drivers/char/mem.c

   1 /*
   2  *  linux/drivers/char/mem.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *
   6  *  Added devfs support.
   7  *    Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
   8  *  Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
   9  */
  10
  11 #include <linux/mm.h>
  12 #include <linux/miscdevice.h>
  13 #include <linux/slab.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/mman.h>
  16 #include <linux/random.h>
  17 #include <linux/init.h>
  18 #include <linux/raw.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
  32 #include <linux/uaccess.h>
  33
  34 #ifdef CONFIG_IA64
  35 # include <linux/efi.h>
  36 #endif
  37
  38 #define DEVPORT_MINOR   4
  39
  40 static inline unsigned long size_inside_page(unsigned long start,
  41                                              unsigned long size)
  42 {
  43         unsigned long sz;
  44
  45         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  46
  47         return min(sz, size);
  48 }
  49
  50 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  51 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
  52 {
  53         return addr + count <= __pa(high_memory);
  54 }
  55
  56 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  57 {
  58         return 1;
  59 }
  60 #endif
  61
  62 #ifdef CONFIG_STRICT_DEVMEM
  63 static inline int page_is_allowed(unsigned long pfn)
  64 {
  65         return devmem_is_allowed(pfn);
  66 }
  67 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  68 {
  69         u64 from = ((u64)pfn) << PAGE_SHIFT;
  70         u64 to = from + size;
  71         u64 cursor = from;
  72
  73         while (cursor < to) {
  74                 if (!devmem_is_allowed(pfn))
  75                         return 0;
  76                 cursor += PAGE_SIZE;
  77                 pfn++;
  78         }
  79         return 1;
  80 }
  81 #else
  82 static inline int page_is_allowed(unsigned long pfn)
  83 {
  84         return 1;
  85 }
  86 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  87 {
  88         return 1;
  89 }
  90 #endif
  91
  92 #ifndef unxlate_dev_mem_ptr
  93 #define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
  94 void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
  95 {
  96 }
  97 #endif
  98
  99 /*
 100  * This funcion reads the *physical* memory. The f_pos points directly to the
 101  * memory location.
 102  */
 103 static ssize_t read_mem(struct file *file, char __user *buf,
 104                         size_t count, loff_t *ppos)
 105 {
 106         phys_addr_t p = *ppos;
 107         ssize_t read, sz;
 108         void *ptr;
 109
 110         if (p != *ppos)
 111                 return 0;
 112
 113         if (!valid_phys_addr_range(p, count))
 114                 return -EFAULT;
 115         read = 0;
 116 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 117         /* we don't have page 0 mapped on sparc and m68k.. */
 118         if (p < PAGE_SIZE) {
 119                 sz = size_inside_page(p, count);
 120                 if (sz > 0) {
 121                         if (clear_user(buf, sz))
 122                                 return -EFAULT;
 123                         buf += sz;
 124                         p += sz;
 125                         count -= sz;
 126                         read += sz;
 127                 }
 128         }
 129 #endif
 130
 131         while (count > 0) {
 132                 unsigned long remaining;
 133                 int allowed;
 134
 135                 sz = size_inside_page(p, count);
 136
 137                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 138                 if (!allowed)
 139                         return -EPERM;
 140                 if (allowed == 2) {
 141                         /* Show zeros for restricted memory. */
 142                         remaining = clear_user(buf, sz);
 143                 } else {
 144                         /*
 145                          * On ia64 if a page has been mapped somewhere as
 146                          * uncached, then it must also be accessed uncached
 147                          * by the kernel or data corruption may occur.
 148                          */
 149                         ptr = xlate_dev_mem_ptr(p);
 150                         if (!ptr)
 151                                 return -EFAULT;
 152
 153                         remaining = copy_to_user(buf, ptr, sz);
 154
 155                         unxlate_dev_mem_ptr(p, ptr);
 156                 }
 157
 158                 if (remaining)
 159                         return -EFAULT;
 160
 161                 buf += sz;
 162                 p += sz;
 163                 count -= sz;
 164                 read += sz;
 165         }
 166
 167         *ppos += read;
 168         return read;
 169 }
 170
 171 static ssize_t write_mem(struct file *file, const char __user *buf,
 172                          size_t count, loff_t *ppos)
 173 {
 174         phys_addr_t p = *ppos;
 175         ssize_t written, sz;
 176         unsigned long copied;
 177         void *ptr;
 178
 179         if (p != *ppos)
 180                 return -EFBIG;
 181
 182         if (!valid_phys_addr_range(p, count))
 183                 return -EFAULT;
 184
 185         written = 0;
 186
 187 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 188         /* we don't have page 0 mapped on sparc and m68k.. */
 189         if (p < PAGE_SIZE) {
 190                 sz = size_inside_page(p, count);
 191                 /* Hmm. Do something? */
 192                 buf += sz;
 193                 p += sz;
 194                 count -= sz;
 195                 written += sz;
 196         }
 197 #endif
 198
 199         while (count > 0) {
 200                 int allowed;
 201
 202                 sz = size_inside_page(p, count);
 203
 204                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 205                 if (!allowed)
 206                         return -EPERM;
 207
 208                 /* Skip actual writing when a page is marked as restricted. */
 209                 if (allowed == 1) {
 210                         /*
 211                          * On ia64 if a page has been mapped somewhere as
 212                          * uncached, then it must also be accessed uncached
 213                          * by the kernel or data corruption may occur.
