SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 /*
 138  * This funcion reads the *physical* memory. The f_pos points directly to the
 139  * memory location.
 140  */
 141 static ssize_t read_mem(struct file * file, char __user * buf,
 142                         size_t count, loff_t *ppos)
 143 {
 144         unsigned long p = *ppos;
 145         ssize_t read, sz;
 146         char *ptr;
 147
 148         if (!valid_phys_addr_range(p, count))
 149                 return -EFAULT;
 150         read = 0;
 151 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 152         /* we don't have page 0 mapped on sparc and m68k.. */
 153         if (p < PAGE_SIZE) {
 154                 sz = PAGE_SIZE - p;
 155                 if (sz > count)
 156                         sz = count;
 157                 if (sz > 0) {
 158                         if (clear_user(buf, sz))
 159                                 return -EFAULT;
 160                         buf += sz;
 161                         p += sz;
 162                         count -= sz;
 163                         read += sz;
 164                 }
 165         }
 166 #endif
 167
 168         while (count > 0) {
 169                 /*
 170                  * Handle first page in case it's not aligned
 171                  */
 172                 if (-p & (PAGE_SIZE - 1))
 173                         sz = -p & (PAGE_SIZE - 1);
 174                 else
 175                         sz = PAGE_SIZE;
 176
 177                 sz = min_t(unsigned long, sz, count);
 178
 179                 /*
 180                  * On ia64 if a page has been mapped somewhere as
 181                  * uncached, then it must also be accessed uncached
 182                  * by the kernel or data corruption may occur
 183                  */
 184                 ptr = xlate_dev_mem_ptr(p);
 185
 186                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 187                         return -EPERM;
 188                 if (copy_to_user(buf, ptr, sz))
 189                         return -EFAULT;
 190                 buf += sz;
 191                 p += sz;
 192                 count -= sz;
 193                 read += sz;
 194         }
 195
 196         *ppos += read;
 197         return read;
 198 }
 199
 200 static ssize_t write_mem(struct file * file, const char __user * buf,
 201                          size_t count, loff_t *ppos)
 202 {
 203         unsigned long p = *ppos;
 204         ssize_t written, sz;
 205         unsigned long copied;
 206         void *ptr;
 207
 208         if (!valid_phys_addr_range(p, count))
 209                 return -EFAULT;
 210
 211         written = 0;
 212
 213 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 214         /* we don't have page 0 mapped on sparc and m68k.. */
 215         if (p < PAGE_SIZE) {
 216                 unsigned long sz = PAGE_SIZE - p;
 217                 if (sz > count)
 218                         sz = count;
 219                 /* Hmm. Do something? */
 220                 buf += sz;
 221                 p += sz;
 222                 count -= sz;
 223                 written += sz;
 224         }
 225 #endif
 226
 227         while (count > 0) {
 228                 /*
 229                  * Handle first page in case it's not aligned
 230                  */
 231                 if (-p & (PAGE_SIZE - 1))
 232                         sz = -p & (PAGE_SIZE - 1);
 233                 else
 234                         sz = PAGE_SIZE;
 235
 236                 sz = min_t(unsigned long, sz, count);
 237
 238                 /*
 239                  * On ia64 if a page has been mapped somewhere as
 240                  * uncached, then it must also be accessed uncached
 241                  * by the kernel or data corruption may occur
 242                  */
 243                 ptr = xlate_dev_mem_ptr(p);
 244
 245                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 246                         return -EPERM;
 247                 copied = copy_from_user(ptr, buf, sz);
 248                 if (copied) {
 249                         written += sz - copied;
 250                         if (written)
 251                                 break;
 252                         return -EFAULT;
 253                 }
 254                 buf += sz;
 255                 p += sz;
 256                 count -= sz;
 257                 written += sz;
 258         }
 259
 260         *ppos += written;
