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/ptrace.h>
  22 #include <linux/device.h>
  23 #include <linux/highmem.h>
  24 #include <linux/crash_dump.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/splice.h>
  28 #include <linux/pfn.h>
  29
  30 #include <asm/uaccess.h>
  31 #include <asm/io.h>
  32
  33 #ifdef CONFIG_IA64
  34 # include <linux/efi.h>
  35 #endif
  36
  37 /*
  38  * Architectures vary in how they handle caching for addresses
  39  * outside of main memory.
  40  *
  41  */
  42 static inline int uncached_access(struct file *file, unsigned long addr)
  43 {
  44 #if defined(__i386__) && !defined(__arch_um__)
  45         /*
  46          * On the PPro and successors, the MTRRs are used to set
  47          * memory types for physical addresses outside main memory,
  48          * so blindly setting PCD or PWT on those pages is wrong.
  49          * For Pentiums and earlier, the surround logic should disable
  50          * caching for the high addresses through the KEN pin, but
  51          * we maintain the tradition of paranoia in this code.
  52          */
  53         if (file->f_flags & O_SYNC)
  54                 return 1;
  55         return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
  56                   test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
  57                   test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
  58                   test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
  59           && addr >= __pa(high_memory);
  60 #elif defined(__x86_64__) && !defined(__arch_um__)
  61         /*
  62          * This is broken because it can generate memory type aliases,
  63          * which can cause cache corruptions
  64          * But it is only available for root and we have to be bug-to-bug
  65          * compatible with i386.
  66          */
  67         if (file->f_flags & O_SYNC)
  68                 return 1;
  69         /* same behaviour as i386. PAT always set to cached and MTRRs control the
  70            caching behaviour.
  71            Hopefully a full PAT implementation will fix that soon. */
  72         return 0;
  73 #elif defined(CONFIG_IA64)
  74         /*
  75          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  76          */
  77         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  78 #elif defined(CONFIG_MIPS)
  79         {
  80                 extern int __uncached_access(struct file *file,
  81                                              unsigned long addr);
  82
  83                 return __uncached_access(file, addr);
  84         }
  85 #else
  86         /*
  87          * Accessing memory above the top the kernel knows about or through a file pointer
  88          * that was marked O_SYNC will be done non-cached.
  89          */
  90         if (file->f_flags & O_SYNC)
  91                 return 1;
  92         return addr >= __pa(high_memory);
  93 #endif
  94 }
  95
  96 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  97 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  98 {
  99         if (addr + count > __pa(high_memory))
 100                 return 0;
 101
 102         return 1;
 103 }
 104
 105 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
 106 {
 107         return 1;
 108 }
 109 #endif
 110
 111 #ifdef CONFIG_NONPROMISC_DEVMEM
 112 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 113 {
 114         u64 from = ((u64)pfn) << PAGE_SHIFT;
 115         u64 to = from + size;
 116         u64 cursor = from;
 117
 118         while (cursor < to) {
 119                 if (!devmem_is_allowed(pfn)) {
 120                         printk(KERN_INFO
 121                 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
 122                                 current->comm, from, to);
 123                         return 0;
 124                 }
 125                 cursor += PAGE_SIZE;
 126                 pfn++;
 127         }
 128         return 1;
 129 }
 130 #else
 131 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 132 {
 133         return 1;
 134 }
 135 #endif
 136
 137 void __attribute__((weak)) unxlate_dev_mem_ptr(unsigned long phys, void *addr)
 138 {
 139 }
 140
 141 /*
 142  * This funcion reads the *physical* memory. The f_pos points directly to the
 143  * memory location.
