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__)
  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 #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 /*
 112  * This funcion reads the *physical* memory. The f_pos points directly to the
 113  * memory location.
 114  */
 115 static ssize_t read_mem(struct file * file, char __user * buf,
 116                         size_t count, loff_t *ppos)
 117 {
 118         unsigned long p = *ppos;
 119         ssize_t read, sz;
 120         char *ptr;
 121
 122         if (!valid_phys_addr_range(p, count))
 123                 return -EFAULT;
 124         read = 0;
 125 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 126         /* we don't have page 0 mapped on sparc and m68k.. */
 127         if (p < PAGE_SIZE) {
 128                 sz = PAGE_SIZE - p;
 129                 if (sz > count)
 130                         sz = count;
 131                 if (sz > 0) {
 132                         if (clear_user(buf, sz))
 133                                 return -EFAULT;
 134                         buf += sz;
 135                         p += sz;
 136                         count -= sz;
 137                         read += sz;
 138                 }
 139         }
 140 #endif
 141
 142         while (count > 0) {
 143                 /*
 144                  * Handle first page in case it's not aligned
 145                  */
 146                 if (-p & (PAGE_SIZE - 1))
 147                         sz = -p & (PAGE_SIZE - 1);
 148                 else
 149                         sz = PAGE_SIZE;
 150
 151                 sz = min_t(unsigned long, sz, count);
 152
 153                 /*
 154                  * On ia64 if a page has been mapped somewhere as
 155                  * uncached, then it must also be accessed uncached
 156                  * by the kernel or data corruption may occur
 157                  */
 158                 ptr = xlate_dev_mem_ptr(p);
 159
 160                 if (copy_to_user(buf, ptr, sz))
 161                         return -EFAULT;
 162                 buf += sz;
 163                 p += sz;
 164                 count -= sz;
 165                 read += sz;
 166         }
 167
 168         *ppos += read;
 169         return read;
 170 }
 171
 172 static ssize_t write_mem(struct file * file, const char __user * buf,
 173                          size_t count, loff_t *ppos)
 174 {
 175         unsigned long p = *ppos;
 176         ssize_t written, sz;
 177         unsigned long copied;
 178         void *ptr;
 179
 180         if (!valid_phys_addr_range(p, count))
 181                 return -EFAULT;
 182
 183         written = 0;
 184
 185 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 186         /* we don't have page 0 mapped on sparc and m68k.. */
 187         if (p < PAGE_SIZE) {
 188                 unsigned long sz = PAGE_SIZE - p;
 189                 if (sz > count)
 190                         sz = count;
 191                 /* Hmm. Do something? */
 192                 buf += sz;
 193                 p += sz;
 194                 count -= sz;
 195                 written += sz;
 196         }
 197 #endif
 198
 199         while (count > 0) {
 200                 /*
 201                  * Handle first page in case it's not aligned
 202                  */
 203                 if (-p & (PAGE_SIZE - 1))
 204                         sz = -p & (PAGE_SIZE - 1);
 205                 else
 206                         sz = PAGE_SIZE;
 207
 208                 sz = min_t(unsigned long, sz, count);
 209
 210                 /*
 211                  * On ia64 if a page has been mapped somewhere as
 212                  * uncached, then it must also be accessed uncached
 213                  * by the kernel or data corruption may occur
 214                  */
 215                 ptr = xlate_dev_mem_ptr(p);
 216
 217                 copied = copy_from_user(ptr, buf, sz);
 218                 if (copied) {
 219                         written += sz - copied;
 220                         if (written)
 221                                 break;
 222                         return -EFAULT;
 223                 }
 224                 buf += sz;
 225                 p += sz;
 226                 count -= sz;
 227                 written += sz;
 228         }
 229
 230         *ppos += written;
 231         return written;
 232 }
 233
 234 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 235 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 236                                      unsigned long size, pgprot_t vma_prot)
 237 {
 238 #ifdef pgprot_noncached
 239         unsigned long offset = pfn << PAGE_SHIFT;
 240
 241         if (uncached_access(file, offset))
 242                 return pgprot_noncached(vma_prot);
