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/config.h>
  12 #include <linux/mm.h>
  13 #include <linux/miscdevice.h>
  14 #include <linux/slab.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/mman.h>
  17 #include <linux/random.h>
  18 #include <linux/init.h>
  19 #include <linux/raw.h>
  20 #include <linux/tty.h>
  21 #include <linux/capability.h>
  22 #include <linux/smp_lock.h>
  23 #include <linux/devfs_fs_kernel.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/device.h>
  26 #include <linux/highmem.h>
  27 #include <linux/crash_dump.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/bootmem.h>
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/io.h>
  33
  34 #ifdef CONFIG_IA64
  35 # include <linux/efi.h>
  36 #endif
  37
  38 /*
  39  * Architectures vary in how they handle caching for addresses
  40  * outside of main memory.
  41  *
  42  */
  43 static inline int uncached_access(struct file *file, unsigned long addr)
  44 {
  45 #if defined(__i386__)
  46         /*
  47          * On the PPro and successors, the MTRRs are used to set
  48          * memory types for physical addresses outside main memory,
  49          * so blindly setting PCD or PWT on those pages is wrong.
  50          * For Pentiums and earlier, the surround logic should disable
  51          * caching for the high addresses through the KEN pin, but
  52          * we maintain the tradition of paranoia in this code.
  53          */
  54         if (file->f_flags & O_SYNC)
  55                 return 1;
  56         return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
  57                   test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
  58                   test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
  59                   test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
  60           && addr >= __pa(high_memory);
  61 #elif defined(__x86_64__)
  62         /*
  63          * This is broken because it can generate memory type aliases,
  64          * which can cause cache corruptions
  65          * But it is only available for root and we have to be bug-to-bug
  66          * compatible with i386.
  67          */
  68         if (file->f_flags & O_SYNC)
  69                 return 1;
  70         /* same behaviour as i386. PAT always set to cached and MTRRs control the
  71            caching behaviour.
  72            Hopefully a full PAT implementation will fix that soon. */
  73         return 0;
  74 #elif defined(CONFIG_IA64)
  75         /*
  76          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  77          */
  78         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  79 #else
  80         /*
  81          * Accessing memory above the top the kernel knows about or through a file pointer
  82          * that was marked O_SYNC will be done non-cached.
  83          */
  84         if (file->f_flags & O_SYNC)
  85                 return 1;
  86         return addr >= __pa(high_memory);
  87 #endif
  88 }
  89
  90 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  91 static inline int valid_phys_addr_range(unsigned long addr, size_t *count)
  92 {
  93         unsigned long end_mem;
  94
  95         end_mem = __pa(high_memory);
  96         if (addr >= end_mem)
  97                 return 0;
  98
  99         if (*count > end_mem - addr)
 100                 *count = end_mem - addr;
 101
 102         return 1;
 103 }
 104
 105 static inline int valid_mmap_phys_addr_range(unsigned long addr, 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 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 249 {
 250         size_t size = vma->vm_end - vma->vm_start;
 251
 252         if (!valid_mmap_phys_addr_range(vma->vm_pgoff << PAGE_SHIFT, &size))
 253                 return -EINVAL;
 254
 255         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 256                                                  size,
 257                                                  vma->vm_page_prot);
 258
 259         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 260         if (remap_pfn_range(vma,
 261                             vma->vm_start,
 262                             vma->vm_pgoff,
 263                             size,
 264                             vma->vm_page_prot))
 265                 return -EAGAIN;
 266         return 0;
 267 }
 268
 269 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 270 {
 271         unsigned long pfn;
 272
 273         /* Turn a kernel-virtual address into a physical page frame */
 274         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 275
 276         /*
 277          * RED-PEN: on some architectures there is more mapped memory
 278          * than available in mem_map which pfn_valid checks
 279          * for. Perhaps should add a new macro here.
 280          *
 281          * RED-PEN: vmalloc is not supported right now.
