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 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 232 {
 233 #if defined(__HAVE_PHYS_MEM_ACCESS_PROT)
 234         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 235                                                  vma->vm_end - vma->vm_start,
 236                                                  vma->vm_page_prot);
 237 #elif defined(pgprot_noncached)
 238         unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
 239         int uncached;
 240
 241         uncached = uncached_access(file, offset);
 242         if (uncached)
 243                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 244 #endif
 245
 246         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 247         if (remap_pfn_range(vma,
 248                             vma->vm_start,
 249                             vma->vm_pgoff,
 250                             vma->vm_end-vma->vm_start,
 251                             vma->vm_page_prot))
 252                 return -EAGAIN;
 253         return 0;
 254 }
 255
 256 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 257 {
 258         unsigned long pfn;
 259
 260         /* Turn a kernel-virtual address into a physical page frame */
 261         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 262
 263         /*
 264          * RED-PEN: on some architectures there is more mapped memory
 265          * than available in mem_map which pfn_valid checks
 266          * for. Perhaps should add a new macro here.
 267          *
 268          * RED-PEN: vmalloc is not supported right now.
 269          */
 270         if (!pfn_valid(pfn))
 271                 return -EIO;
 272
 273         vma->vm_pgoff = pfn;
 274         return mmap_mem(file, vma);
 275 }
 276
 277 #ifdef CONFIG_CRASH_DUMP
 278 /*
 279  * Read memory corresponding to the old kernel.
 280  */
 281 static ssize_t read_oldmem(struct file *file, char __user *buf,
 282                                 size_t count, loff_t *ppos)
 283 {
 284         unsigned long pfn, offset;
 285         size_t read = 0, csize;
 286         int rc = 0;
 287
 288         while (count) {
 289                 pfn = *ppos / PAGE_SIZE;
 290                 if (pfn > saved_max_pfn)
 291                         return read;
 292
 293                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 294                 if (count > PAGE_SIZE - offset)
 295                         csize = PAGE_SIZE - offset;
 296                 else
 297                         csize = count;
 298
 299                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 300                 if (rc < 0)
 301                         return rc;
 302                 buf += csize;
 303                 *ppos += csize;
 304                 read += csize;
 305                 count -= csize;
 306         }
 307         return read;
 308 }
 309 #endif
 310
 311 extern long vread(char *buf, char *addr, unsigned long count);
 312 extern long vwrite(char *buf, char *addr, unsigned long count);
 313
 314 /*
 315  * This function reads the *virtual* memory as seen by the kernel.
 316  */
 317 static ssize_t read_kmem(struct file *file, char __user *buf,
 318                          size_t count, loff_t *ppos)
 319 {
 320         unsigned long p = *ppos;
 321         ssize_t low_count, read, sz;
 322         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 323
 324         read = 0;
 325         if (p < (unsigned long) high_memory) {
 326                 low_count = count;
 327                 if (count > (unsigned long) high_memory - p)
 328                         low_count = (unsigned long) high_memory - p;
 329
 330 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 331                 /* we don't have page 0 mapped on sparc and m68k.. */
 332                 if (p < PAGE_SIZE && low_count > 0) {
 333                         size_t tmp = PAGE_SIZE - p;
 334                         if (tmp > low_count) tmp = low_count;
 335                         if (clear_user(buf, tmp))
 336                                 return -EFAULT;
 337                         buf += tmp;
 338                         p += tmp;
 339                         read += tmp;
 340                         low_count -= tmp;
 341                         count -= tmp;
 342                 }
 343 #endif
 344                 while (low_count > 0) {
 345                         /*
