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
 105 static inline int valid_mmap_phys_addr_range(unsigned long addr, size_t *size)
 106 {
 107         return 1;
 108 }
 109 #endif
 110
 111 /*
 112  * This funcion reads the *physical* memory. The f_pos points directly to the
 113  * memory location.
 114  */
 115 static ssize_t read_mem(struct file * file, char __user * buf,
 116                         size_t count, loff_t *ppos)
 117 {
 118         unsigned long p = *ppos;
 119         ssize_t read, sz;
 120         char *ptr;
 121
 122         if (!valid_phys_addr_range(p, &count))
 123                 return -EFAULT;
 124         read = 0;
 125 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 126         /* we don't have page 0 mapped on sparc and m68k.. */
 127         if (p < PAGE_SIZE) {
 128                 sz = PAGE_SIZE - p;
 129                 if (sz > count)
 130                         sz = count;
 131                 if (sz > 0) {
 132                         if (clear_user(buf, sz))
 133                                 return -EFAULT;
 134                         buf += sz;
 135                         p += sz;
 136                         count -= sz;
 137                         read += sz;
 138                 }
 139         }
 140 #endif
 141
 142         while (count > 0) {
 143                 /*
 144                  * Handle first page in case it's not aligned
 145                  */
 146                 if (-p & (PAGE_SIZE - 1))
 147                         sz = -p & (PAGE_SIZE - 1);
 148                 else
 149                         sz = PAGE_SIZE;
 150
 151                 sz = min_t(unsigned long, sz, count);
 152
 153                 /*
 154                  * On ia64 if a page has been mapped somewhere as
 155                  * uncached, then it must also be accessed uncached
 156                  * by the kernel or data corruption may occur
 157                  */
 158                 ptr = xlate_dev_mem_ptr(p);
 159
 160                 if (copy_to_user(buf, ptr, sz))
 161                         return -EFAULT;
 162                 buf += sz;
 163                 p += sz;
 164                 count -= sz;
 165                 read += sz;
 166         }
 167
 168         *ppos += read;
 169         return read;
 170 }
 171
 172 static ssize_t write_mem(struct file * file, const char __user * buf,
 173                          size_t count, loff_t *ppos)
 174 {
 175         unsigned long p = *ppos;
 176         ssize_t written, sz;
 177         unsigned long copied;
 178         void *ptr;
 179
 180         if (!valid_phys_addr_range(p, &count))
 181                 return -EFAULT;
 182
 183         written = 0;
 184
 185 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 186         /* we don't have page 0 mapped on sparc and m68k.. */
 187         if (p < PAGE_SIZE) {
 188                 unsigned long sz = PAGE_SIZE - p;
 189                 if (sz > count)
 190                         sz = count;
 191                 /* Hmm. Do something? */
 192                 buf += sz;
 193                 p += sz;
 194                 count -= sz;
 195                 written += sz;
 196         }
 197 #endif
 198
 199         while (count > 0) {
 200                 /*
 201                  * Handle first page in case it's not aligned
 202                  */
 203                 if (-p & (PAGE_SIZE - 1))
 204                         sz = -p & (PAGE_SIZE - 1);
 205                 else
 206                         sz = PAGE_SIZE;
 207
 208                 sz = min_t(unsigned long, sz, count);
 209
 210                 /*
 211                  * On ia64 if a page has been mapped somewhere as
 212                  * uncached, then it must also be accessed uncached
 213                  * by the kernel or data corruption may occur
 214                  */
 215                 ptr = xlate_dev_mem_ptr(p);
 216
 217                 copied = copy_from_user(ptr, buf, sz);
 218                 if (copied) {
 219                         ssize_t ret;
 220
 221                         ret = written + (sz - copied);
 222                         if (ret)
 223                                 return ret;
 224                         return -EFAULT;
 225                 }
 226                 buf += sz;
 227                 p += sz;
 228                 count -= sz;
 229                 written += sz;
 230         }
 231
 232         *ppos += written;
 233         return written;
 234 }
 235
 236 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 237 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 238                                      unsigned long size, pgprot_t vma_prot)
 239 {
 240 #ifdef pgprot_noncached
 241         unsigned long offset = pfn << PAGE_SHIFT;
 242
 243         if (uncached_access(file, offset))
 244                 return pgprot_noncached(vma_prot);
 245 #endif
 246         return vma_prot;
 247 }
 248 #endif
 249
 250 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 251 {
 252         size_t size = vma->vm_end - vma->vm_start;
 253
 254         if (!valid_mmap_phys_addr_range(vma->vm_pgoff << PAGE_SHIFT, &size))
 255                 return -EINVAL;
 256
 257         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 258                                                  size,
 259                                                  vma->vm_page_prot);
 260
 261         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 262         if (remap_pfn_range(vma,
 263                             vma->vm_start,
 264                             vma->vm_pgoff,
 265                             size,
 266                             vma->vm_page_prot))
 267                 return -EAGAIN;
 268         return 0;
 269 }
 270
 271 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 272 {
 273         unsigned long pfn;
 274
 275         /* Turn a kernel-virtual address into a physical page frame */
 276         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 277
 278         /*
 279          * RED-PEN: on some architectures there is more mapped memory
 280          * than available in mem_map which pfn_valid checks
 281          * for. Perhaps should add a new macro here.
