drivers/char/mem.c

   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/raw.h>
  20 #include <linux/tty.h>
  21 #include <linux/capability.h>
  22 #include <linux/ptrace.h>
  23 #include <linux/device.h>
  24 #include <linux/highmem.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/shmem_fs.h>
  27 #include <linux/splice.h>
  28 #include <linux/pfn.h>
  29 #include <linux/export.h>
  30 #include <linux/io.h>
  31 #include <linux/uio.h>
  32 #include <linux/uaccess.h>
  33 #include <linux/security.h>
  34
  35 #ifdef CONFIG_IA64
  36 # include <linux/efi.h>
  37 #endif
  38
  39 #define DEVMEM_MINOR    1
  40 #define DEVPORT_MINOR   4
  41
  42 static inline unsigned long size_inside_page(unsigned long start,
  43                                              unsigned long size)
  44 {
  45         unsigned long sz;
  46
  47         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  48
  49         return min(sz, size);
  50 }
  51
  52 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  53 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
  54 {
  55         return addr + count <= __pa(high_memory);
  56 }
  57
  58 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  59 {
  60         return 1;
  61 }
  62 #endif
  63
  64 #ifdef CONFIG_STRICT_DEVMEM
  65 static inline int page_is_allowed(unsigned long pfn)
  66 {
  67         return devmem_is_allowed(pfn);
  68 }
  69 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  70 {
  71         u64 from = ((u64)pfn) << PAGE_SHIFT;
  72         u64 to = from + size;
  73         u64 cursor = from;
  74
  75         while (cursor < to) {
  76                 if (!devmem_is_allowed(pfn))
  77                         return 0;
  78                 cursor += PAGE_SIZE;
  79                 pfn++;
  80         }
  81         return 1;
  82 }
  83 #else
  84 static inline int page_is_allowed(unsigned long pfn)
  85 {
  86         return 1;
  87 }
  88 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  89 {
  90         return 1;
  91 }
  92 #endif
  93
  94 #ifndef unxlate_dev_mem_ptr
  95 #define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
  96 void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
  97 {
  98 }
  99 #endif
 100
 101 static inline bool should_stop_iteration(void)
 102 {
 103         if (need_resched())
 104                 cond_resched();
 105         return fatal_signal_pending(current);
 106 }
 107
 108 /*
 109  * This funcion reads the *physical* memory. The f_pos points directly to the
 110  * memory location.
 111  */
 112 static ssize_t read_mem(struct file *file, char __user *buf,
 113                         size_t count, loff_t *ppos)
 114 {
 115         phys_addr_t p = *ppos;
 116         ssize_t read, sz;
 117         void *ptr;
 118         char *bounce;
 119         int err;
 120
 121         if (p != *ppos)
 122                 return 0;
 123
 124         if (!valid_phys_addr_range(p, count))
 125                 return -EFAULT;
 126         read = 0;
 127 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 128         /* we don't have page 0 mapped on sparc and m68k.. */
 129         if (p < PAGE_SIZE) {
 130                 sz = size_inside_page(p, count);
 131                 if (sz > 0) {
 132                         if (clear_user(buf, sz))
 133                                 return -EFAULT;
 134                         buf += sz;
 135                         p += sz;
 136                         count -= sz;
 137                         read += sz;
 138                 }
 139         }
 140 #endif
 141
 142         bounce = kmalloc(PAGE_SIZE, GFP_KERNEL);
 143         if (!bounce)
 144                 return -ENOMEM;
 145
 146         while (count > 0) {
 147                 unsigned long remaining;
 148                 int allowed, probe;
 149
 150                 sz = size_inside_page(p, count);
 151
 152                 err = -EPERM;
 153                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 154                 if (!allowed)
 155                         goto failed;
 156
 157                 err = -EFAULT;
 158                 if (allowed == 2) {
 159                         /* Show zeros for restricted memory. */
 160                         remaining = clear_user(buf, sz);
 161                 } else {
 162                         /*
 163                          * On ia64 if a page has been mapped somewhere as
 164                          * uncached, then it must also be accessed uncached
 165                          * by the kernel or data corruption may occur.
