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 #if defined(CONFIG_S390_TAPE) && defined(CONFIG_S390_TAPE_CHAR)
  39 extern void tapechar_init(void);
  40 #endif
  41
  42 /*
  43  * Architectures vary in how they handle caching for addresses
  44  * outside of main memory.
  45  *
  46  */
  47 static inline int uncached_access(struct file *file, unsigned long addr)
  48 {
  49 #if defined(__i386__)
  50         /*
  51          * On the PPro and successors, the MTRRs are used to set
  52          * memory types for physical addresses outside main memory,
  53          * so blindly setting PCD or PWT on those pages is wrong.
  54          * For Pentiums and earlier, the surround logic should disable
  55          * caching for the high addresses through the KEN pin, but
  56          * we maintain the tradition of paranoia in this code.
  57          */
  58         if (file->f_flags & O_SYNC)
  59                 return 1;
  60         return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
  61                   test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
  62                   test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
  63                   test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
  64           && addr >= __pa(high_memory);
  65 #elif defined(__x86_64__)
  66         /*
  67          * This is broken because it can generate memory type aliases,
  68          * which can cause cache corruptions
  69          * But it is only available for root and we have to be bug-to-bug
  70          * compatible with i386.
  71          */
  72         if (file->f_flags & O_SYNC)
  73                 return 1;
  74         /* same behaviour as i386. PAT always set to cached and MTRRs control the
  75            caching behaviour.
  76            Hopefully a full PAT implementation will fix that soon. */
  77         return 0;
  78 #elif defined(CONFIG_IA64)
  79         /*
  80          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  81          */
  82         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  83 #else
  84         /*
  85          * Accessing memory above the top the kernel knows about or through a file pointer
  86          * that was marked O_SYNC will be done non-cached.
  87          */
  88         if (file->f_flags & O_SYNC)
  89                 return 1;
  90         return addr >= __pa(high_memory);
  91 #endif
  92 }
  93
  94 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  95 static inline int valid_phys_addr_range(unsigned long addr, size_t *count)
  96 {
  97         unsigned long end_mem;
  98
  99         end_mem = __pa(high_memory);
 100         if (addr >= end_mem)
 101                 return 0;
 102
 103         if (*count > end_mem - addr)
 104                 *count = end_mem - addr;
 105
 106         return 1;
 107 }
 108 #endif
 109
 110 /*
 111  * This funcion reads the *physical* memory. The f_pos points directly to the
 112  * memory location.
 113  */
 114 static ssize_t read_mem(struct file * file, char __user * buf,
 115                         size_t count, loff_t *ppos)
 116 {
 117         unsigned long p = *ppos;
 118         ssize_t read, sz;
 119         char *ptr;
 120
 121         if (!valid_phys_addr_range(p, &count))
 122                 return -EFAULT;
 123         read = 0;
 124 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 125         /* we don't have page 0 mapped on sparc and m68k.. */
 126         if (p < PAGE_SIZE) {
 127                 sz = PAGE_SIZE - p;
 128                 if (sz > count)
 129                         sz = count;
 130                 if (sz > 0) {
 131                         if (clear_user(buf, sz))
 132                                 return -EFAULT;
 133                         buf += sz;
 134                         p += sz;
 135                         count -= sz;
 136                         read += sz;
 137                 }
 138         }
 139 #endif
 140
 141         while (count > 0) {
 142                 /*
 143                  * Handle first page in case it's not aligned
 144                  */
 145                 if (-p & (PAGE_SIZE - 1))
 146                         sz = -p & (PAGE_SIZE - 1);
 147                 else
 148                         sz = PAGE_SIZE;
 149
 150                 sz = min_t(unsigned long, sz, count);
 151
 152                 /*
 153                  * On ia64 if a page has been mapped somewhere as
 154                  * uncached, then it must also be accessed uncached
 155                  * by the kernel or data corruption may occur
 156                  */
 157                 ptr = xlate_dev_mem_ptr(p);
 158
 159                 if (copy_to_user(buf, ptr, sz))
 160                         return -EFAULT;
 161                 buf += sz;
 162                 p += sz;
 163                 count -= sz;
 164                 read += sz;
 165         }
 166
 167         *ppos += read;
 168         return read;
 169 }
 170
 171 static ssize_t write_mem(struct file * file, const char __user * buf,
 172                          size_t count, loff_t *ppos)
 173 {
 174         unsigned long p = *ppos;