 282          *
 283          * RED-PEN: vmalloc is not supported right now.
 284          */
 285         if (!pfn_valid(pfn))
 286                 return -EIO;
 287
 288         vma->vm_pgoff = pfn;
 289         return mmap_mem(file, vma);
 290 }
 291
 292 #ifdef CONFIG_CRASH_DUMP
 293 /*
 294  * Read memory corresponding to the old kernel.
 295  */
 296 static ssize_t read_oldmem(struct file *file, char __user *buf,
 297                                 size_t count, loff_t *ppos)
 298 {
 299         unsigned long pfn, offset;
 300         size_t read = 0, csize;
 301         int rc = 0;
 302
 303         while (count) {
 304                 pfn = *ppos / PAGE_SIZE;
 305                 if (pfn > saved_max_pfn)
 306                         return read;
 307
 308                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 309                 if (count > PAGE_SIZE - offset)
 310                         csize = PAGE_SIZE - offset;
 311                 else
 312                         csize = count;
 313
 314                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 315                 if (rc < 0)
 316                         return rc;
 317                 buf += csize;
 318                 *ppos += csize;
 319                 read += csize;
 320                 count -= csize;
 321         }
 322         return read;
 323 }
 324 #endif
 325
 326 extern long vread(char *buf, char *addr, unsigned long count);
 327 extern long vwrite(char *buf, char *addr, unsigned long count);
 328
 329 /*
 330  * This function reads the *virtual* memory as seen by the kernel.
 331  */
 332 static ssize_t read_kmem(struct file *file, char __user *buf,
 333                          size_t count, loff_t *ppos)
 334 {
 335         unsigned long p = *ppos;
 336         ssize_t low_count, read, sz;
 337         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 338
 339         read = 0;
 340         if (p < (unsigned long) high_memory) {
 341                 low_count = count;
 342                 if (count > (unsigned long) high_memory - p)
 343                         low_count = (unsigned long) high_memory - p;
 344
 345 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 346                 /* we don't have page 0 mapped on sparc and m68k.. */
 347                 if (p < PAGE_SIZE && low_count > 0) {
 348                         size_t tmp = PAGE_SIZE - p;
 349                         if (tmp > low_count) tmp = low_count;
 350                         if (clear_user(buf, tmp))
 351                                 return -EFAULT;
 352                         buf += tmp;
 353                         p += tmp;
 354                         read += tmp;
 355                         low_count -= tmp;
 356                         count -= tmp;
 357                 }
 358 #endif
 359                 while (low_count > 0) {
 360                         /*
 361                          * Handle first page in case it's not aligned
 362                          */
 363                         if (-p & (PAGE_SIZE - 1))
 364                                 sz = -p & (PAGE_SIZE - 1);
 365                         else
 366                                 sz = PAGE_SIZE;
 367
 368                         sz = min_t(unsigned long, sz, low_count);
 369
 370                         /*
 371                          * On ia64 if a page has been mapped somewhere as
 372                          * uncached, then it must also be accessed uncached
 373                          * by the kernel or data corruption may occur
 374                          */
 375                         kbuf = xlate_dev_kmem_ptr((char *)p);
