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
 344         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 345                 return -EINVAL;
 346
 347         if (!private_mapping_ok(vma))
 348                 return -ENOSYS;
 349
 350         if (!range_is_allowed(vma->vm_pgoff, size))
 351                 return -EPERM;
 352
 353         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 354                                                 &vma->vm_page_prot))
 355                 return -EINVAL;
 356
 357         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 358                                                  size,
 359                                                  vma->vm_page_prot);
 360
 361         vma->vm_ops = &mmap_mem_ops;
 362
 363         /* Remap-pfn-range will mark the range VM_IO */
 364         if (remap_pfn_range(vma,
 365                             vma->vm_start,
 366                             vma->vm_pgoff,
 367                             size,
 368                             vma->vm_page_prot)) {
 369                 return -EAGAIN;
 370         }
 371         return 0;
 372 }
 373
 374 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 375 {
 376         unsigned long pfn;
 377
 378         /* Turn a kernel-virtual address into a physical page frame */
 379         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 380
 381         /*
 382          * RED-PEN: on some architectures there is more mapped memory than
 383          * available in mem_map which pfn_valid checks for. Perhaps should add a
 384          * new macro here.
 385          *
 386          * RED-PEN: vmalloc is not supported right now.
 387          */
 388         if (!pfn_valid(pfn))
 389                 return -EIO;
 390
 391         vma->vm_pgoff = pfn;
 392         return mmap_mem(file, vma);
 393 }
 394
 395 /*
 396  * This function reads the *virtual* memory as seen by the kernel.
 397  */
 398 static ssize_t read_kmem(struct file *file, char __user *buf,
 399                          size_t count, loff_t *ppos)
 400 {
 401         unsigned long p = *ppos;
 402         ssize_t low_count, read, sz;
 403         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 404         int err = 0;
 405
 406         read = 0;
 407         if (p < (unsigned long) high_memory) {
 408                 low_count = count;
 409                 if (count > (unsigned long)high_memory - p)
 410                         low_count = (unsigned long)high_memory - p;
 411
 412 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 413                 /* we don't have page 0 mapped on sparc and m68k.. */
 414                 if (p < PAGE_SIZE && low_count > 0) {
 415                         sz = size_inside_page(p, low_count);
 416                         if (clear_user(buf, sz))
 417                                 return -EFAULT;
 418                         buf += sz;
 419                         p += sz;
 420                         read += sz;
 421                         low_count -= sz;
 422                         count -= sz;
 423                 }
 424 #endif
 425                 while (low_count > 0) {
 426                         sz = size_inside_page(p, low_count);
 427
 428                         /*
 429                          * On ia64 if a page has been mapped somewhere as
 430                          * uncached, then it must also be accessed uncached
 431                          * by the kernel or data corruption may occur
 432                          */
 433                         kbuf = xlate_dev_kmem_ptr((void *)p);
 434                         if (!virt_addr_valid(kbuf))
 435                                 return -ENXIO;
 436
 437                         if (copy_to_user(buf, kbuf, sz))
 438                                 return -EFAULT;
 439                         buf += sz;
 440                         p += sz;
 441                         read += sz;
 442                         low_count -= sz;
 443                         count -= sz;
 444                 }
 445         }
 446
 447         if (count > 0) {
 448                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 449                 if (!kbuf)
 450                         return -ENOMEM;
 451                 while (count > 0) {
 452                         sz = size_inside_page(p, count);
 453                         if (!is_vmalloc_or_module_addr((void *)p)) {
 454                                 err = -ENXIO;
 455                                 break;
 456                         }
 457                         sz = vread(kbuf, (char *)p, sz);
 458                         if (!sz)
 459                                 break;
 460                         if (copy_to_user(buf, kbuf, sz)) {
 461                                 err = -EFAULT;
 462                                 break;
 463                         }
 464                         count -= sz;
 465                         buf += sz;
 466                         read += sz;
 467                         p += sz;
 468                 }
 469                 free_page((unsigned long)kbuf);
 470         }
 471         *ppos = p;
 472         return read ? read : err;
 473 }
 474
 475
 476 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 477                                 size_t count, loff_t *ppos)
 478 {
 479         ssize_t written, sz;
 480         unsigned long copied;
 481
 482         written = 0;
 483 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 484         /* we don't have page 0 mapped on sparc and m68k.. */
 485         if (p < PAGE_SIZE) {
 486                 sz = size_inside_page(p, count);
 487                 /* Hmm. Do something? */
 488                 buf += sz;
 489                 p += sz;
 490                 count -= sz;
 491                 written += sz;
 492         }
 493 #endif
 494
 495         while (count > 0) {
 496                 void *ptr;
 497
 498                 sz = size_inside_page(p, count);
 499
 500                 /*
 501                  * On ia64 if a page has been mapped somewhere as uncached, then
 502                  * it must also be accessed uncached by the kernel or data
 503                  * corruption may occur.
