1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/drivers/char/mem.c 4 * 5 * Copyright (C) 1991, 1992 Linus Torvalds 6 * 7 * Added devfs support. 8 * Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu> 9 * Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com> 10 */ 11 12 #include <linux/mm.h> 13 #include <linux/miscdevice.h> 14 #include <linux/slab.h> 15 #include <linux/vmalloc.h> 16 #include <linux/mman.h> 17 #include <linux/random.h> 18 #include <linux/init.h> 19 #include <linux/tty.h> 20 #include <linux/capability.h> 21 #include <linux/ptrace.h> 22 #include <linux/device.h> 23 #include <linux/highmem.h> 24 #include <linux/backing-dev.h> 25 #include <linux/shmem_fs.h> 26 #include <linux/splice.h> 27 #include <linux/pfn.h> 28 #include <linux/export.h> 29 #include <linux/io.h> 30 #include <linux/uio.h> 31 #include <linux/uaccess.h> 32 #include <linux/security.h> 33 34 #ifdef CONFIG_IA64 35 # include <linux/efi.h> 36 #endif 37 38 #define DEVMEM_MINOR 1 39 #define DEVPORT_MINOR 4 40 41 static inline unsigned long size_inside_page(unsigned long start, 42 unsigned long size) 43 { 44 unsigned long sz; 45 46 sz = PAGE_SIZE - (start & (PAGE_SIZE - 1)); 47 48 return min(sz, size); 49 } 50 51 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE 52 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count) 53 { 54 return addr + count <= __pa(high_memory); 55 } 56 57 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size) 58 { 59 return 1; 60 } 61 #endif 62 63 #ifdef CONFIG_STRICT_DEVMEM 64 static inline int page_is_allowed(unsigned long pfn) 65 { 66 return devmem_is_allowed(pfn); 67 } 68 static inline int range_is_allowed(unsigned long pfn, unsigned long size) 69 { 70 u64 from = ((u64)pfn) << PAGE_SHIFT; 71 u64 to = from + size; 72 u64 cursor = from; 73 74 while (cursor < to) { 75 if (!devmem_is_allowed(pfn)) 76 return 0; 77 cursor += PAGE_SIZE; 78 pfn++; 79 } 80 return 1; 81 } 82 #else 83 static inline int page_is_allowed(unsigned long pfn) 84 { 85 return 1; 86 } 87 static inline int range_is_allowed(unsigned long pfn, unsigned long size) 88 { 89 return 1; 90 } 91 #endif 92 93 #ifndef unxlate_dev_mem_ptr 94 #define unxlate_dev_mem_ptr unxlate_dev_mem_ptr 95 void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr) 96 { 97 } 98 #endif 99 100 static inline bool should_stop_iteration(void) 101 { 102 if (need_resched()) 103 cond_resched(); 104 return signal_pending(current); 105 } 106 107 /* 108 * This funcion reads the *physical* memory. The f_pos points directly to the 109 * memory location. 110 */ 111 static ssize_t read_mem(struct file *file, char __user *buf, 112 size_t count, loff_t *ppos) 113 { 114 phys_addr_t p = *ppos; 115 ssize_t read, sz; 116 void *ptr; 117 char *bounce; 118 int err; 119 120 if (p != *ppos) 121 return 0; 122 123 if (!valid_phys_addr_range(p, count)) 124 return -EFAULT; 125 read = 0; 126 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED 127 /* we don't have page 0 mapped on sparc and m68k.. */ 128 if (p < PAGE_SIZE) { 129 sz = size_inside_page(p, count); 130 if (sz > 0) { 131 if (clear_user(buf, sz)) 132 return -EFAULT; 133 buf += sz; 134 p += sz; 135 count -= sz; 136 read += sz; 137 } 138 } 139 #endif 140 141 bounce = kmalloc(PAGE_SIZE, GFP_KERNEL); 142 if (!bounce) 143 return -ENOMEM; 144 145 while (count > 0) { 146 unsigned long remaining; 147 int allowed, probe; 148 149 sz = size_inside_page(p, count); 150 151 err = -EPERM; 152 allowed = page_is_allowed(p >> PAGE_SHIFT); 153 if (!allowed) 154 goto failed; 155 156 err = -EFAULT; 157 if (allowed == 2) { 158 /* Show zeros for restricted memory. */ 159 remaining = clear_user(buf, sz); 160 } else { 161 /* 162 * On ia64 if a page has been mapped somewhere as 163 * uncached, then it must also be accessed uncached 164 * by the kernel or data corruption may occur. 