 175         ssize_t written, sz;
 176         unsigned long copied;
 177         void *ptr;
 178
 179         if (!valid_phys_addr_range(p, &count))
 180                 return -EFAULT;
 181
 182         written = 0;
 183
 184 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 185         /* we don't have page 0 mapped on sparc and m68k.. */
 186         if (p < PAGE_SIZE) {
 187                 unsigned long sz = PAGE_SIZE - p;
 188                 if (sz > count)
 189                         sz = count;
 190                 /* Hmm. Do something? */
 191                 buf += sz;
 192                 p += sz;
 193                 count -= sz;
 194                 written += sz;
 195         }
 196 #endif
 197
 198         while (count > 0) {
 199                 /*
 200                  * Handle first page in case it's not aligned
 201                  */
 202                 if (-p & (PAGE_SIZE - 1))
 203                         sz = -p & (PAGE_SIZE - 1);
 204                 else
 205                         sz = PAGE_SIZE;
 206
 207                 sz = min_t(unsigned long, sz, count);
 208
 209                 /*
 210                  * On ia64 if a page has been mapped somewhere as
 211                  * uncached, then it must also be accessed uncached
 212                  * by the kernel or data corruption may occur
 213                  */
 214                 ptr = xlate_dev_mem_ptr(p);
 215
 216                 copied = copy_from_user(ptr, buf, sz);
 217                 if (copied) {
 218                         ssize_t ret;
 219
 220                         ret = written + (sz - copied);
 221                         if (ret)
 222                                 return ret;
 223                         return -EFAULT;
 224                 }
 225                 buf += sz;
 226                 p += sz;
 227                 count -= sz;
 228                 written += sz;
 229         }
 230
 231         *ppos += written;
 232         return written;
 233 }
 234
 235 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 236 {
 237 #if defined(__HAVE_PHYS_MEM_ACCESS_PROT)
 238         unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
 239
 240         vma->vm_page_prot = phys_mem_access_prot(file, offset,
 241                                                  vma->vm_end - vma->vm_start,
 242                                                  vma->vm_page_prot);
 243 #elif defined(pgprot_noncached)
 244         unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
 245         int uncached;
 246
 247         uncached = uncached_access(file, offset);
 248         if (uncached)
 249                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 250 #endif
 251
 252         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 253         if (remap_pfn_range(vma,
 254                             vma->vm_start,
 255                             vma->vm_pgoff,
 256                             vma->vm_end-vma->vm_start,
 257                             vma->vm_page_prot))
 258                 return -EAGAIN;
 259         return 0;
 260 }
 261
 262 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 263 {
 264         unsigned long pfn;
 265
 266         /* Turn a kernel-virtual address into a physical page frame */
 267         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 268
 269         /*
 270          * RED-PEN: on some architectures there is more mapped memory
 271          * than available in mem_map which pfn_valid checks
 272          * for. Perhaps should add a new macro here.
 273          *
 274          * RED-PEN: vmalloc is not supported right now.
 275          */
 276         if (!pfn_valid(pfn))
 277                 return -EIO;
 278
 279         vma->vm_pgoff = pfn;
 280         return mmap_mem(file, vma);
 281 }
 282
 283 #ifdef CONFIG_CRASH_DUMP
 284 /*
 285  * Read memory corresponding to the old kernel.
 286  */
 287 static ssize_t read_oldmem(struct file *file, char __user *buf,
 288                                 size_t count, loff_t *ppos)
 289 {
 290         unsigned long pfn, offset;
 291         size_t read = 0, csize;
 292         int rc = 0;
 293
 294         while (count) {
 295                 pfn = *ppos / PAGE_SIZE;
 296                 if (pfn > saved_max_pfn)
 297                         return read;
 298
 299                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 300                 if (count > PAGE_SIZE - offset)
 301                         csize = PAGE_SIZE - offset;
 302                 else
 303                         csize = count;
 304
 305                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 306                 if (rc < 0)
 307                         return rc;
 308                 buf += csize;
 309                 *ppos += csize;
 310                 read += csize;
 311                 count -= csize;
 312         }
 313         return read;
 314 }
 315 #endif
 316
 317 extern long vread(char *buf, char *addr, unsigned long count);
 318 extern long vwrite(char *buf, char *addr, unsigned long count);
 319
 320 /*
 321  * This function reads the *virtual* memory as seen by the kernel.
