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/backing-dev.h>
  27
  28 #include <asm/uaccess.h>
  29 #include <asm/io.h>
  30
  31 #ifdef CONFIG_IA64
  32 # include <linux/efi.h>
  33 #endif
  34
  35 #if defined(CONFIG_S390_TAPE) && defined(CONFIG_S390_TAPE_CHAR)
  36 extern void tapechar_init(void);
  37 #endif
  38
  39 static inline int range_is_allowed(unsigned long from, unsigned long to)
  40 {
  41         unsigned long cursor;
  42
  43         cursor = from >> PAGE_SHIFT;
  44         while ((cursor << PAGE_SHIFT) < to) {
  45                 if (!devmem_is_allowed(cursor)) {
  46                         printk ("Program %s tried to read /dev/mem between %lx->%lx."
  47                                         "We stopped at %lx\n", current->comm, from, to, cursor);
  48                         return 0;
  49                 }
  50                 cursor++;
  51         }
  52         return 1;
  53 }
  54
  55 /*
  56  * Architectures vary in how they handle caching for addresses
  57  * outside of main memory.
  58  *
  59  */
  60 static inline int uncached_access(struct file *file, unsigned long addr)
  61 {
  62 #if defined(__i386__)
  63         /*
  64          * On the PPro and successors, the MTRRs are used to set
  65          * memory types for physical addresses outside main memory,
  66          * so blindly setting PCD or PWT on those pages is wrong.
  67          * For Pentiums and earlier, the surround logic should disable
  68          * caching for the high addresses through the KEN pin, but
  69          * we maintain the tradition of paranoia in this code.
  70          */
  71         if (file->f_flags & O_SYNC)
  72                 return 1;
  73         return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
  74                   test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
  75                   test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
  76                   test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
  77           && addr >= __pa(high_memory);
  78 #elif defined(__x86_64__)
  79         /*
  80          * This is broken because it can generate memory type aliases,
  81          * which can cause cache corruptions
  82          * But it is only available for root and we have to be bug-to-bug
  83          * compatible with i386.
  84          */
  85         if (file->f_flags & O_SYNC)
  86                 return 1;
  87         /* same behaviour as i386. PAT always set to cached and MTRRs control the
  88            caching behaviour.
  89            Hopefully a full PAT implementation will fix that soon. */
  90         return 0;
  91 #elif defined(CONFIG_IA64)
  92         /*
  93          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  94          */
  95         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  96 #else
  97         /*
  98          * Accessing memory above the top the kernel knows about or through a file pointer
  99          * that was marked O_SYNC will be done non-cached.
 100          */
 101         if (file->f_flags & O_SYNC)
 102                 return 1;
 103         return addr >= __pa(high_memory);
 104 #endif
 105 }
 106
 107 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
 108 static inline int valid_phys_addr_range(unsigned long addr, size_t *count)
 109 {
 110         unsigned long end_mem;
 111
 112         end_mem = __pa(high_memory);
 113         if (addr >= end_mem)
 114                 return 0;
 115
 116         if (*count > end_mem - addr)
 117                 *count = end_mem - addr;
 118
 119         return 1;
 120 }
 121 #endif
 122
 123 #ifndef ARCH_HAS_DEV_MEM
 124 /*
 125  * This funcion reads the *physical* memory. The f_pos points directly to the
 126  * memory location.
