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