 166                          */
 167                         ptr = xlate_dev_mem_ptr(p);
 168                         if (!ptr)
 169                                 goto failed;
 170
 171                         probe = copy_from_kernel_nofault(bounce, ptr, sz);
 172                         unxlate_dev_mem_ptr(p, ptr);
 173                         if (probe)
 174                                 goto failed;
 175
 176                         remaining = copy_to_user(buf, bounce, sz);
 177                 }
 178
 179                 if (remaining)
 180                         goto failed;
 181
 182                 buf += sz;
 183                 p += sz;
 184                 count -= sz;
 185                 read += sz;
 186                 if (should_stop_iteration())
 187                         break;
 188         }
 189         kfree(bounce);
 190
 191         *ppos += read;
 192         return read;
 193
 194 failed:
 195         kfree(bounce);
 196         return err;
 197 }
 198
 199 static ssize_t write_mem(struct file *file, const char __user *buf,
 200                          size_t count, loff_t *ppos)
 201 {
 202         phys_addr_t p = *ppos;
 203         ssize_t written, sz;
 204         unsigned long copied;
 205         void *ptr;
 206
 207         if (p != *ppos)
 208                 return -EFBIG;
 209
 210         if (!valid_phys_addr_range(p, count))
 211                 return -EFAULT;
 212
 213         written = 0;
 214
 215 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 216         /* we don't have page 0 mapped on sparc and m68k.. */
 217         if (p < PAGE_SIZE) {
 218                 sz = size_inside_page(p, count);
 219                 /* Hmm. Do something? */
 220                 buf += sz;
 221                 p += sz;
 222                 count -= sz;
 223                 written += sz;
 224         }
 225 #endif
 226
 227         while (count > 0) {
 228                 int allowed;
 229
 230                 sz = size_inside_page(p, count);
 231
 232                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 233                 if (!allowed)
 234                         return -EPERM;
 235
 236                 /* Skip actual writing when a page is marked as restricted. */
 237                 if (allowed == 1) {
 238                         /*
 239                          * On ia64 if a page has been mapped somewhere as
 240                          * uncached, then it must also be accessed uncached
 241                          * by the kernel or data corruption may occur.
 242                          */
 243                         ptr = xlate_dev_mem_ptr(p);
 244                         if (!ptr) {
 245                                 if (written)
 246                                         break;
 247                                 return -EFAULT;
 248                         }
 249
 250                         copied = copy_from_user(ptr, buf, sz);
 251                         unxlate_dev_mem_ptr(p, ptr);
 252                         if (copied) {
 253                                 written += sz - copied;
 254                                 if (written)
 255                                         break;
 256                                 return -EFAULT;
 257                         }
 258                 }
 259
 260                 buf += sz;
 261                 p += sz;
 262                 count -= sz;
 263                 written += sz;
 264                 if (should_stop_iteration())
 265                         break;
 266         }
 267
 268         *ppos += written;
 269         return written;
 270 }
 271
 272 int __weak phys_mem_access_prot_allowed(struct file *file,
 273         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 274 {
 275         return 1;
 276 }
 277
 278 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 279
 280 /*
 281  * Architectures vary in how they handle caching for addresses
 282  * outside of main memory.
 283  *
 284  */
 285 #ifdef pgprot_noncached
 286 static int uncached_access(struct file *file, phys_addr_t addr)
 287 {
 288 #if defined(CONFIG_IA64)
 289         /*
 290          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 291          * attribute aliases.
 292          */
 293         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 294 #else
 295         /*
 296          * Accessing memory above the top the kernel knows about or through a
 297          * file pointer
 298          * that was marked O_DSYNC will be done non-cached.