 261         return written;
 262 }
 263
 264 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 265 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 266                                      unsigned long size, pgprot_t vma_prot)
 267 {
 268 #ifdef pgprot_noncached
 269         unsigned long offset = pfn << PAGE_SHIFT;
 270
 271         if (uncached_access(file, offset))
 272                 return pgprot_noncached(vma_prot);
 273 #endif
 274         return vma_prot;
 275 }
 276 #endif
 277
 278 #ifndef CONFIG_MMU
 279 static unsigned long get_unmapped_area_mem(struct file *file,
 280                                            unsigned long addr,
 281                                            unsigned long len,
 282                                            unsigned long pgoff,
 283                                            unsigned long flags)
 284 {
 285         if (!valid_mmap_phys_addr_range(pgoff, len))
 286                 return (unsigned long) -EINVAL;
 287         return pgoff << PAGE_SHIFT;
 288 }
 289
 290 /* can't do an in-place private mapping if there's no MMU */
 291 static inline int private_mapping_ok(struct vm_area_struct *vma)
 292 {
 293         return vma->vm_flags & VM_MAYSHARE;
 294 }
 295 #else
 296 #define get_unmapped_area_mem   NULL
 297
 298 static inline int private_mapping_ok(struct vm_area_struct *vma)
 299 {
 300         return 1;
 301 }
 302 #endif
 303
 304 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 305 {
 306         size_t size = vma->vm_end - vma->vm_start;
 307
 308         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 309                 return -EINVAL;
 310
 311         if (!private_mapping_ok(vma))
 312                 return -ENOSYS;
 313
 314         if (!range_is_allowed(vma->vm_pgoff, size))
 315                 return -EPERM;
 316
 317         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 318                                                  size,
 319                                                  vma->vm_page_prot);
 320
 321         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 322         if (remap_pfn_range(vma,
 323                             vma->vm_start,
 324                             vma->vm_pgoff,
 325                             size,
 326                             vma->vm_page_prot))
 327                 return -EAGAIN;
 328         return 0;
 329 }
 330
 331 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 332 {
 333         unsigned long pfn;
 334
 335         /* Turn a kernel-virtual address into a physical page frame */
 336         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 337
 338         /*
 339          * RED-PEN: on some architectures there is more mapped memory
 340          * than available in mem_map which pfn_valid checks
 341          * for. Perhaps should add a new macro here.
 342          *
 343          * RED-PEN: vmalloc is not supported right now.
 344          */
 345         if (!pfn_valid(pfn))
 346                 return -EIO;
 347
 348         vma->vm_pgoff = pfn;
 349         return mmap_mem(file, vma);
 350 }
 351
 352 #ifdef CONFIG_CRASH_DUMP
 353 /*
 354  * Read memory corresponding to the old kernel.
 355  */
 356 static ssize_t read_oldmem(struct file *file, char __user *buf,
 357                                 size_t count, loff_t *ppos)
 358 {
 359         unsigned long pfn, offset;
 360         size_t read = 0, csize;
 361         int rc = 0;
 362
 363         while (count) {
 364                 pfn = *ppos / PAGE_SIZE;
 365                 if (pfn > saved_max_pfn)
 366                         return read;
 367
 368                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 369                 if (count > PAGE_SIZE - offset)
 370                         csize = PAGE_SIZE - offset;
 371                 else
 372                         csize = count;
 373
 374                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 375                 if (rc < 0)
 376                         return rc;
 377                 buf += csize;
 378                 *ppos += csize;
 379                 read += csize;
 380                 count -= csize;
 381         }
 382         return read;
 383 }
 384 #endif
 385
 386 extern long vread(char *buf, char *addr, unsigned long count);
 387 extern long vwrite(char *buf, char *addr, unsigned long count);
 388
 389 /*
 390  * This function reads the *virtual* memory as seen by the kernel.