 376
 377                         if (copy_to_user(buf, kbuf, sz))
 378                                 return -EFAULT;
 379                         buf += sz;
 380                         p += sz;
 381                         read += sz;
 382                         low_count -= sz;
 383                         count -= sz;
 384                 }
 385         }
 386
 387         if (count > 0) {
 388                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 389                 if (!kbuf)
 390                         return -ENOMEM;
 391                 while (count > 0) {
 392                         int len = count;
 393
 394                         if (len > PAGE_SIZE)
 395                                 len = PAGE_SIZE;
 396                         len = vread(kbuf, (char *)p, len);
 397                         if (!len)
 398                                 break;
 399                         if (copy_to_user(buf, kbuf, len)) {
 400                                 free_page((unsigned long)kbuf);
 401                                 return -EFAULT;
 402                         }
 403                         count -= len;
 404                         buf += len;
 405                         read += len;
 406                         p += len;
 407                 }
 408                 free_page((unsigned long)kbuf);
 409         }
 410         *ppos = p;
 411         return read;
 412 }
 413
 414
 415 static inline ssize_t
 416 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 417               size_t count, loff_t *ppos)
 418 {
 419         ssize_t written, sz;
 420         unsigned long copied;
 421
 422         written = 0;
 423 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 424         /* we don't have page 0 mapped on sparc and m68k.. */
 425         if (realp < PAGE_SIZE) {
 426                 unsigned long sz = PAGE_SIZE - realp;
 427                 if (sz > count)
 428                         sz = count;
 429                 /* Hmm. Do something? */
 430                 buf += sz;
 431                 p += sz;
 432                 realp += sz;
 433                 count -= sz;
 434                 written += sz;
 435         }
 436 #endif
 437
 438         while (count > 0) {
 439                 char *ptr;
 440                 /*
 441                  * Handle first page in case it's not aligned
 442                  */
 443                 if (-realp & (PAGE_SIZE - 1))
 444                         sz = -realp & (PAGE_SIZE - 1);
 445                 else
 446                         sz = PAGE_SIZE;
 447
 448                 sz = min_t(unsigned long, sz, count);
 449
 450                 /*
 451                  * On ia64 if a page has been mapped somewhere as
 452                  * uncached, then it must also be accessed uncached
 453                  * by the kernel or data corruption may occur
 454                  */
 455                 ptr = xlate_dev_kmem_ptr(p);
 456
 457                 copied = copy_from_user(ptr, buf, sz);
 458                 if (copied) {
 459                         ssize_t ret;
 460
 461                         ret = written + (sz - copied);
 462                         if (ret)
 463                                 return ret;
 464                         return -EFAULT;
 465                 }
 466                 buf += sz;
 467                 p += sz;
 468                 realp += sz;
 469                 count -= sz;
 470                 written += sz;
 471         }
 472
 473         *ppos += written;
 474         return written;
 475 }
 476
 477
 478 /*
 479  * This function writes to the *virtual* memory as seen by the kernel.