 504                  */
 505                 ptr = xlate_dev_kmem_ptr((void *)p);
 506                 if (!virt_addr_valid(ptr))
 507                         return -ENXIO;
 508
 509                 copied = copy_from_user(ptr, buf, sz);
 510                 if (copied) {
 511                         written += sz - copied;
 512                         if (written)
 513                                 break;
 514                         return -EFAULT;
 515                 }
 516                 buf += sz;
 517                 p += sz;
 518                 count -= sz;
 519                 written += sz;
 520         }
 521
 522         *ppos += written;
 523         return written;
 524 }
 525
 526 /*
 527  * This function writes to the *virtual* memory as seen by the kernel.
 528  */
 529 static ssize_t write_kmem(struct file *file, const char __user *buf,
 530                           size_t count, loff_t *ppos)
 531 {
 532         unsigned long p = *ppos;
 533         ssize_t wrote = 0;
 534         ssize_t virtr = 0;
 535         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 536         int err = 0;
 537
 538         if (p < (unsigned long) high_memory) {
 539                 unsigned long to_write = min_t(unsigned long, count,
 540                                                (unsigned long)high_memory - p);
 541                 wrote = do_write_kmem(p, buf, to_write, ppos);
 542                 if (wrote != to_write)
 543                         return wrote;
 544                 p += wrote;
 545                 buf += wrote;
 546                 count -= wrote;
 547         }
 548
 549         if (count > 0) {
 550                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 551                 if (!kbuf)
 552                         return wrote ? wrote : -ENOMEM;
 553                 while (count > 0) {
 554                         unsigned long sz = size_inside_page(p, count);
 555                         unsigned long n;
 556
 557                         if (!is_vmalloc_or_module_addr((void *)p)) {
 558                                 err = -ENXIO;
 559                                 break;
 560                         }
 561                         n = copy_from_user(kbuf, buf, sz);
 562                         if (n) {
 563                                 err = -EFAULT;
 564                                 break;
 565                         }
 566                         vwrite(kbuf, (char *)p, sz);
 567                         count -= sz;
 568                         buf += sz;
 569                         virtr += sz;
 570                         p += sz;
 571                 }
 572                 free_page((unsigned long)kbuf);
 573         }
 574
 575         *ppos = p;
 576         return virtr + wrote ? : err;
 577 }
 578
 579 static ssize_t read_port(struct file *file, char __user *buf,
 580                          size_t count, loff_t *ppos)
 581 {
 582         unsigned long i = *ppos;
 583         char __user *tmp = buf;
 584
 585         if (!access_ok(VERIFY_WRITE, buf, count))
 586                 return -EFAULT;
 587         while (count-- > 0 && i < 65536) {
 588                 if (__put_user(inb(i), tmp) < 0)
 589                         return -EFAULT;
 590                 i++;
 591                 tmp++;
 592         }
 593         *ppos = i;
 594         return tmp-buf;
 595 }
 596
 597 static ssize_t write_port(struct file *file, const char __user *buf,
 598                           size_t count, loff_t *ppos)
 599 {
 600         unsigned long i = *ppos;
 601         const char __user *tmp = buf;
 602
 603         if (!access_ok(VERIFY_READ, buf, count))
 604                 return -EFAULT;
 605         while (count-- > 0 && i < 65536) {
 606                 char c;
 607
 608                 if (__get_user(c, tmp)) {
 609                         if (tmp > buf)
 610                                 break;
 611                         return -EFAULT;
 612                 }
 613                 outb(c, i);
 614                 i++;
 615                 tmp++;
 616         }
 617         *ppos = i;
 618         return tmp-buf;
 619 }
 620
 621 static ssize_t read_null(struct file *file, char __user *buf,
 622                          size_t count, loff_t *ppos)
 623 {
 624         return 0;
 625 }
 626
 627 static ssize_t write_null(struct file *file, const char __user *buf,
 628                           size_t count, loff_t *ppos)
 629 {
 630         return count;
 631 }
 632
 633 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
 634 {
 635         return 0;
 636 }
 637
 638 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
 639 {
 640         size_t count = iov_iter_count(from);