 299          */
 300         if (file->f_flags & O_DSYNC)
 301                 return 1;
 302         return addr >= __pa(high_memory);
 303 #endif
 304 }
 305 #endif
 306
 307 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 308                                      unsigned long size, pgprot_t vma_prot)
 309 {
 310 #ifdef pgprot_noncached
 311         phys_addr_t offset = pfn << PAGE_SHIFT;
 312
 313         if (uncached_access(file, offset))
 314                 return pgprot_noncached(vma_prot);
 315 #endif
 316         return vma_prot;
 317 }
 318 #endif
 319
 320 #ifndef CONFIG_MMU
 321 static unsigned long get_unmapped_area_mem(struct file *file,
 322                                            unsigned long addr,
 323                                            unsigned long len,
 324                                            unsigned long pgoff,
 325                                            unsigned long flags)
 326 {
 327         if (!valid_mmap_phys_addr_range(pgoff, len))
 328                 return (unsigned long) -EINVAL;
 329         return pgoff << PAGE_SHIFT;
 330 }
 331
 332 /* permit direct mmap, for read, write or exec */
 333 static unsigned memory_mmap_capabilities(struct file *file)
 334 {
 335         return NOMMU_MAP_DIRECT |
 336                 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
 337 }
 338
 339 static unsigned zero_mmap_capabilities(struct file *file)
 340 {
 341         return NOMMU_MAP_COPY;
 342 }
 343
 344 /* can't do an in-place private mapping if there's no MMU */
 345 static inline int private_mapping_ok(struct vm_area_struct *vma)
 346 {
 347         return vma->vm_flags & VM_MAYSHARE;
 348 }
 349 #else
 350
 351 static inline int private_mapping_ok(struct vm_area_struct *vma)
 352 {
 353         return 1;
 354 }
 355 #endif
 356
 357 static const struct vm_operations_struct mmap_mem_ops = {
 358 #ifdef CONFIG_HAVE_IOREMAP_PROT
 359         .access = generic_access_phys
 360 #endif
 361 };
 362
 363 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 364 {
 365         size_t size = vma->vm_end - vma->vm_start;
 366         phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
 367
 368         /* Does it even fit in phys_addr_t? */
 369         if (offset >> PAGE_SHIFT != vma->vm_pgoff)
 370                 return -EINVAL;
 371
 372         /* It's illegal to wrap around the end of the physical address space. */
 373         if (offset + (phys_addr_t)size - 1 < offset)
 374                 return -EINVAL;
 375
 376         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 377                 return -EINVAL;
 378
 379         if (!private_mapping_ok(vma))
 380                 return -ENOSYS;
 381
 382         if (!range_is_allowed(vma->vm_pgoff, size))
 383                 return -EPERM;
 384
 385         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 386                                                 &vma->vm_page_prot))
 387                 return -EINVAL;
 388
 389         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 390                                                  size,
 391                                                  vma->vm_page_prot);
 392
 393         vma->vm_ops = &mmap_mem_ops;
 394
 395         /* Remap-pfn-range will mark the range VM_IO */
 396         if (remap_pfn_range(vma,
 397                             vma->vm_start,
 398                             vma->vm_pgoff,
 399                             size,
 400                             vma->vm_page_prot)) {
 401                 return -EAGAIN;
 402         }
 403         return 0;
 404 }
 405
 406 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 407 {
 408         unsigned long pfn;
 409
 410         /* Turn a kernel-virtual address into a physical page frame */
 411         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 412
 413         /*
 414          * RED-PEN: on some architectures there is more mapped memory than
 415          * available in mem_map which pfn_valid checks for. Perhaps should add a
 416          * new macro here.
 417          *
 418          * RED-PEN: vmalloc is not supported right now.
 419          */
 420         if (!pfn_valid(pfn))
 421                 return -EIO;
 422
 423         vma->vm_pgoff = pfn;
 424         return mmap_mem(file, vma);
 425 }
 426
 427 /*
 428  * This function reads the *virtual* memory as seen by the kernel.