165 */ 166 ptr = xlate_dev_mem_ptr(p); 167 if (!ptr) 168 goto failed; 169 170 probe = copy_from_kernel_nofault(bounce, ptr, sz); 171 unxlate_dev_mem_ptr(p, ptr); 172 if (probe) 173 goto failed; 174 175 remaining = copy_to_user(buf, bounce, sz); 176 } 177 178 if (remaining) 179 goto failed; 180 181 buf += sz; 182 p += sz; 183 count -= sz; 184 read += sz; 185 if (should_stop_iteration()) 186 break; 187 } 188 kfree(bounce); 189 190 *ppos += read; 191 return read; 192 193 failed: 194 kfree(bounce); 195 return err; 196 } 197 198 static ssize_t write_mem(struct file *file, const char __user *buf, 199 size_t count, loff_t *ppos) 200 { 201 phys_addr_t p = *ppos; 202 ssize_t written, sz; 203 unsigned long copied; 204 void *ptr; 205 206 if (p != *ppos) 207 return -EFBIG; 208 209 if (!valid_phys_addr_range(p, count)) 210 return -EFAULT; 211 212 written = 0; 213 214 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED 215 /* we don't have page 0 mapped on sparc and m68k.. */ 216 if (p < PAGE_SIZE) { 217 sz = size_inside_page(p, count); 218 /* Hmm. Do something? */ 219 buf += sz; 220 p += sz; 221 count -= sz; 222 written += sz; 223 } 224 #endif 225 226 while (count > 0) { 227 int allowed; 228 229 sz = size_inside_page(p, count); 230 231 allowed = page_is_allowed(p >> PAGE_SHIFT); 232 if (!allowed) 233 return -EPERM; 234 235 /* Skip actual writing when a page is marked as restricted. */ 236 if (allowed == 1) { 237 /* 238 * On ia64 if a page has been mapped somewhere as 239 * uncached, then it must also be accessed uncached 240 * by the kernel or data corruption may occur. 241 */ 242 ptr = xlate_dev_mem_ptr(p); 243 if (!ptr) { 244 if (written) 245 break; 246 return -EFAULT; 247 } 248 249 copied = copy_from_user(ptr, buf, sz); 250 unxlate_dev_mem_ptr(p, ptr); 251 if (copied) { 252 written += sz - copied; 253 if (written) 254 break; 255 return -EFAULT; 256 } 257 } 258 259 buf += sz; 260 p += sz; 261 count -= sz; 262 written += sz; 263 if (should_stop_iteration()) 264 break; 265 } 266 267 *ppos += written; 268 return written; 269 } 270 271 int __weak phys_mem_access_prot_allowed(struct file *file, 272 unsigned long pfn, unsigned long size, pgprot_t *vma_prot) 273 { 274 return 1; 275 } 276 277 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT 278 279 /* 280 * Architectures vary in how they handle caching for addresses 281 * outside of main memory. 282 * 283 */ 284 #ifdef pgprot_noncached 285 static int uncached_access(struct file *file, phys_addr_t addr) 286 { 287 #if defined(CONFIG_IA64) 288 /* 289 * On ia64, we ignore O_DSYNC because we cannot tolerate memory 290 * attribute aliases. 291 */ 292 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB); 293 #else 294 /* 295 * Accessing memory above the top the kernel knows about or through a 296 * file pointer 297 * that was marked O_DSYNC will be done non-cached. 298 */ 299 if (file->f_flags & O_DSYNC) 300 return 1; 301 return addr >= __pa(high_memory); 302 #endif 303 } 304 #endif 305 306 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, 307 unsigned long size, pgprot_t vma_prot) 308 { 309 #ifdef pgprot_noncached 310 phys_addr_t offset = pfn << PAGE_SHIFT; 311 312 if (uncached_access(file, offset)) 313 return pgprot_noncached(vma_prot); 314 #endif 315 return vma_prot; 316 } 317 #endif 318 319 #ifndef CONFIG_MMU 320 static unsigned long get_unmapped_area_mem(struct file *file, 321 unsigned long addr, 322 unsigned long len, 323 unsigned long pgoff, 324 unsigned long flags) 325 { 326 if (!valid_mmap_phys_addr_range(pgoff, len)) 327 return (unsigned long) -EINVAL; 328 return pgoff << PAGE_SHIFT; 329 } 330 331 /* permit direct mmap, for read, write or exec */ 332 static