SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 mmaping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
   9  */
  10
  11 #include <linux/config.h>
  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/smp_lock.h>
  23 #include <linux/devfs_fs_kernel.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/device.h>
  26 #include <linux/highmem.h>
  27 #include <linux/crash_dump.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/bootmem.h>
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/io.h>
  33
  34 #ifdef CONFIG_IA64
  35 # include <linux/efi.h>
  36 #endif
  37
  38 /*
  39  * Architectures vary in how they handle caching for addresses
  40  * outside of main memory.
  41  *
  42  */
  43 static inline int uncached_access(struct file *file, unsigned long addr)
  44 {
  45 #if defined(__i386__)
  46         /*
  47          * On the PPro and successors, the MTRRs are used to set
  48          * memory types for physical addresses outside main memory,
  49          * so blindly setting PCD or PWT on those pages is wrong.
  50          * For Pentiums and earlier, the surround logic should disable
  51          * caching for the high addresses through the KEN pin, but
  52          * we maintain the tradition of paranoia in this code.
  53          */
  54         if (file->f_flags & O_SYNC)
  55                 return 1;
  56         return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
  57                   test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
  58                   test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
  59                   test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
  60           && addr >= __pa(high_memory);
  61 #elif defined(__x86_64__)
  62         /*
  63          * This is broken because it can generate memory type aliases,
  64          * which can cause cache corruptions
  65          * But it is only available for root and we have to be bug-to-bug
  66          * compatible with i386.
  67          */
  68         if (file->f_flags & O_SYNC)
  69                 return 1;
  70         /* same behaviour as i386. PAT always set to cached and MTRRs control the
  71            caching behaviour.
  72            Hopefully a full PAT implementation will fix that soon. */
  73         return 0;
  74 #elif defined(CONFIG_IA64)
  75         /*
  76          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  77          */
  78         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  79 #else
  80         /*
  81          * Accessing memory above the top the kernel knows about or through a file pointer
  82          * that was marked O_SYNC will be done non-cached.
  83          */
  84         if (file->f_flags & O_SYNC)
  85                 return 1;
  86         return addr >= __pa(high_memory);
  87 #endif
  88 }
  89
  90 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  91 static inline int valid_phys_addr_range(unsigned long addr, size_t *count)
  92 {
  93         unsigned long end_mem;
  94
  95         end_mem = __pa(high_memory);
  96         if (addr >= end_mem)
  97                 return 0;
  98
  99         if (*count > end_mem - addr)
 100                 *count = end_mem - addr;
 101
 102         return 1;
 103 }
 104 #endif
 105
 106 /*
 107  * This funcion reads the *physical* memory. The f_pos points directly to the
 108  * memory location.
 109  */
 110 static ssize_t read_mem(struct file * file, char __user * buf,
 111                         size_t count, loff_t *ppos)
 112 {
 113         unsigned long p = *ppos;
 114         ssize_t read, sz;
 115         char *ptr;
 116
 117         if (!valid_phys_addr_range(p, &count))
 118                 return -EFAULT;
 119         read = 0;
 120 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 121         /* we don't have page 0 mapped on sparc and m68k.. */
 122         if (p < PAGE_SIZE) {
 123                 sz = PAGE_SIZE - p;
 124                 if (sz > count)
 125                         sz = count;
 126                 if (sz > 0) {
 127                         if (clear_user(buf, sz))
 128                                 return -EFAULT;
 129                         buf += sz;
 130                         p += sz;
 131                         count -= sz;
 132                         read += sz;
 133                 }
 134         }
 135 #endif
 136
 137         while (count > 0) {
 138                 /*
 139                  * Handle first page in case it's not aligned
 140                  */
 141                 if (-p & (PAGE_SIZE - 1))
 142                         sz = -p & (PAGE_SIZE - 1);
 143                 else
 144                         sz = PAGE_SIZE;
 145
 146                 sz = min_t(unsigned long, sz, count);