 641         iov_iter_advance(from, count);
 642         return count;
 643 }
 644
 645 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 646                         struct splice_desc *sd)
 647 {
 648         return sd->len;
 649 }
 650
 651 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 652                                  loff_t *ppos, size_t len, unsigned int flags)
 653 {
 654         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 655 }
 656
 657 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
 658 {
 659         size_t written = 0;
 660
 661         while (iov_iter_count(iter)) {
 662                 size_t chunk = iov_iter_count(iter), n;
 663
 664                 if (chunk > PAGE_SIZE)
 665                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 666                 n = iov_iter_zero(chunk, iter);
 667                 if (!n && iov_iter_count(iter))
 668                         return written ? written : -EFAULT;
 669                 written += n;
 670                 if (signal_pending(current))
 671                         return written ? written : -ERESTARTSYS;
 672                 cond_resched();
 673         }
 674         return written;
 675 }
 676
 677 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 678 {
 679 #ifndef CONFIG_MMU
 680         return -ENOSYS;
 681 #endif
 682         if (vma->vm_flags & VM_SHARED)
 683                 return shmem_zero_setup(vma);
 684         return 0;
 685 }
 686
 687 static unsigned long get_unmapped_area_zero(struct file *file,
 688                                 unsigned long addr, unsigned long len,
 689                                 unsigned long pgoff, unsigned long flags)
 690 {
 691 #ifdef CONFIG_MMU
 692         if (flags & MAP_SHARED) {
 693                 /*
 694                  * mmap_zero() will call shmem_zero_setup() to create a file,
 695                  * so use shmem's get_unmapped_area in case it can be huge;
 696                  * and pass NULL for file as in mmap.c's get_unmapped_area(),
 697                  * so as not to confuse shmem with our handle on "/dev/zero".
 698                  */
 699                 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
 700         }
 701
 702         /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
 703         return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
 704 #else
 705         return -ENOSYS;
 706 #endif
 707 }
 708
 709 static ssize_t write_full(struct file *file, const char __user *buf,
 710                           size_t count, loff_t *ppos)
 711 {
 712         return -ENOSPC;
 713 }
 714
 715 /*
 716  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 717  * can fopen() both devices with "a" now.  This was previously impossible.
 718  * -- SRB.
 719  */
 720 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 721 {
 722         return file->f_pos = 0;
 723 }
 724
 725 /*
 726  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 727  * check against negative addresses: they are ok. The return value is weird,
 728  * though, in that case (0).
 729  *
 730  * also note that seeking relative to the "end of file" isn't supported:
 731  * it has no meaning, so it returns -EINVAL.
 732  */
 733 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 734 {
 735         loff_t ret;
 736
 737         inode_lock(file_inode(file));
 738         switch (orig) {
 739         case SEEK_CUR:
 740                 offset += file->f_pos;
 741         case SEEK_SET:
 742                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 743                 if ((unsigned long long)offset >= -MAX_ERRNO) {
 744                         ret = -EOVERFLOW;
 745                         break;
 746                 }
 747                 file->f_pos = offset;
 748                 ret = file->f_pos;
 749                 force_successful_syscall_return();
 750                 break;
 751         default:
 752                 ret = -EINVAL;
 753         }
 754         inode_unlock(file_inode(file));
 755         return ret;
 756 }
 757
 758 static int open_port(struct inode *inode, struct file *filp)
 759 {
 760         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 761 }
 762
 763 #define zero_lseek      null_lseek
 764 #define full_lseek      null_lseek
 765 #define write_zero      write_null
 766 #define write_iter_zero write_iter_null
 767 #define open_mem        open_port
 768 #define open_kmem       open_mem
 769
 770 static const struct file_operations __maybe_unused mem_fops = {
 771         .llseek         = memory_lseek,
 772         .read           = read_mem,
 773         .write          = write_mem,
 774         .mmap           = mmap_mem,
 775         .open           = open_mem,