 282          */
 283         if (!pfn_valid(pfn))
 284                 return -EIO;
 285
 286         vma->vm_pgoff = pfn;
 287         return mmap_mem(file, vma);
 288 }
 289
 290 #ifdef CONFIG_CRASH_DUMP
 291 /*
 292  * Read memory corresponding to the old kernel.
 293  */
 294 static ssize_t read_oldmem(struct file *file, char __user *buf,
 295                                 size_t count, loff_t *ppos)
 296 {
 297         unsigned long pfn, offset;
 298         size_t read = 0, csize;
 299         int rc = 0;
 300
 301         while (count) {
 302                 pfn = *ppos / PAGE_SIZE;
 303                 if (pfn > saved_max_pfn)
 304                         return read;
 305
 306                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 307                 if (count > PAGE_SIZE - offset)
 308                         csize = PAGE_SIZE - offset;
 309                 else
 310                         csize = count;
 311
 312                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 313                 if (rc < 0)
 314                         return rc;
 315                 buf += csize;
 316                 *ppos += csize;
 317                 read += csize;
 318                 count -= csize;
 319         }
 320         return read;
 321 }
 322 #endif
 323
 324 extern long vread(char *buf, char *addr, unsigned long count);
 325 extern long vwrite(char *buf, char *addr, unsigned long count);
 326
 327 /*
 328  * This function reads the *virtual* memory as seen by the kernel.
 329  */
 330 static ssize_t read_kmem(struct file *file, char __user *buf,
 331                          size_t count, loff_t *ppos)
 332 {
 333         unsigned long p = *ppos;
 334         ssize_t low_count, read, sz;
 335         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 336
 337         read = 0;
 338         if (p < (unsigned long) high_memory) {
 339                 low_count = count;
 340                 if (count > (unsigned long) high_memory - p)
 341                         low_count = (unsigned long) high_memory - p;
 342
 343 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 344                 /* we don't have page 0 mapped on sparc and m68k.. */
 345                 if (p < PAGE_SIZE && low_count > 0) {
 346                         size_t tmp = PAGE_SIZE - p;
 347                         if (tmp > low_count) tmp = low_count;
 348                         if (clear_user(buf, tmp))
 349                                 return -EFAULT;
 350                         buf += tmp;
 351                         p += tmp;
 352                         read += tmp;
 353                         low_count -= tmp;
 354                         count -= tmp;
 355                 }
 356 #endif
 357                 while (low_count > 0) {
 358                         /*
 359                          * Handle first page in case it's not aligned
 360                          */
 361                         if (-p & (PAGE_SIZE - 1))
 362                                 sz = -p & (PAGE_SIZE - 1);
 363                         else
 364                                 sz = PAGE_SIZE;
 365
 366                         sz = min_t(unsigned long, sz, low_count);
 367
 368                         /*
 369                          * On ia64 if a page has been mapped somewhere as
 370                          * uncached, then it must also be accessed uncached
 371                          * by the kernel or data corruption may occur
 372                          */
 373                         kbuf = xlate_dev_kmem_ptr((char *)p);
 374
 375                         if (copy_to_user(buf, kbuf, sz))
 376                                 return -EFAULT;
 377                         buf += sz;
 378                         p += sz;
 379                         read += sz;
 380                         low_count -= sz;
 381                         count -= sz;
 382                 }
 383         }
 384
 385         if (count > 0) {
 386                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 387                 if (!kbuf)
 388                         return -ENOMEM;
 389                 while (count > 0) {
 390                         int len = count;
 391
 392                         if (len > PAGE_SIZE)
 393                                 len = PAGE_SIZE;
 394                         len = vread(kbuf, (char *)p, len);
 395                         if (!len)
 396                                 break;
 397                         if (copy_to_user(buf, kbuf, len)) {
 398                                 free_page((unsigned long)kbuf);
 399                                 return -EFAULT;
 400                         }
 401                         count -= len;
 402                         buf += len;