 144  */
 145 static ssize_t read_mem(struct file * file, char __user * buf,
 146                         size_t count, loff_t *ppos)
 147 {
 148         unsigned long p = *ppos;
 149         ssize_t read, sz;
 150         char *ptr;
 151
 152         if (!valid_phys_addr_range(p, count))
 153                 return -EFAULT;
 154         read = 0;
 155 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 156         /* we don't have page 0 mapped on sparc and m68k.. */
 157         if (p < PAGE_SIZE) {
 158                 sz = PAGE_SIZE - p;
 159                 if (sz > count)
 160                         sz = count;
 161                 if (sz > 0) {
 162                         if (clear_user(buf, sz))
 163                                 return -EFAULT;
 164                         buf += sz;
 165                         p += sz;
 166                         count -= sz;
 167                         read += sz;
 168                 }
 169         }
 170 #endif
 171
 172         while (count > 0) {
 173                 /*
 174                  * Handle first page in case it's not aligned
 175                  */
 176                 if (-p & (PAGE_SIZE - 1))
 177                         sz = -p & (PAGE_SIZE - 1);
 178                 else
 179                         sz = PAGE_SIZE;
 180
 181                 sz = min_t(unsigned long, sz, count);
 182
 183                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 184                         return -EPERM;
 185
 186                 /*
 187                  * On ia64 if a page has been mapped somewhere as
 188                  * uncached, then it must also be accessed uncached
 189                  * by the kernel or data corruption may occur
 190                  */
 191                 ptr = xlate_dev_mem_ptr(p);
 192                 if (!ptr)
 193                         return -EFAULT;
 194
 195                 if (copy_to_user(buf, ptr, sz)) {
 196                         unxlate_dev_mem_ptr(p, ptr);
 197                         return -EFAULT;
 198                 }
 199
 200                 unxlate_dev_mem_ptr(p, ptr);
 201
 202                 buf += sz;
 203                 p += sz;
 204                 count -= sz;
 205                 read += sz;
 206         }
 207
 208         *ppos += read;
 209         return read;
 210 }
 211
 212 static ssize_t write_mem(struct file * file, const char __user * buf,
 213                          size_t count, loff_t *ppos)
 214 {
 215         unsigned long p = *ppos;
 216         ssize_t written, sz;
 217         unsigned long copied;
 218         void *ptr;
 219
 220         if (!valid_phys_addr_range(p, count))
 221                 return -EFAULT;
 222
 223         written = 0;
 224
 225 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 226         /* we don't have page 0 mapped on sparc and m68k.. */
 227         if (p < PAGE_SIZE) {
 228                 unsigned long sz = PAGE_SIZE - p;
 229                 if (sz > count)
 230                         sz = count;
 231                 /* Hmm. Do something? */
 232                 buf += sz;
 233                 p += sz;
 234                 count -= sz;
 235                 written += sz;
 236         }
 237 #endif
 238
 239         while (count > 0) {
 240                 /*
 241                  * Handle first page in case it's not aligned
 242                  */
 243                 if (-p & (PAGE_SIZE - 1))
 244                         sz = -p & (PAGE_SIZE - 1);
 245                 else
 246                         sz = PAGE_SIZE;
 247
 248                 sz = min_t(unsigned long, sz, count);
 249
 250                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 251                         return -EPERM;
 252
 253                 /*
 254                  * On ia64 if a page has been mapped somewhere as
 255                  * uncached, then it must also be accessed uncached
 256                  * by the kernel or data corruption may occur
 257                  */
 258                 ptr = xlate_dev_mem_ptr(p);
 259                 if (!ptr) {
 260                         if (written)
 261                                 break;
 262                         return -EFAULT;
 263                 }
 264
 265                 copied = copy_from_user(ptr, buf, sz);
 266                 if (copied) {
 267                         written += sz - copied;
 268                         unxlate_dev_mem_ptr(p, ptr);
 269                         if (written)
 270                                 break;
 271                         return -EFAULT;
 272                 }
 273
 274                 unxlate_dev_mem_ptr(p, ptr);
