57 static int __init mount_param(char *str) 58 { 59 mount_dev = simple_strtoul(str, NULL, 0); 60 return 1; 61 } 62 __setup("devtmpfs.mount=", mount_param); 63 64 static struct vfsmount *mnt; 65 66 static struct file_system_type internal_fs_type = { 67 .name = "devtmpfs", 68 #ifdef CONFIG_TMPFS 69 .init_fs_context = shmem_init_fs_context, 70 #else 71 .init_fs_context = ramfs_init_fs_context, 72 #endif 73 .kill_sb = kill_litter_super, 74 }; 75 76 /* Simply take a ref on the existing mount */ 77 static int devtmpfs_get_tree(struct fs_context *fc) 78 { 79 struct super_block *sb = mnt->mnt_sb; 80 81 atomic_inc(&sb->s_active); 82 down_write(&sb->s_umount); 83 fc->root = dget(sb->s_root); 84 return 0; 85 } 86 87 /* Ops are filled in during init depending on underlying shmem or ramfs type */ 88 struct fs_context_operations devtmpfs_context_ops = {}; 89 90 /* Call the underlying initialization and set to our ops */ 91 static int devtmpfs_init_fs_context(struct fs_context *fc) 92 { 93 int ret; 94 #ifdef CONFIG_TMPFS 95 ret = shmem_init_fs_context(fc); 96 #else 97 ret = ramfs_init_fs_context(fc); 98 #endif 99 if (ret < 0) 100 return ret; 101 102 fc->ops = &devtmpfs_context_ops; 103 104 return 0; 105 } 106 107 static struct file_system_type dev_fs_type = { 108 .name = "devtmpfs", 109 .init_fs_context = devtmpfs_init_fs_context, 110 }; 111 112 static int devtmpfs_submit_req(struct req *req, const char *tmp) 113 { 114 init_completion(&req->done); 115 116 spin_lock(&req_lock); 117 req->next = requests; 118 requests = req; 119 spin_unlock(&req_lock); 120 121 wake_up_process(thread); 122 wait_for_completion(&req->done); 123 124 kfree(tmp); 125 126 return req->err; 127 } 128 129 int devtmpfs_create_node(struct device *dev) 130 { 131 const char *tmp = NULL; 132 struct req req; 133 134 if (!thread) 135 return 0; 136 137 req.mode = 0; 138 req.uid = GLOBAL_ROOT_UID; 139 req.gid = GLOBAL_ROOT_GID; 140 req.name = device_get_devnode(dev, &req.mode, &req.uid, &req.gid, &tmp); 141 if (!req.name) 142 return -ENOMEM; 143 144 if (req.mode == 0) 145 req.mode = 0600; 146 if (is_blockdev(dev)) 147 req.mode |= S_IFBLK; 148 else 149 req.mode |= S_IFCHR; 150 151 req.dev = dev; 152 153 return devtmpfs_submit_req(&req, tmp); 154 } 155 156 int devtmpfs_delete_node(struct device *dev) 157 { 158 const char *tmp = NULL; 159 struct req req; 160 161 if (!thread) 162 return 0; 163 164 req.name = device_get_devnode(dev, NULL, NULL, NULL, &tmp); 165 if (!req.name) 166 return -ENOMEM; 167 168 req.mode = 0; 169 req.dev = dev; 170 171 return devtmpfs_submit_req(&req, tmp); 172 } 173 174 static int dev_mkdir(const char *name, umode_t mode) 175 { 176 struct dentry *dentry; 177 struct path path; 178 179 dentry = kern_path_create(AT_FDCWD, name, &path, LOOKUP_DIRECTORY); 180 if (IS_ERR(dentry)) 181 return PTR_ERR(dentry); 182 183 dentry = vfs_mkdir(&nop_mnt_idmap, d_inode(path.dentry), dentry, mode); 184 if (!IS_ERR(dentry)) 185 /* mark as kernel-created inode */ 186 d_inode(dentry)->i_private = &thread; 187 done_path_create(&path, dentry); 188 return PTR_ERR_OR_ZERO(dentry); 189 } 190 191 static int create_path(const char *nodepath) 192 { 193 char *path; 194 char *s; 195 int err = 0; 196 197 /* parent directories do not exist, create them */ 198 path = kstrdup(nodepath, GFP_KERNEL); 199 if (!path) 200 return -ENOMEM; 201 202 s = path; 203 for (;;) { 204 s = strchr(s, '/'); 205 if (!s) 206 break; 207 s[0] = '\0'; 208 err = dev_mkdir(path, 0755); 209 if (err && err != -EEXIST) 210 break; 211 s[0] = '/'; 212 s++; 213 } 214 kfree(path); 215 return err; 216 } 217 218 static int handle_create(const char *nodename, umode_t