 403                         read += len;
 404                         p += len;
 405                 }
 406                 free_page((unsigned long)kbuf);
 407         }
 408         *ppos = p;
 409         return read;
 410 }
 411
 412
 413 static inline ssize_t
 414 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 415               size_t count, loff_t *ppos)
 416 {
 417         ssize_t written, sz;
 418         unsigned long copied;
 419
 420         written = 0;
 421 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 422         /* we don't have page 0 mapped on sparc and m68k.. */
 423         if (realp < PAGE_SIZE) {
 424                 unsigned long sz = PAGE_SIZE - realp;
 425                 if (sz > count)
 426                         sz = count;
 427                 /* Hmm. Do something? */
 428                 buf += sz;
 429                 p += sz;
 430                 realp += sz;
 431                 count -= sz;
 432                 written += sz;
 433         }
 434 #endif
 435
 436         while (count > 0) {
 437                 char *ptr;
 438                 /*
 439                  * Handle first page in case it's not aligned
 440                  */
 441                 if (-realp & (PAGE_SIZE - 1))
 442                         sz = -realp & (PAGE_SIZE - 1);
 443                 else
 444                         sz = PAGE_SIZE;
 445
 446                 sz = min_t(unsigned long, sz, count);
 447
 448                 /*
 449                  * On ia64 if a page has been mapped somewhere as
 450                  * uncached, then it must also be accessed uncached
 451                  * by the kernel or data corruption may occur
 452                  */
 453                 ptr = xlate_dev_kmem_ptr(p);
 454
 455                 copied = copy_from_user(ptr, buf, sz);
 456                 if (copied) {
 457                         written += sz - copied;
 458                         if (written)
 459                                 break;
 460                         return -EFAULT;
 461                 }
 462                 buf += sz;
 463                 p += sz;
 464                 realp += sz;
 465                 count -= sz;
 466                 written += sz;
 467         }
 468
 469         *ppos += written;
 470         return written;
 471 }
 472
 473
 474 /*
 475  * This function writes to the *virtual* memory as seen by the kernel.
 476  */
 477 static ssize_t write_kmem(struct file * file, const char __user * buf,
 478                           size_t count, loff_t *ppos)
 479 {
 480         unsigned long p = *ppos;
 481         ssize_t wrote = 0;
 482         ssize_t virtr = 0;
 483         ssize_t written;
 484         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 485
 486         if (p < (unsigned long) high_memory) {
 487
 488                 wrote = count;
 489                 if (count > (unsigned long) high_memory - p)
 490                         wrote = (unsigned long) high_memory - p;
 491
 492                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 493                 if (written != wrote)
 494                         return written;
 495                 wrote = written;
 496                 p += wrote;
 497                 buf += wrote;
 498                 count -= wrote;
 499         }
 500
 501         if (count > 0) {
 502                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 503                 if (!kbuf)
 504                         return wrote ? wrote : -ENOMEM;
 505                 while (count > 0) {
 506                         int len = count;
 507
 508                         if (len > PAGE_SIZE)
 509                                 len = PAGE_SIZE;
 510                         if (len) {
 511                                 written = copy_from_user(kbuf, buf, len);
 512                                 if (written) {
 513                                         if (wrote + virtr)
 514                                                 break;
 515                                         free_page((unsigned long)kbuf);
 516                                         return -EFAULT;
 517                                 }
 518                         }
 519                         len = vwrite(kbuf, (char *)p, len);
 520                         count -= len;
 521                         buf += len;
 522                         virtr += len;
 523                         p += len;
 524                 }
 525                 free_page((unsigned long)kbuf);
 526         }
 527
 528         *ppos = p;
 529         return virtr + wrote;
 530 }
 531
 532 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 533 static ssize_t read_port(struct file * file, char __user * buf,