 275
 276                 buf += sz;
 277                 p += sz;
 278                 count -= sz;
 279                 written += sz;
 280         }
 281
 282         *ppos += written;
 283         return written;
 284 }
 285
 286 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 287 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 288                                      unsigned long size, pgprot_t vma_prot)
 289 {
 290 #ifdef pgprot_noncached
 291         unsigned long offset = pfn << PAGE_SHIFT;
 292
 293         if (uncached_access(file, offset))
 294                 return pgprot_noncached(vma_prot);
 295 #endif
 296         return vma_prot;
 297 }
 298 #endif
 299
 300 #ifndef CONFIG_MMU
 301 static unsigned long get_unmapped_area_mem(struct file *file,
 302                                            unsigned long addr,
 303                                            unsigned long len,
 304                                            unsigned long pgoff,
 305                                            unsigned long flags)
 306 {
 307         if (!valid_mmap_phys_addr_range(pgoff, len))
 308                 return (unsigned long) -EINVAL;
 309         return pgoff << PAGE_SHIFT;
 310 }
 311
 312 /* can't do an in-place private mapping if there's no MMU */
 313 static inline int private_mapping_ok(struct vm_area_struct *vma)
 314 {
 315         return vma->vm_flags & VM_MAYSHARE;
 316 }
 317 #else
 318 #define get_unmapped_area_mem   NULL
 319
 320 static inline int private_mapping_ok(struct vm_area_struct *vma)
 321 {
 322         return 1;
 323 }
 324 #endif
 325
 326 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 327 {
 328         size_t size = vma->vm_end - vma->vm_start;
 329
 330         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 331                 return -EINVAL;
 332
 333         if (!private_mapping_ok(vma))
 334                 return -ENOSYS;
 335
 336         if (!range_is_allowed(vma->vm_pgoff, size))
 337                 return -EPERM;
 338
 339         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 340                                                  size,
 341                                                  vma->vm_page_prot);
 342
 343         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 344         if (remap_pfn_range(vma,
 345                             vma->vm_start,
 346                             vma->vm_pgoff,
 347                             size,
 348                             vma->vm_page_prot))
 349                 return -EAGAIN;
 350         return 0;
 351 }
 352
 353 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 354 {
 355         unsigned long pfn;
 356
 357         /* Turn a kernel-virtual address into a physical page frame */
 358         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 359
 360         /*
 361          * RED-PEN: on some architectures there is more mapped memory
 362          * than available in mem_map which pfn_valid checks
 363          * for. Perhaps should add a new macro here.
 364          *
 365          * RED-PEN: vmalloc is not supported right now.
 366          */
 367         if (!pfn_valid(pfn))
 368                 return -EIO;
 369
 370         vma->vm_pgoff = pfn;
 371         return mmap_mem(file, vma);
 372 }
 373
 374 #ifdef CONFIG_CRASH_DUMP
 375 /*
 376  * Read memory corresponding to the old kernel.
 377  */
 378 static ssize_t read_oldmem(struct file *file, char __user *buf,
 379                                 size_t count, loff_t *ppos)
 380 {
 381         unsigned long pfn, offset;
 382         size_t read = 0, csize;
 383         int rc = 0;
 384
 385         while (count) {
 386                 pfn = *ppos / PAGE_SIZE;
 387                 if (pfn > saved_max_pfn)
 388                         return read;
 389
 390                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 391                 if (count > PAGE_SIZE - offset)
 392                         csize = PAGE_SIZE - offset;
 393                 else
 394                         csize = count;
 395
 396                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 397                 if (rc < 0)
 398                         return rc;
 399                 buf += csize;
 400                 *ppos += csize;
 401                 read += csize;
 402                 count -= csize;
 403         }
 404         return read;
 405 }
 406 #endif
 407
 408 extern long vread(char *buf, char *addr, unsigned long count);
 409 extern long vwrite(char *buf, char *addr, unsigned long count);
 410
 411 /*
 412  * This function reads the *virtual* memory as seen by the kernel.
