drivers/char/mem.c

   1 /*
   2  *  linux/drivers/char/mem.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *
   6  *  Added devfs support.
   7  *    Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
   8  *  Shared /dev/zero mmaping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
   9  */
  10
  11 #include <linux/mm.h>
  12 #include <linux/miscdevice.h>
  13 #include <linux/slab.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/mman.h>
  16 #include <linux/random.h>
  17 #include <linux/init.h>
  18 #include <linux/raw.h>
  19 #include <linux/tty.h>
  20 #include <linux/capability.h>
  21 #include <linux/smp_lock.h>
  22 #include <linux/ptrace.h>
  23 #include <linux/device.h>
  24 #include <linux/highmem.h>
  25 #include <linux/crash_dump.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/bootmem.h>
  28 #include <linux/pipe_fs_i.h>
  29 #include <linux/pfn.h>
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/io.h>
  33
  34 #ifdef CONFIG_IA64
  35 # include <linux/efi.h>
  36 #endif
  37
  38 /*
  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         if (addr + count > __pa(high_memory))
  94                 return 0;
  95
  96         return 1;
  97 }
  98
  99 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
 100 {
 101         return 1;
 102 }
 103 #endif
 104
 105 /*
 106  * This funcion reads the *physical* memory. The f_pos points directly to the
 107  * memory location.
 108  */
 109 static ssize_t read_mem(struct file * file, char __user * buf,
 110                         size_t count, loff_t *ppos)
 111 {
 112         unsigned long p = *ppos;
 113         ssize_t read, sz;
 114         char *ptr;
 115
 116         if (!valid_phys_addr_range(p, count))
 117                 return -EFAULT;
 118         read = 0;
 119 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 120         /* we don't have page 0 mapped on sparc and m68k.. */
 121         if (p < PAGE_SIZE) {
 122                 sz = PAGE_SIZE - p;
 123                 if (sz > count)
 124                         sz = count;
 125                 if (sz > 0) {
 126                         if (clear_user(buf, sz))
 127                                 return -EFAULT;
 128                         buf += sz;
 129                         p += sz;
 130                         count -= sz;
 131                         read += sz;
 132                 }
 133         }
 134 #endif
 135
 136         while (count > 0) {
 137                 /*
 138                  * Handle first page in case it's not aligned
 139                  */
 140                 if (-p & (PAGE_SIZE - 1))
 141                         sz = -p & (PAGE_SIZE - 1);
 142                 else
 143                         sz = PAGE_SIZE;
 144
 145                 sz = min_t(unsigned long, sz, count);
 146
 147                 /*
 148                  * On ia64 if a page has been mapped somewhere as
 149                  * uncached, then it must also be accessed uncached
 150                  * by the kernel or data corruption may occur
 151                  */
 152                 ptr = xlate_dev_mem_ptr(p);
 153
 154                 if (copy_to_user(buf, ptr, sz))
 155                         return -EFAULT;
 156                 buf += sz;
 157                 p += sz;
 158                 count -= sz;
 159                 read += sz;
 160         }
 161
 162         *ppos += read;
 163         return read;
 164 }
 165
 166 static ssize_t write_mem(struct file * file, const char __user * buf,
 167                          size_t count, loff_t *ppos)
 168 {
 169         unsigned long p = *ppos;
 170         ssize_t written, sz;
 171         unsigned long copied;
 172         void *ptr;
 173
 174         if (!valid_phys_addr_range(p, count))
 175                 return -EFAULT;
 176
 177         written = 0;
 178
 179 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 180         /* we don't have page 0 mapped on sparc and m68k.. */
 181         if (p < PAGE_SIZE) {
 182                 unsigned long sz = PAGE_SIZE - p;
 183                 if (sz > count)
 184                         sz = count;
 185                 /* Hmm. Do something? */
 186                 buf += sz;
 187                 p += sz;
 188                 count -= sz;
 189                 written += sz;
