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/smp_lock.h>
  22 #include <linux/ptrace.h>
  23 #include <linux/device.h>
  24 #include <linux/highmem.h>
  25 #include <linux/crash_dump.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/bootmem.h>
  28 #include <linux/pipe_fs_i.h>
  29
  30 #include <asm/uaccess.h>
  31 #include <asm/io.h>
  32
  33 #ifdef CONFIG_IA64
  34 # include <linux/efi.h>
  35 #endif
  36
  37 /*
  38  * Architectures vary in how they handle caching for addresses
  39  * outside of main memory.
  40  *
  41  */
  42 static inline int uncached_access(struct file *file, unsigned long addr)
  43 {
  44 #if defined(__i386__)
  45         /*
  46          * On the PPro and successors, the MTRRs are used to set
  47          * memory types for physical addresses outside main memory,
  48          * so blindly setting PCD or PWT on those pages is wrong.
  49          * For Pentiums and earlier, the surround logic should disable
  50          * caching for the high addresses through the KEN pin, but
  51          * we maintain the tradition of paranoia in this code.
  52          */
  53         if (file->f_flags & O_SYNC)
  54                 return 1;
  55         return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
  56                   test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
  57                   test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
  58                   test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
  59           && addr >= __pa(high_memory);
  60 #elif defined(__x86_64__)
  61         /*
  62          * This is broken because it can generate memory type aliases,
  63          * which can cause cache corruptions
  64          * But it is only available for root and we have to be bug-to-bug
  65          * compatible with i386.
  66          */
  67         if (file->f_flags & O_SYNC)
  68                 return 1;
  69         /* same behaviour as i386. PAT always set to cached and MTRRs control the
  70            caching behaviour.
  71            Hopefully a full PAT implementation will fix that soon. */
  72         return 0;
  73 #elif defined(CONFIG_IA64)
  74         /*
  75          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  76          */
  77         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  78 #else
  79         /*
  80          * Accessing memory above the top the kernel knows about or through a file pointer
  81          * that was marked O_SYNC will be done non-cached.
  82          */
  83         if (file->f_flags & O_SYNC)
  84                 return 1;
  85         return addr >= __pa(high_memory);
  86 #endif
  87 }
  88
  89 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  90 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  91 {
  92         if (addr + count > __pa(high_memory))
  93                 return 0;
  94
  95         return 1;
  96 }
  97
  98 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  99 {
 100         return 1;
 101 }
 102 #endif
 103
 104 /*
 105  * This funcion reads the *physical* memory. The f_pos points directly to the
 106  * memory location.
 107  */
 108 static ssize_t read_mem(struct file * file, char __user * buf,
 109                         size_t count, loff_t *ppos)
 110 {
 111         unsigned long p = *ppos;
 112         ssize_t read, sz;
 113         char *ptr;
 114
 115         if (!valid_phys_addr_range(p, count))
 116                 return -EFAULT;
 117         read = 0;
 118 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 119         /* we don't have page 0 mapped on sparc and m68k.. */
 120         if (p < PAGE_SIZE) {
 121                 sz = PAGE_SIZE - p;
 122                 if (sz > count)
 123                         sz = count;
 124                 if (sz > 0) {
 125                         if (clear_user(buf, sz))
 126                                 return -EFAULT;
 127                         buf += sz;
 128                         p += sz;
 129                         count -= sz;
 130                         read += sz;
 131                 }
 132         }
 133 #endif
 134
 135         while (count > 0) {
 136                 /*
 137                  * Handle first page in case it's not aligned
 138                  */
 139                 if (-p & (PAGE_SIZE - 1))
 140                         sz = -p & (PAGE_SIZE - 1);
 141                 else
 142                         sz = PAGE_SIZE;
 143
 144                 sz = min_t(unsigned long, sz, count);
 145
 146                 /*
 147                  * On ia64 if a page has been mapped somewhere as
 148                  * uncached, then it must also be accessed uncached