 243 #endif
 244         return vma_prot;
 245 }
 246 #endif
 247
 248 #ifndef CONFIG_MMU
 249 static unsigned long get_unmapped_area_mem(struct file *file,
 250                                            unsigned long addr,
 251                                            unsigned long len,
 252                                            unsigned long pgoff,
 253                                            unsigned long flags)
 254 {
 255         if (!valid_mmap_phys_addr_range(pgoff, len))
 256                 return (unsigned long) -EINVAL;
 257         return pgoff << PAGE_SHIFT;
 258 }
 259
 260 /* can't do an in-place private mapping if there's no MMU */
 261 static inline int private_mapping_ok(struct vm_area_struct *vma)
 262 {
 263         return vma->vm_flags & VM_MAYSHARE;
 264 }
 265 #else
 266 #define get_unmapped_area_mem   NULL
 267
 268 static inline int private_mapping_ok(struct vm_area_struct *vma)
 269 {
 270         return 1;
 271 }
 272 #endif
 273
 274 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 275 {
 276         size_t size = vma->vm_end - vma->vm_start;
 277
 278         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 279                 return -EINVAL;
 280
 281         if (!private_mapping_ok(vma))
 282                 return -ENOSYS;
 283
 284         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 285                                                  size,
 286                                                  vma->vm_page_prot);
 287
 288         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 289         if (remap_pfn_range(vma,
 290                             vma->vm_start,
 291                             vma->vm_pgoff,
 292                             size,
 293                             vma->vm_page_prot))
 294                 return -EAGAIN;
 295         return 0;
 296 }
 297
 298 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 299 {
 300         unsigned long pfn;
 301
 302         /* Turn a kernel-virtual address into a physical page frame */
 303         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 304
 305         /*
 306          * RED-PEN: on some architectures there is more mapped memory
 307          * than available in mem_map which pfn_valid checks
 308          * for. Perhaps should add a new macro here.
 309          *
 310          * RED-PEN: vmalloc is not supported right now.
 311          */
 312         if (!pfn_valid(pfn))
 313                 return -EIO;
 314
 315         vma->vm_pgoff = pfn;
 316         return mmap_mem(file, vma);
 317 }
 318
 319 #ifdef CONFIG_CRASH_DUMP
 320 /*
 321  * Read memory corresponding to the old kernel.
 322  */
 323 static ssize_t read_oldmem(struct file *file, char __user *buf,
 324                                 size_t count, loff_t *ppos)
 325 {
 326         unsigned long pfn, offset;
 327         size_t read = 0, csize;
 328         int rc = 0;
 329
 330         while (count) {
 331                 pfn = *ppos / PAGE_SIZE;
 332                 if (pfn > saved_max_pfn)
 333                         return read;
 334
 335                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 336                 if (count > PAGE_SIZE - offset)
 337                         csize = PAGE_SIZE - offset;
 338                 else
 339                         csize = count;
 340
 341                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 342                 if (rc < 0)
 343                         return rc;
 344                 buf += csize;
 345                 *ppos += csize;
 346                 read += csize;
 347                 count -= csize;
 348         }
 349         return read;
 350 }
 351 #endif
 352
 353 extern long vread(char *buf, char *addr, unsigned long count);
 354 extern long vwrite(char *buf, char *addr, unsigned long count);
 355
 356 /*
 357  * This function reads the *virtual* memory as seen by the kernel.
 358  */
 359 static ssize_t read_kmem(struct file *file, char __user *buf,
 360                          size_t count, loff_t *ppos)
 361 {
 362         unsigned long p = *ppos;
 363         ssize_t low_count, read, sz;
 364         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 365
 366         read = 0;
 367         if (p < (unsigned long) high_memory) {
 368                 low_count = count;
 369                 if (count > (unsigned long) high_memory - p)
 370                         low_count = (unsigned long) high_memory - p;
 371
 372 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 373                 /* we don't have page 0 mapped on sparc and m68k.. */
 374                 if (p < PAGE_SIZE && low_count > 0) {
 375                         size_t tmp = PAGE_SIZE - p;
 376                         if (tmp > low_count) tmp = low_count;