 534                          size_t count, loff_t *ppos)
 535 {
 536         unsigned long i = *ppos;
 537         char __user *tmp = buf;
 538
 539         if (!access_ok(VERIFY_WRITE, buf, count))
 540                 return -EFAULT;
 541         while (count-- > 0 && i < 65536) {
 542                 if (__put_user(inb(i),tmp) < 0)
 543                         return -EFAULT;
 544                 i++;
 545                 tmp++;
 546         }
 547         *ppos = i;
 548         return tmp-buf;
 549 }
 550
 551 static ssize_t write_port(struct file * file, const char __user * buf,
 552                           size_t count, loff_t *ppos)
 553 {
 554         unsigned long i = *ppos;
 555         const char __user * tmp = buf;
 556
 557         if (!access_ok(VERIFY_READ,buf,count))
 558                 return -EFAULT;
 559         while (count-- > 0 && i < 65536) {
 560                 char c;
 561                 if (__get_user(c, tmp)) {
 562                         if (tmp > buf)
 563                                 break;
 564                         return -EFAULT;
 565                 }
 566                 outb(c,i);
 567                 i++;
 568                 tmp++;
 569         }
 570         *ppos = i;
 571         return tmp-buf;
 572 }
 573 #endif
 574
 575 static ssize_t read_null(struct file * file, char __user * buf,
 576                          size_t count, loff_t *ppos)
 577 {
 578         return 0;
 579 }
 580
 581 static ssize_t write_null(struct file * file, const char __user * buf,
 582                           size_t count, loff_t *ppos)
 583 {
 584         return count;
 585 }
 586
 587 #ifdef CONFIG_MMU
 588 /*
 589  * For fun, we are using the MMU for this.
 590  */
 591 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
 592 {
 593         struct mm_struct *mm;
 594         struct vm_area_struct * vma;
 595         unsigned long addr=(unsigned long)buf;
 596
 597         mm = current->mm;
 598         /* Oops, this was forgotten before. -ben */
 599         down_read(&mm->mmap_sem);
 600
 601         /* For private mappings, just map in zero pages. */
 602         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 603                 unsigned long count;
 604
 605                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 606                         goto out_up;
 607                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
 608                         break;
 609                 count = vma->vm_end - addr;
 610                 if (count > size)
 611                         count = size;
 612
 613                 zap_page_range(vma, addr, count, NULL);
 614                 zeromap_page_range(vma, addr, count, PAGE_COPY);
 615
 616                 size -= count;
 617                 buf += count;
 618                 addr += count;
 619                 if (size == 0)
 620                         goto out_up;
 621         }
 622
 623         up_read(&mm->mmap_sem);
 624
 625         /* The shared case is hard. Let's do the conventional zeroing. */
 626         do {
 627                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 628                 if (unwritten)
 629                         return size + unwritten - PAGE_SIZE;
 630                 cond_resched();
 631                 buf += PAGE_SIZE;
 632                 size -= PAGE_SIZE;
 633         } while (size);
 634
 635         return size;
 636 out_up:
 637         up_read(&mm->mmap_sem);
 638         return size;
 639 }
 640
 641 static ssize_t read_zero(struct file * file, char __user * buf,
 642                          size_t count, loff_t *ppos)
 643 {
 644         unsigned long left, unwritten, written = 0;
 645
 646         if (!count)
 647                 return 0;
 648
 649         if (!access_ok(VERIFY_WRITE, buf, count))
 650                 return -EFAULT;
 651
 652         left = count;
 653
 654         /* do we want to be clever? Arbitrary cut-off */
 655         if (count >= PAGE_SIZE*4) {
 656                 unsigned long partial;
 657
 658                 /* How much left of the page? */
 659                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 660                 unwritten = clear_user(buf, partial);
 661                 written = partial - unwritten;
 662                 if (unwritten)
 663                         goto out;
 664                 left -= partial;
 665                 buf += partial;
 666                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 667                 written += (left & PAGE_MASK) - unwritten;
 668                 if (unwritten)
 669                         goto out;
 670                 buf += left & PAGE_MASK;
 671                 left &= ~PAGE_MASK;
 672         }
 673         unwritten = clear_user(buf, left);