 149                  * by the kernel or data corruption may occur
 150                  */
 151                 ptr = xlate_dev_mem_ptr(p);
 152
 153                 if (copy_to_user(buf, ptr, sz))
 154                         return -EFAULT;
 155                 buf += sz;
 156                 p += sz;
 157                 count -= sz;
 158                 read += sz;
 159         }
 160
 161         *ppos += read;
 162         return read;
 163 }
 164
 165 static ssize_t write_mem(struct file * file, const char __user * buf,
 166                          size_t count, loff_t *ppos)
 167 {
 168         unsigned long p = *ppos;
 169         ssize_t written, sz;
 170         unsigned long copied;
 171         void *ptr;
 172
 173         if (!valid_phys_addr_range(p, count))
 174                 return -EFAULT;
 175
 176         written = 0;
 177
 178 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 179         /* we don't have page 0 mapped on sparc and m68k.. */
 180         if (p < PAGE_SIZE) {
 181                 unsigned long sz = PAGE_SIZE - p;
 182                 if (sz > count)
 183                         sz = count;
 184                 /* Hmm. Do something? */
 185                 buf += sz;
 186                 p += sz;
 187                 count -= sz;
 188                 written += sz;
 189         }
 190 #endif
 191
 192         while (count > 0) {
 193                 /*
 194                  * Handle first page in case it's not aligned
 195                  */
 196                 if (-p & (PAGE_SIZE - 1))
 197                         sz = -p & (PAGE_SIZE - 1);
 198                 else
 199                         sz = PAGE_SIZE;
 200
 201                 sz = min_t(unsigned long, sz, count);
 202
 203                 /*
 204                  * On ia64 if a page has been mapped somewhere as
 205                  * uncached, then it must also be accessed uncached
 206                  * by the kernel or data corruption may occur
 207                  */
 208                 ptr = xlate_dev_mem_ptr(p);
 209
 210                 copied = copy_from_user(ptr, buf, sz);
 211                 if (copied) {
 212                         written += sz - copied;
 213                         if (written)
 214                                 break;
 215                         return -EFAULT;
 216                 }
 217                 buf += sz;
 218                 p += sz;
 219                 count -= sz;
 220                 written += sz;
 221         }
 222
 223         *ppos += written;
 224         return written;
 225 }
 226
 227 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 228 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 229                                      unsigned long size, pgprot_t vma_prot)
 230 {
 231 #ifdef pgprot_noncached
 232         unsigned long offset = pfn << PAGE_SHIFT;
 233
 234         if (uncached_access(file, offset))
 235                 return pgprot_noncached(vma_prot);
 236 #endif
 237         return vma_prot;
 238 }
 239 #endif
 240
 241 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 242 {
 243         size_t size = vma->vm_end - vma->vm_start;
 244
 245         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 246                 return -EINVAL;
 247
 248         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 249                                                  size,
 250                                                  vma->vm_page_prot);
 251
 252         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 253         if (remap_pfn_range(vma,
 254                             vma->vm_start,
 255                             vma->vm_pgoff,
 256                             size,
 257                             vma->vm_page_prot))
 258                 return -EAGAIN;
 259         return 0;
 260 }
 261
 262 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 263 {
 264         unsigned long pfn;
 265
 266         /* Turn a kernel-virtual address into a physical page frame */
 267         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 268
 269         /*
 270          * RED-PEN: on some architectures there is more mapped memory
 271          * than available in mem_map which pfn_valid checks
 272          * for. Perhaps should add a new macro here.
 273          *
 274          * RED-PEN: vmalloc is not supported right now.
 275          */
 276         if (!pfn_valid(pfn))
 277                 return -EIO;
 278
 279         vma->vm_pgoff = pfn;
 280         return mmap_mem(file, vma);
 281 }
 282
 283 #ifdef CONFIG_CRASH_DUMP
 284 /*
 285  * Read memory corresponding to the old kernel.