 147
 148                 /*
 149                  * On ia64 if a page has been mapped somewhere as
 150                  * uncached, then it must also be accessed uncached
 151                  * by the kernel or data corruption may occur
 152                  */
 153                 ptr = xlate_dev_mem_ptr(p);
 154
 155                 if (copy_to_user(buf, ptr, sz))
 156                         return -EFAULT;
 157                 buf += sz;
 158                 p += sz;
 159                 count -= sz;
 160                 read += sz;
 161         }
 162
 163         *ppos += read;
 164         return read;
 165 }
 166
 167 static ssize_t write_mem(struct file * file, const char __user * buf,
 168                          size_t count, loff_t *ppos)
 169 {
 170         unsigned long p = *ppos;
 171         ssize_t written, sz;
 172         unsigned long copied;
 173         void *ptr;
 174
 175         if (!valid_phys_addr_range(p, &count))
 176                 return -EFAULT;
 177
 178         written = 0;
 179
 180 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 181         /* we don't have page 0 mapped on sparc and m68k.. */
 182         if (p < PAGE_SIZE) {
 183                 unsigned long sz = PAGE_SIZE - p;
 184                 if (sz > count)
 185                         sz = count;
 186                 /* Hmm. Do something? */
 187                 buf += sz;
 188                 p += sz;
 189                 count -= sz;
 190                 written += sz;
 191         }
 192 #endif
 193
 194         while (count > 0) {
 195                 /*
 196                  * Handle first page in case it's not aligned
 197                  */
 198                 if (-p & (PAGE_SIZE - 1))
 199                         sz = -p & (PAGE_SIZE - 1);
 200                 else
 201                         sz = PAGE_SIZE;
 202
 203                 sz = min_t(unsigned long, sz, count);
 204
 205                 /*
 206                  * On ia64 if a page has been mapped somewhere as
 207                  * uncached, then it must also be accessed uncached
 208                  * by the kernel or data corruption may occur
 209                  */
 210                 ptr = xlate_dev_mem_ptr(p);
 211
 212                 copied = copy_from_user(ptr, buf, sz);
 213                 if (copied) {
 214                         ssize_t ret;
 215
 216                         ret = written + (sz - copied);
 217                         if (ret)
 218                                 return ret;
 219                         return -EFAULT;
 220                 }
 221                 buf += sz;
 222                 p += sz;
 223                 count -= sz;
 224                 written += sz;
 225         }
 226
 227         *ppos += written;
 228         return written;
 229 }
 230
 231 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 232 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 233                                      unsigned long size, pgprot_t vma_prot)
 234 {
 235 #ifdef pgprot_noncached
 236         unsigned long offset = pfn << PAGE_SHIFT;
 237
 238         if (uncached_access(file, offset))
 239                 return pgprot_noncached(vma_prot);
 240 #endif
 241         return vma_prot;
 242 }
 243 #endif
 244
 245 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 246 {
 247         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 248                                                  vma->vm_end - vma->vm_start,
 249                                                  vma->vm_page_prot);
 250
 251         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 252         if (remap_pfn_range(vma,
 253                             vma->vm_start,
 254                             vma->vm_pgoff,
 255                             vma->vm_end-vma->vm_start,
 256                             vma->vm_page_prot))
 257                 return -EAGAIN;
 258         return 0;
 259 }
 260
 261 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 262 {
 263         unsigned long pfn;
 264
 265         /* Turn a kernel-virtual address into a physical page frame */
 266         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 267
 268         /*
 269          * RED-PEN: on some architectures there is more mapped memory
 270          * than available in mem_map which pfn_valid checks
 271          * for. Perhaps should add a new macro here.
 272          *
 273          * RED-PEN: vmalloc is not supported right now.
 274          */
 275         if (!pfn_valid(pfn))
 276                 return -EIO;
 277
 278         vma->vm_pgoff = pfn;
 279         return mmap_mem(file, vma);
 280 }
 281
 282 #ifdef CONFIG_CRASH_DUMP
 283 /*
 284  * Read memory corresponding to the old kernel.
