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 #include <linux/pipe_fs_i.h>
  31
  32 #include <asm/uaccess.h>
  33 #include <asm/io.h>
  34
  35 #ifdef CONFIG_IA64
  36 # include <linux/efi.h>
  37 #endif
  38
  39 /*
  40  * Architectures vary in how they handle caching for addresses
  41  * outside of main memory.
  42  *
  43  */
  44 static inline int uncached_access(struct file *file, unsigned long addr)
  45 {
  46 #if defined(__i386__)
  47         /*
  48          * On the PPro and successors, the MTRRs are used to set
  49          * memory types for physical addresses outside main memory,
  50          * so blindly setting PCD or PWT on those pages is wrong.
  51          * For Pentiums and earlier, the surround logic should disable
  52          * caching for the high addresses through the KEN pin, but
  53          * we maintain the tradition of paranoia in this code.
  54          */
  55         if (file->f_flags & O_SYNC)
  56                 return 1;
  57         return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
  58                   test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
  59                   test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
  60                   test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
  61           && addr >= __pa(high_memory);
  62 #elif defined(__x86_64__)
  63         /*
  64          * This is broken because it can generate memory type aliases,
  65          * which can cause cache corruptions
  66          * But it is only available for root and we have to be bug-to-bug
  67          * compatible with i386.
  68          */
  69         if (file->f_flags & O_SYNC)
  70                 return 1;
  71         /* same behaviour as i386. PAT always set to cached and MTRRs control the
  72            caching behaviour.
  73            Hopefully a full PAT implementation will fix that soon. */
  74         return 0;
  75 #elif defined(CONFIG_IA64)
  76         /*
  77          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  78          */
  79         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  80 #else
  81         /*
  82          * Accessing memory above the top the kernel knows about or through a file pointer
  83          * that was marked O_SYNC will be done non-cached.
  84          */
  85         if (file->f_flags & O_SYNC)
  86                 return 1;
  87         return addr >= __pa(high_memory);
  88 #endif
  89 }
  90
  91 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  92 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  93 {
  94         if (addr + count > __pa(high_memory))
  95                 return 0;
  96
  97         return 1;
  98 }
  99
 100 static inline int valid_mmap_phys_addr_range(unsigned long addr, size_t size)
 101 {
 102         return 1;
 103 }
 104 #endif
 105
 106 /*
 107  * This funcion reads the *physical* memory. The f_pos points directly to the
 108  * memory location.
 109  */
 110 static ssize_t read_mem(struct file * file, char __user * buf,
 111                         size_t count, loff_t *ppos)
 112 {
 113         unsigned long p = *ppos;
 114         ssize_t read, sz;
 115         char *ptr;
 116
 117         if (!valid_phys_addr_range(p, count))
 118                 return -EFAULT;
 119         read = 0;
 120 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 121         /* we don't have page 0 mapped on sparc and m68k.. */
 122         if (p < PAGE_SIZE) {
 123                 sz = PAGE_SIZE - p;
 124                 if (sz > count)
 125                         sz = count;
 126                 if (sz > 0) {
 127                         if (clear_user(buf, sz))
 128                                 return -EFAULT;
 129                         buf += sz;
 130                         p += sz;
 131                         count -= sz;
 132                         read += sz;
 133                 }
 134         }
 135 #endif
 136
 137         while (count > 0) {
 138                 /*
 139                  * Handle first page in case it's not aligned
 140                  */
 141                 if (-p & (PAGE_SIZE - 1))
 142                         sz = -p & (PAGE_SIZE - 1);
 143                 else
 144                         sz = PAGE_SIZE;
 145
 146                 sz = min_t(unsigned long, sz, count);
 147
 148                 /*
 149                  * On ia64 if a page has been mapped somewhere as
 150                  * uncached, then it must also be accessed uncached
 151                  * by the kernel or data corruption may occur