 127  */
 128 static ssize_t read_mem(struct file * file, char __user * buf,
 129                         size_t count, loff_t *ppos)
 130 {
 131         unsigned long p = *ppos;
 132         ssize_t read, sz;
 133         char *ptr;
 134
 135         if (!valid_phys_addr_range(p, &count))
 136                 return -EFAULT;
 137         read = 0;
 138 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 139         /* we don't have page 0 mapped on sparc and m68k.. */
 140         if (p < PAGE_SIZE) {
 141                 sz = PAGE_SIZE - p;
 142                 if (sz > count)
 143                         sz = count;
 144                 if (sz > 0) {
 145                         if (clear_user(buf, sz))
 146                                 return -EFAULT;
 147                         buf += sz;
 148                         p += sz;
 149                         count -= sz;
 150                         read += sz;
 151                 }
 152         }
 153 #endif
 154
 155         while (count > 0) {
 156                 /*
 157                  * Handle first page in case it's not aligned
 158                  */
 159                 if (-p & (PAGE_SIZE - 1))
 160                         sz = -p & (PAGE_SIZE - 1);
 161                 else
 162                         sz = PAGE_SIZE;
 163
 164                 sz = min_t(unsigned long, sz, count);
 165
 166                 /*
 167                  * On ia64 if a page has been mapped somewhere as
 168                  * uncached, then it must also be accessed uncached
 169                  * by the kernel or data corruption may occur
 170                  */
 171                 ptr = xlate_dev_mem_ptr(p);
 172
 173                 if (!range_is_allowed(p, p+count))
 174                         return -EPERM;
 175                 if (copy_to_user(buf, ptr, sz))
 176                         return -EFAULT;
 177                 buf += sz;
 178                 p += sz;
 179                 count -= sz;
 180                 read += sz;
 181         }
 182
 183         *ppos += read;
 184         return read;
 185 }
 186
 187 static ssize_t write_mem(struct file * file, const char __user * buf,
 188                          size_t count, loff_t *ppos)
 189 {
 190         unsigned long p = *ppos;
 191         ssize_t written, sz;
 192         unsigned long copied;
 193         void *ptr;
 194
 195         if (!valid_phys_addr_range(p, &count))
 196                 return -EFAULT;
 197
 198         written = 0;
 199
 200 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 201         /* we don't have page 0 mapped on sparc and m68k.. */
 202         if (p < PAGE_SIZE) {
 203                 unsigned long sz = PAGE_SIZE - p;
 204                 if (sz > count)
 205                         sz = count;
 206                 /* Hmm. Do something? */
 207                 buf += sz;
 208                 p += sz;
 209                 count -= sz;
 210                 written += sz;
 211         }
 212 #endif
 213
 214         while (count > 0) {
 215                 /*
 216                  * Handle first page in case it's not aligned
 217                  */
 218                 if (-p & (PAGE_SIZE - 1))
 219                         sz = -p & (PAGE_SIZE - 1);
 220                 else
 221                         sz = PAGE_SIZE;
 222
 223                 sz = min_t(unsigned long, sz, count);
 224
 225                 /*
 226                  * On ia64 if a page has been mapped somewhere as
 227                  * uncached, then it must also be accessed uncached
 228                  * by the kernel or data corruption may occur
 229                  */
 230                 ptr = xlate_dev_mem_ptr(p);
 231
 232                 if (!range_is_allowed(ptr, ptr+sz))
 233                         return -EPERM;
 234                 copied = copy_from_user(ptr, buf, sz);
 235                 if (copied) {
 236                         ssize_t ret;
 237
 238                         ret = written + (sz - copied);
 239                         if (ret)
 240                                 return ret;
 241                         return -EFAULT;
 242                 }
 243                 buf += sz;
 244                 p += sz;
 245                 count -= sz;
 246                 written += sz;
 247         }
 248
 249         *ppos += written;
 250         return written;
 251 }
 252 #endif
 253
 254 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 255 {
 256 #if defined(__HAVE_PHYS_MEM_ACCESS_PROT)
 257         unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
 258
 259         vma->vm_page_prot = phys_mem_access_prot(file, offset,
 260                                                  vma->vm_end - vma->vm_start,
 261                                                  vma->vm_page_prot);
 262 #elif defined(pgprot_noncached)
 263         unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
 264         int uncached;
 265
 266         uncached = uncached_access(file, offset);
 267         if (uncached)
 268                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 269 #endif
 270
 271         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 272         if (remap_pfn_range(vma,
 273                             vma->vm_start,
 274                             vma->vm_pgoff,
 275                             vma->vm_end-vma->vm_start,
 276                             vma->vm_page_prot))
 277                 return -EAGAIN;
 278         return 0;
 279 }
 280
 281 #ifdef CONFIG_CRASH_DUMP
 282 /*
 283  * Read memory corresponding to the old kernel.
 284  * If we are reading from the reserved section, which is
 285  * actually used by the current kernel, we just return zeroes.
 286  * Or if we are reading from the first 640k, we return from the
 287  * backed up area.