 286  */
 287 static ssize_t read_oldmem(struct file *file, char __user *buf,
 288                                 size_t count, loff_t *ppos)
 289 {
 290         unsigned long pfn, offset;
 291         size_t read = 0, csize;
 292         int rc = 0;
 293
 294         while (count) {
 295                 pfn = *ppos / PAGE_SIZE;
 296                 if (pfn > saved_max_pfn)
 297                         return read;
 298
 299                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 300                 if (count > PAGE_SIZE - offset)
 301                         csize = PAGE_SIZE - offset;
 302                 else
 303                         csize = count;
 304
 305                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 306                 if (rc < 0)
 307                         return rc;
 308                 buf += csize;
 309                 *ppos += csize;
 310                 read += csize;
 311                 count -= csize;
 312         }
 313         return read;
 314 }
 315 #endif
 316
 317 extern long vread(char *buf, char *addr, unsigned long count);
 318 extern long vwrite(char *buf, char *addr, unsigned long count);
 319
 320 /*
 321  * This function reads the *virtual* memory as seen by the kernel.
 322  */
 323 static ssize_t read_kmem(struct file *file, char __user *buf,
 324                          size_t count, loff_t *ppos)
 325 {
 326         unsigned long p = *ppos;
 327         ssize_t low_count, read, sz;
 328         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 329
 330         read = 0;
 331         if (p < (unsigned long) high_memory) {
 332                 low_count = count;
 333                 if (count > (unsigned long) high_memory - p)
 334                         low_count = (unsigned long) high_memory - p;
 335
 336 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 337                 /* we don't have page 0 mapped on sparc and m68k.. */
 338                 if (p < PAGE_SIZE && low_count > 0) {
 339                         size_t tmp = PAGE_SIZE - p;
 340                         if (tmp > low_count) tmp = low_count;
 341                         if (clear_user(buf, tmp))
 342                                 return -EFAULT;
 343                         buf += tmp;
 344                         p += tmp;
 345                         read += tmp;
 346                         low_count -= tmp;
 347                         count -= tmp;
 348                 }
 349 #endif
 350                 while (low_count > 0) {
 351                         /*
 352                          * Handle first page in case it's not aligned
 353                          */
 354                         if (-p & (PAGE_SIZE - 1))
 355                                 sz = -p & (PAGE_SIZE - 1);
 356                         else
 357                                 sz = PAGE_SIZE;
 358
 359                         sz = min_t(unsigned long, sz, low_count);
 360
 361                         /*
 362                          * On ia64 if a page has been mapped somewhere as
 363                          * uncached, then it must also be accessed uncached
 364                          * by the kernel or data corruption may occur
 365                          */
 366                         kbuf = xlate_dev_kmem_ptr((char *)p);
 367
 368                         if (copy_to_user(buf, kbuf, sz))
 369                                 return -EFAULT;
 370                         buf += sz;
 371                         p += sz;
 372                         read += sz;
 373                         low_count -= sz;
 374                         count -= sz;
 375                 }
 376         }
 377
 378         if (count > 0) {
 379                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 380                 if (!kbuf)
 381                         return -ENOMEM;
 382                 while (count > 0) {
 383                         int len = count;
 384
 385                         if (len > PAGE_SIZE)
 386                                 len = PAGE_SIZE;
 387                         len = vread(kbuf, (char *)p, len);
 388                         if (!len)
 389                                 break;
 390                         if (copy_to_user(buf, kbuf, len)) {
 391                                 free_page((unsigned long)kbuf);
 392                                 return -EFAULT;
 393                         }
 394                         count -= len;
 395                         buf += len;
 396                         read += len;
 397                         p += len;
 398                 }
 399                 free_page((unsigned long)kbuf);
 400         }
 401         *ppos = p;
 402         return read;
 403 }
 404
 405
 406 static inline ssize_t