 190         }
 191 #endif
 192
 193         while (count > 0) {
 194                 /*
 195                  * Handle first page in case it's not aligned
 196                  */
 197                 if (-p & (PAGE_SIZE - 1))
 198                         sz = -p & (PAGE_SIZE - 1);
 199                 else
 200                         sz = PAGE_SIZE;
 201
 202                 sz = min_t(unsigned long, sz, count);
 203
 204                 /*
 205                  * On ia64 if a page has been mapped somewhere as
 206                  * uncached, then it must also be accessed uncached
 207                  * by the kernel or data corruption may occur
 208                  */
 209                 ptr = xlate_dev_mem_ptr(p);
 210
 211                 copied = copy_from_user(ptr, buf, sz);
 212                 if (copied) {
 213                         written += sz - copied;
 214                         if (written)
 215                                 break;
 216                         return -EFAULT;
 217                 }
 218                 buf += sz;
 219                 p += sz;
 220                 count -= sz;
 221                 written += sz;
 222         }
 223
 224         *ppos += written;
 225         return written;
 226 }
 227
 228 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 229 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 230                                      unsigned long size, pgprot_t vma_prot)
 231 {
 232 #ifdef pgprot_noncached
 233         unsigned long offset = pfn << PAGE_SHIFT;
 234
 235         if (uncached_access(file, offset))
 236                 return pgprot_noncached(vma_prot);
 237 #endif
 238         return vma_prot;
 239 }
 240 #endif
 241
 242 #ifndef CONFIG_MMU
 243 static unsigned long get_unmapped_area_mem(struct file *file,
 244                                            unsigned long addr,
 245                                            unsigned long len,
 246                                            unsigned long pgoff,
 247                                            unsigned long flags)
 248 {
 249         if (!valid_mmap_phys_addr_range(pgoff, len))
 250                 return (unsigned long) -EINVAL;
 251         return pgoff;
 252 }
 253
 254 /* can't do an in-place private mapping if there's no MMU */
 255 static inline int private_mapping_ok(struct vm_area_struct *vma)
 256 {
 257         return vma->vm_flags & VM_MAYSHARE;
 258 }
 259 #else
 260 #define get_unmapped_area_mem   NULL
 261
 262 static inline int private_mapping_ok(struct vm_area_struct *vma)
 263 {
 264         return 1;
 265 }
 266 #endif
 267
 268 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 269 {
 270         size_t size = vma->vm_end - vma->vm_start;
 271
 272         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 273                 return -EINVAL;
 274
 275         if (!private_mapping_ok(vma))
 276                 return -ENOSYS;
 277
 278         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 279                                                  size,
 280                                                  vma->vm_page_prot);
 281
 282         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 283         if (remap_pfn_range(vma,
 284                             vma->vm_start,
 285                             vma->vm_pgoff,
 286                             size,
 287                             vma->vm_page_prot))
 288                 return -EAGAIN;
 289         return 0;
 290 }
 291
 292 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 293 {
 294         unsigned long pfn;
 295
 296         /* Turn a kernel-virtual address into a physical page frame */
 297         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 298
 299         /*
 300          * RED-PEN: on some architectures there is more mapped memory
 301          * than available in mem_map which pfn_valid checks
 302          * for. Perhaps should add a new macro here.
 303          *
 304          * RED-PEN: vmalloc is not supported right now.
 305          */
 306         if (!pfn_valid(pfn))
 307                 return -EIO;
 308
 309         vma->vm_pgoff = pfn;
 310         return mmap_mem(file, vma);
 311 }
 312
 313 #ifdef CONFIG_CRASH_DUMP
 314 /*
 315  * Read memory corresponding to the old kernel.