 346                          * Handle first page in case it's not aligned
 347                          */
 348                         if (-p & (PAGE_SIZE - 1))
 349                                 sz = -p & (PAGE_SIZE - 1);
 350                         else
 351                                 sz = PAGE_SIZE;
 352
 353                         sz = min_t(unsigned long, sz, low_count);
 354
 355                         /*
 356                          * On ia64 if a page has been mapped somewhere as
 357                          * uncached, then it must also be accessed uncached
 358                          * by the kernel or data corruption may occur
 359                          */
 360                         kbuf = xlate_dev_kmem_ptr((char *)p);
 361
 362                         if (copy_to_user(buf, kbuf, sz))
 363                                 return -EFAULT;
 364                         buf += sz;
 365                         p += sz;
 366                         read += sz;
 367                         low_count -= sz;
 368                         count -= sz;
 369                 }
 370         }
 371
 372         if (count > 0) {
 373                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 374                 if (!kbuf)
 375                         return -ENOMEM;
 376                 while (count > 0) {
 377                         int len = count;
 378
 379                         if (len > PAGE_SIZE)
 380                                 len = PAGE_SIZE;
 381                         len = vread(kbuf, (char *)p, len);
 382                         if (!len)
 383                                 break;
 384                         if (copy_to_user(buf, kbuf, len)) {
 385                                 free_page((unsigned long)kbuf);
 386                                 return -EFAULT;
 387                         }
 388                         count -= len;
 389                         buf += len;
 390                         read += len;
 391                         p += len;
 392                 }
 393                 free_page((unsigned long)kbuf);
 394         }
 395         *ppos = p;
 396         return read;
 397 }
 398
 399
 400 static inline ssize_t
 401 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 402               size_t count, loff_t *ppos)
 403 {
 404         ssize_t written, sz;
 405         unsigned long copied;
 406
 407         written = 0;
 408 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 409         /* we don't have page 0 mapped on sparc and m68k.. */
 410         if (realp < PAGE_SIZE) {
 411                 unsigned long sz = PAGE_SIZE - realp;
 412                 if (sz > count)
 413                         sz = count;
 414                 /* Hmm. Do something? */
 415                 buf += sz;
 416                 p += sz;
 417                 realp += sz;
 418                 count -= sz;
 419                 written += sz;
 420         }
 421 #endif
 422
 423         while (count > 0) {
 424                 char *ptr;
 425                 /*
 426                  * Handle first page in case it's not aligned
 427                  */
 428                 if (-realp & (PAGE_SIZE - 1))
 429                         sz = -realp & (PAGE_SIZE - 1);
 430                 else
 431                         sz = PAGE_SIZE;
 432
 433                 sz = min_t(unsigned long, sz, count);
 434
 435                 /*
 436                  * On ia64 if a page has been mapped somewhere as
 437                  * uncached, then it must also be accessed uncached
 438                  * by the kernel or data corruption may occur
 439                  */
 440                 ptr = xlate_dev_kmem_ptr(p);
 441
 442                 copied = copy_from_user(ptr, buf, sz);
 443                 if (copied) {
 444                         ssize_t ret;
 445
 446                         ret = written + (sz - copied);
 447                         if (ret)
 448                                 return ret;
 449                         return -EFAULT;
 450                 }
 451                 buf += sz;
 452                 p += sz;
 453                 realp += sz;
 454                 count -= sz;
 455                 written += sz;
 456         }
 457
 458         *ppos += written;
 459         return written;
 460 }
 461
 462
 463 /*
 464  * This function writes to the *virtual* memory as seen by the kernel.
