git.oblomov.eu Git - 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/pipe_fs_i.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__)
  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__)
  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 #else
  79         /*
  80          * Accessing memory above the top the kernel knows about or through a file pointer
  81          * that was marked O_SYNC will be done non-cached.
  82          */
  83         if (file->f_flags & O_SYNC)
  84                 return 1;
  85         return addr >= __pa(high_memory);
  86 #endif
  87 }
  88
  89 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  90 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  91 {
  92         if (addr + count > __pa(high_memory))
  93                 return 0;
  94
  95         return 1;
  96 }
  97
  98 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  99 {
 100         return 1;
 101 }
 102 #endif
 103
 104 /*
 105  * This funcion reads the *physical* memory. The f_pos points directly to the
 106  * memory location.
 107  */
 108 static ssize_t read_mem(struct file * file, char __user * buf,
 109                         size_t count, loff_t *ppos)
 110 {
 111         unsigned long p = *ppos;
 112         ssize_t read, sz;
 113         char *ptr;
 114
 115         if (!valid_phys_addr_range(p, count))
 116                 return -EFAULT;
 117         read = 0;
 118 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 119         /* we don't have page 0 mapped on sparc and m68k.. */
 120         if (p < PAGE_SIZE) {
 121                 sz = PAGE_SIZE - p;
 122                 if (sz > count)
 123                         sz = count;
 124                 if (sz > 0) {
 125                         if (clear_user(buf, sz))
 126                                 return -EFAULT;
 127                         buf += sz;
 128                         p += sz;
 129                         count -= sz;
 130                         read += sz;
 131                 }
 132         }
 133 #endif
 134
 135         while (count > 0) {
 136                 /*
 137                  * Handle first page in case it's not aligned
 138                  */
 139                 if (-p & (PAGE_SIZE - 1))
 140                         sz = -p & (PAGE_SIZE - 1);
 141                 else
 142                         sz = PAGE_SIZE;
 143
 144                 sz = min_t(unsigned long, sz, count);
 145
 146                 /*
 147                  * On ia64 if a page has been mapped somewhere as
 148                  * uncached, then it must also be accessed uncached
 149                  * by the kernel or data corruption may occur
 150                  */
 151                 ptr = xlate_dev_mem_ptr(p);
 152
 153                 if (copy_to_user(buf, ptr, sz))
 154                         return -EFAULT;
 155                 buf += sz;
 156                 p += sz;
 157                 count -= sz;
 158                 read += sz;
 159         }
 160
 161         *ppos += read;
 162         return read;
 163 }
 164
 165 static ssize_t write_mem(struct file * file, const char __user * buf,
 166                          size_t count, loff_t *ppos)
 167 {
 168         unsigned long p = *ppos;
 169         ssize_t written, sz;
 170         unsigned long copied;
 171         void *ptr;
 172
 173         if (!valid_phys_addr_range(p, count))
 174                 return -EFAULT;
 175
 176         written = 0;
 177
 178 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 179         /* we don't have page 0 mapped on sparc and m68k.. */
 180         if (p < PAGE_SIZE) {
 181                 unsigned long sz = PAGE_SIZE - p;
 182                 if (sz > count)
 183                         sz = count;
 184                 /* Hmm. Do something? */
 185                 buf += sz;
 186                 p += sz;
 187                 count -= sz;
 188                 written += sz;
 189         }
 190 #endif
 191
 192         while (count > 0) {
 193                 /*
 194                  * Handle first page in case it's not aligned
 195                  */
 196                 if (-p & (PAGE_SIZE - 1))
 197                         sz = -p & (PAGE_SIZE - 1);
 198                 else
 199                         sz = PAGE_SIZE;
 200
 201                 sz = min_t(unsigned long, sz, count);
 202
 203                 /*
 204                  * On ia64 if a page has been mapped somewhere as
 205                  * uncached, then it must also be accessed uncached
 206                  * by the kernel or data corruption may occur
 207                  */
 208                 ptr = xlate_dev_mem_ptr(p);
 209
 210                 copied = copy_from_user(ptr, buf, sz);
 211                 if (copied) {