 285  */
 286 static ssize_t read_oldmem(struct file *file, char __user *buf,
 287                                 size_t count, loff_t *ppos)
 288 {
 289         unsigned long pfn, offset;
 290         size_t read = 0, csize;
 291         int rc = 0;
 292
 293         while (count) {
 294                 pfn = *ppos / PAGE_SIZE;
 295                 if (pfn > saved_max_pfn)
 296                         return read;
 297
 298                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 299                 if (count > PAGE_SIZE - offset)
 300                         csize = PAGE_SIZE - offset;
 301                 else
 302                         csize = count;
 303
 304                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 305                 if (rc < 0)
 306                         return rc;
 307                 buf += csize;
 308                 *ppos += csize;
 309                 read += csize;
 310                 count -= csize;
 311         }
 312         return read;
 313 }
 314 #endif
 315
 316 extern long vread(char *buf, char *addr, unsigned long count);
 317 extern long vwrite(char *buf, char *addr, unsigned long count);
 318
 319 /*
 320  * This function reads the *virtual* memory as seen by the kernel.
 321  */
 322 static ssize_t read_kmem(struct file *file, char __user *buf,
 323                          size_t count, loff_t *ppos)
 324 {
 325         unsigned long p = *ppos;
 326         ssize_t low_count, read, sz;
 327         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 328
 329         read = 0;
 330         if (p < (unsigned long) high_memory) {
 331                 low_count = count;
 332                 if (count > (unsigned long) high_memory - p)
 333                         low_count = (unsigned long) high_memory - p;
 334
 335 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 336                 /* we don't have page 0 mapped on sparc and m68k.. */
 337                 if (p < PAGE_SIZE && low_count > 0) {
 338                         size_t tmp = PAGE_SIZE - p;
 339                         if (tmp > low_count) tmp = low_count;
 340                         if (clear_user(buf, tmp))
 341                                 return -EFAULT;
 342                         buf += tmp;
 343                         p += tmp;
 344                         read += tmp;
 345                         low_count -= tmp;
 346                         count -= tmp;
 347                 }
 348 #endif
 349                 while (low_count > 0) {
 350                         /*
 351                          * Handle first page in case it's not aligned
 352                          */
 353                         if (-p & (PAGE_SIZE - 1))
 354                                 sz = -p & (PAGE_SIZE - 1);
 355                         else
 356                                 sz = PAGE_SIZE;
 357
 358                         sz = min_t(unsigned long, sz, low_count);
 359
 360                         /*
 361                          * On ia64 if a page has been mapped somewhere as
 362                          * uncached, then it must also be accessed uncached
 363                          * by the kernel or data corruption may occur
 364                          */
 365                         kbuf = xlate_dev_kmem_ptr((char *)p);
 366
 367                         if (copy_to_user(buf, kbuf, sz))
 368                                 return -EFAULT;
 369                         buf += sz;
 370                         p += sz;
 371                         read += sz;
 372                         low_count -= sz;
 373                         count -= sz;
 374                 }
 375         }
 376
 377         if (count > 0) {
 378                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 379                 if (!kbuf)
 380                         return -ENOMEM;
 381                 while (count > 0) {
 382                         int len = count;
 383
 384                         if (len > PAGE_SIZE)
 385                                 len = PAGE_SIZE;
 386                         len = vread(kbuf, (char *)p, len);
 387                         if (!len)
 388                                 break;
 389                         if (copy_to_user(buf, kbuf, len)) {
 390                                 free_page((unsigned long)kbuf);
 391                                 return -EFAULT;
 392                         }
 393                         count -= len;
 394                         buf += len;
 395                         read += len;
 396                         p += len;
 397                 }
 398                 free_page((unsigned long)kbuf);
 399         }
 400         *ppos = p;