 214                          */
 215                         ptr = xlate_dev_mem_ptr(p);
 216                         if (!ptr) {
 217                                 if (written)
 218                                         break;
 219                                 return -EFAULT;
 220                         }
 221
 222                         copied = copy_from_user(ptr, buf, sz);
 223                         unxlate_dev_mem_ptr(p, ptr);
 224                         if (copied) {
 225                                 written += sz - copied;
 226                                 if (written)
 227                                         break;
 228                                 return -EFAULT;
 229                         }
 230                 }
 231
 232                 buf += sz;
 233                 p += sz;
 234                 count -= sz;
 235                 written += sz;
 236         }
 237
 238         *ppos += written;
 239         return written;
 240 }
 241
 242 int __weak phys_mem_access_prot_allowed(struct file *file,
 243         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 244 {
 245         return 1;
 246 }
 247
 248 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 249
 250 /*
 251  * Architectures vary in how they handle caching for addresses
 252  * outside of main memory.
 253  *
 254  */
 255 #ifdef pgprot_noncached
 256 static int uncached_access(struct file *file, phys_addr_t addr)
 257 {
 258 #if defined(CONFIG_IA64)
 259         /*
 260          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 261          * attribute aliases.
 262          */
 263         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 264 #elif defined(CONFIG_MIPS)
 265         {
 266                 extern int __uncached_access(struct file *file,
 267                                              unsigned long addr);
 268
 269                 return __uncached_access(file, addr);
 270         }
 271 #else
 272         /*
 273          * Accessing memory above the top the kernel knows about or through a
 274          * file pointer
 275          * that was marked O_DSYNC will be done non-cached.
 276          */
 277         if (file->f_flags & O_DSYNC)
 278                 return 1;
 279         return addr >= __pa(high_memory);
 280 #endif
 281 }
 282 #endif
 283
 284 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 285                                      unsigned long size, pgprot_t vma_prot)
 286 {
 287 #ifdef pgprot_noncached
 288         phys_addr_t offset = pfn << PAGE_SHIFT;
 289
 290         if (uncached_access(file, offset))
 291                 return pgprot_noncached(vma_prot);
 292 #endif
 293         return vma_prot;
 294 }
 295 #endif
 296
 297 #ifndef CONFIG_MMU
 298 static unsigned long get_unmapped_area_mem(struct file *file,
 299                                            unsigned long addr,
 300                                            unsigned long len,
 301                                            unsigned long pgoff,
 302                                            unsigned long flags)
 303 {
 304         if (!valid_mmap_phys_addr_range(pgoff, len))
 305                 return (unsigned long) -EINVAL;
 306         return pgoff << PAGE_SHIFT;
 307 }
 308
 309 /* permit direct mmap, for read, write or exec */
 310 static unsigned memory_mmap_capabilities(struct file *file)
 311 {
 312         return NOMMU_MAP_DIRECT |
 313                 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
 314 }
 315
 316 static unsigned zero_mmap_capabilities(struct file *file)
 317 {
 318         return NOMMU_MAP_COPY;
 319 }
 320
 321 /* can't do an in-place private mapping if there's no MMU */
 322 static inline int private_mapping_ok(struct vm_area_struct *vma)
 323 {
 324         return vma->vm_flags & VM_MAYSHARE;
 325 }
 326 #else
 327
 328 static inline int private_mapping_ok(struct vm_area_struct *vma)
 329 {
 330         return 1;
 331 }
 332 #endif
 333
 334 static const struct vm_operations_struct mmap_mem_ops = {
 335 #ifdef CONFIG_HAVE_IOREMAP_PROT
 336         .access = generic_access_phys
 337 #endif
 338 };
 339
 340 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 341 {
 342         size_t size = vma->vm_end - vma->vm_start;
 343         phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
 344
 345         /* It's illegal to wrap around the end of the physical address space. */
 346         if (offset + (phys_addr_t)size - 1 < offset)
 347                 return -EINVAL;
 348
 349         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 350                 return -EINVAL;
 351
 352         if (!private_mapping_ok(vma))
 353                 return -ENOSYS;
 354
 355         if (!range_is_allowed(vma->vm_pgoff, size))
 356                 return -EPERM;
 357
 358         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 359                                                 &vma->vm_page_prot))
 360                 return -EINVAL;
 361
 362         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 363                                                  size,
 364                                                  vma->vm_page_prot);
 365
 366         vma->vm_ops = &mmap_mem_ops;
 367
 368         /* Remap-pfn-range will mark the range VM_IO */
 369         if (remap_pfn_range(vma,