unsigned memory_mmap_capabilities(struct file *file) 333 { 334 return NOMMU_MAP_DIRECT | 335 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC; 336 } 337 338 static unsigned zero_mmap_capabilities(struct file *file) 339 { 340 return NOMMU_MAP_COPY; 341 } 342 343 /* can't do an in-place private mapping if there's no MMU */ 344 static inline int private_mapping_ok(struct vm_area_struct *vma) 345 { 346 return vma->vm_flags & VM_MAYSHARE; 347 } 348 #else 349 350 static inline int private_mapping_ok(struct vm_area_struct *vma) 351 { 352 return 1; 353 } 354 #endif 355 356 static const struct vm_operations_struct mmap_mem_ops = { 357 #ifdef CONFIG_HAVE_IOREMAP_PROT 358 .access = generic_access_phys 359 #endif 360 }; 361 362 static int mmap_mem(struct file *file, struct vm_area_struct *vma) 363 { 364 size_t size = vma->vm_end - vma->vm_start; 365 phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT; 366 367 /* Does it even fit in phys_addr_t? */ 368 if (offset >> PAGE_SHIFT != vma->vm_pgoff) 369 return -EINVAL; 370 371 /* It's illegal to wrap around the end of the physical address space. */ 372 if (offset + (phys_addr_t)size - 1 < offset) 373 return -EINVAL; 374 375 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size)) 376 return -EINVAL; 377 378 if (!private_mapping_ok(vma)) 379 return -ENOSYS; 380 381 if (!range_is_allowed(vma->vm_pgoff, size)) 382 return -EPERM; 383 384 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size, 385 &vma->vm_page_prot)) 386 return -EINVAL; 387 388 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff, 389 size, 390 vma->vm_page_prot); 391 392 vma->vm_ops = &mmap_mem_ops; 393 394 /* Remap-pfn-range will mark the range VM_IO */ 395 if (remap_pfn_range(vma, 396 vma->vm_start, 397 vma->vm_pgoff, 398 size, 399 vma->vm_page_prot)) { 400 return -EAGAIN; 401 } 402 return 0; 403 } 404 405 static ssize_t read_port(struct file *file, char __user *buf, 406 size_t count, loff_t *ppos) 407 { 408 unsigned long i = *ppos; 409 char __user *tmp = buf; 410 411 if (!access_ok(buf, count)) 412 return -EFAULT; 413 while (count-- > 0 && i < 65536) { 414 if (__put_user(inb(i), tmp) < 0) 415 return -EFAULT; 416 i++; 417 tmp++; 418 } 419 *ppos = i; 420 return tmp-buf; 421 } 422 423 static ssize_t write_port(struct file *file, const char __user *buf, 424 size_t count, loff_t *ppos) 425 { 426 unsigned long i = *ppos; 427 const char __user *tmp = buf; 428 429 if (!access_ok(buf, count)) 430 return -EFAULT; 431 while (count-- > 0 && i < 65536) { 432 char c; 433 434 if (__get_user(c, tmp)) { 435 if (tmp > buf) 436 break; 437 return -EFAULT; 438 } 439 outb(c, i); 440 i++; 441 tmp++; 442 } 443 *ppos = i; 444 return tmp-buf; 445 } 446 447 static ssize_t read_null(struct file *file, char __user *buf, 448 size_t count, loff_t *ppos) 449 { 450 return 0; 451 } 452 453 static ssize_t write_null(struct file *file, const char __user *buf, 454 size_t count, loff_t *ppos) 455 { 456 return count; 457 } 458 459 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to) 460 { 461 return 0; 462 } 463 464 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from) 465 { 466 size_t count = iov_iter_count(from); 467 iov_iter_advance(from, count); 468 return count; 469 } 470 471 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf, 472 struct splice_desc *sd) 473 { 474 return sd->len; 475 } 476 477 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out, 478 loff_t *ppos, size_t len, unsigned int flags) 479 { 480 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null); 481 } 482 483 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter) 484 { 485 size_t written = 0; 486 487 while (iov_iter_count(iter)) { 488 size_t chunk = iov_iter_count(iter), n; 489 490 if (chunk > PAGE_SIZE) 