 391  */
 392 static ssize_t read_kmem(struct file *file, char __user *buf,
 393                          size_t count, loff_t *ppos)
 394 {
 395         unsigned long p = *ppos;
 396         ssize_t low_count, read, sz;
 397         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 398
 399         read = 0;
 400         if (p < (unsigned long) high_memory) {
 401                 low_count = count;
 402                 if (count > (unsigned long) high_memory - p)
 403                         low_count = (unsigned long) high_memory - p;
 404
 405 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 406                 /* we don't have page 0 mapped on sparc and m68k.. */
 407                 if (p < PAGE_SIZE && low_count > 0) {
 408                         size_t tmp = PAGE_SIZE - p;
 409                         if (tmp > low_count) tmp = low_count;
 410                         if (clear_user(buf, tmp))
 411                                 return -EFAULT;
 412                         buf += tmp;
 413                         p += tmp;
 414                         read += tmp;
 415                         low_count -= tmp;
 416                         count -= tmp;
 417                 }
 418 #endif
 419                 while (low_count > 0) {
 420                         /*
 421                          * Handle first page in case it's not aligned
 422                          */
 423                         if (-p & (PAGE_SIZE - 1))
 424                                 sz = -p & (PAGE_SIZE - 1);
 425                         else
 426                                 sz = PAGE_SIZE;
 427
 428                         sz = min_t(unsigned long, sz, low_count);
 429
 430                         /*
 431                          * On ia64 if a page has been mapped somewhere as
 432                          * uncached, then it must also be accessed uncached
 433                          * by the kernel or data corruption may occur
 434                          */
 435                         kbuf = xlate_dev_kmem_ptr((char *)p);
 436
 437                         if (copy_to_user(buf, kbuf, sz))
 438                                 return -EFAULT;
 439                         buf += sz;
 440                         p += sz;
 441                         read += sz;
 442                         low_count -= sz;
 443                         count -= sz;
 444                 }
 445         }
 446
 447         if (count > 0) {
 448                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 449                 if (!kbuf)
 450                         return -ENOMEM;
 451                 while (count > 0) {
 452                         int len = count;
 453
 454                         if (len > PAGE_SIZE)
 455                                 len = PAGE_SIZE;
 456                         len = vread(kbuf, (char *)p, len);
 457                         if (!len)
 458                                 break;
 459                         if (copy_to_user(buf, kbuf, len)) {
 460                                 free_page((unsigned long)kbuf);
 461                                 return -EFAULT;
 462                         }
 463                         count -= len;
 464                         buf += len;
 465                         read += len;
 466                         p += len;
 467                 }
 468                 free_page((unsigned long)kbuf);
 469         }
 470         *ppos = p;
 471         return read;
 472 }
 473
 474
 475 static inline ssize_t
 476 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 477               size_t count, loff_t *ppos)
 478 {
 479         ssize_t written, sz;
 480         unsigned long copied;
 481
 482         written = 0;
 483 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 484         /* we don't have page 0 mapped on sparc and m68k.. */
 485         if (realp < PAGE_SIZE) {
 486                 unsigned long sz = PAGE_SIZE - realp;
 487                 if (sz > count)
 488                         sz = count;
 489                 /* Hmm. Do something? */
 490                 buf += sz;
 491                 p += sz;
 492                 realp += sz;
 493                 count -= sz;
 494                 written += sz;
 495         }
 496 #endif
 497
 498         while (count > 0) {
 499                 char *ptr;
 500                 /*
 501                  * Handle first page in case it's not aligned
 502                  */
 503                 if (-realp & (PAGE_SIZE - 1))
 504                         sz = -realp & (PAGE_SIZE - 1);
 505                 else
 506                         sz = PAGE_SIZE;
 507
 508                 sz = min_t(unsigned long, sz, count);
 509
 510                 /*
 511                  * On ia64 if a page has been mapped somewhere as
 512                  * uncached, then it must also be accessed uncached
 513                  * by the kernel or data corruption may occur
 514                  */
 515                 ptr = xlate_dev_kmem_ptr(p);
 516
 517                 copied = copy_from_user(ptr, buf, sz);
 518                 if (copied) {
 519                         written += sz - copied;
 520                         if (written)
 521                                 break;
 522                         return -EFAULT;
 523                 }
 524                 buf += sz;
 525                 p += sz;
 526                 realp += sz;
 527                 count -= sz;
 528                 written += sz;
 529         }
 530
 531         *ppos += written;
 532         return written;
 533 }
 534
 535
 536 /*
 537  * This function writes to the *virtual* memory as seen by the kernel.