 480  */
 481 static ssize_t write_kmem(struct file * file, const char __user * buf,
 482                           size_t count, loff_t *ppos)
 483 {
 484         unsigned long p = *ppos;
 485         ssize_t wrote = 0;
 486         ssize_t virtr = 0;
 487         ssize_t written;
 488         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 489
 490         if (p < (unsigned long) high_memory) {
 491
 492                 wrote = count;
 493                 if (count > (unsigned long) high_memory - p)
 494                         wrote = (unsigned long) high_memory - p;
 495
 496                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 497                 if (written != wrote)
 498                         return written;
 499                 wrote = written;
 500                 p += wrote;
 501                 buf += wrote;
 502                 count -= wrote;
 503         }
 504
 505         if (count > 0) {
 506                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 507                 if (!kbuf)
 508                         return wrote ? wrote : -ENOMEM;
 509                 while (count > 0) {
 510                         int len = count;
 511
 512                         if (len > PAGE_SIZE)
 513                                 len = PAGE_SIZE;
 514                         if (len) {
 515                                 written = copy_from_user(kbuf, buf, len);
 516                                 if (written) {
 517                                         ssize_t ret;
 518
 519                                         free_page((unsigned long)kbuf);
 520                                         ret = wrote + virtr + (len - written);
 521                                         return ret ? ret : -EFAULT;
 522                                 }
 523                         }
 524                         len = vwrite(kbuf, (char *)p, len);
 525                         count -= len;
 526                         buf += len;
 527                         virtr += len;
 528                         p += len;
 529                 }
 530                 free_page((unsigned long)kbuf);
 531         }
 532
 533         *ppos = p;
 534         return virtr + wrote;
 535 }
 536
 537 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 538 static ssize_t read_port(struct file * file, char __user * buf,
 539                          size_t count, loff_t *ppos)
 540 {
 541         unsigned long i = *ppos;
 542         char __user *tmp = buf;
 543
 544         if (!access_ok(VERIFY_WRITE, buf, count))
 545                 return -EFAULT;
 546         while (count-- > 0 && i < 65536) {
 547                 if (__put_user(inb(i),tmp) < 0)
 548                         return -EFAULT;
 549                 i++;
 550                 tmp++;
 551         }
 552         *ppos = i;
 553         return tmp-buf;
 554 }
 555
 556 static ssize_t write_port(struct file * file, const char __user * buf,
 557                           size_t count, loff_t *ppos)
 558 {
 559         unsigned long i = *ppos;
 560         const char __user * tmp = buf;
 561
 562         if (!access_ok(VERIFY_READ,buf,count))
 563                 return -EFAULT;
 564         while (count-- > 0 && i < 65536) {
 565                 char c;
 566                 if (__get_user(c, tmp))
 567                         return -EFAULT;
 568                 outb(c,i);
 569                 i++;
 570                 tmp++;
 571         }
 572         *ppos = i;
 573         return tmp-buf;
 574 }
 575 #endif
 576
 577 static ssize_t read_null(struct file * file, char __user * buf,
 578                          size_t count, loff_t *ppos)
 579 {
 580         return 0;
 581 }
 582
 583 static ssize_t write_null(struct file * file, const char __user * buf,
 584                           size_t count, loff_t *ppos)
 585 {
 586         return count;
 587 }
 588
 589 #ifdef CONFIG_MMU
 590 /*
 591  * For fun, we are using the MMU for this.
 592  */
 593 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
 594 {
 595         struct mm_struct *mm;
 596         struct vm_area_struct * vma;
 597         unsigned long addr=(unsigned long)buf;
 598
 599         mm = current->mm;
 600         /* Oops, this was forgotten before. -ben */
 601         down_read(&mm->mmap_sem);
 602
 603         /* For private mappings, just map in zero pages. */
 604         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 605                 unsigned long count;
 606
 607                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 608                         goto out_up;
 609                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
 610                         break;
 611                 count = vma->vm_end - addr;
 612                 if (count > size)
 613                         count = size;
 614
 615                 zap_page_range(vma, addr, count, NULL);
 616                 zeromap_page_range(vma, addr, count, PAGE_COPY);
 617
 618                 size -= count;
 619                 buf += count;
 620                 addr += count;
 621                 if (size == 0)
 622                         goto out_up;
 623         }
 624
 625         up_read(&mm->mmap_sem);
 626
 627         /* The shared case is hard. Let's do the conventional zeroing. */
 628         do {
 629                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 630                 if (unwritten)
 631                         return size + unwritten - PAGE_SIZE;
 632                 cond_resched();