 212                         written += sz - copied;
 213                         if (written)
 214                                 break;
 215                         return -EFAULT;
 216                 }
 217                 buf += sz;
 218                 p += sz;
 219                 count -= sz;
 220                 written += sz;
 221         }
 222
 223         *ppos += written;
 224         return written;
 225 }
 226
 227 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 228 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 229                                      unsigned long size, pgprot_t vma_prot)
 230 {
 231 #ifdef pgprot_noncached
 232         unsigned long offset = pfn << PAGE_SHIFT;
 233
 234         if (uncached_access(file, offset))
 235                 return pgprot_noncached(vma_prot);
 236 #endif
 237         return vma_prot;
 238 }
 239 #endif
 240
 241 #ifndef CONFIG_MMU
 242 static unsigned long get_unmapped_area_mem(struct file *file,
 243                                            unsigned long addr,
 244                                            unsigned long len,
 245                                            unsigned long pgoff,
 246                                            unsigned long flags)
 247 {
 248         if (!valid_mmap_phys_addr_range(pgoff, len))
 249                 return (unsigned long) -EINVAL;
 250         return pgoff << PAGE_SHIFT;
 251 }
 252
 253 /* can't do an in-place private mapping if there's no MMU */
 254 static inline int private_mapping_ok(struct vm_area_struct *vma)
 255 {
 256         return vma->vm_flags & VM_MAYSHARE;
 257 }
 258 #else
 259 #define get_unmapped_area_mem   NULL
 260
 261 static inline int private_mapping_ok(struct vm_area_struct *vma)
 262 {
 263         return 1;
 264 }
 265 #endif
 266
 267 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 268 {
 269         size_t size = vma->vm_end - vma->vm_start;
 270
 271         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 272                 return -EINVAL;
 273
 274         if (!private_mapping_ok(vma))
 275                 return -ENOSYS;
 276
 277         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 278                                                  size,
 279                                                  vma->vm_page_prot);
 280
 281         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 282         if (remap_pfn_range(vma,
 283                             vma->vm_start,
 284                             vma->vm_pgoff,
 285                             size,
 286                             vma->vm_page_prot))
 287                 return -EAGAIN;
 288         return 0;
 289 }
 290
 291 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 292 {
 293         unsigned long pfn;
 294
 295         /* Turn a kernel-virtual address into a physical page frame */
 296         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 297
 298         /*
 299          * RED-PEN: on some architectures there is more mapped memory
 300          * than available in mem_map which pfn_valid checks
 301          * for. Perhaps should add a new macro here.
 302          *
 303          * RED-PEN: vmalloc is not supported right now.
 304          */
 305         if (!pfn_valid(pfn))
 306                 return -EIO;
 307
 308         vma->vm_pgoff = pfn;
 309         return mmap_mem(file, vma);
 310 }
 311
 312 #ifdef CONFIG_CRASH_DUMP
 313 /*
 314  * Read memory corresponding to the old kernel.
 315  */
 316 static ssize_t read_oldmem(struct file *file, char __user *buf,
 317                                 size_t count, loff_t *ppos)
 318 {
 319         unsigned long pfn, offset;
 320         size_t read = 0, csize;
 321         int rc = 0;
 322
 323         while (count) {
 324                 pfn = *ppos / PAGE_SIZE;
 325                 if (pfn > saved_max_pfn)
 326                         return read;
 327
 328                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 329                 if (count > PAGE_SIZE - offset)
 330                         csize = PAGE_SIZE - offset;
 331                 else
 332                         csize = count;
 333
 334                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 335                 if (rc < 0)
 336                         return rc;
 337                 buf += csize;
 338                 *ppos += csize;
 339                 read += csize;
 340                 count -= csize;
 341         }
 342         return read;
 343 }
 344 #endif
 345
 346 extern long vread(char *buf, char *addr, unsigned long count);
 347 extern long vwrite(char *buf, char *addr, unsigned long count);
 348
 349 /*
 350  * This function reads the *virtual* memory as seen by the kernel.