 152                  */
 153                 ptr = xlate_dev_mem_ptr(p);
 154
 155                 if (copy_to_user(buf, ptr, sz))
 156                         return -EFAULT;
 157                 buf += sz;
 158                 p += sz;
 159                 count -= sz;
 160                 read += sz;
 161         }
 162
 163         *ppos += read;
 164         return read;
 165 }
 166
 167 static ssize_t write_mem(struct file * file, const char __user * buf,
 168                          size_t count, loff_t *ppos)
 169 {
 170         unsigned long p = *ppos;
 171         ssize_t written, sz;
 172         unsigned long copied;
 173         void *ptr;
 174
 175         if (!valid_phys_addr_range(p, count))
 176                 return -EFAULT;
 177
 178         written = 0;
 179
 180 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 181         /* we don't have page 0 mapped on sparc and m68k.. */
 182         if (p < PAGE_SIZE) {
 183                 unsigned long sz = PAGE_SIZE - p;
 184                 if (sz > count)
 185                         sz = count;
 186                 /* Hmm. Do something? */
 187                 buf += sz;
 188                 p += sz;
 189                 count -= sz;
 190                 written += sz;
 191         }
 192 #endif
 193
 194         while (count > 0) {
 195                 /*
 196                  * Handle first page in case it's not aligned
 197                  */
 198                 if (-p & (PAGE_SIZE - 1))
 199                         sz = -p & (PAGE_SIZE - 1);
 200                 else
 201                         sz = PAGE_SIZE;
 202
 203                 sz = min_t(unsigned long, sz, count);
 204
 205                 /*
 206                  * On ia64 if a page has been mapped somewhere as
 207                  * uncached, then it must also be accessed uncached
 208                  * by the kernel or data corruption may occur
 209                  */
 210                 ptr = xlate_dev_mem_ptr(p);
 211
 212                 copied = copy_from_user(ptr, buf, sz);
 213                 if (copied) {
 214                         written += sz - copied;
 215                         if (written)
 216                                 break;
 217                         return -EFAULT;
 218                 }
 219                 buf += sz;
 220                 p += sz;
 221                 count -= sz;
 222                 written += sz;
 223         }
 224
 225         *ppos += written;
 226         return written;
 227 }
 228
 229 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 230 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 231                                      unsigned long size, pgprot_t vma_prot)
 232 {
 233 #ifdef pgprot_noncached
 234         unsigned long offset = pfn << PAGE_SHIFT;
 235
 236         if (uncached_access(file, offset))
 237                 return pgprot_noncached(vma_prot);
 238 #endif
 239         return vma_prot;
 240 }
 241 #endif
 242
 243 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 244 {
 245         size_t size = vma->vm_end - vma->vm_start;
 246
 247         if (!valid_mmap_phys_addr_range(vma->vm_pgoff << PAGE_SHIFT, size))
 248                 return -EINVAL;
 249
 250         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 251                                                  size,
 252                                                  vma->vm_page_prot);
 253
 254         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 255         if (remap_pfn_range(vma,
 256                             vma->vm_start,
 257                             vma->vm_pgoff,
 258                             size,
 259                             vma->vm_page_prot))
 260                 return -EAGAIN;
 261         return 0;
 262 }
 263
 264 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 265 {
 266         unsigned long pfn;
 267
 268         /* Turn a kernel-virtual address into a physical page frame */
 269         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 270
 271         /*
 272          * RED-PEN: on some architectures there is more mapped memory
 273          * than available in mem_map which pfn_valid checks
 274          * for. Perhaps should add a new macro here.
 275          *
 276          * RED-PEN: vmalloc is not supported right now.
 277          */
 278         if (!pfn_valid(pfn))
 279                 return -EIO;
 280
 281         vma->vm_pgoff = pfn;
 282         return mmap_mem(file, vma);
 283 }
 284
 285 #ifdef CONFIG_CRASH_DUMP
 286 /*
 287  * Read memory corresponding to the old kernel.