mode, kuid_t uid, 219 kgid_t gid, struct device *dev) 220 { 221 struct dentry *dentry; 222 struct path path; 223 int err; 224 225 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0); 226 if (dentry == ERR_PTR(-ENOENT)) { 227 create_path(nodename); 228 dentry = kern_path_create(AT_FDCWD, nodename, &path, 0); 229 } 230 if (IS_ERR(dentry)) 231 return PTR_ERR(dentry); 232 233 err = vfs_mknod(&nop_mnt_idmap, d_inode(path.dentry), dentry, mode, 234 dev->devt); 235 if (!err) { 236 struct iattr newattrs; 237 238 newattrs.ia_mode = mode; 239 newattrs.ia_uid = uid; 240 newattrs.ia_gid = gid; 241 newattrs.ia_valid = ATTR_MODE|ATTR_UID|ATTR_GID; 242 inode_lock(d_inode(dentry)); 243 notify_change(&nop_mnt_idmap, dentry, &newattrs, NULL); 244 inode_unlock(d_inode(dentry)); 245 246 /* mark as kernel-created inode */ 247 d_inode(dentry)->i_private = &thread; 248 } 249 done_path_create(&path, dentry); 250 return err; 251 } 252 253 static int dev_rmdir(const char *name) 254 { 255 struct path parent; 256 struct dentry *dentry; 257 int err; 258 259 dentry = kern_path_locked(name, &parent); 260 if (IS_ERR(dentry)) 261 return PTR_ERR(dentry); 262 if (d_inode(dentry)->i_private == &thread) 263 err = vfs_rmdir(&nop_mnt_idmap, d_inode(parent.dentry), 264 dentry); 265 else 266 err = -EPERM; 267 268 dput(dentry); 269 inode_unlock(d_inode(parent.dentry)); 270 path_put(&parent); 271 return err; 272 } 273 274 static int delete_path(const char *nodepath) 275 { 276 char *path; 277 int err = 0; 278 279 path = kstrdup(nodepath, GFP_KERNEL); 280 if (!path) 281 return -ENOMEM; 282 283 for (;;) { 284 char *base; 285 286 base = strrchr(path, '/'); 287 if (!base) 288 break; 289 base[0] = '\0'; 290 err = dev_rmdir(path); 291 if (err) 292 break; 293 } 294 295 kfree(path); 296 return err; 297 } 298 299 static int dev_mynode(struct device *dev, struct inode *inode) 300 { 301 /* did we create it */ 302 if (inode->i_private != &thread) 303 return 0; 304 305 /* does the dev_t match */ 306 if (is_blockdev(dev)) { 307 if (!S_ISBLK(inode->i_mode)) 308 return 0; 309 } else { 310 if (!S_ISCHR(inode->i_mode)) 311 return 0; 312 } 313 if (inode->i_rdev != dev->devt) 314 return 0; 315 316 /* ours */ 317 return 1; 318 } 319 320 static int handle_remove(const char *nodename, struct device *dev) 321 { 322 struct path parent; 323 struct dentry *dentry; 324 struct inode *inode; 325 int deleted = 0; 326 int err = 0; 327 328 dentry = kern_path_locked(nodename, &parent); 329 if (IS_ERR(dentry)) 330 return PTR_ERR(dentry); 331 332 inode = d_inode(dentry); 333 if (dev_mynode(dev, inode)) { 334 struct iattr newattrs; 335 /* 336 * before unlinking this node, reset permissions 337 * of possible references like hardlinks 338 */ 339 newattrs.ia_uid = GLOBAL_ROOT_UID; 340 newattrs.ia_gid = GLOBAL_ROOT_GID; 341 newattrs.ia_mode = inode->i_mode & ~0777; 342 newattrs.ia_valid = 343 ATTR_UID|ATTR_GID|ATTR_MODE; 344 inode_lock(d_inode(dentry)); 345 notify_change(&nop_mnt_idmap, dentry, &newattrs, NULL); 346 inode_unlock(d_inode(dentry)); 347 err = vfs_unlink(&nop_mnt_idmap, d_inode(parent.dentry), 348 dentry, NULL); 349 if (!err || err == -ENOENT) 350 deleted = 1; 351 } 352 dput(dentry); 353 inode_unlock(d_inode(parent.dentry)); 354 355 path_put(&parent); 356 if (deleted && strchr(nodename, '/')) 357 delete_path(nodename); 358 return err; 359 } 360 361 /* 362 * If configured, or requested by the commandline, devtmpfs will be 363 * auto-mounted after the kernel mounted the root filesystem. 