 351  */
 352 static ssize_t read_kmem(struct file *file, char __user *buf,
 353                          size_t count, loff_t *ppos)
 354 {
 355         unsigned long p = *ppos;
 356         ssize_t low_count, read, sz;
 357         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 358
 359         read = 0;
 360         if (p < (unsigned long) high_memory) {
 361                 low_count = count;
 362                 if (count > (unsigned long) high_memory - p)
 363                         low_count = (unsigned long) high_memory - p;
 364
 365 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 366                 /* we don't have page 0 mapped on sparc and m68k.. */
 367                 if (p < PAGE_SIZE && low_count > 0) {
 368                         size_t tmp = PAGE_SIZE - p;
 369                         if (tmp > low_count) tmp = low_count;
 370                         if (clear_user(buf, tmp))
 371                                 return -EFAULT;
 372                         buf += tmp;
 373                         p += tmp;
 374                         read += tmp;
 375                         low_count -= tmp;
 376                         count -= tmp;
 377                 }
 378 #endif
 379                 while (low_count > 0) {
 380                         /*
 381                          * Handle first page in case it's not aligned
 382                          */
 383                         if (-p & (PAGE_SIZE - 1))
 384                                 sz = -p & (PAGE_SIZE - 1);
 385                         else
 386                                 sz = PAGE_SIZE;
 387
 388                         sz = min_t(unsigned long, sz, low_count);
 389
 390                         /*
 391                          * On ia64 if a page has been mapped somewhere as
 392                          * uncached, then it must also be accessed uncached
 393                          * by the kernel or data corruption may occur
 394                          */
 395                         kbuf = xlate_dev_kmem_ptr((char *)p);
 396
 397                         if (copy_to_user(buf, kbuf, sz))
 398                                 return -EFAULT;
 399                         buf += sz;
 400                         p += sz;
 401                         read += sz;
 402                         low_count -= sz;
 403                         count -= sz;
 404                 }
 405         }
 406
 407         if (count > 0) {
 408                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 409                 if (!kbuf)
 410                         return -ENOMEM;
 411                 while (count > 0) {
 412                         int len = count;
 413
 414                         if (len > PAGE_SIZE)
 415                                 len = PAGE_SIZE;
 416                         len = vread(kbuf, (char *)p, len);
 417                         if (!len)
 418                                 break;
 419                         if (copy_to_user(buf, kbuf, len)) {
 420                                 free_page((unsigned long)kbuf);
 421                                 return -EFAULT;
 422                         }
 423                         count -= len;
 424                         buf += len;
 425                         read += len;
 426                         p += len;
 427                 }
 428                 free_page((unsigned long)kbuf);
 429         }
 430         *ppos = p;
 431         return read;
 432 }
 433
 434
 435 static inline ssize_t
 436 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 437               size_t count, loff_t *ppos)
 438 {
 439         ssize_t written, sz;
 440         unsigned long copied;
 441
 442         written = 0;
 443 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 444         /* we don't have page 0 mapped on sparc and m68k.. */
 445         if (realp < PAGE_SIZE) {
 446                 unsigned long sz = PAGE_SIZE - realp;
 447                 if (sz > count)
 448                         sz = count;
 449                 /* Hmm. Do something? */
 450                 buf += sz;
 451                 p += sz;
 452                 realp += sz;
 453                 count -= sz;
 454                 written += sz;
 455         }
 456 #endif
 457
 458         while (count > 0) {
 459                 char *ptr;
 460                 /*
 461                  * Handle first page in case it's not aligned
 462                  */
 463                 if (-realp & (PAGE_SIZE - 1))
 464                         sz = -realp & (PAGE_SIZE - 1);
 465                 else
 466                         sz = PAGE_SIZE;
 467
 468                 sz = min_t(unsigned long, sz, count);
 469
 470                 /*
 471                  * On ia64 if a page has been mapped somewhere as
 472                  * uncached, then it must also be accessed uncached
 473                  * by the kernel or data corruption may occur
 474                  */
 475                 ptr = xlate_dev_kmem_ptr(p);
 476
 477                 copied = copy_from_user(ptr, buf, sz);
 478                 if (copied) {
 479                         written += sz - copied;
 480                         if (written)
 481                                 break;
 482                         return -EFAULT;
 483                 }
 484                 buf += sz;
 485                 p += sz;
 486                 realp += sz;
 487                 count -= sz;
 488                 written += sz;
 489         }
 490
 491         *ppos += written;
 492         return written;
 493 }
 494
 495
 496 /*
 497  * This function writes to the *virtual* memory as seen by the kernel.