 288  */
 289 static ssize_t read_oldmem(struct file * file, char * buf,
 290                                 size_t count, loff_t *ppos)
 291 {
 292         unsigned long pfn;
 293         unsigned backup_start, backup_end, relocate_start;
 294         size_t read=0, csize;
 295
 296         backup_start = CRASH_BACKUP_BASE / PAGE_SIZE;
 297         backup_end = backup_start + (CRASH_BACKUP_SIZE / PAGE_SIZE);
 298         relocate_start = (CRASH_BACKUP_BASE + CRASH_BACKUP_SIZE) / PAGE_SIZE;
 299
 300         while(count) {
 301                 pfn = *ppos / PAGE_SIZE;
 302
 303                 csize = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 304
 305                 /* Perform translation (see comment above) */
 306                 if ((pfn >= backup_start) && (pfn < backup_end)) {
 307                         if (clear_user(buf, csize)) {
 308                                 read = -EFAULT;
 309                                 goto done;
 310                         }
 311
 312                         goto copy_done;
 313                 } else if (pfn < (CRASH_RELOCATE_SIZE / PAGE_SIZE))
 314                         pfn += relocate_start;
 315
 316                 if (pfn > saved_max_pfn) {
 317                         read = 0;
 318                         goto done;
 319                 }
 320
 321                 if (copy_oldmem_page(pfn, buf, csize, 1)) {
 322                         read = -EFAULT;
 323                         goto done;
 324                 }
 325
 326 copy_done:
 327                 buf += csize;
 328                 *ppos += csize;
 329                 read += csize;
 330                 count -= csize;
 331         }
 332 done:
 333         return read;
 334 }
 335 #endif
 336
 337 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 338 {
 339         unsigned long long val;
 340         /*
 341          * RED-PEN: on some architectures there is more mapped memory
 342          * than available in mem_map which pfn_valid checks
 343          * for. Perhaps should add a new macro here.
 344          *
 345          * RED-PEN: vmalloc is not supported right now.
 346          */
 347         if (!pfn_valid(vma->vm_pgoff))
 348                 return -EIO;
 349         val = (u64)vma->vm_pgoff << PAGE_SHIFT;
 350         vma->vm_pgoff = __pa(val) >> PAGE_SHIFT;
 351         return mmap_mem(file, vma);
 352 }
 353
 354 extern long vread(char *buf, char *addr, unsigned long count);
 355 extern long vwrite(char *buf, char *addr, unsigned long count);
 356
 357 /*
 358  * This function reads the *virtual* memory as seen by the kernel.
 359  */
 360 static ssize_t read_kmem(struct file *file, char __user *buf,
 361                          size_t count, loff_t *ppos)
 362 {
 363         unsigned long p = *ppos;
 364         ssize_t low_count, read, sz;
 365         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 366
 367         return -EPERM;
 368
 369         read = 0;
 370         if (p < (unsigned long) high_memory) {
 371                 low_count = count;
 372                 if (count > (unsigned long) high_memory - p)
 373                         low_count = (unsigned long) high_memory - p;
 374
 375 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 376                 /* we don't have page 0 mapped on sparc and m68k.. */
 377                 if (p < PAGE_SIZE && low_count > 0) {
 378                         size_t tmp = PAGE_SIZE - p;
 379                         if (tmp > low_count) tmp = low_count;
 380                         if (clear_user(buf, tmp))
 381                                 return -EFAULT;
 382                         buf += tmp;
 383                         p += tmp;
 384                         read += tmp;
 385                         low_count -= tmp;
 386                         count -= tmp;
 387                 }
 388 #endif
 389                 while (low_count > 0) {
 390                         /*
 391                          * Handle first page in case it's not aligned
 392                          */
 393                         if (-p & (PAGE_SIZE - 1))
 394                                 sz = -p & (PAGE_SIZE - 1);
 395                         else
 396                                 sz = PAGE_SIZE;
 397
 398                         sz = min_t(unsigned long, sz, low_count);
 399
 400                         /*
 401                          * On ia64 if a page has been mapped somewhere as
 402                          * uncached, then it must also be accessed uncached
 403                          * by the kernel or data corruption may occur
 404                          */
 405                         kbuf = xlate_dev_kmem_ptr((char *)p);
 406
 407                         if (copy_to_user(buf, kbuf, sz))
 408                                 return -EFAULT;
 409                         buf += sz;
 410                         p += sz;
 411                         read += sz;
 412                         low_count -= sz;
 413                         count -= sz;
 414                 }
 415         }
 416