 316  */
 317 static ssize_t read_oldmem(struct file *file, char __user *buf,
 318                                 size_t count, loff_t *ppos)
 319 {
 320         unsigned long pfn, offset;
 321         size_t read = 0, csize;
 322         int rc = 0;
 323
 324         while (count) {
 325                 pfn = *ppos / PAGE_SIZE;
 326                 if (pfn > saved_max_pfn)
 327                         return read;
 328
 329                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 330                 if (count > PAGE_SIZE - offset)
 331                         csize = PAGE_SIZE - offset;
 332                 else
 333                         csize = count;
 334
 335                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 336                 if (rc < 0)
 337                         return rc;
 338                 buf += csize;
 339                 *ppos += csize;
 340                 read += csize;
 341                 count -= csize;
 342         }
 343         return read;
 344 }
 345 #endif
 346
 347 extern long vread(char *buf, char *addr, unsigned long count);
 348 extern long vwrite(char *buf, char *addr, unsigned long count);
 349
 350 /*
 351  * This function reads the *virtual* memory as seen by the kernel.
 352  */
 353 static ssize_t read_kmem(struct file *file, char __user *buf,
 354                          size_t count, loff_t *ppos)
 355 {
 356         unsigned long p = *ppos;
 357         ssize_t low_count, read, sz;
 358         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 359
 360         read = 0;
 361         if (p < (unsigned long) high_memory) {
 362                 low_count = count;
 363                 if (count > (unsigned long) high_memory - p)
 364                         low_count = (unsigned long) high_memory - p;
 365
 366 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 367                 /* we don't have page 0 mapped on sparc and m68k.. */
 368                 if (p < PAGE_SIZE && low_count > 0) {
 369                         size_t tmp = PAGE_SIZE - p;
 370                         if (tmp > low_count) tmp = low_count;
 371                         if (clear_user(buf, tmp))
 372                                 return -EFAULT;
 373                         buf += tmp;
 374                         p += tmp;
 375                         read += tmp;
 376                         low_count -= tmp;
 377                         count -= tmp;
 378                 }
 379 #endif
 380                 while (low_count > 0) {
 381                         /*
 382                          * Handle first page in case it's not aligned
 383                          */
 384                         if (-p & (PAGE_SIZE - 1))
 385                                 sz = -p & (PAGE_SIZE - 1);
 386                         else
 387                                 sz = PAGE_SIZE;
 388
 389                         sz = min_t(unsigned long, sz, low_count);
 390
 391                         /*
 392                          * On ia64 if a page has been mapped somewhere as
 393                          * uncached, then it must also be accessed uncached
 394                          * by the kernel or data corruption may occur
 395                          */
 396                         kbuf = xlate_dev_kmem_ptr((char *)p);
 397
 398                         if (copy_to_user(buf, kbuf, sz))
 399                                 return -EFAULT;
 400                         buf += sz;
 401                         p += sz;
 402                         read += sz;
 403                         low_count -= sz;
 404                         count -= sz;
 405                 }
 406         }
 407
 408         if (count > 0) {
 409                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 410                 if (!kbuf)
 411                         return -ENOMEM;
 412                 while (count > 0) {
 413                         int len = count;
 414
 415                         if (len > PAGE_SIZE)
 416                                 len = PAGE_SIZE;
 417                         len = vread(kbuf, (char *)p, len);
 418                         if (!len)
 419                                 break;
 420                         if (copy_to_user(buf, kbuf, len)) {
 421                                 free_page((unsigned long)kbuf);
 422                                 return -EFAULT;
 423                         }
 424                         count -= len;
 425                         buf += len;
 426                         read += len;
 427                         p += len;
 428                 }
 429                 free_page((unsigned long)kbuf);
 430         }
 431         *ppos = p;
 432         return read;
 433 }
 434
 435
 436 static inline ssize_t
 437 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 438               size_t count, loff_t *ppos)
 439 {
 440         ssize_t written, sz;
 441         unsigned long copied;
 442
 443         written = 0;
 444 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 445         /* we don't have page 0 mapped on sparc and m68k.. */
 446         if (realp < PAGE_SIZE) {
 447                 unsigned long sz = PAGE_SIZE - realp;
 448                 if (sz > count)
 449                         sz = count;
 450                 /* Hmm. Do something? */
 451                 buf += sz;
 452                 p += sz;
 453                 realp += sz;
 454                 count -= sz;
 455                 written += sz;
 456         }
 457 #endif
 458
 459         while (count > 0) {
 460                 char *ptr;
 461                 /*
 462                  * Handle first page in case it's not aligned
 463                  */
 464                 if (-realp & (PAGE_SIZE - 1))
 465                         sz = -realp & (PAGE_SIZE - 1);
 466                 else
 467                         sz = PAGE_SIZE;
 468
 469                 sz = min_t(unsigned long, sz, count);
 470
 471                 /*
 472                  * On ia64 if a page has been mapped somewhere as
 473                  * uncached, then it must also be accessed uncached
 474                  * by the kernel or data corruption may occur
 475                  */
 476                 ptr = xlate_dev_kmem_ptr(p);
 477
 478                 copied = copy_from_user(ptr, buf, sz);
 479                 if (copied) {
 480                         written += sz - copied;
 481                         if (written)
 482                                 break;
 483                         return -EFAULT;
 484                 }
 485                 buf += sz;
 486                 p += sz;
 487                 realp += sz;
 488                 count -= sz;
 489                 written += sz;
 490         }
 491
 492         *ppos += written;
 493         return written;
 494 }
 495
 496
 497 /*
 498  * This function writes to the *virtual* memory as seen by the kernel.