 465  */
 466 static ssize_t write_kmem(struct file * file, const char __user * buf,
 467                           size_t count, loff_t *ppos)
 468 {
 469         unsigned long p = *ppos;
 470         ssize_t wrote = 0;
 471         ssize_t virtr = 0;
 472         ssize_t written;
 473         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 474
 475         if (p < (unsigned long) high_memory) {
 476
 477                 wrote = count;
 478                 if (count > (unsigned long) high_memory - p)
 479                         wrote = (unsigned long) high_memory - p;
 480
 481                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 482                 if (written != wrote)
 483                         return written;
 484                 wrote = written;
 485                 p += wrote;
 486                 buf += wrote;
 487                 count -= wrote;
 488         }
 489
 490         if (count > 0) {
 491                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 492                 if (!kbuf)
 493                         return wrote ? wrote : -ENOMEM;
 494                 while (count > 0) {
 495                         int len = count;
 496
 497                         if (len > PAGE_SIZE)
 498                                 len = PAGE_SIZE;
 499                         if (len) {
 500                                 written = copy_from_user(kbuf, buf, len);
 501                                 if (written) {
 502                                         ssize_t ret;
 503
 504                                         free_page((unsigned long)kbuf);
 505                                         ret = wrote + virtr + (len - written);
 506                                         return ret ? ret : -EFAULT;
 507                                 }
 508                         }
 509                         len = vwrite(kbuf, (char *)p, len);
 510                         count -= len;
 511                         buf += len;
 512                         virtr += len;
 513                         p += len;
 514                 }
 515                 free_page((unsigned long)kbuf);
 516         }
 517
 518         *ppos = p;
 519         return virtr + wrote;
 520 }
 521
 522 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 523 static ssize_t read_port(struct file * file, char __user * buf,
 524                          size_t count, loff_t *ppos)
 525 {
 526         unsigned long i = *ppos;
 527         char __user *tmp = buf;
 528
 529         if (!access_ok(VERIFY_WRITE, buf, count))
 530                 return -EFAULT;
 531         while (count-- > 0 && i < 65536) {
 532                 if (__put_user(inb(i),tmp) < 0)
 533                         return -EFAULT;
 534                 i++;
 535                 tmp++;
 536         }
 537         *ppos = i;
 538         return tmp-buf;
 539 }
 540
 541 static ssize_t write_port(struct file * file, const char __user * buf,
 542                           size_t count, loff_t *ppos)
 543 {
 544         unsigned long i = *ppos;
 545         const char __user * tmp = buf;
 546
 547         if (!access_ok(VERIFY_READ,buf,count))
 548                 return -EFAULT;
 549         while (count-- > 0 && i < 65536) {
 550                 char c;
 551                 if (__get_user(c, tmp))
 552                         return -EFAULT;
 553                 outb(c,i);
 554                 i++;
 555                 tmp++;
 556         }
 557         *ppos = i;
 558         return tmp-buf;
 559 }
 560 #endif
 561
 562 static ssize_t read_null(struct file * file, char __user * buf,
 563                          size_t count, loff_t *ppos)
 564 {
 565         return 0;
 566 }
 567
 568 static ssize_t write_null(struct file * file, const char __user * buf,
 569                           size_t count, loff_t *ppos)
 570 {
 571         return count;
 572 }
 573
 574 #ifdef CONFIG_MMU
 575 /*
 576  * For fun, we are using the MMU for this.
 577  */
 578 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
 579 {
 580         struct mm_struct *mm;
 581         struct vm_area_struct * vma;
 582         unsigned long addr=(unsigned long)buf;
 583
 584         mm = current->mm;
 585         /* Oops, this was forgotten before. -ben */
 586         down_read(&mm->mmap_sem);
 587
 588         /* For private mappings, just map in zero pages. */
 589         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 590                 unsigned long count;
 591
 592                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 593                         goto out_up;
 594                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
 595                         break;
 596                 count = vma->vm_end - addr;
 597                 if (count > size)
 598                         count = size;
 599
 600                 zap_page_range(vma, addr, count, NULL);
 601                 zeromap_page_range(vma, addr, count, PAGE_COPY);
 602
 603                 size -= count;
 604                 buf += count;
 605                 addr += count;
 606                 if (size == 0)
 607                         goto out_up;
 608         }
 609
 610         up_read(&mm->mmap_sem);
 611
 612         /* The shared case is hard. Let's do the conventional zeroing. */
 613         do {
 614                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 615                 if (unwritten)