 417         if (count > 0) {
 418                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 419                 if (!kbuf)
 420                         return -ENOMEM;
 421                 while (count > 0) {
 422                         int len = count;
 423
 424                         if (len > PAGE_SIZE)
 425                                 len = PAGE_SIZE;
 426                         len = vread(kbuf, (char *)p, len);
 427                         if (!len)
 428                                 break;
 429                         if (copy_to_user(buf, kbuf, len)) {
 430                                 free_page((unsigned long)kbuf);
 431                                 return -EFAULT;
 432                         }
 433                         count -= len;
 434                         buf += len;
 435                         read += len;
 436                         p += len;
 437                 }
 438                 free_page((unsigned long)kbuf);
 439         }
 440         *ppos = p;
 441         return read;
 442 }
 443
 444
 445 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 446 static ssize_t read_port(struct file * file, char __user * buf,
 447                          size_t count, loff_t *ppos)
 448 {
 449         unsigned long i = *ppos;
 450         char __user *tmp = buf;
 451
 452         if (!access_ok(VERIFY_WRITE, buf, count))
 453                 return -EFAULT;
 454         while (count-- > 0 && i < 65536) {
 455                 if (__put_user(inb(i),tmp) < 0)
 456                         return -EFAULT;
 457                 i++;
 458                 tmp++;
 459         }
 460         *ppos = i;
 461         return tmp-buf;
 462 }
 463
 464 static ssize_t write_port(struct file * file, const char __user * buf,
 465                           size_t count, loff_t *ppos)
 466 {
 467         unsigned long i = *ppos;
 468         const char __user * tmp = buf;
 469
 470         if (!access_ok(VERIFY_READ,buf,count))
 471                 return -EFAULT;
 472         while (count-- > 0 && i < 65536) {
 473                 char c;
 474                 if (__get_user(c, tmp))
 475                         return -EFAULT;
 476                 outb(c,i);
 477                 i++;
 478                 tmp++;
 479         }
 480         *ppos = i;
 481         return tmp-buf;
 482 }
 483 #endif
 484
 485 static ssize_t read_null(struct file * file, char __user * buf,
 486                          size_t count, loff_t *ppos)
 487 {
 488         return 0;
 489 }
 490
 491 static ssize_t write_null(struct file * file, const char __user * buf,
 492                           size_t count, loff_t *ppos)
 493 {
 494         return count;
 495 }
 496
 497 #ifdef CONFIG_MMU
 498 /*
 499  * For fun, we are using the MMU for this.
 500  */
 501 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
 502 {
 503         struct mm_struct *mm;
 504         struct vm_area_struct * vma;
 505         unsigned long addr=(unsigned long)buf;
 506
 507         mm = current->mm;
 508         /* Oops, this was forgotten before. -ben */
 509         down_read(&mm->mmap_sem);
 510
 511         /* For private mappings, just map in zero pages. */
 512         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 513                 unsigned long count;
 514
 515                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 516                         goto out_up;
 517                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
 518                         break;
 519                 count = vma->vm_end - addr;
 520                 if (count > size)
 521                         count = size;
 522
 523                 zap_page_range(vma, addr, count, NULL);
 524                 zeromap_page_range(vma, addr, count, PAGE_COPY);
 525
 526                 size -= count;
 527                 buf += count;
 528                 addr += count;
 529                 if (size == 0)
 530                         goto out_up;
 531         }
 532
 533         up_read(&mm->mmap_sem);
 534
 535         /* The shared case is hard. Let's do the conventional zeroing. */
 536         do {
 537                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 538                 if (unwritten)
 539                         return size + unwritten - PAGE_SIZE;
 540                 cond_resched();
 541                 buf += PAGE_SIZE;
 542                 size -= PAGE_SIZE;
 543         } while (size);
 544
 545         return size;
 546 out_up:
 547         up_read(&mm->mmap_sem);
 548         return size;
 549 }
 550
 551 static ssize_t read_zero(struct file * file, char __user * buf,
 552                          size_t count, loff_t *ppos)
 553 {
 554         unsigned long left, unwritten, written = 0;
 555
 556         if (!count)
 557                 return 0;
 558
 559         if (!access_ok(VERIFY_WRITE, buf, count))
 560                 return -EFAULT;
 561
 562         left = count;
 563
 564         /* do we want to be clever? Arbitrary cut-off */
 565         if (count >= PAGE_SIZE*4) {