 370                             vma->vm_start,
 371                             vma->vm_pgoff,
 372                             size,
 373                             vma->vm_page_prot)) {
 374                 return -EAGAIN;
 375         }
 376         return 0;
 377 }
 378
 379 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 380 {
 381         unsigned long pfn;
 382
 383         /* Turn a kernel-virtual address into a physical page frame */
 384         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 385
 386         /*
 387          * RED-PEN: on some architectures there is more mapped memory than
 388          * available in mem_map which pfn_valid checks for. Perhaps should add a
 389          * new macro here.
 390          *
 391          * RED-PEN: vmalloc is not supported right now.
 392          */
 393         if (!pfn_valid(pfn))
 394                 return -EIO;
 395
 396         vma->vm_pgoff = pfn;
 397         return mmap_mem(file, vma);
 398 }
 399
 400 /*
 401  * This function reads the *virtual* memory as seen by the kernel.
 402  */
 403 static ssize_t read_kmem(struct file *file, char __user *buf,
 404                          size_t count, loff_t *ppos)
 405 {
 406         unsigned long p = *ppos;
 407         ssize_t low_count, read, sz;
 408         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 409         int err = 0;
 410
 411         read = 0;
 412         if (p < (unsigned long) high_memory) {
 413                 low_count = count;
 414                 if (count > (unsigned long)high_memory - p)
 415                         low_count = (unsigned long)high_memory - p;
 416
 417 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 418                 /* we don't have page 0 mapped on sparc and m68k.. */
 419                 if (p < PAGE_SIZE && low_count > 0) {
 420                         sz = size_inside_page(p, low_count);
 421                         if (clear_user(buf, sz))
 422                                 return -EFAULT;
 423                         buf += sz;
 424                         p += sz;
 425                         read += sz;
 426                         low_count -= sz;
 427                         count -= sz;
 428                 }
 429 #endif
 430                 while (low_count > 0) {
 431                         sz = size_inside_page(p, low_count);
 432
 433                         /*
 434                          * On ia64 if a page has been mapped somewhere as
 435                          * uncached, then it must also be accessed uncached
 436                          * by the kernel or data corruption may occur
 437                          */
 438                         kbuf = xlate_dev_kmem_ptr((void *)p);
 439                         if (!virt_addr_valid(kbuf))
 440                                 return -ENXIO;
 441
 442                         if (copy_to_user(buf, kbuf, sz))
 443                                 return -EFAULT;
 444                         buf += sz;
 445                         p += sz;
 446                         read += sz;
 447                         low_count -= sz;
 448                         count -= sz;
 449                 }
 450         }
 451
 452         if (count > 0) {
 453                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 454                 if (!kbuf)
 455                         return -ENOMEM;
 456                 while (count > 0) {
 457                         sz = size_inside_page(p, count);
 458                         if (!is_vmalloc_or_module_addr((void *)p)) {
 459                                 err = -ENXIO;
 460                                 break;
 461                         }
 462                         sz = vread(kbuf, (char *)p, sz);
 463                         if (!sz)
 464                                 break;
 465                         if (copy_to_user(buf, kbuf, sz)) {
 466                                 err = -EFAULT;
 467                                 break;
 468                         }
 469                         count -= sz;
 470                         buf += sz;
 471                         read += sz;
 472                         p += sz;
 473                 }
 474                 free_page((unsigned long)kbuf);
 475         }
 476         *ppos = p;
 477         return read ? read : err;
 478 }
 479
 480
 481 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 482                                 size_t count, loff_t *ppos)
 483 {
 484         ssize_t written, sz;
 485         unsigned long copied;
 486
 487         written = 0;
 488 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 489         /* we don't have page 0 mapped on sparc and m68k.. */
 490         if (p < PAGE_SIZE) {
 491                 sz = size_inside_page(p, count);
 492                 /* Hmm. Do something? */
 493                 buf += sz;
 494                 p += sz;
 495                 count -= sz;
 496                 written += sz;
 497         }
 498 #endif
 499
 500         while (count > 0) {
 501                 void *ptr;
 502
 503                 sz = size_inside_page(p, count);
 504
 505                 /*
 506                  * On ia64 if a page has been mapped somewhere as uncached, then
 507                  * it must also be accessed uncached by the kernel or data
 508                  * corruption may occur.