 407 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 408               size_t count, loff_t *ppos)
 409 {
 410         ssize_t written, sz;
 411         unsigned long copied;
 412
 413         written = 0;
 414 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 415         /* we don't have page 0 mapped on sparc and m68k.. */
 416         if (realp < PAGE_SIZE) {
 417                 unsigned long sz = PAGE_SIZE - realp;
 418                 if (sz > count)
 419                         sz = count;
 420                 /* Hmm. Do something? */
 421                 buf += sz;
 422                 p += sz;
 423                 realp += sz;
 424                 count -= sz;
 425                 written += sz;
 426         }
 427 #endif
 428
 429         while (count > 0) {
 430                 char *ptr;
 431                 /*
 432                  * Handle first page in case it's not aligned
 433                  */
 434                 if (-realp & (PAGE_SIZE - 1))
 435                         sz = -realp & (PAGE_SIZE - 1);
 436                 else
 437                         sz = PAGE_SIZE;
 438
 439                 sz = min_t(unsigned long, sz, count);
 440
 441                 /*
 442                  * On ia64 if a page has been mapped somewhere as
 443                  * uncached, then it must also be accessed uncached
 444                  * by the kernel or data corruption may occur
 445                  */
 446                 ptr = xlate_dev_kmem_ptr(p);
 447
 448                 copied = copy_from_user(ptr, buf, sz);
 449                 if (copied) {
 450                         written += sz - copied;
 451                         if (written)
 452                                 break;
 453                         return -EFAULT;
 454                 }
 455                 buf += sz;
 456                 p += sz;
 457                 realp += sz;
 458                 count -= sz;
 459                 written += sz;
 460         }
 461
 462         *ppos += written;
 463         return written;
 464 }
 465
 466
 467 /*
 468  * This function writes to the *virtual* memory as seen by the kernel.
 469  */
 470 static ssize_t write_kmem(struct file * file, const char __user * buf,
 471                           size_t count, loff_t *ppos)
 472 {
 473         unsigned long p = *ppos;
 474         ssize_t wrote = 0;
 475         ssize_t virtr = 0;
 476         ssize_t written;
 477         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 478
 479         if (p < (unsigned long) high_memory) {
 480
 481                 wrote = count;
 482                 if (count > (unsigned long) high_memory - p)
 483                         wrote = (unsigned long) high_memory - p;
 484
 485                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 486                 if (written != wrote)
 487                         return written;
 488                 wrote = written;
 489                 p += wrote;
 490                 buf += wrote;
 491                 count -= wrote;
 492         }
 493
 494         if (count > 0) {
 495                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 496                 if (!kbuf)
 497                         return wrote ? wrote : -ENOMEM;
 498                 while (count > 0) {
 499                         int len = count;
 500
 501                         if (len > PAGE_SIZE)
 502                                 len = PAGE_SIZE;
 503                         if (len) {
 504                                 written = copy_from_user(kbuf, buf, len);
 505                                 if (written) {
 506                                         if (wrote + virtr)
 507                                                 break;
 508                                         free_page((unsigned long)kbuf);
 509                                         return -EFAULT;
 510                                 }
 511                         }
 512                         len = vwrite(kbuf, (char *)p, len);
 513                         count -= len;
 514                         buf += len;
 515                         virtr += len;
 516                         p += len;
 517                 }
 518                 free_page((unsigned long)kbuf);
 519         }
 520
 521         *ppos = p;
 522         return virtr + wrote;
 523 }
 524
 525 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 526 static ssize_t read_port(struct file * file, char __user * buf,
 527                          size_t count, loff_t *ppos)
 528 {
 529         unsigned long i = *ppos;
 530         char __user *tmp = buf;
 531
 532         if (!access_ok(VERIFY_WRITE, buf, count))
 533                 return -EFAULT;
 534         while (count-- > 0 && i < 65536) {
 535                 if (__put_user(inb(i),tmp) < 0)