 566                 unsigned long partial;
 567
 568                 /* How much left of the page? */
 569                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 570                 unwritten = clear_user(buf, partial);
 571                 written = partial - unwritten;
 572                 if (unwritten)
 573                         goto out;
 574                 left -= partial;
 575                 buf += partial;
 576                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 577                 written += (left & PAGE_MASK) - unwritten;
 578                 if (unwritten)
 579                         goto out;
 580                 buf += left & PAGE_MASK;
 581                 left &= ~PAGE_MASK;
 582         }
 583         unwritten = clear_user(buf, left);
 584         written += left - unwritten;
 585 out:
 586         return written ? written : -EFAULT;
 587 }
 588
 589 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 590 {
 591         if (vma->vm_flags & VM_SHARED)
 592                 return shmem_zero_setup(vma);
 593         if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
 594                 return -EAGAIN;
 595         return 0;
 596 }
 597 #else /* CONFIG_MMU */
 598 static ssize_t read_zero(struct file * file, char * buf,
 599                          size_t count, loff_t *ppos)
 600 {
 601         size_t todo = count;
 602
 603         while (todo) {
 604                 size_t chunk = todo;
 605
 606                 if (chunk > 4096)
 607                         chunk = 4096;   /* Just for latency reasons */
 608                 if (clear_user(buf, chunk))
 609                         return -EFAULT;
 610                 buf += chunk;
 611                 todo -= chunk;
 612                 cond_resched();
 613         }
 614         return count;
 615 }
 616
 617 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 618 {
 619         return -ENOSYS;
 620 }
 621 #endif /* CONFIG_MMU */
 622
 623 static ssize_t write_full(struct file * file, const char __user * buf,
 624                           size_t count, loff_t *ppos)
 625 {
 626         return -ENOSPC;
 627 }
 628
 629 /*
 630  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 631  * can fopen() both devices with "a" now.  This was previously impossible.
 632  * -- SRB.
 633  */
 634
 635 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 636 {
 637         return file->f_pos = 0;
 638 }
 639
 640 /*
 641  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 642  * check against negative addresses: they are ok. The return value is weird,
 643  * though, in that case (0).
 644  *
 645  * also note that seeking relative to the "end of file" isn't supported:
 646  * it has no meaning, so it returns -EINVAL.
 647  */
 648 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 649 {
 650         loff_t ret;
 651
 652         down(&file->f_dentry->d_inode->i_sem);
 653         switch (orig) {
 654                 case 0:
 655                         file->f_pos = offset;
 656                         ret = file->f_pos;
 657                         force_successful_syscall_return();
 658                         break;
 659                 case 1:
 660                         file->f_pos += offset;
 661                         ret = file->f_pos;
 662                         force_successful_syscall_return();
 663                         break;
 664                 default:
 665                         ret = -EINVAL;
 666         }
 667         up(&file->f_dentry->d_inode->i_sem);
 668         return ret;
 669 }
 670
 671 static int open_port(struct inode * inode, struct file * filp)
 672 {
 673         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 674 }
 675
 676 #define zero_lseek      null_lseek
 677 #define full_lseek      null_lseek
 678 #define write_zero      write_null
 679 #define read_full       read_zero
 680 #define open_mem        open_port
 681 #define open_kmem       open_mem
 682 #define open_oldmem     open_mem
 683
 684 #ifndef ARCH_HAS_DEV_MEM
 685 static struct file_operations mem_fops = {
 686         .llseek         = memory_lseek,
 687         .read           = read_mem,
 688         .write          = write_mem,
 689         .mmap           = mmap_mem,
 690         .open           = open_mem,
 691 };
 692 #else
 693 extern struct file_operations mem_fops;
 694 #endif
 695
 696 static struct file_operations kmem_fops = {
 697         .llseek         = memory_lseek,
 698         .read           = read_kmem,
 699         .mmap           = mmap_kmem,
 700         .open           = open_kmem,
 701 };
 702
 703 static struct file_operations null_fops = {
 704         .llseek         = null_lseek,
 705         .read           = read_null,
 706         .write          = write_null,
 707 };
 708
 709 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 710 static struct file_operations port_fops = {
 711         .llseek         = memory_lseek,
 712         .read           = read_port,
 713         .write          = write_port,
 714         .open           = open_port,