 509                  */
 510                 ptr = xlate_dev_kmem_ptr((void *)p);
 511                 if (!virt_addr_valid(ptr))
 512                         return -ENXIO;
 513
 514                 copied = copy_from_user(ptr, buf, sz);
 515                 if (copied) {
 516                         written += sz - copied;
 517                         if (written)
 518                                 break;
 519                         return -EFAULT;
 520                 }
 521                 buf += sz;
 522                 p += sz;
 523                 count -= sz;
 524                 written += sz;
 525         }
 526
 527         *ppos += written;
 528         return written;
 529 }
 530
 531 /*
 532  * This function writes to the *virtual* memory as seen by the kernel.
 533  */
 534 static ssize_t write_kmem(struct file *file, const char __user *buf,
 535                           size_t count, loff_t *ppos)
 536 {
 537         unsigned long p = *ppos;
 538         ssize_t wrote = 0;
 539         ssize_t virtr = 0;
 540         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 541         int err = 0;
 542
 543         if (p < (unsigned long) high_memory) {
 544                 unsigned long to_write = min_t(unsigned long, count,
 545                                                (unsigned long)high_memory - p);
 546                 wrote = do_write_kmem(p, buf, to_write, ppos);
 547                 if (wrote != to_write)
 548                         return wrote;
 549                 p += wrote;
 550                 buf += wrote;
 551                 count -= wrote;
 552         }
 553
 554         if (count > 0) {
 555                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 556                 if (!kbuf)
 557                         return wrote ? wrote : -ENOMEM;
 558                 while (count > 0) {
 559                         unsigned long sz = size_inside_page(p, count);
 560                         unsigned long n;
 561
 562                         if (!is_vmalloc_or_module_addr((void *)p)) {
 563                                 err = -ENXIO;
 564                                 break;
 565                         }
 566                         n = copy_from_user(kbuf, buf, sz);
 567                         if (n) {
 568                                 err = -EFAULT;
 569                                 break;
 570                         }
 571                         vwrite(kbuf, (char *)p, sz);
 572                         count -= sz;
 573                         buf += sz;
 574                         virtr += sz;
 575                         p += sz;
 576                 }
 577                 free_page((unsigned long)kbuf);
 578         }
 579
 580         *ppos = p;
 581         return virtr + wrote ? : err;
 582 }
 583
 584 static ssize_t read_port(struct file *file, char __user *buf,
 585                          size_t count, loff_t *ppos)
 586 {
 587         unsigned long i = *ppos;
 588         char __user *tmp = buf;
 589
 590         if (!access_ok(VERIFY_WRITE, buf, count))
 591                 return -EFAULT;
 592         while (count-- > 0 && i < 65536) {
 593                 if (__put_user(inb(i), tmp) < 0)
 594                         return -EFAULT;
 595                 i++;
 596                 tmp++;
 597         }
 598         *ppos = i;
 599         return tmp-buf;
 600 }
 601
 602 static ssize_t write_port(struct file *file, const char __user *buf,
 603                           size_t count, loff_t *ppos)
 604 {
 605         unsigned long i = *ppos;
 606         const char __user *tmp = buf;
 607
 608         if (!access_ok(VERIFY_READ, buf, count))
 609                 return -EFAULT;
 610         while (count-- > 0 && i < 65536) {
 611                 char c;
 612
 613                 if (__get_user(c, tmp)) {
 614                         if (tmp > buf)
 615                                 break;
 616                         return -EFAULT;
 617                 }
 618                 outb(c, i);
 619                 i++;
 620                 tmp++;
 621         }
 622         *ppos = i;
 623         return tmp-buf;
 624 }
 625
 626 static ssize_t read_null(struct file *file, char __user *buf,
 627                          size_t count, loff_t *ppos)
 628 {
 629         return 0;
 630 }
 631
 632 static ssize_t write_null(struct file *file, const char __user *buf,
 633                           size_t count, loff_t *ppos)
 634 {
 635         return count;
 636 }
 637
 638 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
 639 {
 640         return 0;
 641 }
 642
 643 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
 644 {
 645         size_t count = iov_iter_count(from);