 538  */
 539 static ssize_t write_kmem(struct file * file, const char __user * buf,
 540                           size_t count, loff_t *ppos)
 541 {
 542         unsigned long p = *ppos;
 543         ssize_t wrote = 0;
 544         ssize_t virtr = 0;
 545         ssize_t written;
 546         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 547
 548         if (p < (unsigned long) high_memory) {
 549
 550                 wrote = count;
 551                 if (count > (unsigned long) high_memory - p)
 552                         wrote = (unsigned long) high_memory - p;
 553
 554                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 555                 if (written != wrote)
 556                         return written;
 557                 wrote = written;
 558                 p += wrote;
 559                 buf += wrote;
 560                 count -= wrote;
 561         }
 562
 563         if (count > 0) {
 564                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 565                 if (!kbuf)
 566                         return wrote ? wrote : -ENOMEM;
 567                 while (count > 0) {
 568                         int len = count;
 569
 570                         if (len > PAGE_SIZE)
 571                                 len = PAGE_SIZE;
 572                         if (len) {
 573                                 written = copy_from_user(kbuf, buf, len);
 574                                 if (written) {
 575                                         if (wrote + virtr)
 576                                                 break;
 577                                         free_page((unsigned long)kbuf);
 578                                         return -EFAULT;
 579                                 }
 580                         }
 581                         len = vwrite(kbuf, (char *)p, len);
 582                         count -= len;
 583                         buf += len;
 584                         virtr += len;
 585                         p += len;
 586                 }
 587                 free_page((unsigned long)kbuf);
 588         }
 589
 590         *ppos = p;
 591         return virtr + wrote;
 592 }
 593
 594 #ifdef CONFIG_DEVPORT
 595 static ssize_t read_port(struct file * file, char __user * buf,
 596                          size_t count, loff_t *ppos)
 597 {
 598         unsigned long i = *ppos;
 599         char __user *tmp = buf;
 600
 601         if (!access_ok(VERIFY_WRITE, buf, count))
 602                 return -EFAULT;
 603         while (count-- > 0 && i < 65536) {
 604                 if (__put_user(inb(i),tmp) < 0)
 605                         return -EFAULT;
 606                 i++;
 607                 tmp++;
 608         }
 609         *ppos = i;
 610         return tmp-buf;
 611 }
 612
 613 static ssize_t write_port(struct file * file, const char __user * buf,
 614                           size_t count, loff_t *ppos)
 615 {
 616         unsigned long i = *ppos;
 617         const char __user * tmp = buf;
 618
 619         if (!access_ok(VERIFY_READ,buf,count))
 620                 return -EFAULT;
 621         while (count-- > 0 && i < 65536) {
 622                 char c;
 623                 if (__get_user(c, tmp)) {
 624                         if (tmp > buf)
 625                                 break;
 626                         return -EFAULT;
 627                 }
 628                 outb(c,i);
 629                 i++;
 630                 tmp++;
 631         }
 632         *ppos = i;
 633         return tmp-buf;
 634 }
 635 #endif
 636
 637 static ssize_t read_null(struct file * file, char __user * buf,
 638                          size_t count, loff_t *ppos)
 639 {
 640         return 0;
 641 }
 642
 643 static ssize_t write_null(struct file * file, const char __user * buf,
 644                           size_t count, loff_t *ppos)
 645 {
 646         return count;
 647 }
 648
 649 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 650                         struct splice_desc *sd)
 651 {
 652         return sd->len;
 653 }
 654
 655 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 656                                  loff_t *ppos, size_t len, unsigned int flags)
 657 {
 658         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 659 }
 660
 661 static ssize_t read_zero(struct file * file, char __user * buf,
 662                          size_t count, loff_t *ppos)
 663 {
 664         size_t written;
 665
 666         if (!count)
 667                 return 0;
 668
 669         if (!access_ok(VERIFY_WRITE, buf, count))
 670                 return -EFAULT;
 671
 672         written = 0;
 673         while (count) {
 674                 unsigned long unwritten;
 675                 size_t chunk = count;
 676
 677                 if (chunk > PAGE_SIZE)
 678                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 679                 unwritten = clear_user(buf, chunk);
 680                 written += chunk - unwritten;
 681                 if (unwritten)
 682                         break;
 683                 buf += chunk;
 684                 count -= chunk;
 685                 cond_resched();
 686         }
 687         return written ? written : -EFAULT;
 688 }
 689
 690 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 691 {
 692 #ifndef CONFIG_MMU
 693         return -ENOSYS;
 694 #endif
 695         if (vma->vm_flags & VM_SHARED)
 696                 return shmem_zero_setup(vma);
 697         return 0;
 698 }
 699
 700 static ssize_t write_full(struct file * file, const char __user * buf,
 701                           size_t count, loff_t *ppos)
 702 {
 703         return -ENOSPC;
 704 }
 705
 706 /*
 707  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 708  * can fopen() both devices with "a" now.  This was previously impossible.