 413  */
 414 static ssize_t read_kmem(struct file *file, char __user *buf,
 415                          size_t count, loff_t *ppos)
 416 {
 417         unsigned long p = *ppos;
 418         ssize_t low_count, read, sz;
 419         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 420
 421         read = 0;
 422         if (p < (unsigned long) high_memory) {
 423                 low_count = count;
 424                 if (count > (unsigned long) high_memory - p)
 425                         low_count = (unsigned long) high_memory - p;
 426
 427 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 428                 /* we don't have page 0 mapped on sparc and m68k.. */
 429                 if (p < PAGE_SIZE && low_count > 0) {
 430                         size_t tmp = PAGE_SIZE - p;
 431                         if (tmp > low_count) tmp = low_count;
 432                         if (clear_user(buf, tmp))
 433                                 return -EFAULT;
 434                         buf += tmp;
 435                         p += tmp;
 436                         read += tmp;
 437                         low_count -= tmp;
 438                         count -= tmp;
 439                 }
 440 #endif
 441                 while (low_count > 0) {
 442                         /*
 443                          * Handle first page in case it's not aligned
 444                          */
 445                         if (-p & (PAGE_SIZE - 1))
 446                                 sz = -p & (PAGE_SIZE - 1);
 447                         else
 448                                 sz = PAGE_SIZE;
 449
 450                         sz = min_t(unsigned long, sz, low_count);
 451
 452                         /*
 453                          * On ia64 if a page has been mapped somewhere as
 454                          * uncached, then it must also be accessed uncached
 455                          * by the kernel or data corruption may occur
 456                          */
 457                         kbuf = xlate_dev_kmem_ptr((char *)p);
 458
 459                         if (copy_to_user(buf, kbuf, sz))
 460                                 return -EFAULT;
 461                         buf += sz;
 462                         p += sz;
 463                         read += sz;
 464                         low_count -= sz;
 465                         count -= sz;
 466                 }
 467         }
 468
 469         if (count > 0) {
 470                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 471                 if (!kbuf)
 472                         return -ENOMEM;
 473                 while (count > 0) {
 474                         int len = count;
 475
 476                         if (len > PAGE_SIZE)
 477                                 len = PAGE_SIZE;
 478                         len = vread(kbuf, (char *)p, len);
 479                         if (!len)
 480                                 break;
 481                         if (copy_to_user(buf, kbuf, len)) {
 482                                 free_page((unsigned long)kbuf);
 483                                 return -EFAULT;
 484                         }
 485                         count -= len;
 486                         buf += len;
 487                         read += len;
 488                         p += len;
 489                 }
 490                 free_page((unsigned long)kbuf);
 491         }
 492         *ppos = p;
 493         return read;
 494 }
 495
 496
 497 static inline ssize_t
 498 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 499               size_t count, loff_t *ppos)
 500 {
 501         ssize_t written, sz;
 502         unsigned long copied;
 503
 504         written = 0;
 505 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 506         /* we don't have page 0 mapped on sparc and m68k.. */
 507         if (realp < PAGE_SIZE) {
 508                 unsigned long sz = PAGE_SIZE - realp;
 509                 if (sz > count)
 510                         sz = count;
 511                 /* Hmm. Do something? */
 512                 buf += sz;
 513                 p += sz;
 514                 realp += sz;
 515                 count -= sz;
 516                 written += sz;
 517         }
 518 #endif
 519
 520         while (count > 0) {
 521                 char *ptr;
 522                 /*
 523                  * Handle first page in case it's not aligned
 524                  */
 525                 if (-realp & (PAGE_SIZE - 1))
 526                         sz = -realp & (PAGE_SIZE - 1);
 527                 else
 528                         sz = PAGE_SIZE;
 529
 530                 sz = min_t(unsigned long, sz, count);
 531
 532                 /*
 533                  * On ia64 if a page has been mapped somewhere as
 534                  * uncached, then it must also be accessed uncached
 535                  * by the kernel or data corruption may occur
 536                  */
 537                 ptr = xlate_dev_kmem_ptr(p);
 538
 539                 copied = copy_from_user(ptr, buf, sz);
 540                 if (copied) {
 541                         written += sz - copied;
 542                         if (written)
 543                                 break;
 544                         return -EFAULT;
 545                 }
 546                 buf += sz;
 547                 p += sz;
 548                 realp += sz;
 549                 count -= sz;
 550                 written += sz;
 551         }
 552
 553         *ppos += written;
 554         return written;
 555 }
 556
 557
 558 /*
 559  * This function writes to the *virtual* memory as seen by the kernel.