 377                         if (clear_user(buf, tmp))
 378                                 return -EFAULT;
 379                         buf += tmp;
 380                         p += tmp;
 381                         read += tmp;
 382                         low_count -= tmp;
 383                         count -= tmp;
 384                 }
 385 #endif
 386                 while (low_count > 0) {
 387                         /*
 388                          * Handle first page in case it's not aligned
 389                          */
 390                         if (-p & (PAGE_SIZE - 1))
 391                                 sz = -p & (PAGE_SIZE - 1);
 392                         else
 393                                 sz = PAGE_SIZE;
 394
 395                         sz = min_t(unsigned long, sz, low_count);
 396
 397                         /*
 398                          * On ia64 if a page has been mapped somewhere as
 399                          * uncached, then it must also be accessed uncached
 400                          * by the kernel or data corruption may occur
 401                          */
 402                         kbuf = xlate_dev_kmem_ptr((char *)p);
 403
 404                         if (copy_to_user(buf, kbuf, sz))
 405                                 return -EFAULT;
 406                         buf += sz;
 407                         p += sz;
 408                         read += sz;
 409                         low_count -= sz;
 410                         count -= sz;
 411                 }
 412         }
 413
 414         if (count > 0) {
 415                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 416                 if (!kbuf)
 417                         return -ENOMEM;
 418                 while (count > 0) {
 419                         int len = count;
 420
 421                         if (len > PAGE_SIZE)
 422                                 len = PAGE_SIZE;
 423                         len = vread(kbuf, (char *)p, len);
 424                         if (!len)
 425                                 break;
 426                         if (copy_to_user(buf, kbuf, len)) {
 427                                 free_page((unsigned long)kbuf);
 428                                 return -EFAULT;
 429                         }
 430                         count -= len;
 431                         buf += len;
 432                         read += len;
 433                         p += len;
 434                 }
 435                 free_page((unsigned long)kbuf);
 436         }
 437         *ppos = p;
 438         return read;
 439 }
 440
 441
 442 static inline ssize_t
 443 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 444               size_t count, loff_t *ppos)
 445 {
 446         ssize_t written, sz;
 447         unsigned long copied;
 448
 449         written = 0;
 450 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 451         /* we don't have page 0 mapped on sparc and m68k.. */
 452         if (realp < PAGE_SIZE) {
 453                 unsigned long sz = PAGE_SIZE - realp;
 454                 if (sz > count)
 455                         sz = count;
 456                 /* Hmm. Do something? */
 457                 buf += sz;
 458                 p += sz;
 459                 realp += sz;
 460                 count -= sz;
 461                 written += sz;
 462         }
 463 #endif
 464
 465         while (count > 0) {
 466                 char *ptr;
 467                 /*
 468                  * Handle first page in case it's not aligned
 469                  */
 470                 if (-realp & (PAGE_SIZE - 1))
 471                         sz = -realp & (PAGE_SIZE - 1);
 472                 else
 473                         sz = PAGE_SIZE;
 474
 475                 sz = min_t(unsigned long, sz, count);
 476
 477                 /*
 478                  * On ia64 if a page has been mapped somewhere as
 479                  * uncached, then it must also be accessed uncached
 480                  * by the kernel or data corruption may occur
 481                  */
 482                 ptr = xlate_dev_kmem_ptr(p);
 483
 484                 copied = copy_from_user(ptr, buf, sz);
 485                 if (copied) {
 486                         written += sz - copied;
 487                         if (written)
 488                                 break;
 489                         return -EFAULT;
 490                 }
 491                 buf += sz;
 492                 p += sz;
 493                 realp += sz;
 494                 count -= sz;
 495                 written += sz;
 496         }
 497
 498         *ppos += written;
 499         return written;
 500 }
 501
 502
 503 /*
 504  * This function writes to the *virtual* memory as seen by the kernel.