 709  * -- SRB.
 710  */
 711
 712 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 713 {
 714         return file->f_pos = 0;
 715 }
 716
 717 /*
 718  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 719  * check against negative addresses: they are ok. The return value is weird,
 720  * though, in that case (0).
 721  *
 722  * also note that seeking relative to the "end of file" isn't supported:
 723  * it has no meaning, so it returns -EINVAL.
 724  */
 725 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 726 {
 727         loff_t ret;
 728
 729         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 730         switch (orig) {
 731                 case 0:
 732                         file->f_pos = offset;
 733                         ret = file->f_pos;
 734                         force_successful_syscall_return();
 735                         break;
 736                 case 1:
 737                         file->f_pos += offset;
 738                         ret = file->f_pos;
 739                         force_successful_syscall_return();
 740                         break;
 741                 default:
 742                         ret = -EINVAL;
 743         }
 744         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 745         return ret;
 746 }
 747
 748 static int open_port(struct inode * inode, struct file * filp)
 749 {
 750         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 751 }
 752
 753 #define zero_lseek      null_lseek
 754 #define full_lseek      null_lseek
 755 #define write_zero      write_null
 756 #define read_full       read_zero
 757 #define open_mem        open_port
 758 #define open_kmem       open_mem
 759 #define open_oldmem     open_mem
 760
 761 static const struct file_operations mem_fops = {
 762         .llseek         = memory_lseek,
 763         .read           = read_mem,
 764         .write          = write_mem,
 765         .mmap           = mmap_mem,
 766         .open           = open_mem,
 767         .get_unmapped_area = get_unmapped_area_mem,
 768 };
 769
 770 static const struct file_operations kmem_fops = {
 771         .llseek         = memory_lseek,
 772         .read           = read_kmem,
 773         .write          = write_kmem,
 774         .mmap           = mmap_kmem,
 775         .open           = open_kmem,
 776         .get_unmapped_area = get_unmapped_area_mem,
 777 };
 778
 779 static const struct file_operations null_fops = {
 780         .llseek         = null_lseek,
 781         .read           = read_null,
 782         .write          = write_null,
 783         .splice_write   = splice_write_null,
 784 };
 785
 786 #ifdef CONFIG_DEVPORT
 787 static const struct file_operations port_fops = {
 788         .llseek         = memory_lseek,
 789         .read           = read_port,
 790         .write          = write_port,
 791         .open           = open_port,
 792 };
 793 #endif
 794
 795 static const struct file_operations zero_fops = {
 796         .llseek         = zero_lseek,
 797         .read           = read_zero,
 798         .write          = write_zero,
 799         .mmap           = mmap_zero,
 800 };
 801
 802 /*
 803  * capabilities for /dev/zero
 804  * - permits private mappings, "copies" are taken of the source of zeros
 805  */
 806 static struct backing_dev_info zero_bdi = {
 807         .capabilities   = BDI_CAP_MAP_COPY,
 808 };
 809
 810 static const struct file_operations full_fops = {
 811         .llseek         = full_lseek,
 812         .read           = read_full,
 813         .write          = write_full,
 814 };
 815
 816 #ifdef CONFIG_CRASH_DUMP
 817 static const struct file_operations oldmem_fops = {
 818         .read   = read_oldmem,
 819         .open   = open_oldmem,
 820 };
 821 #endif
 822
 823 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 824                           size_t count, loff_t *ppos)
 825 {
 826         char *tmp;
 827         ssize_t ret;
 828
 829         tmp = kmalloc(count + 1, GFP_KERNEL);
 830         if (tmp == NULL)
 831                 return -ENOMEM;
 832         ret = -EFAULT;
 833         if (!copy_from_user(tmp, buf, count)) {
 834                 tmp[count] = 0;