 401         return read;
 402 }
 403
 404
 405 static inline ssize_t
 406 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 407               size_t count, loff_t *ppos)
 408 {
 409         ssize_t written, sz;
 410         unsigned long copied;
 411
 412         written = 0;
 413 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 414         /* we don't have page 0 mapped on sparc and m68k.. */
 415         if (realp < PAGE_SIZE) {
 416                 unsigned long sz = PAGE_SIZE - realp;
 417                 if (sz > count)
 418                         sz = count;
 419                 /* Hmm. Do something? */
 420                 buf += sz;
 421                 p += sz;
 422                 realp += sz;
 423                 count -= sz;
 424                 written += sz;
 425         }
 426 #endif
 427
 428         while (count > 0) {
 429                 char *ptr;
 430                 /*
 431                  * Handle first page in case it's not aligned
 432                  */
 433                 if (-realp & (PAGE_SIZE - 1))
 434                         sz = -realp & (PAGE_SIZE - 1);
 435                 else
 436                         sz = PAGE_SIZE;
 437
 438                 sz = min_t(unsigned long, sz, count);
 439
 440                 /*
 441                  * On ia64 if a page has been mapped somewhere as
 442                  * uncached, then it must also be accessed uncached
 443                  * by the kernel or data corruption may occur
 444                  */
 445                 ptr = xlate_dev_kmem_ptr(p);
 446
 447                 copied = copy_from_user(ptr, buf, sz);
 448                 if (copied) {
 449                         ssize_t ret;
 450
 451                         ret = written + (sz - copied);
 452                         if (ret)
 453                                 return ret;
 454                         return -EFAULT;
 455                 }
 456                 buf += sz;
 457                 p += sz;
 458                 realp += sz;
 459                 count -= sz;
 460                 written += sz;
 461         }
 462
 463         *ppos += written;
 464         return written;
 465 }
 466
 467
 468 /*
 469  * This function writes to the *virtual* memory as seen by the kernel.
 470  */
 471 static ssize_t write_kmem(struct file * file, const char __user * buf,
 472                           size_t count, loff_t *ppos)
 473 {
 474         unsigned long p = *ppos;
 475         ssize_t wrote = 0;
 476         ssize_t virtr = 0;
 477         ssize_t written;
 478         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 479
 480         if (p < (unsigned long) high_memory) {
 481
 482                 wrote = count;
 483                 if (count > (unsigned long) high_memory - p)
 484                         wrote = (unsigned long) high_memory - p;
 485
 486                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 487                 if (written != wrote)
 488                         return written;
 489                 wrote = written;
 490                 p += wrote;
 491                 buf += wrote;
 492                 count -= wrote;
 493         }
 494
 495         if (count > 0) {
 496                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 497                 if (!kbuf)
 498                         return wrote ? wrote : -ENOMEM;
 499                 while (count > 0) {
 500                         int len = count;
 501
 502                         if (len > PAGE_SIZE)
 503                                 len = PAGE_SIZE;
 504                         if (len) {
 505                                 written = copy_from_user(kbuf, buf, len);
 506                                 if (written) {
 507                                         ssize_t ret;
 508
 509                                         free_page((unsigned long)kbuf);
 510                                         ret = wrote + virtr + (len - written);
 511                                         return ret ? ret : -EFAULT;
 512                                 }
 513                         }
 514                         len = vwrite(kbuf, (char *)p, len);
 515                         count -= len;
 516                         buf += len;
 517                         virtr += len;
 518                         p += len;
 519                 }
 520                 free_page((unsigned long)kbuf);
 521         }
 522
 523         *ppos = p;
 524         return virtr + wrote;
 525 }
 526
 527 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 528 static ssize_t read_port(struct file * file, char __user * buf,
 529                          size_t count, loff_t *ppos)
 530 {