 499  */
 500 static ssize_t write_kmem(struct file * file, const char __user * buf,
 501                           size_t count, loff_t *ppos)
 502 {
 503         unsigned long p = *ppos;
 504         ssize_t wrote = 0;
 505         ssize_t virtr = 0;
 506         ssize_t written;
 507         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 508
 509         if (p < (unsigned long) high_memory) {
 510
 511                 wrote = count;
 512                 if (count > (unsigned long) high_memory - p)
 513                         wrote = (unsigned long) high_memory - p;
 514
 515                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 516                 if (written != wrote)
 517                         return written;
 518                 wrote = written;
 519                 p += wrote;
 520                 buf += wrote;
 521                 count -= wrote;
 522         }
 523
 524         if (count > 0) {
 525                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 526                 if (!kbuf)
 527                         return wrote ? wrote : -ENOMEM;
 528                 while (count > 0) {
 529                         int len = count;
 530
 531                         if (len > PAGE_SIZE)
 532                                 len = PAGE_SIZE;
 533                         if (len) {
 534                                 written = copy_from_user(kbuf, buf, len);
 535                                 if (written) {
 536                                         if (wrote + virtr)
 537                                                 break;
 538                                         free_page((unsigned long)kbuf);
 539                                         return -EFAULT;
 540                                 }
 541                         }
 542                         len = vwrite(kbuf, (char *)p, len);
 543                         count -= len;
 544                         buf += len;
 545                         virtr += len;
 546                         p += len;
 547                 }
 548                 free_page((unsigned long)kbuf);
 549         }
 550
 551         *ppos = p;
 552         return virtr + wrote;
 553 }
 554
 555 #if (defined(CONFIG_ISA) || defined(CONFIG_PCI)) && !defined(__mc68000__)
 556 static ssize_t read_port(struct file * file, char __user * buf,
 557                          size_t count, loff_t *ppos)
 558 {
 559         unsigned long i = *ppos;
 560         char __user *tmp = buf;
 561
 562         if (!access_ok(VERIFY_WRITE, buf, count))
 563                 return -EFAULT;
 564         while (count-- > 0 && i < 65536) {
 565                 if (__put_user(inb(i),tmp) < 0)
 566                         return -EFAULT;
 567                 i++;
 568                 tmp++;
 569         }
 570         *ppos = i;
 571         return tmp-buf;
 572 }
 573
 574 static ssize_t write_port(struct file * file, const char __user * buf,
 575                           size_t count, loff_t *ppos)
 576 {
 577         unsigned long i = *ppos;
 578         const char __user * tmp = buf;
 579
 580         if (!access_ok(VERIFY_READ,buf,count))
 581                 return -EFAULT;
 582         while (count-- > 0 && i < 65536) {
 583                 char c;
 584                 if (__get_user(c, tmp)) {
 585                         if (tmp > buf)
 586                                 break;
 587                         return -EFAULT;
 588                 }
 589                 outb(c,i);
 590                 i++;
 591                 tmp++;
 592         }
 593         *ppos = i;
 594         return tmp-buf;
 595 }
 596 #endif
 597
 598 static ssize_t read_null(struct file * file, char __user * buf,
 599                          size_t count, loff_t *ppos)
 600 {
 601         return 0;
 602 }
 603
 604 static ssize_t write_null(struct file * file, const char __user * buf,
 605                           size_t count, loff_t *ppos)
 606 {
 607         return count;
 608 }
 609
 610 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 611                         struct splice_desc *sd)
 612 {
 613         return sd->len;
 614 }
 615
 616 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 617                                  loff_t *ppos, size_t len, unsigned int flags)
 618 {
 619         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 620 }
 621
 622 #ifdef CONFIG_MMU
 623 /*
 624  * For fun, we are using the MMU for this.