 498  */
 499 static ssize_t write_kmem(struct file * file, const char __user * buf,
 500                           size_t count, loff_t *ppos)
 501 {
 502         unsigned long p = *ppos;
 503         ssize_t wrote = 0;
 504         ssize_t virtr = 0;
 505         ssize_t written;
 506         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 507
 508         if (p < (unsigned long) high_memory) {
 509
 510                 wrote = count;
 511                 if (count > (unsigned long) high_memory - p)
 512                         wrote = (unsigned long) high_memory - p;
 513
 514                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 515                 if (written != wrote)
 516                         return written;
 517                 wrote = written;
 518                 p += wrote;
 519                 buf += wrote;
 520                 count -= wrote;
 521         }
 522
 523         if (count > 0) {
 524                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 525                 if (!kbuf)
 526                         return wrote ? wrote : -ENOMEM;
 527                 while (count > 0) {
 528                         int len = count;
 529
 530                         if (len > PAGE_SIZE)
 531                                 len = PAGE_SIZE;
 532                         if (len) {
 533                                 written = copy_from_user(kbuf, buf, len);
 534                                 if (written) {
 535                                         if (wrote + virtr)
 536                                                 break;
 537                                         free_page((unsigned long)kbuf);
 538                                         return -EFAULT;
 539                                 }
 540                         }
 541                         len = vwrite(kbuf, (char *)p, len);
 542                         count -= len;
 543                         buf += len;
 544                         virtr += len;
 545                         p += len;
 546                 }
 547                 free_page((unsigned long)kbuf);
 548         }
 549
 550         *ppos = p;
 551         return virtr + wrote;
 552 }
 553
 554 #ifdef CONFIG_DEVPORT
 555 static ssize_t read_port(struct file * file, char __user * buf,
 556                          size_t count, loff_t *ppos)
 557 {
 558         unsigned long i = *ppos;
 559         char __user *tmp = buf;
 560
 561         if (!access_ok(VERIFY_WRITE, buf, count))
 562                 return -EFAULT;
 563         while (count-- > 0 && i < 65536) {
 564                 if (__put_user(inb(i),tmp) < 0)
 565                         return -EFAULT;
 566                 i++;
 567                 tmp++;
 568         }
 569         *ppos = i;
 570         return tmp-buf;
 571 }
 572
 573 static ssize_t write_port(struct file * file, const char __user * buf,
 574                           size_t count, loff_t *ppos)
 575 {
 576         unsigned long i = *ppos;
 577         const char __user * tmp = buf;
 578
 579         if (!access_ok(VERIFY_READ,buf,count))
 580                 return -EFAULT;
 581         while (count-- > 0 && i < 65536) {
 582                 char c;
 583                 if (__get_user(c, tmp)) {
 584                         if (tmp > buf)
 585                                 break;
 586                         return -EFAULT;
 587                 }
 588                 outb(c,i);
 589                 i++;
 590                 tmp++;
 591         }
 592         *ppos = i;
 593         return tmp-buf;
 594 }
 595 #endif
 596
 597 static ssize_t read_null(struct file * file, char __user * buf,
 598                          size_t count, loff_t *ppos)
 599 {
 600         return 0;
 601 }
 602
 603 static ssize_t write_null(struct file * file, const char __user * buf,
 604                           size_t count, loff_t *ppos)
 605 {
 606         return count;
 607 }
 608
 609 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 610                         struct splice_desc *sd)
 611 {
 612         return sd->len;
 613 }
 614
 615 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 616                                  loff_t *ppos, size_t len, unsigned int flags)
 617 {
 618         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 619 }
 620
 621 #ifdef CONFIG_MMU
 622 /*
 623  * For fun, we are using the MMU for this.
