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