491 chunk = PAGE_SIZE; /* Just for latency reasons */ 492 n = iov_iter_zero(chunk, iter); 493 if (!n && iov_iter_count(iter)) 494 return written ? written : -EFAULT; 495 written += n; 496 if (signal_pending(current)) 497 return written ? written : -ERESTARTSYS; 498 if (!need_resched()) 499 continue; 500 if (iocb->ki_flags & IOCB_NOWAIT) 501 return written ? written : -EAGAIN; 502 cond_resched(); 503 } 504 return written; 505 } 506 507 static ssize_t read_zero(struct file *file, char __user *buf, 508 size_t count, loff_t *ppos) 509 { 510 size_t cleared = 0; 511 512 while (count) { 513 size_t chunk = min_t(size_t, count, PAGE_SIZE); 514 size_t left; 515 516 left = clear_user(buf + cleared, chunk); 517 if (unlikely(left)) { 518 cleared += (chunk - left); 519 if (!cleared) 520 return -EFAULT; 521 break; 522 } 523 cleared += chunk; 524 count -= chunk; 525 526 if (signal_pending(current)) 527 break; 528 cond_resched(); 529 } 530 531 return cleared; 532 } 533 534 static int mmap_zero(struct file *file, struct vm_area_struct *vma) 535 { 536 #ifndef CONFIG_MMU 537 return -ENOSYS; 538 #endif 539 if (vma->vm_flags & VM_SHARED) 540 return shmem_zero_setup(vma); 541 vma_set_anonymous(vma); 542 return 0; 543 } 544 545 static unsigned long get_unmapped_area_zero(struct file *file, 546 unsigned long addr, unsigned long len, 547 unsigned long pgoff, unsigned long flags) 548 { 549 #ifdef CONFIG_MMU 550 if (flags & MAP_SHARED) { 551 /* 552 * mmap_zero() will call shmem_zero_setup() to create a file, 553 * so use shmem's get_unmapped_area in case it can be huge; 554 * and pass NULL for file as in mmap.c's get_unmapped_area(), 555 * so as not to confuse shmem with our handle on "/dev/zero". 556 */ 557 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags); 558 } 559 560 /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */ 561 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); 562 #else 563 return -ENOSYS; 564 #endif 565 } 566 567 static ssize_t write_full(struct file *file, const char __user *buf, 568 size_t count, loff_t *ppos) 569 { 570 return -ENOSPC; 571 } 572 573 /* 574 * Special lseek() function for /dev/null and /dev/zero. Most notably, you 575 * can fopen() both devices with "a" now. This was previously impossible. 576 * -- SRB. 577 */ 578 static loff_t null_lseek(struct file *file, loff_t offset, int orig) 579 { 580 return file->f_pos = 0; 581 } 582 583 /* 584 * The memory devices use the full 32/64 bits of the offset, and so we cannot 585 * check against negative addresses: they are ok. The return value is weird, 586 * though, in that case (0). 587 * 588 * also note that seeking relative to the "end of file" isn't supported: 589 * it has no meaning, so it returns -EINVAL. 590 */ 591 static loff_t memory_lseek(struct file *file, loff_t offset, int orig) 592 { 593 loff_t ret; 594 595 inode_lock(file_inode(file)); 596 switch (orig) { 597 case SEEK_CUR: 598 offset += file->f_pos; 599 fallthrough; 600 case SEEK_SET: 601 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */ 602 if ((unsigned long long)offset >= -MAX_ERRNO) { 603 ret = -EOVERFLOW; 604 break; 605 } 606 file->f_pos = offset; 607 ret = file->f_pos; 608 force_successful_syscall_return(); 609 break; 610 default: 611 ret = -EINVAL; 612 } 613 inode_unlock(file_inode(file)); 614 return ret; 615 } 616 617 static int open_port(struct inode *inode, struct file *filp) 618 { 619 int rc; 620 621 if (!capable(CAP_SYS_RAWIO)) 622 return -EPERM; 623 624 rc = security_locked_down(LOCKDOWN_DEV_MEM); 625 if (rc) 626 return rc; 627 628 if (iminor(inode) != DEVMEM_MINOR) 629 return 0; 630 631 /* 632 * Use a unified address space to have a single point to manage 633 * revocations when drivers want to take over a /dev/mem mapped 634 * range. 