 429  */
 430 static ssize_t read_kmem(struct file *file, char __user *buf,
 431                          size_t count, loff_t *ppos)
 432 {
 433         unsigned long p = *ppos;
 434         ssize_t low_count, read, sz;
 435         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 436         int err = 0;
 437
 438         read = 0;
 439         if (p < (unsigned long) high_memory) {
 440                 low_count = count;
 441                 if (count > (unsigned long)high_memory - p)
 442                         low_count = (unsigned long)high_memory - p;
 443
 444 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 445                 /* we don't have page 0 mapped on sparc and m68k.. */
 446                 if (p < PAGE_SIZE && low_count > 0) {
 447                         sz = size_inside_page(p, low_count);
 448                         if (clear_user(buf, sz))
 449                                 return -EFAULT;
 450                         buf += sz;
 451                         p += sz;
 452                         read += sz;
 453                         low_count -= sz;
 454                         count -= sz;
 455                 }
 456 #endif
 457                 while (low_count > 0) {
 458                         sz = size_inside_page(p, low_count);
 459
 460                         /*
 461                          * On ia64 if a page has been mapped somewhere as
 462                          * uncached, then it must also be accessed uncached
 463                          * by the kernel or data corruption may occur
 464                          */
 465                         kbuf = xlate_dev_kmem_ptr((void *)p);
 466                         if (!virt_addr_valid(kbuf))
 467                                 return -ENXIO;
 468
 469                         if (copy_to_user(buf, kbuf, sz))
 470                                 return -EFAULT;
 471                         buf += sz;
 472                         p += sz;
 473                         read += sz;
 474                         low_count -= sz;
 475                         count -= sz;
 476                         if (should_stop_iteration()) {
 477                                 count = 0;
 478                                 break;
 479                         }
 480                 }
 481         }
 482
 483         if (count > 0) {
 484                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 485                 if (!kbuf)
 486                         return -ENOMEM;
 487                 while (count > 0) {
 488                         sz = size_inside_page(p, count);
 489                         if (!is_vmalloc_or_module_addr((void *)p)) {
 490                                 err = -ENXIO;
 491                                 break;
 492                         }
 493                         sz = vread(kbuf, (char *)p, sz);
 494                         if (!sz)
 495                                 break;
 496                         if (copy_to_user(buf, kbuf, sz)) {
 497                                 err = -EFAULT;
 498                                 break;
 499                         }
 500                         count -= sz;
 501                         buf += sz;
 502                         read += sz;
 503                         p += sz;
 504                         if (should_stop_iteration())
 505                                 break;
 506                 }
 507                 free_page((unsigned long)kbuf);
 508         }
 509         *ppos = p;
 510         return read ? read : err;
 511 }
 512
 513
 514 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 515                                 size_t count, loff_t *ppos)
 516 {
 517         ssize_t written, sz;
 518         unsigned long copied;
 519
 520         written = 0;
 521 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 522         /* we don't have page 0 mapped on sparc and m68k.. */
 523         if (p < PAGE_SIZE) {
 524                 sz = size_inside_page(p, count);
 525                 /* Hmm. Do something? */
 526                 buf += sz;
 527                 p += sz;
 528                 count -= sz;
 529                 written += sz;
 530         }
 531 #endif
 532
 533         while (count > 0) {
 534                 void *ptr;
 535
 536                 sz = size_inside_page(p, count);
 537
 538                 /*
 539                  * On ia64 if a page has been mapped somewhere as uncached, then
 540                  * it must also be accessed uncached by the kernel or data
 541                  * corruption may occur.
 542                  */
 543                 ptr = xlate_dev_kmem_ptr((void *)p);
 544                 if (!virt_addr_valid(ptr))
 545                         return -ENXIO;
 546
 547                 copied = copy_from_user(ptr, buf, sz);
 548                 if (copied) {
 549                         written += sz - copied;
 550                         if (written)
 551                                 break;
 552                         return -EFAULT;
 553                 }
 554                 buf += sz;
 555                 p += sz;
 556                 count -= sz;
 557                 written += sz;
 558                 if (should_stop_iteration())
 559                         break;
 560         }
 561
 562         *ppos += written;
 563         return written;
 564 }
 565
 566 /*
 567  * This function writes to the *virtual* memory as seen by the kernel.