 536                         return -EFAULT;
 537                 i++;
 538                 tmp++;
 539         }
 540         *ppos = i;
 541         return tmp-buf;
 542 }
 543
 544 static ssize_t write_port(struct file * file, const char __user * buf,
 545                           size_t count, loff_t *ppos)
 546 {
 547         unsigned long i = *ppos;
 548         const char __user * tmp = buf;
 549
 550         if (!access_ok(VERIFY_READ,buf,count))
 551                 return -EFAULT;
 552         while (count-- > 0 && i < 65536) {
 553                 char c;
 554                 if (__get_user(c, tmp)) {
 555                         if (tmp > buf)
 556                                 break;
 557                         return -EFAULT;
 558                 }
 559                 outb(c,i);
 560                 i++;
 561                 tmp++;
 562         }
 563         *ppos = i;
 564         return tmp-buf;
 565 }
 566 #endif
 567
 568 static ssize_t read_null(struct file * file, char __user * buf,
 569                          size_t count, loff_t *ppos)
 570 {
 571         return 0;
 572 }
 573
 574 static ssize_t write_null(struct file * file, const char __user * buf,
 575                           size_t count, loff_t *ppos)
 576 {
 577         return count;
 578 }
 579
 580 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 581                         struct splice_desc *sd)
 582 {
 583         return sd->len;
 584 }
 585
 586 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 587                                  loff_t *ppos, size_t len, unsigned int flags)
 588 {
 589         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 590 }
 591
 592 #ifdef CONFIG_MMU
 593 /*
 594  * For fun, we are using the MMU for this.
 595  */
 596 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
 597 {
 598         struct mm_struct *mm;
 599         struct vm_area_struct * vma;
 600         unsigned long addr=(unsigned long)buf;
 601
 602         mm = current->mm;
 603         /* Oops, this was forgotten before. -ben */
 604         down_read(&mm->mmap_sem);
 605
 606         /* For private mappings, just map in zero pages. */
 607         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 608                 unsigned long count;
 609
 610                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 611                         goto out_up;
 612                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
 613                         break;
 614                 count = vma->vm_end - addr;
 615                 if (count > size)
 616                         count = size;
 617
 618                 zap_page_range(vma, addr, count, NULL);
 619                 if (zeromap_page_range(vma, addr, count, PAGE_COPY))
 620                         break;
 621
 622                 size -= count;
 623                 buf += count;
 624                 addr += count;
 625                 if (size == 0)
 626                         goto out_up;
 627         }
 628
 629         up_read(&mm->mmap_sem);
 630
 631         /* The shared case is hard. Let's do the conventional zeroing. */
 632         do {
 633                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 634                 if (unwritten)
 635                         return size + unwritten - PAGE_SIZE;
 636                 cond_resched();
 637                 buf += PAGE_SIZE;
 638                 size -= PAGE_SIZE;
 639         } while (size);
 640
 641         return size;
 642 out_up:
 643         up_read(&mm->mmap_sem);
 644         return size;
 645 }
 646
 647 static ssize_t read_zero(struct file * file, char __user * buf,
 648                          size_t count, loff_t *ppos)
 649 {
 650         unsigned long left, unwritten, written = 0;
 651
 652         if (!count)
 653                 return 0;
 654
 655         if (!access_ok(VERIFY_WRITE, buf, count))
 656                 return -EFAULT;
 657
 658         left = count;
 659
 660         /* do we want to be clever? Arbitrary cut-off */
 661         if (count >= PAGE_SIZE*4) {
 662                 unsigned long partial;
 663
 664                 /* How much left of the page? */
 665                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 666                 unwritten = clear_user(buf, partial);
 667                 written = partial - unwritten;
 668                 if (unwritten)
 669                         goto out;
 670                 left -= partial;
 671                 buf += partial;
 672                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 673                 written += (left & PAGE_MASK) - unwritten;
 674                 if (unwritten)
 675                         goto out;
 676                 buf += left & PAGE_MASK;
 677                 left &= ~PAGE_MASK;