 616                         return size + unwritten - PAGE_SIZE;
 617                 cond_resched();
 618                 buf += PAGE_SIZE;
 619                 size -= PAGE_SIZE;
 620         } while (size);
 621
 622         return size;
 623 out_up:
 624         up_read(&mm->mmap_sem);
 625         return size;
 626 }
 627
 628 static ssize_t read_zero(struct file * file, char __user * buf,
 629                          size_t count, loff_t *ppos)
 630 {
 631         unsigned long left, unwritten, written = 0;
 632
 633         if (!count)
 634                 return 0;
 635
 636         if (!access_ok(VERIFY_WRITE, buf, count))
 637                 return -EFAULT;
 638
 639         left = count;
 640
 641         /* do we want to be clever? Arbitrary cut-off */
 642         if (count >= PAGE_SIZE*4) {
 643                 unsigned long partial;
 644
 645                 /* How much left of the page? */
 646                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 647                 unwritten = clear_user(buf, partial);
 648                 written = partial - unwritten;
 649                 if (unwritten)
 650                         goto out;
 651                 left -= partial;
 652                 buf += partial;
 653                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 654                 written += (left & PAGE_MASK) - unwritten;
 655                 if (unwritten)
 656                         goto out;
 657                 buf += left & PAGE_MASK;
 658                 left &= ~PAGE_MASK;
 659         }
 660         unwritten = clear_user(buf, left);
 661         written += left - unwritten;
 662 out:
 663         return written ? written : -EFAULT;
 664 }
 665
 666 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 667 {
 668         if (vma->vm_flags & VM_SHARED)
 669                 return shmem_zero_setup(vma);
 670         if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
 671                 return -EAGAIN;
 672         return 0;
 673 }
 674 #else /* CONFIG_MMU */
 675 static ssize_t read_zero(struct file * file, char * buf,
 676                          size_t count, loff_t *ppos)
 677 {
 678         size_t todo = count;
 679
 680         while (todo) {
 681                 size_t chunk = todo;
 682
 683                 if (chunk > 4096)
 684                         chunk = 4096;   /* Just for latency reasons */
 685                 if (clear_user(buf, chunk))
 686                         return -EFAULT;
 687                 buf += chunk;
 688                 todo -= chunk;
 689                 cond_resched();
 690         }
 691         return count;
 692 }
 693
 694 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 695 {
 696         return -ENOSYS;
 697 }
 698 #endif /* CONFIG_MMU */
 699
 700 static ssize_t write_full(struct file * file, const char __user * buf,
 701                           size_t count, loff_t *ppos)
 702 {
 703         return -ENOSPC;
 704 }
 705
 706 /*
 707  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 708  * can fopen() both devices with "a" now.  This was previously impossible.
 709  * -- SRB.
 710  */
 711
 712 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 713 {
 714         return file->f_pos = 0;
 715 }
 716
 717 /*
 718  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 719  * check against negative addresses: they are ok. The return value is weird,
 720  * though, in that case (0).
 721  *
 722  * also note that seeking relative to the "end of file" isn't supported:
 723  * it has no meaning, so it returns -EINVAL.
 724  */
 725 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 726 {
 727         loff_t ret;
 728
 729         down(&file->f_dentry->d_inode->i_sem);
 730         switch (orig) {
 731                 case 0:
 732                         file->f_pos = offset;
 733                         ret = file->f_pos;
 734                         force_successful_syscall_return();
 735                         break;
 736                 case 1:
 737                         file->f_pos += offset;
 738                         ret = file->f_pos;
 739                         force_successful_syscall_return();
 740                         break;
 741                 default:
 742                         ret = -EINVAL;
 743         }
 744         up(&file->f_dentry->d_inode->i_sem);
 745         return ret;
 746 }
 747
 748 static int open_port(struct inode * inode, struct file * filp)
 749 {
 750         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 751 }
 752
 753 #define zero_lseek      null_lseek
 754 #define full_lseek      null_lseek
 755 #define write_zero      write_null
 756 #define read_full       read_zero
 757 #define open_mem        open_port
 758 #define open_kmem       open_mem
 759 #define open_oldmem     open_mem
 760
 761 static struct file_operations mem_fops = {
 762         .llseek         = memory_lseek,
 763         .read           = read_mem,
 764         .write          = write_mem,
 765         .mmap           = mmap_mem,
 766         .open           = open_mem,
 767 };
 768
 769 static struct file_operations kmem_fops = {
 770         .llseek         = memory_lseek,
 771         .read           = read_kmem,
 772         .write          = write_kmem,
 773         .mmap           = mmap_kmem,
 774         .open           = open_kmem,
 775 };
 776
 777 static struct file_operations null_fops = {
 778         .llseek         = null_lseek,
 779         .read           = read_null,