 624  */
 625 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
 626 {
 627         struct mm_struct *mm;
 628         struct vm_area_struct * vma;
 629         unsigned long addr=(unsigned long)buf;
 630
 631         mm = current->mm;
 632         /* Oops, this was forgotten before. -ben */
 633         down_read(&mm->mmap_sem);
 634
 635         /* For private mappings, just map in zero pages. */
 636         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 637                 unsigned long count;
 638
 639                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 640                         goto out_up;
 641                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
 642                         break;
 643                 count = vma->vm_end - addr;
 644                 if (count > size)
 645                         count = size;
 646
 647                 zap_page_range(vma, addr, count, NULL);
 648                 if (zeromap_page_range(vma, addr, count, PAGE_COPY))
 649                         break;
 650
 651                 size -= count;
 652                 buf += count;
 653                 addr += count;
 654                 if (size == 0)
 655                         goto out_up;
 656         }
 657
 658         up_read(&mm->mmap_sem);
 659
 660         /* The shared case is hard. Let's do the conventional zeroing. */
 661         do {
 662                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 663                 if (unwritten)
 664                         return size + unwritten - PAGE_SIZE;
 665                 cond_resched();
 666                 buf += PAGE_SIZE;
 667                 size -= PAGE_SIZE;
 668         } while (size);
 669
 670         return size;
 671 out_up:
 672         up_read(&mm->mmap_sem);
 673         return size;
 674 }
 675
 676 static ssize_t read_zero(struct file * file, char __user * buf,
 677                          size_t count, loff_t *ppos)
 678 {
 679         unsigned long left, unwritten, written = 0;
 680
 681         if (!count)
 682                 return 0;
 683
 684         if (!access_ok(VERIFY_WRITE, buf, count))
 685                 return -EFAULT;
 686
 687         left = count;
 688
 689         /* do we want to be clever? Arbitrary cut-off */
 690         if (count >= PAGE_SIZE*4) {
 691                 unsigned long partial;
 692
 693                 /* How much left of the page? */
 694                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 695                 unwritten = clear_user(buf, partial);
 696                 written = partial - unwritten;
 697                 if (unwritten)
 698                         goto out;
 699                 left -= partial;
 700                 buf += partial;
 701                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 702                 written += (left & PAGE_MASK) - unwritten;
 703                 if (unwritten)
 704                         goto out;
 705                 buf += left & PAGE_MASK;
 706                 left &= ~PAGE_MASK;
 707         }
 708         unwritten = clear_user(buf, left);
 709         written += left - unwritten;
 710 out:
 711         return written ? written : -EFAULT;
 712 }
 713
 714 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 715 {
 716         int err;
 717
 718         if (vma->vm_flags & VM_SHARED)
 719                 return shmem_zero_setup(vma);
 720         err = zeromap_page_range(vma, vma->vm_start,
 721                         vma->vm_end - vma->vm_start, vma->vm_page_prot);
 722         BUG_ON(err == -EEXIST);
 723         return err;
 724 }
 725 #else /* CONFIG_MMU */
 726 static ssize_t read_zero(struct file * file, char * buf,
 727                          size_t count, loff_t *ppos)
 728 {
 729         size_t todo = count;
 730
 731         while (todo) {
 732                 size_t chunk = todo;
 733
 734                 if (chunk > 4096)
 735                         chunk = 4096;   /* Just for latency reasons */
 736                 if (clear_user(buf, chunk))
 737                         return -EFAULT;
 738                 buf += chunk;
 739                 todo -= chunk;
 740                 cond_resched();
 741         }
 742         return count;
 743 }
 744
 745 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 746 {
 747         return -ENOSYS;
 748 }
 749 #endif /* CONFIG_MMU */
 750
 751 static ssize_t write_full(struct file * file, const char __user * buf,
 752                           size_t count, loff_t *ppos)
 753 {
 754         return -ENOSPC;
 755 }
 756
 757 /*
 758  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 759  * can fopen() both devices with "a" now.  This was previously impossible.