 531         unsigned long i = *ppos;
 532         char __user *tmp = buf;
 533
 534         if (!access_ok(VERIFY_WRITE, buf, count))
 535                 return -EFAULT;
 536         while (count-- > 0 && i < 65536) {
 537                 if (__put_user(inb(i),tmp) < 0)
 538                         return -EFAULT;
 539                 i++;
 540                 tmp++;
 541         }
 542         *ppos = i;
 543         return tmp-buf;
 544 }
 545
 546 static ssize_t write_port(struct file * file, const char __user * buf,
 547                           size_t count, loff_t *ppos)
 548 {
 549         unsigned long i = *ppos;
 550         const char __user * tmp = buf;
 551
 552         if (!access_ok(VERIFY_READ,buf,count))
 553                 return -EFAULT;
 554         while (count-- > 0 && i < 65536) {
 555                 char c;
 556                 if (__get_user(c, tmp))
 557                         return -EFAULT;
 558                 outb(c,i);
 559                 i++;
 560                 tmp++;
 561         }
 562         *ppos = i;
 563         return tmp-buf;
 564 }
 565 #endif
 566
 567 static ssize_t read_null(struct file * file, char __user * buf,
 568                          size_t count, loff_t *ppos)
 569 {
 570         return 0;
 571 }
 572
 573 static ssize_t write_null(struct file * file, const char __user * buf,
 574                           size_t count, loff_t *ppos)
 575 {
 576         return count;
 577 }
 578
 579 #ifdef CONFIG_MMU
 580 /*
 581  * For fun, we are using the MMU for this.
 582  */
 583 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
 584 {
 585         struct mm_struct *mm;
 586         struct vm_area_struct * vma;
 587         unsigned long addr=(unsigned long)buf;
 588
 589         mm = current->mm;
 590         /* Oops, this was forgotten before. -ben */
 591         down_read(&mm->mmap_sem);
 592
 593         /* For private mappings, just map in zero pages. */
 594         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 595                 unsigned long count;
 596
 597                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 598                         goto out_up;
 599                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
 600                         break;
 601                 count = vma->vm_end - addr;
 602                 if (count > size)
 603                         count = size;
 604
 605                 zap_page_range(vma, addr, count, NULL);
 606                 zeromap_page_range(vma, addr, count, PAGE_COPY);
 607
 608                 size -= count;
 609                 buf += count;
 610                 addr += count;
 611                 if (size == 0)
 612                         goto out_up;
 613         }
 614
 615         up_read(&mm->mmap_sem);
 616
 617         /* The shared case is hard. Let's do the conventional zeroing. */
 618         do {
 619                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 620                 if (unwritten)
 621                         return size + unwritten - PAGE_SIZE;
 622                 cond_resched();
 623                 buf += PAGE_SIZE;
 624                 size -= PAGE_SIZE;
 625         } while (size);
 626
 627         return size;
 628 out_up:
 629         up_read(&mm->mmap_sem);
 630         return size;
 631 }
 632
 633 static ssize_t read_zero(struct file * file, char __user * buf,
 634                          size_t count, loff_t *ppos)
 635 {
 636         unsigned long left, unwritten, written = 0;
 637
 638         if (!count)
 639                 return 0;
 640
 641         if (!access_ok(VERIFY_WRITE, buf, count))
 642                 return -EFAULT;
 643
 644         left = count;
 645
 646         /* do we want to be clever? Arbitrary cut-off */
 647         if (count >= PAGE_SIZE*4) {
 648                 unsigned long partial;
 649
 650                 /* How much left of the page? */
 651                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 652                 unwritten = clear_user(buf, partial);
 653                 written = partial - unwritten;
 654                 if (unwritten)
 655                         goto out;
 656                 left -= partial;
 657                 buf += partial;
 658                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 659                 written += (left & PAGE_MASK) - unwritten;
 660                 if (unwritten)
 661                         goto out;
 662                 buf += left & PAGE_MASK;
 663                 left &= ~PAGE_MASK;
 664         }
 665         unwritten = clear_user(buf, left);
 666         written += left - unwritten;
 667 out:
 668         return written ? written : -EFAULT;