 322  */
 323 static ssize_t read_kmem(struct file *file, char __user *buf,
 324                          size_t count, loff_t *ppos)
 325 {
 326         unsigned long p = *ppos;
 327         ssize_t low_count, read, sz;
 328         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 329
 330         read = 0;
 331         if (p < (unsigned long) high_memory) {
 332                 low_count = count;
 333                 if (count > (unsigned long) high_memory - p)
 334                         low_count = (unsigned long) high_memory - p;
 335
 336 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 337                 /* we don't have page 0 mapped on sparc and m68k.. */
 338                 if (p < PAGE_SIZE && low_count > 0) {
 339                         size_t tmp = PAGE_SIZE - p;
 340                         if (tmp > low_count) tmp = low_count;
 341                         if (clear_user(buf, tmp))
 342                                 return -EFAULT;
 343                         buf += tmp;
 344                         p += tmp;
 345                         read += tmp;
 346                         low_count -= tmp;
 347                         count -= tmp;
 348                 }
 349 #endif
 350                 while (low_count > 0) {
 351                         /*
 352                          * Handle first page in case it's not aligned
 353                          */
 354                         if (-p & (PAGE_SIZE - 1))
 355                                 sz = -p & (PAGE_SIZE - 1);
 356                         else
 357                                 sz = PAGE_SIZE;
 358
 359                         sz = min_t(unsigned long, sz, low_count);
 360
 361                         /*
 362                          * On ia64 if a page has been mapped somewhere as
 363                          * uncached, then it must also be accessed uncached
 364                          * by the kernel or data corruption may occur
 365                          */
 366                         kbuf = xlate_dev_kmem_ptr((char *)p);
 367
 368                         if (copy_to_user(buf, kbuf, sz))
 369                                 return -EFAULT;
 370                         buf += sz;
 371                         p += sz;
 372                         read += sz;
 373                         low_count -= sz;
 374                         count -= sz;
 375                 }
 376         }
 377
 378         if (count > 0) {
 379                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 380                 if (!kbuf)
 381                         return -ENOMEM;
 382                 while (count > 0) {
 383                         int len = count;
 384
 385                         if (len > PAGE_SIZE)
 386                                 len = PAGE_SIZE;
 387                         len = vread(kbuf, (char *)p, len);
 388                         if (!len)
 389                                 break;
 390                         if (copy_to_user(buf, kbuf, len)) {
 391                                 free_page((unsigned long)kbuf);
 392                                 return -EFAULT;
 393                         }
 394                         count -= len;
 395                         buf += len;
 396                         read += len;
 397                         p += len;
 398                 }
 399                 free_page((unsigned long)kbuf);
 400         }
 401         *ppos = p;
 402         return read;
 403 }
 404
 405
 406 static inline ssize_t
 407 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 408               size_t count, loff_t *ppos)
 409 {
 410         ssize_t written, sz;
 411         unsigned long copied;
 412
 413         written = 0;
 414 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 415         /* we don't have page 0 mapped on sparc and m68k.. */
 416         if (realp < PAGE_SIZE) {
 417                 unsigned long sz = PAGE_SIZE - realp;
 418                 if (sz > count)
 419                         sz = count;
 420                 /* Hmm. Do something? */
 421                 buf += sz;
 422                 p += sz;
 423                 realp += sz;
 424                 count -= sz;
 425                 written += sz;
 426         }
 427 #endif
 428
 429         while (count > 0) {
 430                 char *ptr;
 431                 /*
 432                  * Handle first page in case it's not aligned
 433                  */
 434                 if (-realp & (PAGE_SIZE - 1))
 435                         sz = -realp & (PAGE_SIZE - 1);
 436                 else
 437                         sz = PAGE_SIZE;
 438
 439                 sz = min_t(unsigned long, sz, count);
 440
 441                 /*
 442                  * On ia64 if a page has been mapped somewhere as
 443                  * uncached, then it must also be accessed uncached
 444                  * by the kernel or data corruption may occur
 445                  */
 446                 ptr = xlate_dev_kmem_ptr(p);
 447
 448                 copied = copy_from_user(ptr, buf, sz);
 449                 if (copied) {
 450                         ssize_t ret;
 451
 452                         ret = written + (sz - copied);
 453                         if (ret)
 454                                 return ret;
 455                         return -EFAULT;
 456                 }
 457                 buf += sz;
 458                 p += sz;
 459                 realp += sz;
 460                 count -= sz;
 461                 written += sz;
 462         }
 463
 464         *ppos += written;
 465         return written;
 466 }
 467
 468
 469 /*
 470  * This function writes to the *virtual* memory as seen by the kernel.