 760  * -- SRB.
 761  */
 762
 763 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 764 {
 765         return file->f_pos = 0;
 766 }
 767
 768 /*
 769  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 770  * check against negative addresses: they are ok. The return value is weird,
 771  * though, in that case (0).
 772  *
 773  * also note that seeking relative to the "end of file" isn't supported:
 774  * it has no meaning, so it returns -EINVAL.
 775  */
 776 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 777 {
 778         loff_t ret;
 779
 780         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 781         switch (orig) {
 782                 case 0:
 783                         file->f_pos = offset;
 784                         ret = file->f_pos;
 785                         force_successful_syscall_return();
 786                         break;
 787                 case 1:
 788                         file->f_pos += offset;
 789                         ret = file->f_pos;
 790                         force_successful_syscall_return();
 791                         break;
 792                 default:
 793                         ret = -EINVAL;
 794         }
 795         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 796         return ret;
 797 }
 798
 799 static int open_port(struct inode * inode, struct file * filp)
 800 {
 801         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 802 }
 803
 804 #define zero_lseek      null_lseek
 805 #define full_lseek      null_lseek
 806 #define write_zero      write_null
 807 #define read_full       read_zero
 808 #define open_mem        open_port
 809 #define open_kmem       open_mem
 810 #define open_oldmem     open_mem
 811
 812 static const struct file_operations mem_fops = {
 813         .llseek         = memory_lseek,
 814         .read           = read_mem,
 815         .write          = write_mem,
 816         .mmap           = mmap_mem,
 817         .open           = open_mem,
 818         .get_unmapped_area = get_unmapped_area_mem,
 819 };
 820
 821 static const struct file_operations kmem_fops = {
 822         .llseek         = memory_lseek,
 823         .read           = read_kmem,
 824         .write          = write_kmem,
 825         .mmap           = mmap_kmem,
 826         .open           = open_kmem,
 827         .get_unmapped_area = get_unmapped_area_mem,
 828 };
 829
 830 static const struct file_operations null_fops = {
 831         .llseek         = null_lseek,
 832         .read           = read_null,
 833         .write          = write_null,
 834         .splice_write   = splice_write_null,
 835 };
 836
 837 #ifdef CONFIG_DEVPORT
 838 static const struct file_operations port_fops = {
 839         .llseek         = memory_lseek,
 840         .read           = read_port,
 841         .write          = write_port,
 842         .open           = open_port,
 843 };
 844 #endif
 845
 846 static const struct file_operations zero_fops = {
 847         .llseek         = zero_lseek,
 848         .read           = read_zero,
 849         .write          = write_zero,
 850         .mmap           = mmap_zero,
 851 };
 852
 853 /*
 854  * capabilities for /dev/zero
 855  * - permits private mappings, "copies" are taken of the source of zeros
 856  */
 857 static struct backing_dev_info zero_bdi = {
 858         .capabilities   = BDI_CAP_MAP_COPY,
 859 };
 860
 861 static const struct file_operations full_fops = {
 862         .llseek         = full_lseek,
 863         .read           = read_full,
 864         .write          = write_full,
 865 };
 866
 867 #ifdef CONFIG_CRASH_DUMP
 868 static const struct file_operations oldmem_fops = {
 869         .read   = read_oldmem,
 870         .open   = open_oldmem,
 871 };
 872 #endif
 873
 874 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 875                           size_t count, loff_t *ppos)
 876 {
 877         char *tmp;
 878         ssize_t ret;
 879
 880         tmp = kmalloc(count + 1, GFP_KERNEL);
 881         if (tmp == NULL)
 882                 return -ENOMEM;
 883         ret = -EFAULT;
 884         if (!copy_from_user(tmp, buf, count)) {