 633                 buf += PAGE_SIZE;
 634                 size -= PAGE_SIZE;
 635         } while (size);
 636
 637         return size;
 638 out_up:
 639         up_read(&mm->mmap_sem);
 640         return size;
 641 }
 642
 643 static ssize_t read_zero(struct file * file, char __user * buf,
 644                          size_t count, loff_t *ppos)
 645 {
 646         unsigned long left, unwritten, written = 0;
 647
 648         if (!count)
 649                 return 0;
 650
 651         if (!access_ok(VERIFY_WRITE, buf, count))
 652                 return -EFAULT;
 653
 654         left = count;
 655
 656         /* do we want to be clever? Arbitrary cut-off */
 657         if (count >= PAGE_SIZE*4) {
 658                 unsigned long partial;
 659
 660                 /* How much left of the page? */
 661                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 662                 unwritten = clear_user(buf, partial);
 663                 written = partial - unwritten;
 664                 if (unwritten)
 665                         goto out;
 666                 left -= partial;
 667                 buf += partial;
 668                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 669                 written += (left & PAGE_MASK) - unwritten;
 670                 if (unwritten)
 671                         goto out;
 672                 buf += left & PAGE_MASK;
 673                 left &= ~PAGE_MASK;
 674         }
 675         unwritten = clear_user(buf, left);
 676         written += left - unwritten;
 677 out:
 678         return written ? written : -EFAULT;
 679 }
 680
 681 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 682 {
 683         if (vma->vm_flags & VM_SHARED)
 684                 return shmem_zero_setup(vma);
 685         if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
 686                 return -EAGAIN;
 687         return 0;
 688 }
 689 #else /* CONFIG_MMU */
 690 static ssize_t read_zero(struct file * file, char * buf,
 691                          size_t count, loff_t *ppos)
 692 {
 693         size_t todo = count;
 694
 695         while (todo) {
 696                 size_t chunk = todo;
 697
 698                 if (chunk > 4096)
 699                         chunk = 4096;   /* Just for latency reasons */
 700                 if (clear_user(buf, chunk))
 701                         return -EFAULT;
 702                 buf += chunk;
 703                 todo -= chunk;
 704                 cond_resched();
 705         }
 706         return count;
 707 }
 708
 709 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 710 {
 711         return -ENOSYS;
 712 }
 713 #endif /* CONFIG_MMU */
 714
 715 static ssize_t write_full(struct file * file, const char __user * buf,
 716                           size_t count, loff_t *ppos)
 717 {
 718         return -ENOSPC;
 719 }
 720
 721 /*
 722  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 723  * can fopen() both devices with "a" now.  This was previously impossible.
 724  * -- SRB.
 725  */
 726
 727 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 728 {
 729         return file->f_pos = 0;
 730 }
 731
 732 /*
 733  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 734  * check against negative addresses: they are ok. The return value is weird,
 735  * though, in that case (0).
 736  *
 737  * also note that seeking relative to the "end of file" isn't supported:
 738  * it has no meaning, so it returns -EINVAL.
 739  */
 740 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 741 {
 742         loff_t ret;
 743
 744         mutex_lock(&file->f_dentry->d_inode->i_mutex);
 745         switch (orig) {
 746                 case 0:
 747                         file->f_pos = offset;
 748                         ret = file->f_pos;
 749                         force_successful_syscall_return();
 750                         break;
 751                 case 1:
 752                         file->f_pos += offset;
 753                         ret = file->f_pos;
 754                         force_successful_syscall_return();
 755                         break;
 756                 default:
 757                         ret = -EINVAL;
 758         }
 759         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 760         return ret;
 761 }
 762
 763 static int open_port(struct inode * inode, struct file * filp)
 764 {
 765         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 766 }
 767
 768 #define zero_lseek      null_lseek
 769 #define full_lseek      null_lseek
 770 #define write_zero      write_null
 771 #define read_full       read_zero
 772 #define open_mem        open_port
 773 #define open_kmem       open_mem
 774 #define open_oldmem     open_mem
 775
 776 static struct file_operations mem_fops = {
 777         .llseek         = memory_lseek,
 778         .read           = read_mem,
 779         .write          = write_mem,
 780         .mmap           = mmap_mem,
 781         .open           = open_mem,
 782 };
 783
 784 static struct file_operations kmem_fops = {
 785         .llseek         = memory_lseek,
 786         .read           = read_kmem,
 787         .write          = write_kmem,
 788         .mmap           = mmap_kmem,
 789         .open           = open_kmem,
 790 };
 791
 792 static struct file_operations null_fops = {
 793         .llseek         = null_lseek,
 794         .read           = read_null,
 795         .write          = write_null,
 796 };
 797