 288  */
 289 static ssize_t read_oldmem(struct file *file, char __user *buf,
 290                                 size_t count, loff_t *ppos)
 291 {
 292         unsigned long pfn, offset;
 293         size_t read = 0, csize;
 294         int rc = 0;
 295
 296         while (count) {
 297                 pfn = *ppos / PAGE_SIZE;
 298                 if (pfn > saved_max_pfn)
 299                         return read;
 300
 301                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 302                 if (count > PAGE_SIZE - offset)
 303                         csize = PAGE_SIZE - offset;
 304                 else
 305                         csize = count;
 306
 307                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 308                 if (rc < 0)
 309                         return rc;
 310                 buf += csize;
 311                 *ppos += csize;
 312                 read += csize;
 313                 count -= csize;
 314         }
 315         return read;
 316 }
 317 #endif
 318
 319 extern long vread(char *buf, char *addr, unsigned long count);
 320 extern long vwrite(char *buf, char *addr, unsigned long count);
 321
 322 /*
 323  * This function reads the *virtual* memory as seen by the kernel.
 324  */
 325 static ssize_t read_kmem(struct file *file, char __user *buf,
 326                          size_t count, loff_t *ppos)
 327 {
 328         unsigned long p = *ppos;
 329         ssize_t low_count, read, sz;
 330         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 331
 332         read = 0;
 333         if (p < (unsigned long) high_memory) {
 334                 low_count = count;
 335                 if (count > (unsigned long) high_memory - p)
 336                         low_count = (unsigned long) high_memory - p;
 337
 338 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 339                 /* we don't have page 0 mapped on sparc and m68k.. */
 340                 if (p < PAGE_SIZE && low_count > 0) {
 341                         size_t tmp = PAGE_SIZE - p;
 342                         if (tmp > low_count) tmp = low_count;
 343                         if (clear_user(buf, tmp))
 344                                 return -EFAULT;
 345                         buf += tmp;
 346                         p += tmp;
 347                         read += tmp;
 348                         low_count -= tmp;
 349                         count -= tmp;
 350                 }
 351 #endif
 352                 while (low_count > 0) {
 353                         /*
 354                          * Handle first page in case it's not aligned
 355                          */
 356                         if (-p & (PAGE_SIZE - 1))
 357                                 sz = -p & (PAGE_SIZE - 1);
 358                         else
 359                                 sz = PAGE_SIZE;
 360
 361                         sz = min_t(unsigned long, sz, low_count);
 362
 363                         /*
 364                          * On ia64 if a page has been mapped somewhere as
 365                          * uncached, then it must also be accessed uncached
 366                          * by the kernel or data corruption may occur
 367                          */
 368                         kbuf = xlate_dev_kmem_ptr((char *)p);
 369
 370                         if (copy_to_user(buf, kbuf, sz))
 371                                 return -EFAULT;
 372                         buf += sz;
 373                         p += sz;
 374                         read += sz;
 375                         low_count -= sz;
 376                         count -= sz;
 377                 }
 378         }
 379
 380         if (count > 0) {
 381                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 382                 if (!kbuf)
 383                         return -ENOMEM;
 384                 while (count > 0) {
 385                         int len = count;
 386
 387                         if (len > PAGE_SIZE)
 388                                 len = PAGE_SIZE;
 389                         len = vread(kbuf, (char *)p, len);
 390                         if (!len)
 391                                 break;
 392                         if (copy_to_user(buf, kbuf, len)) {
 393                                 free_page((unsigned long)kbuf);
 394                                 return -EFAULT;
 395                         }
 396                         count -= len;
 397                         buf += len;
 398                         read += len;
 399                         p += len;
 400                 }
 401                 free_page((unsigned long)kbuf);
 402         }
 403         *ppos = p;
 404         return read;
 405 }
 406
 407
 408 static inline ssize_t