 505  */
 506 static ssize_t write_kmem(struct file * file, const char __user * buf,
 507                           size_t count, loff_t *ppos)
 508 {
 509         unsigned long p = *ppos;
 510         ssize_t wrote = 0;
 511         ssize_t virtr = 0;
 512         ssize_t written;
 513         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 514
 515         if (p < (unsigned long) high_memory) {
 516
 517                 wrote = count;
 518                 if (count > (unsigned long) high_memory - p)
 519                         wrote = (unsigned long) high_memory - p;
 520
 521                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 522                 if (written != wrote)
 523                         return written;
 524                 wrote = written;
 525                 p += wrote;
 526                 buf += wrote;
 527                 count -= wrote;
 528         }
 529
 530         if (count > 0) {
 531                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 532                 if (!kbuf)
 533                         return wrote ? wrote : -ENOMEM;
 534                 while (count > 0) {
 535                         int len = count;
 536
 537                         if (len > PAGE_SIZE)
 538                                 len = PAGE_SIZE;
 539                         if (len) {
 540                                 written = copy_from_user(kbuf, buf, len);
 541                                 if (written) {
 542                                         if (wrote + virtr)
 543                                                 break;
 544                                         free_page((unsigned long)kbuf);
 545                                         return -EFAULT;
 546                                 }
 547                         }
 548                         len = vwrite(kbuf, (char *)p, len);
 549                         count -= len;
 550                         buf += len;
 551                         virtr += len;
 552                         p += len;
 553                 }
 554                 free_page((unsigned long)kbuf);
 555         }
 556
 557         *ppos = p;
 558         return virtr + wrote;
 559 }
 560
 561 #ifdef CONFIG_DEVPORT
 562 static ssize_t read_port(struct file * file, char __user * buf,
 563                          size_t count, loff_t *ppos)
 564 {
 565         unsigned long i = *ppos;
 566         char __user *tmp = buf;
 567
 568         if (!access_ok(VERIFY_WRITE, buf, count))
 569                 return -EFAULT;
 570         while (count-- > 0 && i < 65536) {
 571                 if (__put_user(inb(i),tmp) < 0)
 572                         return -EFAULT;
 573                 i++;
 574                 tmp++;
 575         }
 576         *ppos = i;
 577         return tmp-buf;
 578 }
 579
 580 static ssize_t write_port(struct file * file, const char __user * buf,
 581                           size_t count, loff_t *ppos)
 582 {
 583         unsigned long i = *ppos;
 584         const char __user * tmp = buf;
 585
 586         if (!access_ok(VERIFY_READ,buf,count))
 587                 return -EFAULT;
 588         while (count-- > 0 && i < 65536) {
 589                 char c;
 590                 if (__get_user(c, tmp)) {
 591                         if (tmp > buf)
 592                                 break;
 593                         return -EFAULT;
 594                 }
 595                 outb(c,i);
 596                 i++;
 597                 tmp++;
 598         }
 599         *ppos = i;
 600         return tmp-buf;
 601 }
 602 #endif
 603
 604 static ssize_t read_null(struct file * file, char __user * buf,
 605                          size_t count, loff_t *ppos)
 606 {
 607         return 0;
 608 }
 609
 610 static ssize_t write_null(struct file * file, const char __user * buf,
 611                           size_t count, loff_t *ppos)
 612 {
 613         return count;
 614 }
 615
 616 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 617                         struct splice_desc *sd)
 618 {
 619         return sd->len;
 620 }
 621
 622 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 623                                  loff_t *ppos, size_t len, unsigned int flags)
 624 {
 625         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 626 }
 627
 628 #ifdef CONFIG_MMU
 629 /*
 630  * For fun, we are using the MMU for this.
 631  */
 632 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
 633 {
 634         struct mm_struct *mm;
 635         struct vm_area_struct * vma;
 636         unsigned long addr=(unsigned long)buf;
 637
 638         mm = current->mm;
 639         /* Oops, this was forgotten before. -ben */
 640         down_read(&mm->mmap_sem);
 641
 642         /* For private mappings, just map in zero pages. */
 643         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 644                 unsigned long count;
 645
 646                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 647                         goto out_up;
 648                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
 649                         break;
 650                 count = vma->vm_end - addr;
 651                 if (count > size)
 652                         count = size;
 653
 654                 zap_page_range(vma, addr, count, NULL);
 655                 if (zeromap_page_range(vma, addr, count, PAGE_COPY))
 656                         break;
 657
 658                 size -= count;
 659                 buf += count;
 660                 addr += count;
 661                 if (size == 0)
 662                         goto out_up;
 663         }
 664
 665         up_read(&mm->mmap_sem);
 666
 667         /* The shared case is hard. Let's do the conventional zeroing. */
 668         do {