 568  */
 569 static ssize_t write_kmem(struct file *file, const char __user *buf,
 570                           size_t count, loff_t *ppos)
 571 {
 572         unsigned long p = *ppos;
 573         ssize_t wrote = 0;
 574         ssize_t virtr = 0;
 575         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 576         int err = 0;
 577
 578         if (p < (unsigned long) high_memory) {
 579                 unsigned long to_write = min_t(unsigned long, count,
 580                                                (unsigned long)high_memory - p);
 581                 wrote = do_write_kmem(p, buf, to_write, ppos);
 582                 if (wrote != to_write)
 583                         return wrote;
 584                 p += wrote;
 585                 buf += wrote;
 586                 count -= wrote;
 587         }
 588
 589         if (count > 0) {
 590                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 591                 if (!kbuf)
 592                         return wrote ? wrote : -ENOMEM;
 593                 while (count > 0) {
 594                         unsigned long sz = size_inside_page(p, count);
 595                         unsigned long n;
 596
 597                         if (!is_vmalloc_or_module_addr((void *)p)) {
 598                                 err = -ENXIO;
 599                                 break;
 600                         }
 601                         n = copy_from_user(kbuf, buf, sz);
 602                         if (n) {
 603                                 err = -EFAULT;
 604                                 break;
 605                         }
 606                         vwrite(kbuf, (char *)p, sz);
 607                         count -= sz;
 608                         buf += sz;
 609                         virtr += sz;
 610                         p += sz;
 611                         if (should_stop_iteration())
 612                                 break;
 613                 }
 614                 free_page((unsigned long)kbuf);
 615         }
 616
 617         *ppos = p;
 618         return virtr + wrote ? : err;
 619 }
 620
 621 static ssize_t read_port(struct file *file, char __user *buf,
 622                          size_t count, loff_t *ppos)
 623 {
 624         unsigned long i = *ppos;
 625         char __user *tmp = buf;
 626
 627         if (!access_ok(buf, count))
 628                 return -EFAULT;
 629         while (count-- > 0 && i < 65536) {
 630                 if (__put_user(inb(i), tmp) < 0)
 631                         return -EFAULT;
 632                 i++;
 633                 tmp++;
 634         }
 635         *ppos = i;
 636         return tmp-buf;
 637 }
 638
 639 static ssize_t write_port(struct file *file, const char __user *buf,
 640                           size_t count, loff_t *ppos)
 641 {
 642         unsigned long i = *ppos;
 643         const char __user *tmp = buf;
 644
 645         if (!access_ok(buf, count))
 646                 return -EFAULT;
 647         while (count-- > 0 && i < 65536) {
 648                 char c;
 649
 650                 if (__get_user(c, tmp)) {
 651                         if (tmp > buf)
 652                                 break;
 653                         return -EFAULT;
 654                 }
 655                 outb(c, i);
 656                 i++;
 657                 tmp++;
 658         }
 659         *ppos = i;
 660         return tmp-buf;
 661 }
 662
 663 static ssize_t read_null(struct file *file, char __user *buf,
 664                          size_t count, loff_t *ppos)
 665 {
 666         return 0;
 667 }
 668
 669 static ssize_t write_null(struct file *file, const char __user *buf,
 670                           size_t count, loff_t *ppos)
 671 {
 672         return count;
 673 }
 674
 675 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
 676 {
 677         return 0;
 678 }
 679
 680 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
 681 {
 682         size_t count = iov_iter_count(from);
 683         iov_iter_advance(from, count);
 684         return count;
 685 }
 686
 687 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 688                         struct splice_desc *sd)
 689 {
 690         return sd->len;
 691 }
 692
 693 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 694                                  loff_t *ppos, size_t len, unsigned int flags)
 695 {
 696         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 697 }
 698
 699 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
 700 {
 701         size_t written = 0;
 702
 703         while (iov_iter_count(iter)) {
 704                 size_t chunk = iov_iter_count(iter), n;
 705
 706                 if (chunk > PAGE_SIZE)
 707                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 708                 n = iov_iter_zero(chunk, iter);
 709                 if (!n && iov_iter_count(iter))
 710                         return written ? written : -EFAULT;
 711                 written += n;
 712                 if (signal_pending(current))
 713                         return written ? written : -ERESTARTSYS;
 714                 cond_resched();
 715         }
 716         return written;
 717 }
 718
 719 static ssize_t read_zero(struct file *file, char __user *buf,
 720                          size_t count, loff_t *ppos)
 721 {
 722         size_t cleared = 0;
 723
 724         while (count) {
 725                 size_t chunk = min_t(size_t, count, PAGE_SIZE);
 726                 size_t left;
 727
 728                 left = clear_user(buf + cleared, chunk);
 729                 if (unlikely(left)) {
 730                         cleared += (chunk - left);
 731                         if (!cleared)
 732                                 return -EFAULT;
 733                         break;
 734                 }
 735                 cleared += chunk;
 736                 count -= chunk;
 737
 738                 if (signal_pending(current))
 739                         break;
 740                 cond_resched();
 741         }
 742
 743         return cleared;
 744 }
 745
 746 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 747 {
 748 #ifndef CONFIG_MMU
 749         return -ENOSYS;
 750 #endif
 751         if (vma->vm_flags & VM_SHARED)
 752                 return shmem_zero_setup(vma);
 753         vma_set_anonymous(vma);
 754         return 0;
 755 }
 756
 757 static unsigned long get_unmapped_area_zero(struct file *file,
 758                                 unsigned long addr, unsigned long len,
 759                                 unsigned long pgoff, unsigned long flags)
 760 {
 761 #ifdef CONFIG_MMU
 762         if (flags & MAP_SHARED) {
 763                 /*
 764                  * mmap_zero() will call shmem_zero_setup() to create a file,
 765                  * so use shmem's get_unmapped_area in case it can be huge;
 766                  * and pass NULL for file as in mmap.c's get_unmapped_area(),
 767                  * so as not to confuse shmem with our handle on "/dev/zero".