 625  */
 626 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
 627 {
 628         struct mm_struct *mm;
 629         struct vm_area_struct * vma;
 630         unsigned long addr=(unsigned long)buf;
 631
 632         mm = current->mm;
 633         /* Oops, this was forgotten before. -ben */
 634         down_read(&mm->mmap_sem);
 635
 636         /* For private mappings, just map in zero pages. */
 637         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 638                 unsigned long count;
 639
 640                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 641                         goto out_up;
 642                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
 643                         break;
 644                 count = vma->vm_end - addr;
 645                 if (count > size)
 646                         count = size;
 647
 648                 zap_page_range(vma, addr, count, NULL);
 649                 if (zeromap_page_range(vma, addr, count, PAGE_COPY))
 650                         break;
 651
 652                 size -= count;
 653                 buf += count;
 654                 addr += count;
 655                 if (size == 0)
 656                         goto out_up;
 657         }
 658
 659         up_read(&mm->mmap_sem);
 660
 661         /* The shared case is hard. Let's do the conventional zeroing. */
 662         do {
 663                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 664                 if (unwritten)
 665                         return size + unwritten - PAGE_SIZE;
 666                 cond_resched();
 667                 buf += PAGE_SIZE;
 668                 size -= PAGE_SIZE;
 669         } while (size);
 670
 671         return size;
 672 out_up:
 673         up_read(&mm->mmap_sem);
 674         return size;
 675 }
 676
 677 static ssize_t read_zero(struct file * file, char __user * buf,
 678                          size_t count, loff_t *ppos)
 679 {
 680         unsigned long left, unwritten, written = 0;
 681
 682         if (!count)
 683                 return 0;
 684
 685         if (!access_ok(VERIFY_WRITE, buf, count))
 686                 return -EFAULT;
 687
 688         left = count;
 689
 690         /* do we want to be clever? Arbitrary cut-off */
 691         if (count >= PAGE_SIZE*4) {
 692                 unsigned long partial;
 693
 694                 /* How much left of the page? */
 695                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 696                 unwritten = clear_user(buf, partial);
 697                 written = partial - unwritten;
 698                 if (unwritten)
 699                         goto out;
 700                 left -= partial;
 701                 buf += partial;
 702                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 703                 written += (left & PAGE_MASK) - unwritten;
 704                 if (unwritten)
 705                         goto out;
 706                 buf += left & PAGE_MASK;
 707                 left &= ~PAGE_MASK;
 708         }
 709         unwritten = clear_user(buf, left);
 710         written += left - unwritten;
 711 out:
 712         return written ? written : -EFAULT;
 713 }
 714
 715 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 716 {
 717         int err;
 718
 719         if (vma->vm_flags & VM_SHARED)
 720                 return shmem_zero_setup(vma);
 721         err = zeromap_page_range(vma, vma->vm_start,
 722                         vma->vm_end - vma->vm_start, vma->vm_page_prot);
 723         BUG_ON(err == -EEXIST);
 724         return err;
 725 }
 726 #else /* CONFIG_MMU */
 727 static ssize_t read_zero(struct file * file, char * buf,
 728                          size_t count, loff_t *ppos)
 729 {
 730         size_t todo = count;
 731
 732         while (todo) {
 733                 size_t chunk = todo;
 734
 735                 if (chunk > 4096)
 736                         chunk = 4096;   /* Just for latency reasons */
 737                 if (clear_user(buf, chunk))
 738                         return -EFAULT;
 739                 buf += chunk;
 740                 todo -= chunk;
 741                 cond_resched();
 742         }
 743         return count;
 744 }
 745
 746 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 747 {
 748         return -ENOSYS;
 749 }
 750 #endif /* CONFIG_MMU */
 751
 752 static ssize_t write_full(struct file * file, const char __user * buf,
 753                           size_t count, loff_t *ppos)
 754 {
 755         return -ENOSPC;
 756 }
 757
 758 /*
 759  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 760  * can fopen() both devices with "a" now.  This was previously impossible.