 646         iov_iter_advance(from, count);
 647         return count;
 648 }
 649
 650 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 651                         struct splice_desc *sd)
 652 {
 653         return sd->len;
 654 }
 655
 656 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 657                                  loff_t *ppos, size_t len, unsigned int flags)
 658 {
 659         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 660 }
 661
 662 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
 663 {
 664         size_t written = 0;
 665
 666         while (iov_iter_count(iter)) {
 667                 size_t chunk = iov_iter_count(iter), n;
 668
 669                 if (chunk > PAGE_SIZE)
 670                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 671                 n = iov_iter_zero(chunk, iter);
 672                 if (!n && iov_iter_count(iter))
 673                         return written ? written : -EFAULT;
 674                 written += n;
 675                 if (signal_pending(current))
 676                         return written ? written : -ERESTARTSYS;
 677                 cond_resched();
 678         }
 679         return written;
 680 }
 681
 682 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 683 {
 684 #ifndef CONFIG_MMU
 685         return -ENOSYS;
 686 #endif
 687         if (vma->vm_flags & VM_SHARED)
 688                 return shmem_zero_setup(vma);
 689         return 0;
 690 }
 691
 692 static unsigned long get_unmapped_area_zero(struct file *file,
 693                                 unsigned long addr, unsigned long len,
 694                                 unsigned long pgoff, unsigned long flags)
 695 {
 696 #ifdef CONFIG_MMU
 697         if (flags & MAP_SHARED) {
 698                 /*
 699                  * mmap_zero() will call shmem_zero_setup() to create a file,
 700                  * so use shmem's get_unmapped_area in case it can be huge;
 701                  * and pass NULL for file as in mmap.c's get_unmapped_area(),
 702                  * so as not to confuse shmem with our handle on "/dev/zero".
 703                  */
 704                 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
 705         }
 706
 707         /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
 708         return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
 709 #else
 710         return -ENOSYS;
 711 #endif
 712 }
 713
 714 static ssize_t write_full(struct file *file, const char __user *buf,
 715                           size_t count, loff_t *ppos)
 716 {
 717         return -ENOSPC;
 718 }
 719
 720 /*
 721  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 722  * can fopen() both devices with "a" now.  This was previously impossible.
 723  * -- SRB.
 724  */
 725 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 726 {
 727         return file->f_pos = 0;
 728 }
 729
 730 /*
 731  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 732  * check against negative addresses: they are ok. The return value is weird,
 733  * though, in that case (0).
 734  *
 735  * also note that seeking relative to the "end of file" isn't supported:
 736  * it has no meaning, so it returns -EINVAL.
 737  */
 738 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 739 {
 740         loff_t ret;
 741
 742         inode_lock(file_inode(file));
 743         switch (orig) {
 744         case SEEK_CUR:
 745                 offset += file->f_pos;
 746         case SEEK_SET:
 747                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 748                 if ((unsigned long long)offset >= -MAX_ERRNO) {
 749                         ret = -EOVERFLOW;
 750                         break;
 751                 }
 752                 file->f_pos = offset;
 753                 ret = file->f_pos;
 754                 force_successful_syscall_return();
 755                 break;
 756         default:
 757                 ret = -EINVAL;
 758         }
 759         inode_unlock(file_inode(file));
 760         return ret;
 761 }
 762
 763 static int open_port(struct inode *inode, struct file *filp)
 764 {
 765         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 766 }
 767
 768 #define zero_lseek      null_lseek
 769 #define full_lseek      null_lseek
 770 #define write_zero      write_null
 771 #define write_iter_zero write_iter_null
 772 #define open_mem        open_port
 773 #define open_kmem       open_mem
 774
 775 static const struct file_operations __maybe_unused mem_fops = {
 776         .llseek         = memory_lseek,
 777         .read           = read_mem,
 778         .write          = write_mem,
 779         .mmap           = mmap_mem,
 780         .open           = open_mem,