 835                 ret = printk("%s", tmp);
 836                 if (ret > count)
 837                         /* printk can add a prefix */
 838                         ret = count;
 839         }
 840         kfree(tmp);
 841         return ret;
 842 }
 843
 844 static const struct file_operations kmsg_fops = {
 845         .write =        kmsg_write,
 846 };
 847
 848 static int memory_open(struct inode * inode, struct file * filp)
 849 {
 850         switch (iminor(inode)) {
 851                 case 1:
 852                         filp->f_op = &mem_fops;
 853                         filp->f_mapping->backing_dev_info =
 854                                 &directly_mappable_cdev_bdi;
 855                         break;
 856                 case 2:
 857                         filp->f_op = &kmem_fops;
 858                         filp->f_mapping->backing_dev_info =
 859                                 &directly_mappable_cdev_bdi;
 860                         break;
 861                 case 3:
 862                         filp->f_op = &null_fops;
 863                         break;
 864 #ifdef CONFIG_DEVPORT
 865                 case 4:
 866                         filp->f_op = &port_fops;
 867                         break;
 868 #endif
 869                 case 5:
 870                         filp->f_mapping->backing_dev_info = &zero_bdi;
 871                         filp->f_op = &zero_fops;
 872                         break;
 873                 case 7:
 874                         filp->f_op = &full_fops;
 875                         break;
 876                 case 8:
 877                         filp->f_op = &random_fops;
 878                         break;
 879                 case 9:
 880                         filp->f_op = &urandom_fops;
 881                         break;
 882                 case 11:
 883                         filp->f_op = &kmsg_fops;
 884                         break;
 885 #ifdef CONFIG_CRASH_DUMP
 886                 case 12:
 887                         filp->f_op = &oldmem_fops;
 888                         break;
 889 #endif
 890                 default:
 891                         return -ENXIO;
 892         }
 893         if (filp->f_op && filp->f_op->open)
 894                 return filp->f_op->open(inode,filp);
 895         return 0;
 896 }
 897
 898 static const struct file_operations memory_fops = {
 899         .open           = memory_open,  /* just a selector for the real open */
 900 };
 901
 902 static const struct {
 903         unsigned int            minor;
 904         char                    *name;
 905         umode_t                 mode;
 906         const struct file_operations    *fops;
 907 } devlist[] = { /* list of minor devices */
 908         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 909         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 910         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 911 #ifdef CONFIG_DEVPORT
 912         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 913 #endif
 914         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 915         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 916         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 917         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 918         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 919 #ifdef CONFIG_CRASH_DUMP
 920         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 921 #endif
 922 };
 923
 924 static struct class *mem_class;
 925
 926 static int __init chr_dev_init(void)
 927 {
 928         int i;
 929         int err;
 930
 931         err = bdi_init(&zero_bdi);
 932         if (err)
 933                 return err;
 934
 935         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 936                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 937
 938         mem_class = class_create(THIS_MODULE, "mem");
 939         for (i = 0; i < ARRAY_SIZE(devlist); i++)
 940                 device_create(mem_class, NULL,
 941                               MKDEV(MEM_MAJOR, devlist[i].minor),
 942                               devlist[i].name);
 943
 944         return 0;
 945 }
 946
 947 fs_initcall(chr_dev_init);