 409 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 410               size_t count, loff_t *ppos)
 411 {
 412         ssize_t written, sz;
 413         unsigned long copied;
 414
 415         written = 0;
 416 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 417         /* we don't have page 0 mapped on sparc and m68k.. */
 418         if (realp < PAGE_SIZE) {
 419                 unsigned long sz = PAGE_SIZE - realp;
 420                 if (sz > count)
 421                         sz = count;
 422                 /* Hmm. Do something? */
 423                 buf += sz;
 424                 p += sz;
 425                 realp += sz;
 426                 count -= sz;
 427                 written += sz;
 428         }
 429 #endif
 430
 431         while (count > 0) {
 432                 char *ptr;
 433                 /*
 434                  * Handle first page in case it's not aligned
 435                  */
 436                 if (-realp & (PAGE_SIZE - 1))
 437                         sz = -realp & (PAGE_SIZE - 1);
 438                 else
 439                         sz = PAGE_SIZE;
 440
 441                 sz = min_t(unsigned long, sz, count);
 442
 443                 /*
 444                  * On ia64 if a page has been mapped somewhere as
 445                  * uncached, then it must also be accessed uncached
 446                  * by the kernel or data corruption may occur
 447                  */
 448                 ptr = xlate_dev_kmem_ptr(p);
 449
 450                 copied = copy_from_user(ptr, buf, sz);
 451                 if (copied) {
 452                         written += sz - copied;
 453                         if (written)
 454                                 break;
 455                         return -EFAULT;
 456                 }
 457                 buf += sz;
 458                 p += sz;
 459                 realp += sz;
 460                 count -= sz;
 461                 written += sz;
 462         }
 463
 464         *ppos += written;
 465         return written;
 466 }
 467
 468
 469 /*
 470  * This function writes to the *virtual* memory as seen by the kernel.
 471  */
 472 static ssize_t write_kmem(struct file * file, const char __user * buf,
 473                           size_t count, loff_t *ppos)
 474 {
 475         unsigned long p = *ppos;
 476         ssize_t wrote = 0;
 477         ssize_t virtr = 0;
 478         ssize_t written;
 479         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 480
 481         if (p < (unsigned long) high_memory) {
 482
 483                 wrote = count;
 484                 if (count > (unsigned long) high_memory - p)
 485                         wrote = (unsigned long) high_memory - p;
 486
 487                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 488                 if (written != wrote)
 489                         return written;
 490                 wrote = written;
 491                 p += wrote;
 492                 buf += wrote;
 493                 count -= wrote;
 494         }
 495
 496         if (count > 0) {
 497                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 498                 if (!kbuf)
 499                         return wrote ? wrote : -ENOMEM;
 500                 while (count > 0) {
 501                         int len = count;
 502
 503                         if (len > PAGE_SIZE)
 504                                 len = PAGE_SIZE;
 505                         if (len) {
 506                                 written = copy_from_user(kbuf, buf, len);
 507                                 if (written) {
 508                                         if (wrote + virtr)
 509                                                 break;
 510                                         free_page((unsigned long)kbuf);
 511                                         return -EFAULT;
 512                                 }
 513                         }
 514                         len = vwrite(kbuf, (char *)p, len);
 515                         count -= len;
 516                         buf += len;
 517                         virtr += len;
 518                         p += len;
 519                 }
 520                 free_page((unsigned long)kbuf);
 521         }
 522
 523         *ppos = p;
 524         return virtr + wrote;
 525 }
 526
 527 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 528 static ssize_t read_port(struct file * file, char __user * buf,
 529                          size_t count, loff_t *ppos)
 530 {
 531         unsigned long i = *ppos;
 532         char __user *tmp = buf;
 533
 534         if (!access_ok(VERIFY_WRITE, buf, count))
 535                 return -EFAULT;
 536         while (count-- > 0 && i < 65536) {
 537                 if (__put_user(inb(i),tmp) < 0)