 669 }
 670
 671 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 672 {
 673         if (vma->vm_flags & VM_SHARED)
 674                 return shmem_zero_setup(vma);
 675         if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
 676                 return -EAGAIN;
 677         return 0;
 678 }
 679 #else /* CONFIG_MMU */
 680 static ssize_t read_zero(struct file * file, char * buf,
 681                          size_t count, loff_t *ppos)
 682 {
 683         size_t todo = count;
 684
 685         while (todo) {
 686                 size_t chunk = todo;
 687
 688                 if (chunk > 4096)
 689                         chunk = 4096;   /* Just for latency reasons */
 690                 if (clear_user(buf, chunk))
 691                         return -EFAULT;
 692                 buf += chunk;
 693                 todo -= chunk;
 694                 cond_resched();
 695         }
 696         return count;
 697 }
 698
 699 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 700 {
 701         return -ENOSYS;
 702 }
 703 #endif /* CONFIG_MMU */
 704
 705 static ssize_t write_full(struct file * file, const char __user * buf,
 706                           size_t count, loff_t *ppos)
 707 {
 708         return -ENOSPC;
 709 }
 710
 711 /*
 712  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 713  * can fopen() both devices with "a" now.  This was previously impossible.
 714  * -- SRB.
 715  */
 716
 717 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 718 {
 719         return file->f_pos = 0;
 720 }
 721
 722 /*
 723  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 724  * check against negative addresses: they are ok. The return value is weird,
 725  * though, in that case (0).
 726  *
 727  * also note that seeking relative to the "end of file" isn't supported:
 728  * it has no meaning, so it returns -EINVAL.
 729  */
 730 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 731 {
 732         loff_t ret;
 733
 734         down(&file->f_dentry->d_inode->i_sem);
 735         switch (orig) {
 736                 case 0:
 737                         file->f_pos = offset;
 738                         ret = file->f_pos;
 739                         force_successful_syscall_return();
 740                         break;
 741                 case 1:
 742                         file->f_pos += offset;
 743                         ret = file->f_pos;
 744                         force_successful_syscall_return();
 745                         break;
 746                 default:
 747                         ret = -EINVAL;
 748         }
 749         up(&file->f_dentry->d_inode->i_sem);
 750         return ret;
 751 }
 752
 753 static int open_port(struct inode * inode, struct file * filp)
 754 {
 755         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 756 }
 757
 758 #define zero_lseek      null_lseek
 759 #define full_lseek      null_lseek
 760 #define write_zero      write_null
 761 #define read_full       read_zero
 762 #define open_mem        open_port
 763 #define open_kmem       open_mem
 764 #define open_oldmem     open_mem
 765
 766 static struct file_operations mem_fops = {
 767         .llseek         = memory_lseek,
 768         .read           = read_mem,
 769         .write          = write_mem,
 770         .mmap           = mmap_mem,
 771         .open           = open_mem,
 772 };
 773
 774 static struct file_operations kmem_fops = {
 775         .llseek         = memory_lseek,
 776         .read           = read_kmem,
 777         .write          = write_kmem,
 778         .mmap           = mmap_kmem,
 779         .open           = open_kmem,
 780 };
 781
 782 static struct file_operations null_fops = {
 783         .llseek         = null_lseek,
 784         .read           = read_null,
 785         .write          = write_null,
 786 };
 787
 788 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 789 static struct file_operations port_fops = {
 790         .llseek         = memory_lseek,
 791         .read           = read_port,
 792         .write          = write_port,
 793         .open           = open_port,
 794 };
 795 #endif
 796
 797 static struct file_operations zero_fops = {
 798         .llseek         = zero_lseek,
 799         .read           = read_zero,
 800         .write          = write_zero,
 801         .mmap           = mmap_zero,
 802 };
 803
 804 static struct backing_dev_info zero_bdi = {
 805         .capabilities   = BDI_CAP_MAP_COPY,
 806 };
 807
 808 static struct file_operations full_fops = {
 809         .llseek         = full_lseek,
 810         .read           = read_full,
 811         .write          = write_full,
 812 };