 761  * -- SRB.
 762  */
 763
 764 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 765 {
 766         return file->f_pos = 0;
 767 }
 768
 769 /*
 770  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 771  * check against negative addresses: they are ok. The return value is weird,
 772  * though, in that case (0).
 773  *
 774  * also note that seeking relative to the "end of file" isn't supported:
 775  * it has no meaning, so it returns -EINVAL.
 776  */
 777 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 778 {
 779         loff_t ret;
 780
 781         mutex_lock(&file->f_dentry->d_inode->i_mutex);
 782         switch (orig) {
 783                 case 0:
 784                         file->f_pos = offset;
 785                         ret = file->f_pos;
 786                         force_successful_syscall_return();
 787                         break;
 788                 case 1:
 789                         file->f_pos += offset;
 790                         ret = file->f_pos;
 791                         force_successful_syscall_return();
 792                         break;
 793                 default:
 794                         ret = -EINVAL;
 795         }
 796         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 797         return ret;
 798 }
 799
 800 static int open_port(struct inode * inode, struct file * filp)
 801 {
 802         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 803 }
 804
 805 #define zero_lseek      null_lseek
 806 #define full_lseek      null_lseek
 807 #define write_zero      write_null
 808 #define read_full       read_zero
 809 #define open_mem        open_port
 810 #define open_kmem       open_mem
 811 #define open_oldmem     open_mem
 812
 813 static const struct file_operations mem_fops = {
 814         .llseek         = memory_lseek,
 815         .read           = read_mem,
 816         .write          = write_mem,
 817         .mmap           = mmap_mem,
 818         .open           = open_mem,
 819         .get_unmapped_area = get_unmapped_area_mem,
 820 };
 821
 822 static const struct file_operations kmem_fops = {
 823         .llseek         = memory_lseek,
 824         .read           = read_kmem,
 825         .write          = write_kmem,
 826         .mmap           = mmap_kmem,
 827         .open           = open_kmem,
 828         .get_unmapped_area = get_unmapped_area_mem,
 829 };
 830
 831 static const struct file_operations null_fops = {
 832         .llseek         = null_lseek,
 833         .read           = read_null,
 834         .write          = write_null,
 835         .splice_write   = splice_write_null,
 836 };
 837
 838 #if (defined(CONFIG_ISA) || defined(CONFIG_PCI)) && !defined(__mc68000__)
 839 static const struct file_operations port_fops = {
 840         .llseek         = memory_lseek,
 841         .read           = read_port,
 842         .write          = write_port,
 843         .open           = open_port,
 844 };
 845 #endif
 846
 847 static const struct file_operations zero_fops = {
 848         .llseek         = zero_lseek,
 849         .read           = read_zero,
 850         .write          = write_zero,
 851         .mmap           = mmap_zero,
 852 };
 853
 854 /*
 855  * capabilities for /dev/zero
 856  * - permits private mappings, "copies" are taken of the source of zeros
 857  */
 858 static struct backing_dev_info zero_bdi = {
 859         .capabilities   = BDI_CAP_MAP_COPY,
 860 };
 861
 862 static const struct file_operations full_fops = {
 863         .llseek         = full_lseek,
 864         .read           = read_full,
 865         .write          = write_full,
 866 };
 867
 868 #ifdef CONFIG_CRASH_DUMP
 869 static const struct file_operations oldmem_fops = {
 870         .read   = read_oldmem,
 871         .open   = open_oldmem,
 872 };
 873 #endif
 874
 875 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 876                           size_t count, loff_t *ppos)
 877 {
 878         char *tmp;
 879         ssize_t ret;
 880
 881         tmp = kmalloc(count + 1, GFP_KERNEL);
 882         if (tmp == NULL)
 883                 return -ENOMEM;
 884         ret = -EFAULT;
 885         if (!copy_from_user(tmp, buf, count)) {
 886                 tmp[count] = 0;