 678         }
 679         unwritten = clear_user(buf, left);
 680         written += left - unwritten;
 681 out:
 682         return written ? written : -EFAULT;
 683 }
 684
 685 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 686 {
 687         int err;
 688
 689         if (vma->vm_flags & VM_SHARED)
 690                 return shmem_zero_setup(vma);
 691         err = zeromap_page_range(vma, vma->vm_start,
 692                         vma->vm_end - vma->vm_start, vma->vm_page_prot);
 693         BUG_ON(err == -EEXIST);
 694         return err;
 695 }
 696 #else /* CONFIG_MMU */
 697 static ssize_t read_zero(struct file * file, char * buf,
 698                          size_t count, loff_t *ppos)
 699 {
 700         size_t todo = count;
 701
 702         while (todo) {
 703                 size_t chunk = todo;
 704
 705                 if (chunk > 4096)
 706                         chunk = 4096;   /* Just for latency reasons */
 707                 if (clear_user(buf, chunk))
 708                         return -EFAULT;
 709                 buf += chunk;
 710                 todo -= chunk;
 711                 cond_resched();
 712         }
 713         return count;
 714 }
 715
 716 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 717 {
 718         return -ENOSYS;
 719 }
 720 #endif /* CONFIG_MMU */
 721
 722 static ssize_t write_full(struct file * file, const char __user * buf,
 723                           size_t count, loff_t *ppos)
 724 {
 725         return -ENOSPC;
 726 }
 727
 728 /*
 729  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 730  * can fopen() both devices with "a" now.  This was previously impossible.
 731  * -- SRB.
 732  */
 733
 734 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 735 {
 736         return file->f_pos = 0;
 737 }
 738
 739 /*
 740  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 741  * check against negative addresses: they are ok. The return value is weird,
 742  * though, in that case (0).
 743  *
 744  * also note that seeking relative to the "end of file" isn't supported:
 745  * it has no meaning, so it returns -EINVAL.
 746  */
 747 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 748 {
 749         loff_t ret;
 750
 751         mutex_lock(&file->f_dentry->d_inode->i_mutex);
 752         switch (orig) {
 753                 case 0:
 754                         file->f_pos = offset;
 755                         ret = file->f_pos;
 756                         force_successful_syscall_return();
 757                         break;
 758                 case 1:
 759                         file->f_pos += offset;
 760                         ret = file->f_pos;
 761                         force_successful_syscall_return();
 762                         break;
 763                 default:
 764                         ret = -EINVAL;
 765         }
 766         mutex_unlock(&file->f_dentry->d_inode->i_mutex);
 767         return ret;
 768 }
 769
 770 static int open_port(struct inode * inode, struct file * filp)
 771 {
 772         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 773 }
 774
 775 #define zero_lseek      null_lseek
 776 #define full_lseek      null_lseek
 777 #define write_zero      write_null
 778 #define read_full       read_zero
 779 #define open_mem        open_port
 780 #define open_kmem       open_mem
 781 #define open_oldmem     open_mem
 782
 783 static const struct file_operations mem_fops = {
 784         .llseek         = memory_lseek,
 785         .read           = read_mem,
 786         .write          = write_mem,
 787         .mmap           = mmap_mem,
 788         .open           = open_mem,
 789 };
 790
 791 static const struct file_operations kmem_fops = {
 792         .llseek         = memory_lseek,
 793         .read           = read_kmem,
 794         .write          = write_kmem,
 795         .mmap           = mmap_kmem,
 796         .open           = open_kmem,
 797 };
 798
 799 static const struct file_operations null_fops = {
 800         .llseek         = null_lseek,
 801         .read           = read_null,
 802         .write          = write_null,
 803         .splice_write   = splice_write_null,
 804 };
 805
 806 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 807 static const struct file_operations port_fops = {
 808         .llseek         = memory_lseek,
 809         .read           = read_port,
 810         .write          = write_port,
 811         .open           = open_port,
 812 };
 813 #endif
 814
 815 static const struct file_operations zero_fops = {
 816         .llseek         = zero_lseek,
 817         .read           = read_zero,
 818         .write          = write_zero,
 819         .mmap           = mmap_zero,
 820 };
 821
 822 static struct backing_dev_info zero_bdi = {