 560  */
 561 static ssize_t write_kmem(struct file * file, const char __user * buf,
 562                           size_t count, loff_t *ppos)
 563 {
 564         unsigned long p = *ppos;
 565         ssize_t wrote = 0;
 566         ssize_t virtr = 0;
 567         ssize_t written;
 568         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 569
 570         if (p < (unsigned long) high_memory) {
 571
 572                 wrote = count;
 573                 if (count > (unsigned long) high_memory - p)
 574                         wrote = (unsigned long) high_memory - p;
 575
 576                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 577                 if (written != wrote)
 578                         return written;
 579                 wrote = written;
 580                 p += wrote;
 581                 buf += wrote;
 582                 count -= wrote;
 583         }
 584
 585         if (count > 0) {
 586                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 587                 if (!kbuf)
 588                         return wrote ? wrote : -ENOMEM;
 589                 while (count > 0) {
 590                         int len = count;
 591
 592                         if (len > PAGE_SIZE)
 593                                 len = PAGE_SIZE;
 594                         if (len) {
 595                                 written = copy_from_user(kbuf, buf, len);
 596                                 if (written) {
 597                                         if (wrote + virtr)
 598                                                 break;
 599                                         free_page((unsigned long)kbuf);
 600                                         return -EFAULT;
 601                                 }
 602                         }
 603                         len = vwrite(kbuf, (char *)p, len);
 604                         count -= len;
 605                         buf += len;
 606                         virtr += len;
 607                         p += len;
 608                 }
 609                 free_page((unsigned long)kbuf);
 610         }
 611
 612         *ppos = p;
 613         return virtr + wrote;
 614 }
 615
 616 #ifdef CONFIG_DEVPORT
 617 static ssize_t read_port(struct file * file, char __user * buf,
 618                          size_t count, loff_t *ppos)
 619 {
 620         unsigned long i = *ppos;
 621         char __user *tmp = buf;
 622
 623         if (!access_ok(VERIFY_WRITE, buf, count))
 624                 return -EFAULT;
 625         while (count-- > 0 && i < 65536) {
 626                 if (__put_user(inb(i),tmp) < 0)
 627                         return -EFAULT;
 628                 i++;
 629                 tmp++;
 630         }
 631         *ppos = i;
 632         return tmp-buf;
 633 }
 634
 635 static ssize_t write_port(struct file * file, const char __user * buf,
 636                           size_t count, loff_t *ppos)
 637 {
 638         unsigned long i = *ppos;
 639         const char __user * tmp = buf;
 640
 641         if (!access_ok(VERIFY_READ,buf,count))
 642                 return -EFAULT;
 643         while (count-- > 0 && i < 65536) {
 644                 char c;
 645                 if (__get_user(c, tmp)) {
 646                         if (tmp > buf)
 647                                 break;
 648                         return -EFAULT;
 649                 }
 650                 outb(c,i);
 651                 i++;
 652                 tmp++;
 653         }
 654         *ppos = i;
 655         return tmp-buf;
 656 }
 657 #endif
 658
 659 static ssize_t read_null(struct file * file, char __user * buf,
 660                          size_t count, loff_t *ppos)
 661 {
 662         return 0;
 663 }
 664
 665 static ssize_t write_null(struct file * file, const char __user * buf,
 666                           size_t count, loff_t *ppos)
 667 {
 668         return count;
 669 }
 670
 671 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 672                         struct splice_desc *sd)
 673 {
 674         return sd->len;
 675 }
 676
 677 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 678                                  loff_t *ppos, size_t len, unsigned int flags)
 679 {
 680         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 681 }
 682
 683 static ssize_t read_zero(struct file * file, char __user * buf,
 684                          size_t count, loff_t *ppos)
 685 {
 686         size_t written;
 687
 688         if (!count)
 689                 return 0;
 690
 691         if (!access_ok(VERIFY_WRITE, buf, count))
 692                 return -EFAULT;
 693
 694         written = 0;
 695         while (count) {
 696                 unsigned long unwritten;
 697                 size_t chunk = count;
 698
 699                 if (chunk > PAGE_SIZE)
 700                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 701                 unwritten = clear_user(buf, chunk);
 702                 written += chunk - unwritten;
 703                 if (unwritten)
 704                         break;
 705                 buf += chunk;
 706                 count -= chunk;
 707                 cond_resched();
 708         }
 709         return written ? written : -EFAULT;
 710 }
 711
 712 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 713 {
 714 #ifndef CONFIG_MMU
 715         return -ENOSYS;
 716 #endif
 717         if (vma->vm_flags & VM_SHARED)
 718                 return shmem_zero_setup(vma);
 719         return 0;
 720 }
 721
 722 static ssize_t write_full(struct file * file, const char __user * buf,
 723                           size_t count, loff_t *ppos)
 724 {
 725         return -ENOSPC;
 726 }
 727
 728 /*
 729  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 730  * can fopen() both devices with "a" now.  This was previously impossible.