 715 };
 716 #endif
 717
 718 static struct file_operations zero_fops = {
 719         .llseek         = zero_lseek,
 720         .read           = read_zero,
 721         .write          = write_zero,
 722         .mmap           = mmap_zero,
 723 };
 724
 725 static struct backing_dev_info zero_bdi = {
 726         .capabilities   = BDI_CAP_MAP_COPY,
 727 };
 728
 729 static struct file_operations full_fops = {
 730         .llseek         = full_lseek,
 731         .read           = read_full,
 732         .write          = write_full,
 733 };
 734
 735 #ifdef CONFIG_CRASH_DUMP
 736 static struct file_operations oldmem_fops = {
 737         .read   = read_oldmem,
 738         .open   = open_oldmem,
 739 };
 740 #endif
 741
 742 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 743                           size_t count, loff_t *ppos)
 744 {
 745         char *tmp;
 746         int ret;
 747
 748         tmp = kmalloc(count + 1, GFP_KERNEL);
 749         if (tmp == NULL)
 750                 return -ENOMEM;
 751         ret = -EFAULT;
 752         if (!copy_from_user(tmp, buf, count)) {
 753                 tmp[count] = 0;
 754                 ret = printk("%s", tmp);
 755         }
 756         kfree(tmp);
 757         return ret;
 758 }
 759
 760 static struct file_operations kmsg_fops = {
 761         .write =        kmsg_write,
 762 };
 763
 764 static int memory_open(struct inode * inode, struct file * filp)
 765 {
 766         switch (iminor(inode)) {
 767                 case 1:
 768                         filp->f_op = &mem_fops;
 769                         break;
 770                 case 2:
 771                         filp->f_op = &kmem_fops;
 772                         break;
 773                 case 3:
 774                         filp->f_op = &null_fops;
 775                         break;
 776 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 777                 case 4:
 778                         filp->f_op = &port_fops;
 779                         break;
 780 #endif
 781                 case 5:
 782                         filp->f_mapping->backing_dev_info = &zero_bdi;
 783                         filp->f_op = &zero_fops;
 784                         break;
 785                 case 7:
 786                         filp->f_op = &full_fops;
 787                         break;
 788                 case 8:
 789                         filp->f_op = &random_fops;
 790                         break;
 791                 case 9:
 792                         filp->f_op = &urandom_fops;
 793                         break;
 794                 case 11:
 795                         filp->f_op = &kmsg_fops;
 796                         break;
 797 #ifdef CONFIG_CRASH_DUMP
 798                 case 12:
 799                         filp->f_op = &oldmem_fops;
 800                         break;
 801 #endif
 802                 default:
 803                         return -ENXIO;
 804         }
 805         if (filp->f_op && filp->f_op->open)
 806                 return filp->f_op->open(inode,filp);
 807         return 0;
 808 }
 809
 810 static struct file_operations memory_fops = {
 811         .open           = memory_open,  /* just a selector for the real open */
 812 };
 813
 814 static const struct {
 815         unsigned int            minor;
 816         char                    *name;
 817         umode_t                 mode;
 818         struct file_operations  *fops;
 819 } devlist[] = { /* list of minor devices */
 820         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 821         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 822 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 823         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 824 #endif
 825         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 826         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 827         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 828         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 829         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 830 #ifdef CONFIG_CRASH_DUMP
 831         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 832 #endif
 833 };
 834
 835 static struct class_simple *mem_class;
 836
 837 static int __init chr_dev_init(void)
 838 {
 839         int i;
 840
 841         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 842                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 843
 844         mem_class = class_simple_create(THIS_MODULE, "mem");
 845         for (i = 0; i < ARRAY_SIZE(devlist); i++) {
 846                 class_simple_device_add(mem_class,
 847                                         MKDEV(MEM_MAJOR, devlist[i].minor),
 848                                         NULL, devlist[i].name);
 849                 devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor),
 850                                 S_IFCHR | devlist[i].mode, devlist[i].name);
 851         }
 852
 853         return 0;
 854 }
 855
 856 fs_initcall(chr_dev_init);