 780         .write          = write_null,
 781 };
 782
 783 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 784 static struct file_operations port_fops = {
 785         .llseek         = memory_lseek,
 786         .read           = read_port,
 787         .write          = write_port,
 788         .open           = open_port,
 789 };
 790 #endif
 791
 792 static struct file_operations zero_fops = {
 793         .llseek         = zero_lseek,
 794         .read           = read_zero,
 795         .write          = write_zero,
 796         .mmap           = mmap_zero,
 797 };
 798
 799 static struct backing_dev_info zero_bdi = {
 800         .capabilities   = BDI_CAP_MAP_COPY,
 801 };
 802
 803 static struct file_operations full_fops = {
 804         .llseek         = full_lseek,
 805         .read           = read_full,
 806         .write          = write_full,
 807 };
 808
 809 #ifdef CONFIG_CRASH_DUMP
 810 static struct file_operations oldmem_fops = {
 811         .read   = read_oldmem,
 812         .open   = open_oldmem,
 813 };
 814 #endif
 815
 816 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 817                           size_t count, loff_t *ppos)
 818 {
 819         char *tmp;
 820         int ret;
 821
 822         tmp = kmalloc(count + 1, GFP_KERNEL);
 823         if (tmp == NULL)
 824                 return -ENOMEM;
 825         ret = -EFAULT;
 826         if (!copy_from_user(tmp, buf, count)) {
 827                 tmp[count] = 0;
 828                 ret = printk("%s", tmp);
 829         }
 830         kfree(tmp);
 831         return ret;
 832 }
 833
 834 static struct file_operations kmsg_fops = {
 835         .write =        kmsg_write,
 836 };
 837
 838 static int memory_open(struct inode * inode, struct file * filp)
 839 {
 840         switch (iminor(inode)) {
 841                 case 1:
 842                         filp->f_op = &mem_fops;
 843                         break;
 844                 case 2:
 845                         filp->f_op = &kmem_fops;
 846                         break;
 847                 case 3:
 848                         filp->f_op = &null_fops;
 849                         break;
 850 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 851                 case 4:
 852                         filp->f_op = &port_fops;
 853                         break;
 854 #endif
 855                 case 5:
 856                         filp->f_mapping->backing_dev_info = &zero_bdi;
 857                         filp->f_op = &zero_fops;
 858                         break;
 859                 case 7:
 860                         filp->f_op = &full_fops;
 861                         break;
 862                 case 8:
 863                         filp->f_op = &random_fops;
 864                         break;
 865                 case 9:
 866                         filp->f_op = &urandom_fops;
 867                         break;
 868                 case 11:
 869                         filp->f_op = &kmsg_fops;
 870                         break;
 871 #ifdef CONFIG_CRASH_DUMP
 872                 case 12:
 873                         filp->f_op = &oldmem_fops;
 874                         break;
 875 #endif
 876                 default:
 877                         return -ENXIO;
 878         }
 879         if (filp->f_op && filp->f_op->open)
 880                 return filp->f_op->open(inode,filp);
 881         return 0;
 882 }
 883
 884 static struct file_operations memory_fops = {
 885         .open           = memory_open,  /* just a selector for the real open */
 886 };
 887
 888 static const struct {
 889         unsigned int            minor;
 890         char                    *name;
 891         umode_t                 mode;
 892         struct file_operations  *fops;
 893 } devlist[] = { /* list of minor devices */
 894         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 895         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 896         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 897 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 898         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 899 #endif
 900         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 901         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 902         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 903         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 904         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 905 #ifdef CONFIG_CRASH_DUMP
 906         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 907 #endif
 908 };
 909
 910 static struct class *mem_class;
 911
 912 static int __init chr_dev_init(void)
 913 {
 914         int i;
 915
 916         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 917                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 918
 919         mem_class = class_create(THIS_MODULE, "mem");
 920         for (i = 0; i < ARRAY_SIZE(devlist); i++) {
 921                 class_device_create(mem_class, NULL,
 922                                         MKDEV(MEM_MAJOR, devlist[i].minor),
 923                                         NULL, devlist[i].name);
 924                 devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor),
 925                                 S_IFCHR | devlist[i].mode, devlist[i].name);
 926         }
 927
 928         return 0;
 929 }
 930
 931 fs_initcall(chr_dev_init);