 669                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 670                 if (unwritten)
 671                         return size + unwritten - PAGE_SIZE;
 672                 cond_resched();
 673                 buf += PAGE_SIZE;
 674                 size -= PAGE_SIZE;
 675         } while (size);
 676
 677         return size;
 678 out_up:
 679         up_read(&mm->mmap_sem);
 680         return size;
 681 }
 682
 683 static ssize_t read_zero(struct file * file, char __user * buf,
 684                          size_t count, loff_t *ppos)
 685 {
 686         unsigned long left, unwritten, written = 0;
 687
 688         if (!count)
 689                 return 0;
 690
 691         if (!access_ok(VERIFY_WRITE, buf, count))
 692                 return -EFAULT;
 693
 694         left = count;
 695
 696         /* do we want to be clever? Arbitrary cut-off */
 697         if (count >= PAGE_SIZE*4) {
 698                 unsigned long partial;
 699
 700                 /* How much left of the page? */
 701                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 702                 unwritten = clear_user(buf, partial);
 703                 written = partial - unwritten;
 704                 if (unwritten)
 705                         goto out;
 706                 left -= partial;
 707                 buf += partial;
 708                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 709                 written += (left & PAGE_MASK) - unwritten;
 710                 if (unwritten)
 711                         goto out;
 712                 buf += left & PAGE_MASK;
 713                 left &= ~PAGE_MASK;
 714         }
 715         unwritten = clear_user(buf, left);
 716         written += left - unwritten;
 717 out:
 718         return written ? written : -EFAULT;
 719 }
 720
 721 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 722 {
 723         int err;
 724
 725         if (vma->vm_flags & VM_SHARED)
 726                 return shmem_zero_setup(vma);
 727         err = zeromap_page_range(vma, vma->vm_start,
 728                         vma->vm_end - vma->vm_start, vma->vm_page_prot);
 729         BUG_ON(err == -EEXIST);
 730         return err;
 731 }
 732 #else /* CONFIG_MMU */
 733 static ssize_t read_zero(struct file * file, char * buf,
 734                          size_t count, loff_t *ppos)
 735 {
 736         size_t todo = count;
 737
 738         while (todo) {
 739                 size_t chunk = todo;
 740
 741                 if (chunk > 4096)
 742                         chunk = 4096;   /* Just for latency reasons */
 743                 if (clear_user(buf, chunk))
 744                         return -EFAULT;
 745                 buf += chunk;
 746                 todo -= chunk;
 747                 cond_resched();
 748         }
 749         return count;
 750 }
 751
 752 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 753 {
 754         return -ENOSYS;
 755 }
 756 #endif /* CONFIG_MMU */
 757
 758 static ssize_t write_full(struct file * file, const char __user * buf,
 759                           size_t count, loff_t *ppos)
 760 {
 761         return -ENOSPC;
 762 }
 763
 764 /*
 765  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 766  * can fopen() both devices with "a" now.  This was previously impossible.
 767  * -- SRB.
 768  */
 769
 770 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 771 {
 772         return file->f_pos = 0;
 773 }
 774
 775 /*
 776  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 777  * check against negative addresses: they are ok. The return value is weird,
 778  * though, in that case (0).
 779  *
 780  * also note that seeking relative to the "end of file" isn't supported:
 781  * it has no meaning, so it returns -EINVAL.
 782  */
 783 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 784 {
 785         loff_t ret;
 786
 787         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 788         switch (orig) {
 789                 case 0:
 790                         file->f_pos = offset;
 791                         ret = file->f_pos;
 792                         force_successful_syscall_return();
 793                         break;
 794                 case 1:
 795                         file->f_pos += offset;
 796                         ret = file->f_pos;
 797                         force_successful_syscall_return();
 798                         break;
 799                 default:
 800                         ret = -EINVAL;
 801         }
 802         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 803         return ret;
 804 }
 805
 806 static int open_port(struct inode * inode, struct file * filp)
 807 {
 808         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 809 }
 810
 811 #define zero_lseek      null_lseek
 812 #define full_lseek      null_lseek
 813 #define write_zero      write_null
 814 #define read_full       read_zero
 815 #define open_mem        open_port
 816 #define open_kmem       open_mem
 817 #define open_oldmem     open_mem
 818
 819 static const struct file_operations mem_fops = {
 820         .llseek         = memory_lseek,
 821         .read           = read_mem,
 822         .write          = write_mem,
 823         .mmap           = mmap_mem,
 824         .open           = open_mem,
 825         .get_unmapped_area = get_unmapped_area_mem,
 826 };
 827
 828 static const struct file_operations kmem_fops = {
 829         .llseek         = memory_lseek,
 830         .read           = read_kmem,
 831         .write          = write_kmem,
 832         .mmap           = mmap_kmem,
 833         .open           = open_kmem,
 834         .get_unmapped_area = get_unmapped_area_mem,
 835 };
 836
 837 static const struct file_operations null_fops = {
 838         .llseek         = null_lseek,