 885                 tmp[count] = 0;
 886                 ret = printk("%s", tmp);
 887                 if (ret > count)
 888                         /* printk can add a prefix */
 889                         ret = count;
 890         }
 891         kfree(tmp);
 892         return ret;
 893 }
 894
 895 static const struct file_operations kmsg_fops = {
 896         .write =        kmsg_write,
 897 };
 898
 899 static int memory_open(struct inode * inode, struct file * filp)
 900 {
 901         switch (iminor(inode)) {
 902                 case 1:
 903                         filp->f_op = &mem_fops;
 904                         filp->f_mapping->backing_dev_info =
 905                                 &directly_mappable_cdev_bdi;
 906                         break;
 907                 case 2:
 908                         filp->f_op = &kmem_fops;
 909                         filp->f_mapping->backing_dev_info =
 910                                 &directly_mappable_cdev_bdi;
 911                         break;
 912                 case 3:
 913                         filp->f_op = &null_fops;
 914                         break;
 915 #ifdef CONFIG_DEVPORT
 916                 case 4:
 917                         filp->f_op = &port_fops;
 918                         break;
 919 #endif
 920                 case 5:
 921                         filp->f_mapping->backing_dev_info = &zero_bdi;
 922                         filp->f_op = &zero_fops;
 923                         break;
 924                 case 7:
 925                         filp->f_op = &full_fops;
 926                         break;
 927                 case 8:
 928                         filp->f_op = &random_fops;
 929                         break;
 930                 case 9:
 931                         filp->f_op = &urandom_fops;
 932                         break;
 933                 case 11:
 934                         filp->f_op = &kmsg_fops;
 935                         break;
 936 #ifdef CONFIG_CRASH_DUMP
 937                 case 12:
 938                         filp->f_op = &oldmem_fops;
 939                         break;
 940 #endif
 941                 default:
 942                         return -ENXIO;
 943         }
 944         if (filp->f_op && filp->f_op->open)
 945                 return filp->f_op->open(inode,filp);
 946         return 0;
 947 }
 948
 949 static const struct file_operations memory_fops = {
 950         .open           = memory_open,  /* just a selector for the real open */
 951 };
 952
 953 static const struct {
 954         unsigned int            minor;
 955         char                    *name;
 956         umode_t                 mode;
 957         const struct file_operations    *fops;
 958 } devlist[] = { /* list of minor devices */
 959         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 960         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 961         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 962 #ifdef CONFIG_DEVPORT
 963         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 964 #endif
 965         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 966         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 967         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 968         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 969         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 970 #ifdef CONFIG_CRASH_DUMP
 971         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 972 #endif
 973 };
 974
 975 static struct class *mem_class;
 976
 977 static int __init chr_dev_init(void)
 978 {
 979         int i;
 980
 981         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 982                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 983
 984         mem_class = class_create(THIS_MODULE, "mem");
 985         for (i = 0; i < ARRAY_SIZE(devlist); i++)
 986                 device_create(mem_class, NULL,
 987                               MKDEV(MEM_MAJOR, devlist[i].minor),
 988                               devlist[i].name);
 989
 990         return 0;
 991 }
 992
 993 fs_initcall(chr_dev_init);