 471  */
 472 static ssize_t write_kmem(struct file * file, const char __user * buf,
 473                           size_t count, loff_t *ppos)
 474 {
 475         unsigned long p = *ppos;
 476         ssize_t wrote = 0;
 477         ssize_t virtr = 0;
 478         ssize_t written;
 479         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 480
 481         if (p < (unsigned long) high_memory) {
 482
 483                 wrote = count;
 484                 if (count > (unsigned long) high_memory - p)
 485                         wrote = (unsigned long) high_memory - p;
 486
 487                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 488                 if (written != wrote)
 489                         return written;
 490                 wrote = written;
 491                 p += wrote;
 492                 buf += wrote;
 493                 count -= wrote;
 494         }
 495
 496         if (count > 0) {
 497                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 498                 if (!kbuf)
 499                         return wrote ? wrote : -ENOMEM;
 500                 while (count > 0) {
 501                         int len = count;
 502
 503                         if (len > PAGE_SIZE)
 504                                 len = PAGE_SIZE;
 505                         if (len) {
 506                                 written = copy_from_user(kbuf, buf, len);
 507                                 if (written) {
 508                                         ssize_t ret;
 509
 510                                         free_page((unsigned long)kbuf);
 511                                         ret = wrote + virtr + (len - written);
 512                                         return ret ? ret : -EFAULT;
 513                                 }
 514                         }
 515                         len = vwrite(kbuf, (char *)p, len);
 516                         count -= len;
 517                         buf += len;
 518                         virtr += len;
 519                         p += len;
 520                 }
 521                 free_page((unsigned long)kbuf);
 522         }
 523
 524         *ppos = p;
 525         return virtr + wrote;
 526 }
 527
 528 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 529 static ssize_t read_port(struct file * file, char __user * buf,
 530                          size_t count, loff_t *ppos)
 531 {
 532         unsigned long i = *ppos;
 533         char __user *tmp = buf;
 534
 535         if (!access_ok(VERIFY_WRITE, buf, count))
 536                 return -EFAULT;
 537         while (count-- > 0 && i < 65536) {
 538                 if (__put_user(inb(i),tmp) < 0)
 539                         return -EFAULT;
 540                 i++;
 541                 tmp++;
 542         }
 543         *ppos = i;
 544         return tmp-buf;
 545 }
 546
 547 static ssize_t write_port(struct file * file, const char __user * buf,
 548                           size_t count, loff_t *ppos)
 549 {
 550         unsigned long i = *ppos;
 551         const char __user * tmp = buf;
 552
 553         if (!access_ok(VERIFY_READ,buf,count))
 554                 return -EFAULT;
 555         while (count-- > 0 && i < 65536) {
 556                 char c;
 557                 if (__get_user(c, tmp))
 558                         return -EFAULT;
 559                 outb(c,i);
 560                 i++;
 561                 tmp++;
 562         }
 563         *ppos = i;
 564         return tmp-buf;
 565 }
 566 #endif
 567
 568 static ssize_t read_null(struct file * file, char __user * buf,
 569                          size_t count, loff_t *ppos)
 570 {
 571         return 0;
 572 }
 573
 574 static ssize_t write_null(struct file * file, const char __user * buf,
 575                           size_t count, loff_t *ppos)
 576 {
 577         return count;
 578 }
 579
 580 #ifdef CONFIG_MMU
 581 /*
 582  * For fun, we are using the MMU for this.