 839         .read           = read_null,
 840         .write          = write_null,
 841         .splice_write   = splice_write_null,
 842 };
 843
 844 #ifdef CONFIG_DEVPORT
 845 static const struct file_operations port_fops = {
 846         .llseek         = memory_lseek,
 847         .read           = read_port,
 848         .write          = write_port,
 849         .open           = open_port,
 850 };
 851 #endif
 852
 853 static const struct file_operations zero_fops = {
 854         .llseek         = zero_lseek,
 855         .read           = read_zero,
 856         .write          = write_zero,
 857         .mmap           = mmap_zero,
 858 };
 859
 860 /*
 861  * capabilities for /dev/zero
 862  * - permits private mappings, "copies" are taken of the source of zeros
 863  */
 864 static struct backing_dev_info zero_bdi = {
 865         .capabilities   = BDI_CAP_MAP_COPY,
 866 };
 867
 868 static const struct file_operations full_fops = {
 869         .llseek         = full_lseek,
 870         .read           = read_full,
 871         .write          = write_full,
 872 };
 873
 874 #ifdef CONFIG_CRASH_DUMP
 875 static const struct file_operations oldmem_fops = {
 876         .read   = read_oldmem,
 877         .open   = open_oldmem,
 878 };
 879 #endif
 880
 881 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 882                           size_t count, loff_t *ppos)
 883 {
 884         char *tmp;
 885         ssize_t ret;
 886
 887         tmp = kmalloc(count + 1, GFP_KERNEL);
 888         if (tmp == NULL)
 889                 return -ENOMEM;
 890         ret = -EFAULT;
 891         if (!copy_from_user(tmp, buf, count)) {
 892                 tmp[count] = 0;
 893                 ret = printk("%s", tmp);
 894                 if (ret > count)
 895                         /* printk can add a prefix */
 896                         ret = count;
 897         }
 898         kfree(tmp);
 899         return ret;
 900 }
 901
 902 static const struct file_operations kmsg_fops = {
 903         .write =        kmsg_write,
 904 };
 905
 906 static int memory_open(struct inode * inode, struct file * filp)
 907 {
 908         switch (iminor(inode)) {
 909                 case 1:
 910                         filp->f_op = &mem_fops;
 911                         filp->f_mapping->backing_dev_info =
 912                                 &directly_mappable_cdev_bdi;
 913                         break;
 914                 case 2:
 915                         filp->f_op = &kmem_fops;
 916                         filp->f_mapping->backing_dev_info =
 917                                 &directly_mappable_cdev_bdi;
 918                         break;
 919                 case 3:
 920                         filp->f_op = &null_fops;
 921                         break;
 922 #ifdef CONFIG_DEVPORT
 923                 case 4:
 924                         filp->f_op = &port_fops;
 925                         break;
 926 #endif
 927                 case 5:
 928                         filp->f_mapping->backing_dev_info = &zero_bdi;
 929                         filp->f_op = &zero_fops;
 930                         break;
 931                 case 7:
 932                         filp->f_op = &full_fops;
 933                         break;
 934                 case 8:
 935                         filp->f_op = &random_fops;
 936                         break;
 937                 case 9:
 938                         filp->f_op = &urandom_fops;
 939                         break;
 940                 case 11:
 941                         filp->f_op = &kmsg_fops;
 942                         break;
 943 #ifdef CONFIG_CRASH_DUMP
 944                 case 12:
 945                         filp->f_op = &oldmem_fops;
 946                         break;
 947 #endif
 948                 default:
 949                         return -ENXIO;
 950         }
 951         if (filp->f_op && filp->f_op->open)
 952                 return filp->f_op->open(inode,filp);
 953         return 0;
 954 }
 955
 956 static const struct file_operations memory_fops = {
 957         .open           = memory_open,  /* just a selector for the real open */
 958 };
 959
 960 static const struct {
 961         unsigned int            minor;
 962         char                    *name;
 963         umode_t                 mode;
 964         const struct file_operations    *fops;
 965 } devlist[] = { /* list of minor devices */
 966         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 967         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 968         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 969 #ifdef CONFIG_DEVPORT
 970         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 971 #endif
 972         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 973         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 974         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 975         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 976         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 977 #ifdef CONFIG_CRASH_DUMP
 978         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 979 #endif
 980 };
 981
 982 static struct class *mem_class;
 983
 984 static int __init chr_dev_init(void)
 985 {
 986         int i;
 987
 988         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 989                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 990
 991         mem_class = class_create(THIS_MODULE, "mem");
 992         for (i = 0; i < ARRAY_SIZE(devlist); i++)
 993                 device_create(mem_class, NULL,
 994                               MKDEV(MEM_MAJOR, devlist[i].minor),
 995                               devlist[i].name);
 996
 997         return 0;
 998 }
 999
1000 fs_initcall(chr_dev_init);