 538                         return -EFAULT;
 539                 i++;
 540                 tmp++;
 541         }
 542         *ppos = i;
 543         return tmp-buf;
 544 }
 545
 546 static ssize_t write_port(struct file * file, const char __user * buf,
 547                           size_t count, loff_t *ppos)
 548 {
 549         unsigned long i = *ppos;
 550         const char __user * tmp = buf;
 551
 552         if (!access_ok(VERIFY_READ,buf,count))
 553                 return -EFAULT;
 554         while (count-- > 0 && i < 65536) {
 555                 char c;
 556                 if (__get_user(c, tmp)) {
 557                         if (tmp > buf)
 558                                 break;
 559                         return -EFAULT;
 560                 }
 561                 outb(c,i);
 562                 i++;
 563                 tmp++;
 564         }
 565         *ppos = i;
 566         return tmp-buf;
 567 }
 568 #endif
 569
 570 static ssize_t read_null(struct file * file, char __user * buf,
 571                          size_t count, loff_t *ppos)
 572 {
 573         return 0;
 574 }
 575
 576 static ssize_t write_null(struct file * file, const char __user * buf,
 577                           size_t count, loff_t *ppos)
 578 {
 579         return count;
 580 }
 581
 582 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 583                         struct splice_desc *sd)
 584 {
 585         return sd->len;
 586 }
 587
 588 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 589                                  loff_t *ppos, size_t len, unsigned int flags)
 590 {
 591         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 592 }
 593
 594 #ifdef CONFIG_MMU
 595 /*
 596  * For fun, we are using the MMU for this.
 597  */
 598 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
 599 {
 600         struct mm_struct *mm;
 601         struct vm_area_struct * vma;
 602         unsigned long addr=(unsigned long)buf;
 603
 604         mm = current->mm;
 605         /* Oops, this was forgotten before. -ben */
 606         down_read(&mm->mmap_sem);
 607
 608         /* For private mappings, just map in zero pages. */
 609         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 610                 unsigned long count;
 611
 612                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 613                         goto out_up;
 614                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
 615                         break;
 616                 count = vma->vm_end - addr;
 617                 if (count > size)
 618                         count = size;
 619
 620                 zap_page_range(vma, addr, count, NULL);
 621                 zeromap_page_range(vma, addr, count, PAGE_COPY);
 622
 623                 size -= count;
 624                 buf += count;
 625                 addr += count;
 626                 if (size == 0)
 627                         goto out_up;
 628         }
 629
 630         up_read(&mm->mmap_sem);
 631
 632         /* The shared case is hard. Let's do the conventional zeroing. */
 633         do {
 634                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 635                 if (unwritten)
 636                         return size + unwritten - PAGE_SIZE;
 637                 cond_resched();
 638                 buf += PAGE_SIZE;
 639                 size -= PAGE_SIZE;
 640         } while (size);
 641
 642         return size;
 643 out_up:
 644         up_read(&mm->mmap_sem);
 645         return size;
 646 }
 647
 648 static ssize_t read_zero(struct file * file, char __user * buf,
 649                          size_t count, loff_t *ppos)
 650 {
 651         unsigned long left, unwritten, written = 0;
 652
 653         if (!count)
 654                 return 0;
 655
 656         if (!access_ok(VERIFY_WRITE, buf, count))
 657                 return -EFAULT;
 658
 659         left = count;
 660
 661         /* do we want to be clever? Arbitrary cut-off */
 662         if (count >= PAGE_SIZE*4) {
 663                 unsigned long partial;
 664
 665                 /* How much left of the page? */
 666                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 667                 unwritten = clear_user(buf, partial);
 668                 written = partial - unwritten;
 669                 if (unwritten)
 670                         goto out;
 671                 left -= partial;
 672                 buf += partial;
 673                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 674                 written += (left & PAGE_MASK) - unwritten;
 675                 if (unwritten)
 676                         goto out;
 677                 buf += left & PAGE_MASK;
 678                 left &= ~PAGE_MASK;
 679         }