 674         written += left - unwritten;
 675 out:
 676         return written ? written : -EFAULT;
 677 }
 678
 679 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 680 {
 681         if (vma->vm_flags & VM_SHARED)
 682                 return shmem_zero_setup(vma);
 683         if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
 684                 return -EAGAIN;
 685         return 0;
 686 }
 687 #else /* CONFIG_MMU */
 688 static ssize_t read_zero(struct file * file, char * buf,
 689                          size_t count, loff_t *ppos)
 690 {
 691         size_t todo = count;
 692
 693         while (todo) {
 694                 size_t chunk = todo;
 695
 696                 if (chunk > 4096)
 697                         chunk = 4096;   /* Just for latency reasons */
 698                 if (clear_user(buf, chunk))
 699                         return -EFAULT;
 700                 buf += chunk;
 701                 todo -= chunk;
 702                 cond_resched();
 703         }
 704         return count;
 705 }
 706
 707 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 708 {
 709         return -ENOSYS;
 710 }
 711 #endif /* CONFIG_MMU */
 712
 713 static ssize_t write_full(struct file * file, const char __user * buf,
 714                           size_t count, loff_t *ppos)
 715 {
 716         return -ENOSPC;
 717 }
 718
 719 /*
 720  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 721  * can fopen() both devices with "a" now.  This was previously impossible.
 722  * -- SRB.
 723  */
 724
 725 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 726 {
 727         return file->f_pos = 0;
 728 }
 729
 730 /*
 731  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 732  * check against negative addresses: they are ok. The return value is weird,
 733  * though, in that case (0).
 734  *
 735  * also note that seeking relative to the "end of file" isn't supported:
 736  * it has no meaning, so it returns -EINVAL.
 737  */
 738 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 739 {
 740         loff_t ret;
 741
 742         mutex_lock(&file->f_dentry->d_inode->i_mutex);
 743         switch (orig) {
 744                 case 0:
 745                         file->f_pos = offset;
 746                         ret = file->f_pos;
 747                         force_successful_syscall_return();
 748                         break;
 749                 case 1:
 750                         file->f_pos += offset;
 751                         ret = file->f_pos;
 752                         force_successful_syscall_return();
 753                         break;
 754                 default:
 755                         ret = -EINVAL;
 756         }
 757         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 758         return ret;
 759 }
 760
 761 static int open_port(struct inode * inode, struct file * filp)
 762 {
 763         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 764 }
 765
 766 #define zero_lseek      null_lseek
 767 #define full_lseek      null_lseek
 768 #define write_zero      write_null
 769 #define read_full       read_zero
 770 #define open_mem        open_port
 771 #define open_kmem       open_mem
 772 #define open_oldmem     open_mem
 773
 774 static struct file_operations mem_fops = {
 775         .llseek         = memory_lseek,
 776         .read           = read_mem,
 777         .write          = write_mem,
 778         .mmap           = mmap_mem,
 779         .open           = open_mem,
 780 };
 781
 782 static struct file_operations kmem_fops = {
 783         .llseek         = memory_lseek,
 784         .read           = read_kmem,
 785         .write          = write_kmem,
 786         .mmap           = mmap_kmem,
 787         .open           = open_kmem,
 788 };
 789
 790 static struct file_operations null_fops = {
 791         .llseek         = null_lseek,
 792         .read           = read_null,
 793         .write          = write_null,
 794 };
 795
 796 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 797 static struct file_operations port_fops = {
 798         .llseek         = memory_lseek,
 799         .read           = read_port,
 800         .write          = write_port,
 801         .open           = open_port,
 802 };
 803 #endif
 804
 805 static struct file_operations zero_fops = {
 806         .llseek         = zero_lseek,
 807         .read           = read_zero,
 808         .write          = write_zero,
 809         .mmap           = mmap_zero,
 810 };
 811
 812 static struct backing_dev_info zero_bdi = {
 813         .capabilities   = BDI_CAP_MAP_COPY,
 814 };
 815
 816 static struct file_operations full_fops = {
 817         .llseek         = full_lseek,
 818         .read           = read_full,