 823         .capabilities   = BDI_CAP_MAP_COPY,
 824 };
 825
 826 static const struct file_operations full_fops = {
 827         .llseek         = full_lseek,
 828         .read           = read_full,
 829         .write          = write_full,
 830 };
 831
 832 #ifdef CONFIG_CRASH_DUMP
 833 static const struct file_operations oldmem_fops = {
 834         .read   = read_oldmem,
 835         .open   = open_oldmem,
 836 };
 837 #endif
 838
 839 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 840                           size_t count, loff_t *ppos)
 841 {
 842         char *tmp;
 843         ssize_t ret;
 844
 845         tmp = kmalloc(count + 1, GFP_KERNEL);
 846         if (tmp == NULL)
 847                 return -ENOMEM;
 848         ret = -EFAULT;
 849         if (!copy_from_user(tmp, buf, count)) {
 850                 tmp[count] = 0;
 851                 ret = printk("%s", tmp);
 852                 if (ret > count)
 853                         /* printk can add a prefix */
 854                         ret = count;
 855         }
 856         kfree(tmp);
 857         return ret;
 858 }
 859
 860 static const struct file_operations kmsg_fops = {
 861         .write =        kmsg_write,
 862 };
 863
 864 static int memory_open(struct inode * inode, struct file * filp)
 865 {
 866         switch (iminor(inode)) {
 867                 case 1:
 868                         filp->f_op = &mem_fops;
 869                         break;
 870                 case 2:
 871                         filp->f_op = &kmem_fops;
 872                         break;
 873                 case 3:
 874                         filp->f_op = &null_fops;
 875                         break;
 876 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 877                 case 4:
 878                         filp->f_op = &port_fops;
 879                         break;
 880 #endif
 881                 case 5:
 882                         filp->f_mapping->backing_dev_info = &zero_bdi;
 883                         filp->f_op = &zero_fops;
 884                         break;
 885                 case 7:
 886                         filp->f_op = &full_fops;
 887                         break;
 888                 case 8:
 889                         filp->f_op = &random_fops;
 890                         break;
 891                 case 9:
 892                         filp->f_op = &urandom_fops;
 893                         break;
 894                 case 11:
 895                         filp->f_op = &kmsg_fops;
 896                         break;
 897 #ifdef CONFIG_CRASH_DUMP
 898                 case 12:
 899                         filp->f_op = &oldmem_fops;
 900                         break;
 901 #endif
 902                 default:
 903                         return -ENXIO;
 904         }
 905         if (filp->f_op && filp->f_op->open)
 906                 return filp->f_op->open(inode,filp);
 907         return 0;
 908 }
 909
 910 static const struct file_operations memory_fops = {
 911         .open           = memory_open,  /* just a selector for the real open */
 912 };
 913
 914 static const struct {
 915         unsigned int            minor;
 916         char                    *name;
 917         umode_t                 mode;
 918         const struct file_operations    *fops;
 919 } devlist[] = { /* list of minor devices */
 920         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 921         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 922         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 923 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 924         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 925 #endif
 926         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 927         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 928         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 929         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 930         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 931 #ifdef CONFIG_CRASH_DUMP
 932         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 933 #endif
 934 };
 935
 936 static struct class *mem_class;
 937
 938 static int __init chr_dev_init(void)
 939 {
 940         int i;
 941
 942         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 943                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 944
 945         mem_class = class_create(THIS_MODULE, "mem");
 946         for (i = 0; i < ARRAY_SIZE(devlist); i++)
 947                 class_device_create(mem_class, NULL,
 948                                         MKDEV(MEM_MAJOR, devlist[i].minor),
 949                                         NULL, devlist[i].name);
 950
 951         return 0;
 952 }
 953
 954 fs_initcall(chr_dev_init);