 680         unwritten = clear_user(buf, left);
 681         written += left - unwritten;
 682 out:
 683         return written ? written : -EFAULT;
 684 }
 685
 686 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 687 {
 688         if (vma->vm_flags & VM_SHARED)
 689                 return shmem_zero_setup(vma);
 690         if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
 691                 return -EAGAIN;
 692         return 0;
 693 }
 694 #else /* CONFIG_MMU */
 695 static ssize_t read_zero(struct file * file, char * buf,
 696                          size_t count, loff_t *ppos)
 697 {
 698         size_t todo = count;
 699
 700         while (todo) {
 701                 size_t chunk = todo;
 702
 703                 if (chunk > 4096)
 704                         chunk = 4096;   /* Just for latency reasons */
 705                 if (clear_user(buf, chunk))
 706                         return -EFAULT;
 707                 buf += chunk;
 708                 todo -= chunk;
 709                 cond_resched();
 710         }
 711         return count;
 712 }
 713
 714 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 715 {
 716         return -ENOSYS;
 717 }
 718 #endif /* CONFIG_MMU */
 719
 720 static ssize_t write_full(struct file * file, const char __user * buf,
 721                           size_t count, loff_t *ppos)
 722 {
 723         return -ENOSPC;
 724 }
 725
 726 /*
 727  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 728  * can fopen() both devices with "a" now.  This was previously impossible.
 729  * -- SRB.
 730  */
 731
 732 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 733 {
 734         return file->f_pos = 0;
 735 }
 736
 737 /*
 738  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 739  * check against negative addresses: they are ok. The return value is weird,
 740  * though, in that case (0).
 741  *
 742  * also note that seeking relative to the "end of file" isn't supported:
 743  * it has no meaning, so it returns -EINVAL.
 744  */
 745 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 746 {
 747         loff_t ret;
 748
 749         mutex_lock(&file->f_dentry->d_inode->i_mutex);
 750         switch (orig) {
 751                 case 0:
 752                         file->f_pos = offset;
 753                         ret = file->f_pos;
 754                         force_successful_syscall_return();
 755                         break;
 756                 case 1:
 757                         file->f_pos += offset;
 758                         ret = file->f_pos;
 759                         force_successful_syscall_return();
 760                         break;
 761                 default:
 762                         ret = -EINVAL;
 763         }
 764         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 765         return ret;
 766 }
 767
 768 static int open_port(struct inode * inode, struct file * filp)
 769 {
 770         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 771 }
 772
 773 #define zero_lseek      null_lseek
 774 #define full_lseek      null_lseek
 775 #define write_zero      write_null
 776 #define read_full       read_zero
 777 #define open_mem        open_port
 778 #define open_kmem       open_mem
 779 #define open_oldmem     open_mem
 780
 781 static struct file_operations mem_fops = {
 782         .llseek         = memory_lseek,
 783         .read           = read_mem,
 784         .write          = write_mem,
 785         .mmap           = mmap_mem,
 786         .open           = open_mem,
 787 };
 788
 789 static struct file_operations kmem_fops = {
 790         .llseek         = memory_lseek,
 791         .read           = read_kmem,
 792         .write          = write_kmem,
 793         .mmap           = mmap_kmem,
 794         .open           = open_kmem,
 795 };
 796
 797 static struct file_operations null_fops = {
 798         .llseek         = null_lseek,
 799         .read           = read_null,
 800         .write          = write_null,
 801         .splice_write   = splice_write_null,
 802 };
 803
 804 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 805 static struct file_operations port_fops = {
 806         .llseek         = memory_lseek,
 807         .read           = read_port,
 808         .write          = write_port,
 809         .open           = open_port,
 810 };
 811 #endif
 812
 813 static struct file_operations zero_fops = {
 814         .llseek         = zero_lseek,
 815         .read           = read_zero,
 816         .write          = write_zero,
 817         .mmap           = mmap_zero,
 818 };
 819
 820 static struct backing_dev_info zero_bdi = {
 821         .capabilities   = BDI_CAP_MAP_COPY,
 822 };
 823
 824 static struct file_operations full_fops = {
 825         .llseek         = full_lseek,