 731  * -- SRB.
 732  */
 733
 734 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 735 {
 736         return file->f_pos = 0;
 737 }
 738
 739 /*
 740  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 741  * check against negative addresses: they are ok. The return value is weird,
 742  * though, in that case (0).
 743  *
 744  * also note that seeking relative to the "end of file" isn't supported:
 745  * it has no meaning, so it returns -EINVAL.
 746  */
 747 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 748 {
 749         loff_t ret;
 750
 751         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 752         switch (orig) {
 753                 case 0:
 754                         file->f_pos = offset;
 755                         ret = file->f_pos;
 756                         force_successful_syscall_return();
 757                         break;
 758                 case 1:
 759                         file->f_pos += offset;
 760                         ret = file->f_pos;
 761                         force_successful_syscall_return();
 762                         break;
 763                 default:
 764                         ret = -EINVAL;
 765         }
 766         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 767         return ret;
 768 }
 769
 770 static int open_port(struct inode * inode, struct file * filp)
 771 {
 772         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 773 }
 774
 775 #define zero_lseek      null_lseek
 776 #define full_lseek      null_lseek
 777 #define write_zero      write_null
 778 #define read_full       read_zero
 779 #define open_mem        open_port
 780 #define open_kmem       open_mem
 781 #define open_oldmem     open_mem
 782
 783 static const struct file_operations mem_fops = {
 784         .llseek         = memory_lseek,
 785         .read           = read_mem,
 786         .write          = write_mem,
 787         .mmap           = mmap_mem,
 788         .open           = open_mem,
 789         .get_unmapped_area = get_unmapped_area_mem,
 790 };
 791
 792 static const struct file_operations kmem_fops = {
 793         .llseek         = memory_lseek,
 794         .read           = read_kmem,
 795         .write          = write_kmem,
 796         .mmap           = mmap_kmem,
 797         .open           = open_kmem,
 798         .get_unmapped_area = get_unmapped_area_mem,
 799 };
 800
 801 static const struct file_operations null_fops = {
 802         .llseek         = null_lseek,
 803         .read           = read_null,
 804         .write          = write_null,
 805         .splice_write   = splice_write_null,
 806 };
 807
 808 #ifdef CONFIG_DEVPORT
 809 static const struct file_operations port_fops = {
 810         .llseek         = memory_lseek,
 811         .read           = read_port,
 812         .write          = write_port,
 813         .open           = open_port,
 814 };
 815 #endif
 816
 817 static const struct file_operations zero_fops = {
 818         .llseek         = zero_lseek,
 819         .read           = read_zero,
 820         .write          = write_zero,
 821         .mmap           = mmap_zero,
 822 };
 823
 824 /*
 825  * capabilities for /dev/zero
 826  * - permits private mappings, "copies" are taken of the source of zeros
 827  */
 828 static struct backing_dev_info zero_bdi = {
 829         .capabilities   = BDI_CAP_MAP_COPY,
 830 };
 831
 832 static const struct file_operations full_fops = {
 833         .llseek         = full_lseek,
 834         .read           = read_full,
 835         .write          = write_full,
 836 };
 837
 838 #ifdef CONFIG_CRASH_DUMP
 839 static const struct file_operations oldmem_fops = {
 840         .read   = read_oldmem,
 841         .open   = open_oldmem,
 842 };
 843 #endif
 844
 845 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 846                           size_t count, loff_t *ppos)
 847 {
 848         char *tmp;
 849         ssize_t ret;
 850
 851         tmp = kmalloc(count + 1, GFP_KERNEL);
 852         if (tmp == NULL)
 853                 return -ENOMEM;
 854         ret = -EFAULT;
 855         if (!copy_from_user(tmp, buf, count)) {
 856                 tmp[count] = 0;