 781 #ifndef CONFIG_MMU
 782         .get_unmapped_area = get_unmapped_area_mem,
 783         .mmap_capabilities = memory_mmap_capabilities,
 784 #endif
 785 };
 786
 787 static const struct file_operations __maybe_unused kmem_fops = {
 788         .llseek         = memory_lseek,
 789         .read           = read_kmem,
 790         .write          = write_kmem,
 791         .mmap           = mmap_kmem,
 792         .open           = open_kmem,
 793 #ifndef CONFIG_MMU
 794         .get_unmapped_area = get_unmapped_area_mem,
 795         .mmap_capabilities = memory_mmap_capabilities,
 796 #endif
 797 };
 798
 799 static const struct file_operations null_fops = {
 800         .llseek         = null_lseek,
 801         .read           = read_null,
 802         .write          = write_null,
 803         .read_iter      = read_iter_null,
 804         .write_iter     = write_iter_null,
 805         .splice_write   = splice_write_null,
 806 };
 807
 808 static const struct file_operations __maybe_unused port_fops = {
 809         .llseek         = memory_lseek,
 810         .read           = read_port,
 811         .write          = write_port,
 812         .open           = open_port,
 813 };
 814
 815 static const struct file_operations zero_fops = {
 816         .llseek         = zero_lseek,
 817         .write          = write_zero,
 818         .read_iter      = read_iter_zero,
 819         .write_iter     = write_iter_zero,
 820         .mmap           = mmap_zero,
 821         .get_unmapped_area = get_unmapped_area_zero,
 822 #ifndef CONFIG_MMU
 823         .mmap_capabilities = zero_mmap_capabilities,
 824 #endif
 825 };
 826
 827 static const struct file_operations full_fops = {
 828         .llseek         = full_lseek,
 829         .read_iter      = read_iter_zero,
 830         .write          = write_full,
 831 };
 832
 833 static const struct memdev {
 834         const char *name;
 835         umode_t mode;
 836         const struct file_operations *fops;
 837         fmode_t fmode;
 838 } devlist[] = {
 839 #ifdef CONFIG_DEVMEM
 840          [1] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
 841 #endif
 842 #ifdef CONFIG_DEVKMEM
 843          [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
 844 #endif
 845          [3] = { "null", 0666, &null_fops, 0 },
 846 #ifdef CONFIG_DEVPORT
 847          [4] = { "port", 0, &port_fops, 0 },
 848 #endif
 849          [5] = { "zero", 0666, &zero_fops, 0 },
 850          [7] = { "full", 0666, &full_fops, 0 },
 851          [8] = { "random", 0666, &random_fops, 0 },
 852          [9] = { "urandom", 0666, &urandom_fops, 0 },
 853 #ifdef CONFIG_PRINTK
 854         [11] = { "kmsg", 0644, &kmsg_fops, 0 },
 855 #endif
 856 };
 857
 858 static int memory_open(struct inode *inode, struct file *filp)
 859 {
 860         int minor;
 861         const struct memdev *dev;
 862
 863         minor = iminor(inode);
 864         if (minor >= ARRAY_SIZE(devlist))
 865                 return -ENXIO;
 866
 867         dev = &devlist[minor];
 868         if (!dev->fops)
 869                 return -ENXIO;
 870
 871         filp->f_op = dev->fops;
 872         filp->f_mode |= dev->fmode;
 873
 874         if (dev->fops->open)
 875                 return dev->fops->open(inode, filp);
 876
 877         return 0;
 878 }
 879
 880 static const struct file_operations memory_fops = {
 881         .open = memory_open,
 882         .llseek = noop_llseek,
 883 };
 884
 885 static char *mem_devnode(struct device *dev, umode_t *mode)
 886 {
 887         if (mode && devlist[MINOR(dev->devt)].mode)
 888                 *mode = devlist[MINOR(dev->devt)].mode;
 889         return NULL;
 890 }
 891
 892 static struct class *mem_class;
 893
 894 static int __init chr_dev_init(void)
 895 {
 896         int minor;
 897
 898         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 899                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 900
 901         mem_class = class_create(THIS_MODULE, "mem");
 902         if (IS_ERR(mem_class))
 903                 return PTR_ERR(mem_class);
 904
 905         mem_class->devnode = mem_devnode;
 906         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 907                 if (!devlist[minor].name)
 908                         continue;
 909
 910                 /*
 911                  * Create /dev/port?
 912                  */
 913                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 914                         continue;
 915
 916                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 917                               NULL, devlist[minor].name);
 918         }
 919
 920         return tty_init();
 921 }
 922
 923 fs_initcall(chr_dev_init);