635 */ 636 filp->f_mapping = iomem_get_mapping(); 637 638 return 0; 639 } 640 641 #define zero_lseek null_lseek 642 #define full_lseek null_lseek 643 #define write_zero write_null 644 #define write_iter_zero write_iter_null 645 #define open_mem open_port 646 647 static const struct file_operations __maybe_unused mem_fops = { 648 .llseek = memory_lseek, 649 .read = read_mem, 650 .write = write_mem, 651 .mmap = mmap_mem, 652 .open = open_mem, 653 #ifndef CONFIG_MMU 654 .get_unmapped_area = get_unmapped_area_mem, 655 .mmap_capabilities = memory_mmap_capabilities, 656 #endif 657 }; 658 659 static const struct file_operations null_fops = { 660 .llseek = null_lseek, 661 .read = read_null, 662 .write = write_null, 663 .read_iter = read_iter_null, 664 .write_iter = write_iter_null, 665 .splice_write = splice_write_null, 666 }; 667 668 static const struct file_operations __maybe_unused port_fops = { 669 .llseek = memory_lseek, 670 .read = read_port, 671 .write = write_port, 672 .open = open_port, 673 }; 674 675 static const struct file_operations zero_fops = { 676 .llseek = zero_lseek, 677 .write = write_zero, 678 .read_iter = read_iter_zero, 679 .read = read_zero, 680 .write_iter = write_iter_zero, 681 .mmap = mmap_zero, 682 .get_unmapped_area = get_unmapped_area_zero, 683 #ifndef CONFIG_MMU 684 .mmap_capabilities = zero_mmap_capabilities, 685 #endif 686 }; 687 688 static const struct file_operations full_fops = { 689 .llseek = full_lseek, 690 .read_iter = read_iter_zero, 691 .write = write_full, 692 }; 693 694 static const struct memdev { 695 const char *name; 696 umode_t mode; 697 const struct file_operations *fops; 698 fmode_t fmode; 699 } devlist[] = { 700 #ifdef CONFIG_DEVMEM 701 [DEVMEM_MINOR] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET }, 702 #endif 703 [3] = { "null", 0666, &null_fops, FMODE_NOWAIT }, 704 #ifdef CONFIG_DEVPORT 705 [4] = { "port", 0, &port_fops, 0 }, 706 #endif 707 [5] = { "zero", 0666, &zero_fops, FMODE_NOWAIT }, 708 [7] = { "full", 0666, &full_fops, 0 }, 709 [8] = { "random", 0666, &random_fops, 0 }, 710 [9] = { "urandom", 0666, &urandom_fops, 0 }, 711 #ifdef CONFIG_PRINTK 712 [11] = { "kmsg", 0644, &kmsg_fops, 0 }, 713 #endif 714 }; 715 716 static int memory_open(struct inode *inode, struct file *filp) 717 { 718 int minor; 719 const struct memdev *dev; 720 721 minor = iminor(inode); 722 if (minor >= ARRAY_SIZE(devlist)) 723 return -ENXIO; 724 725 dev = &devlist[minor]; 726 if (!dev->fops) 727 return -ENXIO; 728 729 filp->f_op = dev->fops; 730 filp->f_mode |= dev->fmode; 731 732 if (dev->fops->open) 733 return dev->fops->open(inode, filp); 734 735 return 0; 736 } 737 738 static const struct file_operations memory_fops = { 739 .open = memory_open, 740 .llseek = noop_llseek, 741 }; 742 743 static char *mem_devnode(struct device *dev, umode_t *mode) 744 { 745 if (mode && devlist[MINOR(dev->devt)].mode) 746 *mode = devlist[MINOR(dev->devt)].mode; 747 return NULL; 748 } 749 750 static struct class *mem_class; 751 752 static int __init chr_dev_init(void) 753 { 754 int minor; 755 756 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops)) 757 printk("unable to get major %d for memory devs\n", MEM_MAJOR); 758 759 mem_class = class_create(THIS_MODULE, "mem"); 760 if (IS_ERR(mem_class)) 761 return PTR_ERR(mem_class); 762 763 mem_class->devnode = mem_devnode; 764 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) { 765 if (!devlist[minor].name) 766 continue; 767 768 /* 769 * Create /dev/port? 770 */ 771 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port()) 772 continue; 773 774 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor), 775 NULL, devlist[minor].name); 776 } 777 778 return tty_init(); 779 } 780 781 fs_initcall(chr_dev_init); 782