 826         .read           = read_full,
 827         .write          = write_full,
 828 };
 829
 830 #ifdef CONFIG_CRASH_DUMP
 831 static struct file_operations oldmem_fops = {
 832         .read   = read_oldmem,
 833         .open   = open_oldmem,
 834 };
 835 #endif
 836
 837 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 838                           size_t count, loff_t *ppos)
 839 {
 840         char *tmp;
 841         ssize_t ret;
 842
 843         tmp = kmalloc(count + 1, GFP_KERNEL);
 844         if (tmp == NULL)
 845                 return -ENOMEM;
 846         ret = -EFAULT;
 847         if (!copy_from_user(tmp, buf, count)) {
 848                 tmp[count] = 0;
 849                 ret = printk("%s", tmp);
 850                 if (ret > count)
 851                         /* printk can add a prefix */
 852                         ret = count;
 853         }
 854         kfree(tmp);
 855         return ret;
 856 }
 857
 858 static struct file_operations kmsg_fops = {
 859         .write =        kmsg_write,
 860 };
 861
 862 static int memory_open(struct inode * inode, struct file * filp)
 863 {
 864         switch (iminor(inode)) {
 865                 case 1:
 866                         filp->f_op = &mem_fops;
 867                         break;
 868                 case 2:
 869                         filp->f_op = &kmem_fops;
 870                         break;
 871                 case 3:
 872                         filp->f_op = &null_fops;
 873                         break;
 874 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 875                 case 4:
 876                         filp->f_op = &port_fops;
 877                         break;
 878 #endif
 879                 case 5:
 880                         filp->f_mapping->backing_dev_info = &zero_bdi;
 881                         filp->f_op = &zero_fops;
 882                         break;
 883                 case 7:
 884                         filp->f_op = &full_fops;
 885                         break;
 886                 case 8:
 887                         filp->f_op = &random_fops;
 888                         break;
 889                 case 9:
 890                         filp->f_op = &urandom_fops;
 891                         break;
 892                 case 11:
 893                         filp->f_op = &kmsg_fops;
 894                         break;
 895 #ifdef CONFIG_CRASH_DUMP
 896                 case 12:
 897                         filp->f_op = &oldmem_fops;
 898                         break;
 899 #endif
 900                 default:
 901                         return -ENXIO;
 902         }
 903         if (filp->f_op && filp->f_op->open)
 904                 return filp->f_op->open(inode,filp);
 905         return 0;
 906 }
 907
 908 static struct file_operations memory_fops = {
 909         .open           = memory_open,  /* just a selector for the real open */
 910 };
 911
 912 static const struct {
 913         unsigned int            minor;
 914         char                    *name;
 915         umode_t                 mode;
 916         const struct file_operations    *fops;
 917 } devlist[] = { /* list of minor devices */
 918         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 919         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 920         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 921 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 922         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 923 #endif
 924         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 925         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 926         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 927         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 928         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 929 #ifdef CONFIG_CRASH_DUMP
 930         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 931 #endif
 932 };
 933
 934 static struct class *mem_class;
 935
 936 static int __init chr_dev_init(void)
 937 {
 938         int i;
 939
 940         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 941                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 942
 943         mem_class = class_create(THIS_MODULE, "mem");
 944         for (i = 0; i < ARRAY_SIZE(devlist); i++) {
 945                 class_device_create(mem_class, NULL,
 946                                         MKDEV(MEM_MAJOR, devlist[i].minor),
 947                                         NULL, devlist[i].name);
 948                 devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor),
 949                                 S_IFCHR | devlist[i].mode, devlist[i].name);
 950         }
 951
 952         return 0;
 953 }
 954
 955 fs_initcall(chr_dev_init);