 583  */
 584 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
 585 {
 586         struct mm_struct *mm;
 587         struct vm_area_struct * vma;
 588         unsigned long addr=(unsigned long)buf;
 589
 590         mm = current->mm;
 591         /* Oops, this was forgotten before. -ben */
 592         down_read(&mm->mmap_sem);
 593
 594         /* For private mappings, just map in zero pages. */
 595         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 596                 unsigned long count;
 597
 598                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 599                         goto out_up;
 600                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
 601                         break;
 602                 count = vma->vm_end - addr;
 603                 if (count > size)
 604                         count = size;
 605
 606                 zap_page_range(vma, addr, count, NULL);
 607                 zeromap_page_range(vma, addr, count, PAGE_COPY);
 608
 609                 size -= count;
 610                 buf += count;
 611                 addr += count;
 612                 if (size == 0)
 613                         goto out_up;
 614         }
 615
 616         up_read(&mm->mmap_sem);
 617
 618         /* The shared case is hard. Let's do the conventional zeroing. */
 619         do {
 620                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 621                 if (unwritten)
 622                         return size + unwritten - PAGE_SIZE;
 623                 cond_resched();
 624                 buf += PAGE_SIZE;
 625                 size -= PAGE_SIZE;
 626         } while (size);
 627
 628         return size;
 629 out_up:
 630         up_read(&mm->mmap_sem);
 631         return size;
 632 }
 633
 634 static ssize_t read_zero(struct file * file, char __user * buf,
 635                          size_t count, loff_t *ppos)
 636 {
 637         unsigned long left, unwritten, written = 0;
 638
 639         if (!count)
 640                 return 0;
 641
 642         if (!access_ok(VERIFY_WRITE, buf, count))
 643                 return -EFAULT;
 644
 645         left = count;
 646
 647         /* do we want to be clever? Arbitrary cut-off */
 648         if (count >= PAGE_SIZE*4) {
 649                 unsigned long partial;
 650
 651                 /* How much left of the page? */
 652                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 653                 unwritten = clear_user(buf, partial);
 654                 written = partial - unwritten;
 655                 if (unwritten)
 656                         goto out;
 657                 left -= partial;
 658                 buf += partial;
 659                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 660                 written += (left & PAGE_MASK) - unwritten;
 661                 if (unwritten)
 662                         goto out;
 663                 buf += left & PAGE_MASK;
 664                 left &= ~PAGE_MASK;
 665         }
 666         unwritten = clear_user(buf, left);
 667         written += left - unwritten;
 668 out:
 669         return written ? written : -EFAULT;
 670 }
 671
 672 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 673 {
 674         if (vma->vm_flags & VM_SHARED)
 675                 return shmem_zero_setup(vma);
 676         if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
 677                 return -EAGAIN;
 678         return 0;
 679 }
 680 #else /* CONFIG_MMU */
 681 static ssize_t read_zero(struct file * file, char * buf,
 682                          size_t count, loff_t *ppos)
 683 {
 684         size_t todo = count;
 685
 686         while (todo) {
 687                 size_t chunk = todo;
 688
 689                 if (chunk > 4096)
 690                         chunk = 4096;   /* Just for latency reasons */
 691                 if (clear_user(buf, chunk))
 692                         return -EFAULT;
 693                 buf += chunk;
 694                 todo -= chunk;
 695                 cond_resched();
 696         }
 697         return count;
 698 }
 699
 700 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 701 {
 702         return -ENOSYS;
 703 }
 704 #endif /* CONFIG_MMU */
 705
 706 static ssize_t write_full(struct file * file, const char __user * buf,
 707                           size_t count, loff_t *ppos)
 708 {
 709         return -ENOSPC;
 710 }
 711
 712 /*
 713  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 714  * can fopen() both devices with "a" now.  This was previously impossible.
 715  * -- SRB.
 716  */
 717
 718 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 719 {
 720         return file->f_pos = 0;
 721 }
 722
 723 /*
 724  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 725  * check against negative addresses: they are ok. The return value is weird,
 726  * though, in that case (0).
 727  *
 728  * also note that seeking relative to the "end of file" isn't supported:
 729  * it has no meaning, so it returns -EINVAL.
 730  */
 731 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 732 {
 733         loff_t ret;
 734
 735         down(&file->f_dentry->d_inode->i_sem);
 736         switch (orig) {
 737                 case 0:
 738                         file->f_pos = offset;
 739                         ret = file->f_pos;
 740                         force_successful_syscall_return();
 741                         break;
 742                 case 1:
 743                         file->f_pos += offset;
 744                         ret = file->f_pos;
 745                         force_successful_syscall_return();
 746                         break;
 747                 default:
 748                         ret = -EINVAL;
 749         }
 750         up(&file->f_dentry->d_inode->i_sem);
 751         return ret;
 752 }
 753
 754 static int open_port(struct inode * inode, struct file * filp)
 755 {
 756         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 757 }
 758
 759 #define zero_lseek      null_lseek
 760 #define full_lseek      null_lseek
 761 #define write_zero      write_null
 762 #define read_full       read_zero