 768                  */
 769                 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
 770         }
 771
 772         /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
 773         return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
 774 #else
 775         return -ENOSYS;
 776 #endif
 777 }
 778
 779 static ssize_t write_full(struct file *file, const char __user *buf,
 780                           size_t count, loff_t *ppos)
 781 {
 782         return -ENOSPC;
 783 }
 784
 785 /*
 786  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 787  * can fopen() both devices with "a" now.  This was previously impossible.
 788  * -- SRB.
 789  */
 790 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 791 {
 792         return file->f_pos = 0;
 793 }
 794
 795 /*
 796  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 797  * check against negative addresses: they are ok. The return value is weird,
 798  * though, in that case (0).
 799  *
 800  * also note that seeking relative to the "end of file" isn't supported:
 801  * it has no meaning, so it returns -EINVAL.
 802  */
 803 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 804 {
 805         loff_t ret;
 806
 807         inode_lock(file_inode(file));
 808         switch (orig) {
 809         case SEEK_CUR:
 810                 offset += file->f_pos;
 811                 fallthrough;
 812         case SEEK_SET:
 813                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 814                 if ((unsigned long long)offset >= -MAX_ERRNO) {
 815                         ret = -EOVERFLOW;
 816                         break;
 817                 }
 818                 file->f_pos = offset;
 819                 ret = file->f_pos;
 820                 force_successful_syscall_return();
 821                 break;
 822         default:
 823                 ret = -EINVAL;
 824         }
 825         inode_unlock(file_inode(file));
 826         return ret;
 827 }
 828
 829 static int open_port(struct inode *inode, struct file *filp)
 830 {
 831         int rc;
 832
 833         if (!capable(CAP_SYS_RAWIO))
 834                 return -EPERM;
 835
 836         rc = security_locked_down(LOCKDOWN_DEV_MEM);
 837         if (rc)
 838                 return rc;
 839
 840         if (iminor(inode) != DEVMEM_MINOR)
 841                 return 0;
 842
 843         /*
 844          * Use a unified address space to have a single point to manage
 845          * revocations when drivers want to take over a /dev/mem mapped
 846          * range.