 887                 ret = printk("%s", tmp);
 888                 if (ret > count)
 889                         /* printk can add a prefix */
 890                         ret = count;
 891         }
 892         kfree(tmp);
 893         return ret;
 894 }
 895
 896 static const struct file_operations kmsg_fops = {
 897         .write =        kmsg_write,
 898 };
 899
 900 static int memory_open(struct inode * inode, struct file * filp)
 901 {
 902         switch (iminor(inode)) {
 903                 case 1:
 904                         filp->f_op = &mem_fops;
 905                         filp->f_mapping->backing_dev_info =
 906                                 &directly_mappable_cdev_bdi;
 907                         break;
 908                 case 2:
 909                         filp->f_op = &kmem_fops;
 910                         filp->f_mapping->backing_dev_info =
 911                                 &directly_mappable_cdev_bdi;
 912                         break;
 913                 case 3:
 914                         filp->f_op = &null_fops;
 915                         break;
 916 #if (defined(CONFIG_ISA) || defined(CONFIG_PCI)) && !defined(__mc68000__)
 917                 case 4:
 918                         filp->f_op = &port_fops;
 919                         break;
 920 #endif
 921                 case 5:
 922                         filp->f_mapping->backing_dev_info = &zero_bdi;
 923                         filp->f_op = &zero_fops;
 924                         break;
 925                 case 7:
 926                         filp->f_op = &full_fops;
 927                         break;
 928                 case 8:
 929                         filp->f_op = &random_fops;
 930                         break;
 931                 case 9:
 932                         filp->f_op = &urandom_fops;
 933                         break;
 934                 case 11:
 935                         filp->f_op = &kmsg_fops;
 936                         break;
 937 #ifdef CONFIG_CRASH_DUMP
 938                 case 12:
 939                         filp->f_op = &oldmem_fops;
 940                         break;
 941 #endif
 942                 default:
 943                         return -ENXIO;
 944         }
 945         if (filp->f_op && filp->f_op->open)
 946                 return filp->f_op->open(inode,filp);
 947         return 0;
 948 }
 949
 950 static const struct file_operations memory_fops = {
 951         .open           = memory_open,  /* just a selector for the real open */
 952 };
 953
 954 static const struct {
 955         unsigned int            minor;
 956         char                    *name;
 957         umode_t                 mode;
 958         const struct file_operations    *fops;
 959 } devlist[] = { /* list of minor devices */
 960         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 961         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 962         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 963 #if (defined(CONFIG_ISA) || defined(CONFIG_PCI)) && !defined(__mc68000__)
 964         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 965 #endif
 966         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 967         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 968         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 969         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 970         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 971 #ifdef CONFIG_CRASH_DUMP
 972         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 973 #endif
 974 };
 975
 976 static struct class *mem_class;
 977
 978 static int __init chr_dev_init(void)
 979 {
 980         int i;
 981
 982         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 983                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 984
 985         mem_class = class_create(THIS_MODULE, "mem");
 986         for (i = 0; i < ARRAY_SIZE(devlist); i++)
 987                 class_device_create(mem_class, NULL,
 988                                         MKDEV(MEM_MAJOR, devlist[i].minor),
 989                                         NULL, devlist[i].name);
 990
 991         return 0;
 992 }
 993
 994 fs_initcall(chr_dev_init);