364 */ 365 int __init devtmpfs_mount(void) 366 { 367 int err; 368 369 if (!mount_dev) 370 return 0; 371 372 if (!thread) 373 return 0; 374 375 err = init_mount("devtmpfs", "dev", "devtmpfs", DEVTMPFS_MFLAGS, NULL); 376 if (err) 377 pr_info("error mounting %d\n", err); 378 else 379 pr_info("mounted\n"); 380 return err; 381 } 382 383 static __initdata DECLARE_COMPLETION(setup_done); 384 385 static int handle(const char *name, umode_t mode, kuid_t uid, kgid_t gid, 386 struct device *dev) 387 { 388 if (mode) 389 return handle_create(name, mode, uid, gid, dev); 390 else 391 return handle_remove(name, dev); 392 } 393 394 static void __noreturn devtmpfs_work_loop(void) 395 { 396 while (1) { 397 spin_lock(&req_lock); 398 while (requests) { 399 struct req *req = requests; 400 requests = NULL; 401 spin_unlock(&req_lock); 402 while (req) { 403 struct req *next = req->next; 404 req->err = handle(req->name, req->mode, 405 req->uid, req->gid, req->dev); 406 complete(&req->done); 407 req = next; 408 } 409 spin_lock(&req_lock); 410 } 411 __set_current_state(TASK_INTERRUPTIBLE); 412 spin_unlock(&req_lock); 413 schedule(); 414 } 415 } 416 417 static noinline int __init devtmpfs_setup(void *p) 418 { 419 int err; 420 421 err = ksys_unshare(CLONE_NEWNS); 422 if (err) 423 goto out; 424 err = init_mount("devtmpfs", "/", "devtmpfs", DEVTMPFS_MFLAGS, NULL); 425 if (err) 426 goto out; 427 init_chdir("/.."); /* will traverse into overmounted root */ 428 init_chroot("."); 429 out: 430 *(int *)p = err; 431 return err; 432 } 433 434 /* 435 * The __ref is because devtmpfs_setup needs to be __init for the routines it 436 * calls. That call is done while devtmpfs_init, which is marked __init, 437 * synchronously waits for it to complete. 438 */ 439 static int __ref devtmpfsd(void *p) 440 { 441 int err = devtmpfs_setup(p); 442 443 complete(&setup_done); 444 if (err) 445 return err; 446 devtmpfs_work_loop(); 447 return 0; 448 } 449 450 /* 451 * Get the underlying (shmem/ramfs) context ops to build ours 452 */ 453 static int devtmpfs_configure_context(void) 454 { 455 struct fs_context *fc; 456 457 fc = fs_context_for_reconfigure(mnt->mnt_root, mnt->mnt_sb->s_flags, 458 MS_RMT_MASK); 459 if (IS_ERR(fc)) 460 return PTR_ERR(fc); 461 462 /* Set up devtmpfs_context_ops based on underlying type */ 463 devtmpfs_context_ops.free = fc->ops->free; 464 devtmpfs_context_ops.dup = fc->ops->dup; 465 devtmpfs_context_ops.parse_param = fc->ops->parse_param; 466 devtmpfs_context_ops.parse_monolithic = fc->ops->parse_monolithic; 467 devtmpfs_context_ops.get_tree = &devtmpfs_get_tree; 468 devtmpfs_context_ops.reconfigure = fc->ops->reconfigure; 469 470 put_fs_context(fc); 471 472 return 0; 473 } 474 475 /* 476 * Create devtmpfs instance, driver-core devices will add their device 477 * nodes here. 478 */ 479 int __init devtmpfs_init(void) 480 { 481 char opts[] = "mode=0755"; 482 int err; 483 484 mnt = vfs_kern_mount(&internal_fs_type, 0, "devtmpfs", opts); 485 if (IS_ERR(mnt)) { 486 pr_err("unable to create devtmpfs %ld\n", PTR_ERR(mnt)); 487 return PTR_ERR(mnt); 488 } 489 490 err = devtmpfs_configure_context(); 491 if (err) { 492 pr_err("unable to configure devtmpfs type %d\n", err); 493 return err; 494 } 495 496 err = register_filesystem(&dev_fs_type); 497 if (err) { 498 pr_err("unable to register devtmpfs type %d\n", err); 499 return err; 500 } 501 502 thread = kthread_run(devtmpfsd, &err, "kdevtmpfs"); 503 if (!IS_ERR(thread)) { 504 wait_for_completion(&setup_done); 505 } else { 506 err = PTR_ERR(thread); 507 thread = NULL; 508 } 509 510 if (err) { 511 pr_err("unable to create devtmpfs %d\n", err); 512 unregister_filesystem(&dev_fs_type); 513 thread = NULL; 514 return err; 515 } 516 517 pr_info("initialized\n"); 518 return 0; 519 } 520