 847          */
 848         filp->f_mapping = iomem_get_mapping();
 849
 850         return 0;
 851 }
 852
 853 #define zero_lseek      null_lseek
 854 #define full_lseek      null_lseek
 855 #define write_zero      write_null
 856 #define write_iter_zero write_iter_null
 857 #define open_mem        open_port
 858 #define open_kmem       open_mem
 859
 860 static const struct file_operations __maybe_unused mem_fops = {
 861         .llseek         = memory_lseek,
 862         .read           = read_mem,
 863         .write          = write_mem,
 864         .mmap           = mmap_mem,
 865         .open           = open_mem,
 866 #ifndef CONFIG_MMU
 867         .get_unmapped_area = get_unmapped_area_mem,
 868         .mmap_capabilities = memory_mmap_capabilities,
 869 #endif
 870 };
 871
 872 static const struct file_operations __maybe_unused kmem_fops = {
 873         .llseek         = memory_lseek,
 874         .read           = read_kmem,
 875         .write          = write_kmem,
 876         .mmap           = mmap_kmem,
 877         .open           = open_kmem,
 878 #ifndef CONFIG_MMU
 879         .get_unmapped_area = get_unmapped_area_mem,
 880         .mmap_capabilities = memory_mmap_capabilities,
 881 #endif
 882 };
 883
 884 static const struct file_operations null_fops = {
 885         .llseek         = null_lseek,
 886         .read           = read_null,
 887         .write          = write_null,
 888         .read_iter      = read_iter_null,
 889         .write_iter     = write_iter_null,
 890         .splice_write   = splice_write_null,
 891 };
 892
 893 static const struct file_operations __maybe_unused port_fops = {
 894         .llseek         = memory_lseek,
 895         .read           = read_port,
 896         .write          = write_port,
 897         .open           = open_port,
 898 };
 899
 900 static const struct file_operations zero_fops = {
 901         .llseek         = zero_lseek,
 902         .write          = write_zero,
 903         .read_iter      = read_iter_zero,
 904         .read           = read_zero,
 905         .write_iter     = write_iter_zero,
 906         .mmap           = mmap_zero,
 907         .get_unmapped_area = get_unmapped_area_zero,
 908 #ifndef CONFIG_MMU
 909         .mmap_capabilities = zero_mmap_capabilities,
 910 #endif
 911 };
 912
 913 static const struct file_operations full_fops = {
 914         .llseek         = full_lseek,
 915         .read_iter      = read_iter_zero,
 916         .write          = write_full,
 917 };
 918
 919 static const struct memdev {
 920         const char *name;
 921         umode_t mode;
 922         const struct file_operations *fops;
 923         fmode_t fmode;
 924 } devlist[] = {
 925 #ifdef CONFIG_DEVMEM
 926          [DEVMEM_MINOR] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
 927 #endif
 928 #ifdef CONFIG_DEVKMEM
 929          [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
 930 #endif
 931          [3] = { "null", 0666, &null_fops, 0 },
 932 #ifdef CONFIG_DEVPORT
 933          [4] = { "port", 0, &port_fops, 0 },
 934 #endif
 935          [5] = { "zero", 0666, &zero_fops, 0 },
 936          [7] = { "full", 0666, &full_fops, 0 },
 937          [8] = { "random", 0666, &random_fops, 0 },
 938          [9] = { "urandom", 0666, &urandom_fops, 0 },
 939 #ifdef CONFIG_PRINTK
 940         [11] = { "kmsg", 0644, &kmsg_fops, 0 },
 941 #endif
 942 };
 943
 944 static int memory_open(struct inode *inode, struct file *filp)
 945 {
 946         int minor;
 947         const struct memdev *dev;
 948
 949         minor = iminor(inode);
 950         if (minor >= ARRAY_SIZE(devlist))
 951                 return -ENXIO;
 952
 953         dev = &devlist[minor];
 954         if (!dev->fops)
 955                 return -ENXIO;
 956
 957         filp->f_op = dev->fops;
 958         filp->f_mode |= dev->fmode;
 959
 960         if (dev->fops->open)
 961                 return dev->fops->open(inode, filp);
 962
 963         return 0;
 964 }
 965
 966 static const struct file_operations memory_fops = {
 967         .open = memory_open,
 968         .llseek = noop_llseek,
 969 };
 970
 971 static char *mem_devnode(struct device *dev, umode_t *mode)
 972 {
 973         if (mode && devlist[MINOR(dev->devt)].mode)
 974                 *mode = devlist[MINOR(dev->devt)].mode;
 975         return NULL;
 976 }
 977
 978 static struct class *mem_class;
 979
 980 static int __init chr_dev_init(void)
 981 {
 982         int minor;
 983
 984         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 985                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 986
 987         mem_class = class_create(THIS_MODULE, "mem");
 988         if (IS_ERR(mem_class))
 989                 return PTR_ERR(mem_class);
 990
 991         mem_class->devnode = mem_devnode;
 992         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 993                 if (!devlist[minor].name)
 994                         continue;
 995
 996                 /*
 997                  * Create /dev/port?
 998                  */
 999                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
1000                         continue;
1001
1002                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
1003                               NULL, devlist[minor].name);
1004         }
1005
1006         return tty_init();
1007 }
1008
1009 fs_initcall(chr_dev_init);