 798 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 799 static struct file_operations port_fops = {
 800         .llseek         = memory_lseek,
 801         .read           = read_port,
 802         .write          = write_port,
 803         .open           = open_port,
 804 };
 805 #endif
 806
 807 static struct file_operations zero_fops = {
 808         .llseek         = zero_lseek,
 809         .read           = read_zero,
 810         .write          = write_zero,
 811         .mmap           = mmap_zero,
 812 };
 813
 814 static struct backing_dev_info zero_bdi = {
 815         .capabilities   = BDI_CAP_MAP_COPY,
 816 };
 817
 818 static struct file_operations full_fops = {
 819         .llseek         = full_lseek,
 820         .read           = read_full,
 821         .write          = write_full,
 822 };
 823
 824 #ifdef CONFIG_CRASH_DUMP
 825 static struct file_operations oldmem_fops = {
 826         .read   = read_oldmem,
 827         .open   = open_oldmem,
 828 };
 829 #endif
 830
 831 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 832                           size_t count, loff_t *ppos)
 833 {
 834         char *tmp;
 835         ssize_t ret;
 836
 837         tmp = kmalloc(count + 1, GFP_KERNEL);
 838         if (tmp == NULL)
 839                 return -ENOMEM;
 840         ret = -EFAULT;
 841         if (!copy_from_user(tmp, buf, count)) {
 842                 tmp[count] = 0;
 843                 ret = printk("%s", tmp);
 844                 if (ret > count)
 845                         /* printk can add a prefix */
 846                         ret = count;
 847         }
 848         kfree(tmp);
 849         return ret;
 850 }
 851
 852 static struct file_operations kmsg_fops = {
 853         .write =        kmsg_write,
 854 };
 855
 856 static int memory_open(struct inode * inode, struct file * filp)
 857 {
 858         switch (iminor(inode)) {
 859                 case 1:
 860                         filp->f_op = &mem_fops;
 861                         break;
 862                 case 2:
 863                         filp->f_op = &kmem_fops;
 864                         break;
 865                 case 3:
 866                         filp->f_op = &null_fops;
 867                         break;
 868 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 869                 case 4:
 870                         filp->f_op = &port_fops;
 871                         break;
 872 #endif
 873                 case 5:
 874                         filp->f_mapping->backing_dev_info = &zero_bdi;
 875                         filp->f_op = &zero_fops;
 876                         break;
 877                 case 7:
 878                         filp->f_op = &full_fops;
 879                         break;
 880                 case 8:
 881                         filp->f_op = &random_fops;
 882                         break;
 883                 case 9:
 884                         filp->f_op = &urandom_fops;
 885                         break;
 886                 case 11:
 887                         filp->f_op = &kmsg_fops;
 888                         break;
 889 #ifdef CONFIG_CRASH_DUMP
 890                 case 12:
 891                         filp->f_op = &oldmem_fops;
 892                         break;
 893 #endif
 894                 default:
 895                         return -ENXIO;
 896         }
 897         if (filp->f_op && filp->f_op->open)
 898                 return filp->f_op->open(inode,filp);
 899         return 0;
 900 }
 901
 902 static struct file_operations memory_fops = {
 903         .open           = memory_open,  /* just a selector for the real open */
 904 };
 905
 906 static const struct {
 907         unsigned int            minor;
 908         char                    *name;
 909         umode_t                 mode;
 910         struct file_operations  *fops;
 911 } devlist[] = { /* list of minor devices */
 912         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 913         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 914         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 915 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 916         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 917 #endif
 918         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 919         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 920         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 921         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 922         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 923 #ifdef CONFIG_CRASH_DUMP
 924         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 925 #endif
 926 };
 927
 928 static struct class *mem_class;
 929
 930 static int __init chr_dev_init(void)
 931 {
 932         int i;
 933
 934         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 935                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 936
 937         mem_class = class_create(THIS_MODULE, "mem");
 938         for (i = 0; i < ARRAY_SIZE(devlist); i++) {
 939                 class_device_create(mem_class, NULL,
 940                                         MKDEV(MEM_MAJOR, devlist[i].minor),
 941                                         NULL, devlist[i].name);
 942                 devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor),
 943                                 S_IFCHR | devlist[i].mode, devlist[i].name);
 944         }
 945
 946         return 0;
 947 }
 948
 949 fs_initcall(chr_dev_init);