 813
 814 #ifdef CONFIG_CRASH_DUMP
 815 static struct file_operations oldmem_fops = {
 816         .read   = read_oldmem,
 817         .open   = open_oldmem,
 818 };
 819 #endif
 820
 821 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 822                           size_t count, loff_t *ppos)
 823 {
 824         char *tmp;
 825         ssize_t ret;
 826
 827         tmp = kmalloc(count + 1, GFP_KERNEL);
 828         if (tmp == NULL)
 829                 return -ENOMEM;
 830         ret = -EFAULT;
 831         if (!copy_from_user(tmp, buf, count)) {
 832                 tmp[count] = 0;
 833                 ret = printk("%s", tmp);
 834                 if (ret > count)
 835                         /* printk can add a prefix */
 836                         ret = count;
 837         }
 838         kfree(tmp);
 839         return ret;
 840 }
 841
 842 static struct file_operations kmsg_fops = {
 843         .write =        kmsg_write,
 844 };
 845
 846 static int memory_open(struct inode * inode, struct file * filp)
 847 {
 848         switch (iminor(inode)) {
 849                 case 1:
 850                         filp->f_op = &mem_fops;
 851                         break;
 852                 case 2:
 853                         filp->f_op = &kmem_fops;
 854                         break;
 855                 case 3:
 856                         filp->f_op = &null_fops;
 857                         break;
 858 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 859                 case 4:
 860                         filp->f_op = &port_fops;
 861                         break;
 862 #endif
 863                 case 5:
 864                         filp->f_mapping->backing_dev_info = &zero_bdi;
 865                         filp->f_op = &zero_fops;
 866                         break;
 867                 case 7:
 868                         filp->f_op = &full_fops;
 869                         break;
 870                 case 8:
 871                         filp->f_op = &random_fops;
 872                         break;
 873                 case 9:
 874                         filp->f_op = &urandom_fops;
 875                         break;
 876                 case 11:
 877                         filp->f_op = &kmsg_fops;
 878                         break;
 879 #ifdef CONFIG_CRASH_DUMP
 880                 case 12:
 881                         filp->f_op = &oldmem_fops;
 882                         break;
 883 #endif
 884                 default:
 885                         return -ENXIO;
 886         }
 887         if (filp->f_op && filp->f_op->open)
 888                 return filp->f_op->open(inode,filp);
 889         return 0;
 890 }
 891
 892 static struct file_operations memory_fops = {
 893         .open           = memory_open,  /* just a selector for the real open */
 894 };
 895
 896 static const struct {
 897         unsigned int            minor;
 898         char                    *name;
 899         umode_t                 mode;
 900         struct file_operations  *fops;
 901 } devlist[] = { /* list of minor devices */
 902         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 903         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 904         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 905 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 906         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 907 #endif
 908         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 909         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 910         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 911         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 912         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 913 #ifdef CONFIG_CRASH_DUMP
 914         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 915 #endif
 916 };
 917
 918 static struct class *mem_class;
 919
 920 static int __init chr_dev_init(void)
 921 {
 922         int i;
 923
 924         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 925                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 926
 927         mem_class = class_create(THIS_MODULE, "mem");
 928         for (i = 0; i < ARRAY_SIZE(devlist); i++) {
 929                 class_device_create(mem_class, NULL,
 930                                         MKDEV(MEM_MAJOR, devlist[i].minor),
 931                                         NULL, devlist[i].name);
 932                 devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor),
 933                                 S_IFCHR | devlist[i].mode, devlist[i].name);
 934         }
 935
 936         return 0;
 937 }
 938
 939 fs_initcall(chr_dev_init);