 776 #ifndef CONFIG_MMU
 777         .get_unmapped_area = get_unmapped_area_mem,
 778         .mmap_capabilities = memory_mmap_capabilities,
 779 #endif
 780 };
 781
 782 static const struct file_operations __maybe_unused kmem_fops = {
 783         .llseek         = memory_lseek,
 784         .read           = read_kmem,
 785         .write          = write_kmem,
 786         .mmap           = mmap_kmem,
 787         .open           = open_kmem,
 788 #ifndef CONFIG_MMU
 789         .get_unmapped_area = get_unmapped_area_mem,
 790         .mmap_capabilities = memory_mmap_capabilities,
 791 #endif
 792 };
 793
 794 static const struct file_operations null_fops = {
 795         .llseek         = null_lseek,
 796         .read           = read_null,
 797         .write          = write_null,
 798         .read_iter      = read_iter_null,
 799         .write_iter     = write_iter_null,
 800         .splice_write   = splice_write_null,
 801 };
 802
 803 static const struct file_operations __maybe_unused port_fops = {
 804         .llseek         = memory_lseek,
 805         .read           = read_port,
 806         .write          = write_port,
 807         .open           = open_port,
 808 };
 809
 810 static const struct file_operations zero_fops = {
 811         .llseek         = zero_lseek,
 812         .write          = write_zero,
 813         .read_iter      = read_iter_zero,
 814         .write_iter     = write_iter_zero,
 815         .mmap           = mmap_zero,
 816         .get_unmapped_area = get_unmapped_area_zero,
 817 #ifndef CONFIG_MMU
 818         .mmap_capabilities = zero_mmap_capabilities,
 819 #endif
 820 };
 821
 822 static const struct file_operations full_fops = {
 823         .llseek         = full_lseek,
 824         .read_iter      = read_iter_zero,
 825         .write          = write_full,
 826 };
 827
 828 static const struct memdev {
 829         const char *name;
 830         umode_t mode;
 831         const struct file_operations *fops;
 832         fmode_t fmode;
 833 } devlist[] = {
 834 #ifdef CONFIG_DEVMEM
 835          [1] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
 836 #endif
 837 #ifdef CONFIG_DEVKMEM
 838          [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
 839 #endif
 840          [3] = { "null", 0666, &null_fops, 0 },
 841 #ifdef CONFIG_DEVPORT
 842          [4] = { "port", 0, &port_fops, 0 },
 843 #endif
 844          [5] = { "zero", 0666, &zero_fops, 0 },
 845          [7] = { "full", 0666, &full_fops, 0 },
 846          [8] = { "random", 0666, &random_fops, 0 },
 847          [9] = { "urandom", 0666, &urandom_fops, 0 },
 848 #ifdef CONFIG_PRINTK
 849         [11] = { "kmsg", 0644, &kmsg_fops, 0 },
 850 #endif
 851 };
 852
 853 static int memory_open(struct inode *inode, struct file *filp)
 854 {
 855         int minor;
 856         const struct memdev *dev;
 857
 858         minor = iminor(inode);
 859         if (minor >= ARRAY_SIZE(devlist))
 860                 return -ENXIO;
 861
 862         dev = &devlist[minor];
 863         if (!dev->fops)
 864                 return -ENXIO;
 865
 866         filp->f_op = dev->fops;
 867         filp->f_mode |= dev->fmode;
 868
 869         if (dev->fops->open)
 870                 return dev->fops->open(inode, filp);
 871
 872         return 0;
 873 }
 874
 875 static const struct file_operations memory_fops = {
 876         .open = memory_open,
 877         .llseek = noop_llseek,
 878 };
 879
 880 static char *mem_devnode(struct device *dev, umode_t *mode)
 881 {
 882         if (mode && devlist[MINOR(dev->devt)].mode)
 883                 *mode = devlist[MINOR(dev->devt)].mode;
 884         return NULL;
 885 }
 886
 887 static struct class *mem_class;
 888
 889 static int __init chr_dev_init(void)
 890 {
 891         int minor;
 892
 893         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 894                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 895
 896         mem_class = class_create(THIS_MODULE, "mem");
 897         if (IS_ERR(mem_class))
 898                 return PTR_ERR(mem_class);
 899
 900         mem_class->devnode = mem_devnode;
 901         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 902                 if (!devlist[minor].name)
 903                         continue;
 904
 905                 /*
 906                  * Create /dev/port?
 907                  */
 908                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 909                         continue;
 910
 911                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 912                               NULL, devlist[minor].name);
 913         }
 914
 915         return tty_init();
 916 }
 917
 918 fs_initcall(chr_dev_init);