 763 #define open_mem        open_port
 764 #define open_kmem       open_mem
 765 #define open_oldmem     open_mem
 766
 767 static struct file_operations mem_fops = {
 768         .llseek         = memory_lseek,
 769         .read           = read_mem,
 770         .write          = write_mem,
 771         .mmap           = mmap_mem,
 772         .open           = open_mem,
 773 };
 774
 775 static struct file_operations kmem_fops = {
 776         .llseek         = memory_lseek,
 777         .read           = read_kmem,
 778         .write          = write_kmem,
 779         .mmap           = mmap_kmem,
 780         .open           = open_kmem,
 781 };
 782
 783 static struct file_operations null_fops = {
 784         .llseek         = null_lseek,
 785         .read           = read_null,
 786         .write          = write_null,
 787 };
 788
 789 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 790 static struct file_operations port_fops = {
 791         .llseek         = memory_lseek,
 792         .read           = read_port,
 793         .write          = write_port,
 794         .open           = open_port,
 795 };
 796 #endif
 797
 798 static struct file_operations zero_fops = {
 799         .llseek         = zero_lseek,
 800         .read           = read_zero,
 801         .write          = write_zero,
 802         .mmap           = mmap_zero,
 803 };
 804
 805 static struct backing_dev_info zero_bdi = {
 806         .capabilities   = BDI_CAP_MAP_COPY,
 807 };
 808
 809 static struct file_operations full_fops = {
 810         .llseek         = full_lseek,
 811         .read           = read_full,
 812         .write          = write_full,
 813 };
 814
 815 #ifdef CONFIG_CRASH_DUMP
 816 static struct file_operations oldmem_fops = {
 817         .read   = read_oldmem,
 818         .open   = open_oldmem,
 819 };
 820 #endif
 821
 822 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 823                           size_t count, loff_t *ppos)
 824 {
 825         char *tmp;
 826         int ret;
 827
 828         tmp = kmalloc(count + 1, GFP_KERNEL);
 829         if (tmp == NULL)
 830                 return -ENOMEM;
 831         ret = -EFAULT;
 832         if (!copy_from_user(tmp, buf, count)) {
 833                 tmp[count] = 0;
 834                 ret = printk("%s", tmp);
 835         }
 836         kfree(tmp);
 837         return ret;
 838 }
 839
 840 static struct file_operations kmsg_fops = {
 841         .write =        kmsg_write,
 842 };
 843
 844 static int memory_open(struct inode * inode, struct file * filp)
 845 {
 846         switch (iminor(inode)) {
 847                 case 1:
 848                         filp->f_op = &mem_fops;
 849                         break;
 850                 case 2:
 851                         filp->f_op = &kmem_fops;
 852                         break;
 853                 case 3:
 854                         filp->f_op = &null_fops;
 855                         break;
 856 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 857                 case 4:
 858                         filp->f_op = &port_fops;
 859                         break;
 860 #endif
 861                 case 5:
 862                         filp->f_mapping->backing_dev_info = &zero_bdi;
 863                         filp->f_op = &zero_fops;
 864                         break;
 865                 case 7:
 866                         filp->f_op = &full_fops;
 867                         break;
 868                 case 8:
 869                         filp->f_op = &random_fops;
 870                         break;
 871                 case 9:
 872                         filp->f_op = &urandom_fops;
 873                         break;
 874                 case 11:
 875                         filp->f_op = &kmsg_fops;
 876                         break;
 877 #ifdef CONFIG_CRASH_DUMP
 878                 case 12:
 879                         filp->f_op = &oldmem_fops;
 880                         break;
 881 #endif
 882                 default:
 883                         return -ENXIO;
 884         }
 885         if (filp->f_op && filp->f_op->open)
 886                 return filp->f_op->open(inode,filp);
 887         return 0;
 888 }
 889
 890 static struct file_operations memory_fops = {
 891         .open           = memory_open,  /* just a selector for the real open */
 892 };
 893
 894 static const struct {
 895         unsigned int            minor;
 896         char                    *name;
 897         umode_t                 mode;
 898         struct file_operations  *fops;
 899 } devlist[] = { /* list of minor devices */
 900         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 901         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 902         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 903 #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI))
 904         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 905 #endif
 906         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 907         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 908         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 909         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 910         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 911 #ifdef CONFIG_CRASH_DUMP
 912         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 913 #endif
 914 };
 915
 916 static struct class *mem_class;
 917
 918 static int __init chr_dev_init(void)
 919 {
 920         int i;
 921
 922         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 923                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 924
 925         mem_class = class_create(THIS_MODULE, "mem");
 926         for (i = 0; i < ARRAY_SIZE(devlist); i++) {
 927                 class_device_create(mem_class, MKDEV(MEM_MAJOR, devlist[i].minor),
 928                                         NULL, devlist[i].name);
 929                 devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor),
 930                                 S_IFCHR | devlist[i].mode, devlist[i].name);
 931         }
 932
 933         return 0;
 934 }
 935
 936 fs_initcall(chr_dev_init);