 819         .write          = write_full,
 820 };
 821
 822 #ifdef CONFIG_CRASH_DUMP
 823 static struct file_operations oldmem_fops = {
 824         .read   = read_oldmem,
 825         .open   = open_oldmem,
 826 };
 827 #endif
 828
 829 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 830                           size_t count, loff_t *ppos)
 831 {
 832         char *tmp;
 833         ssize_t ret;
 834
 835         tmp = kmalloc(count + 1, GFP_KERNEL);
 836         if (tmp == NULL)
 837                 return -ENOMEM;
 838         ret = -EFAULT;
 839         if (!copy_from_user(tmp, buf, count)) {
 840                 tmp[count] = 0;
 841                 ret = printk("%s", tmp);
 842                 if (ret > count)
 843                         /* printk can add a prefix */
 844                         ret = count;
 845         }
 846         kfree(tmp);
 847         return ret;
 848 }
 849
 850 static struct file_operations kmsg_fops = {
 851         .write =        kmsg_write,
 852 };
 853
 854 static int memory_open(struct inode * inode, struct file * filp)
 855 {
 856         switch (iminor(inode)) {
 857                 case 1:
 858                         filp->f_op = &mem_fops;
 859                         break;
 860                 case 2:
 861                         filp->f_op = &kmem_fops;
 862                         break;
 863                 case 3:
 864                         filp->f_op = &null_fops;
 865                         break;
 866 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 867                 case 4:
 868                         filp->f_op = &port_fops;
 869                         break;
 870 #endif
 871                 case 5:
 872                         filp->f_mapping->backing_dev_info = &zero_bdi;
 873                         filp->f_op = &zero_fops;
 874                         break;
 875                 case 7:
 876                         filp->f_op = &full_fops;
 877                         break;
 878                 case 8:
 879                         filp->f_op = &random_fops;
 880                         break;
 881                 case 9:
 882                         filp->f_op = &urandom_fops;
 883                         break;
 884                 case 11:
 885                         filp->f_op = &kmsg_fops;
 886                         break;
 887 #ifdef CONFIG_CRASH_DUMP
 888                 case 12:
 889                         filp->f_op = &oldmem_fops;
 890                         break;
 891 #endif
 892                 default:
 893                         return -ENXIO;
 894         }
 895         if (filp->f_op && filp->f_op->open)
 896                 return filp->f_op->open(inode,filp);
 897         return 0;
 898 }
 899
 900 static struct file_operations memory_fops = {
 901         .open           = memory_open,  /* just a selector for the real open */
 902 };
 903
 904 static const struct {
 905         unsigned int            minor;
 906         char                    *name;
 907         umode_t                 mode;
 908         struct file_operations  *fops;
 909 } devlist[] = { /* list of minor devices */
 910         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 911         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 912         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 913 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 914         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 915 #endif
 916         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 917         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 918         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 919         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 920         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 921 #ifdef CONFIG_CRASH_DUMP
 922         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 923 #endif
 924 };
 925
 926 static struct class *mem_class;
 927
 928 static int __init chr_dev_init(void)
 929 {
 930         int i;
 931
 932         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 933                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 934
 935         mem_class = class_create(THIS_MODULE, "mem");
 936         for (i = 0; i < ARRAY_SIZE(devlist); i++) {
 937                 class_device_create(mem_class, NULL,
 938                                         MKDEV(MEM_MAJOR, devlist[i].minor),
 939                                         NULL, devlist[i].name);
 940                 devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor),
 941                                 S_IFCHR | devlist[i].mode, devlist[i].name);
 942         }
 943
 944         return 0;
 945 }
 946
 947 fs_initcall(chr_dev_init);