 857                 ret = printk("%s", tmp);
 858                 if (ret > count)
 859                         /* printk can add a prefix */
 860                         ret = count;
 861         }
 862         kfree(tmp);
 863         return ret;
 864 }
 865
 866 static const struct file_operations kmsg_fops = {
 867         .write =        kmsg_write,
 868 };
 869
 870 static int memory_open(struct inode * inode, struct file * filp)
 871 {
 872         switch (iminor(inode)) {
 873                 case 1:
 874                         filp->f_op = &mem_fops;
 875                         filp->f_mapping->backing_dev_info =
 876                                 &directly_mappable_cdev_bdi;
 877                         break;
 878                 case 2:
 879                         filp->f_op = &kmem_fops;
 880                         filp->f_mapping->backing_dev_info =
 881                                 &directly_mappable_cdev_bdi;
 882                         break;
 883                 case 3:
 884                         filp->f_op = &null_fops;
 885                         break;
 886 #ifdef CONFIG_DEVPORT
 887                 case 4:
 888                         filp->f_op = &port_fops;
 889                         break;
 890 #endif
 891                 case 5:
 892                         filp->f_mapping->backing_dev_info = &zero_bdi;
 893                         filp->f_op = &zero_fops;
 894                         break;
 895                 case 7:
 896                         filp->f_op = &full_fops;
 897                         break;
 898                 case 8:
 899                         filp->f_op = &random_fops;
 900                         break;
 901                 case 9:
 902                         filp->f_op = &urandom_fops;
 903                         break;
 904                 case 11:
 905                         filp->f_op = &kmsg_fops;
 906                         break;
 907 #ifdef CONFIG_CRASH_DUMP
 908                 case 12:
 909                         filp->f_op = &oldmem_fops;
 910                         break;
 911 #endif
 912                 default:
 913                         return -ENXIO;
 914         }
 915         if (filp->f_op && filp->f_op->open)
 916                 return filp->f_op->open(inode,filp);
 917         return 0;
 918 }
 919
 920 static const struct file_operations memory_fops = {
 921         .open           = memory_open,  /* just a selector for the real open */
 922 };
 923
 924 static const struct {
 925         unsigned int            minor;
 926         char                    *name;
 927         umode_t                 mode;
 928         const struct file_operations    *fops;
 929 } devlist[] = { /* list of minor devices */
 930         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 931         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 932         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 933 #ifdef CONFIG_DEVPORT
 934         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 935 #endif
 936         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 937         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 938         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 939         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 940         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 941 #ifdef CONFIG_CRASH_DUMP
 942         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 943 #endif
 944 };
 945
 946 static struct class *mem_class;
 947
 948 static int __init chr_dev_init(void)
 949 {
 950         int i;
 951         int err;
 952
 953         err = bdi_init(&zero_bdi);
 954         if (err)
 955                 return err;
 956
 957         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 958                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 959
 960         mem_class = class_create(THIS_MODULE, "mem");
 961         for (i = 0; i < ARRAY_SIZE(devlist); i++)
 962                 device_create(mem_class, NULL,
 963                               MKDEV(MEM_MAJOR, devlist[i].minor),
 964                               devlist[i].name);
 965
 966         return 0;
 967 }
 968
 969 fs_initcall(chr_dev_init);