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 mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
   9  */
  10
  11 #include <linux/mm.h>
  12 #include <linux/miscdevice.h>
  13 #include <linux/slab.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/mman.h>
  16 #include <linux/random.h>
  17 #include <linux/init.h>
  18 #include <linux/raw.h>
  19 #include <linux/tty.h>
  20 #include <linux/capability.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/device.h>
  23 #include <linux/highmem.h>
  24 #include <linux/crash_dump.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/splice.h>
  28 #include <linux/pfn.h>
  29
  30 #include <asm/uaccess.h>
  31 #include <asm/io.h>
  32
  33 #ifdef CONFIG_IA64
  34 # include <linux/efi.h>
  35 #endif
  36
  37 static inline unsigned long size_inside_page(unsigned long start,
  38                                              unsigned long size)
  39 {
  40         unsigned long sz;
  41
  42         if (-start & (PAGE_SIZE - 1))
  43                 sz = -start & (PAGE_SIZE - 1);
  44         else
  45                 sz = PAGE_SIZE;
  46
  47         return min_t(unsigned long, sz, size);
  48 }
  49
  50 /*
  51  * Architectures vary in how they handle caching for addresses
  52  * outside of main memory.
  53  *
  54  */
  55 static inline int uncached_access(struct file *file, unsigned long addr)
  56 {
  57 #if defined(CONFIG_IA64)
  58         /*
  59          * On ia64, we ignore O_DSYNC because we cannot tolerate memory attribute aliases.
  60          */
  61         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  62 #elif defined(CONFIG_MIPS)
  63         {
  64                 extern int __uncached_access(struct file *file,
  65                                              unsigned long addr);
  66
  67                 return __uncached_access(file, addr);
  68         }
  69 #else
  70         /*
  71          * Accessing memory above the top the kernel knows about or through a file pointer
  72          * that was marked O_DSYNC will be done non-cached.
  73          */
  74         if (file->f_flags & O_DSYNC)
  75                 return 1;
  76         return addr >= __pa(high_memory);
  77 #endif
  78 }
  79
  80 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  81 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  82 {
  83         if (addr + count > __pa(high_memory))
  84                 return 0;
  85
  86         return 1;
  87 }
  88
  89 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  90 {
  91         return 1;
  92 }
  93 #endif
  94
  95 #ifdef CONFIG_STRICT_DEVMEM
  96 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  97 {
  98         u64 from = ((u64)pfn) << PAGE_SHIFT;
  99         u64 to = from + size;
 100         u64 cursor = from;
 101
 102         while (cursor < to) {
 103                 if (!devmem_is_allowed(pfn)) {
 104                         printk(KERN_INFO
 105                 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
 106                                 current->comm, from, to);
 107                         return 0;
 108                 }
 109                 cursor += PAGE_SIZE;
 110                 pfn++;
 111         }
 112         return 1;
 113 }
 114 #else
 115 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 116 {
 117         return 1;
 118 }
 119 #endif
 120
 121 void __attribute__((weak)) unxlate_dev_mem_ptr(unsigned long phys, void *addr)
 122 {
 123 }
 124
 125 /*
 126  * This funcion reads the *physical* memory. The f_pos points directly to the
 127  * memory location.
 128  */
 129 static ssize_t read_mem(struct file * file, char __user * buf,
 130                         size_t count, loff_t *ppos)
 131 {
 132         unsigned long p = *ppos;
 133         ssize_t read, sz;
 134         char *ptr;
 135
 136         if (!valid_phys_addr_range(p, count))
 137                 return -EFAULT;
 138         read = 0;
 139 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 140         /* we don't have page 0 mapped on sparc and m68k.. */
 141         if (p < PAGE_SIZE) {
 142                 sz = PAGE_SIZE - p;
 143                 if (sz > count)
 144                         sz = count;
 145                 if (sz > 0) {
 146                         if (clear_user(buf, sz))
 147                                 return -EFAULT;
 148                         buf += sz;
 149                         p += sz;
 150                         count -= sz;
 151                         read += sz;
 152                 }
 153         }
 154 #endif
 155
 156         while (count > 0) {
 157                 unsigned long remaining;
 158
 159                 sz = size_inside_page(p, count);
 160
 161                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 162                         return -EPERM;
 163
 164                 /*
 165                  * On ia64 if a page has been mapped somewhere as
 166                  * uncached, then it must also be accessed uncached
 167                  * by the kernel or data corruption may occur
 168                  */
 169                 ptr = xlate_dev_mem_ptr(p);
 170                 if (!ptr)
 171                         return -EFAULT;
 172
 173                 remaining = copy_to_user(buf, ptr, sz);
 174                 unxlate_dev_mem_ptr(p, ptr);
 175                 if (remaining)
 176                         return -EFAULT;
 177
 178                 buf += sz;
 179                 p += sz;
 180                 count -= sz;
 181                 read += sz;
 182         }
 183
 184         *ppos += read;
 185         return read;
 186 }
 187
 188 static ssize_t write_mem(struct file * file, const char __user * buf,
 189                          size_t count, loff_t *ppos)
 190 {
 191         unsigned long p = *ppos;
 192         ssize_t written, sz;
 193         unsigned long copied;
 194         void *ptr;
 195
 196         if (!valid_phys_addr_range(p, count))
 197                 return -EFAULT;
 198
 199         written = 0;
 200
 201 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 202         /* we don't have page 0 mapped on sparc and m68k.. */
 203         if (p < PAGE_SIZE) {
 204                 unsigned long sz = PAGE_SIZE - p;
 205                 if (sz > count)
 206                         sz = count;
 207                 /* Hmm. Do something? */
 208                 buf += sz;
 209                 p += sz;
 210                 count -= sz;
 211                 written += sz;
 212         }
 213 #endif
 214
 215         while (count > 0) {
 216                 sz = size_inside_page(p, count);
 217
 218                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 219                         return -EPERM;
 220
 221                 /*
 222                  * On ia64 if a page has been mapped somewhere as
 223                  * uncached, then it must also be accessed uncached
 224                  * by the kernel or data corruption may occur
 225                  */
 226                 ptr = xlate_dev_mem_ptr(p);
 227                 if (!ptr) {
 228                         if (written)
 229                                 break;
 230                         return -EFAULT;
 231                 }
 232
 233                 copied = copy_from_user(ptr, buf, sz);
 234                 unxlate_dev_mem_ptr(p, ptr);
 235                 if (copied) {
 236                         written += sz - copied;
 237                         if (written)
 238                                 break;
 239                         return -EFAULT;
 240                 }
 241
 242                 buf += sz;
 243                 p += sz;
 244                 count -= sz;
 245                 written += sz;
 246         }
 247
 248         *ppos += written;
 249         return written;
 250 }
 251
 252 int __attribute__((weak)) phys_mem_access_prot_allowed(struct file *file,
 253         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 254 {
 255         return 1;
 256 }
 257
 258 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 259 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 260                                      unsigned long size, pgprot_t vma_prot)
 261 {
 262 #ifdef pgprot_noncached
 263         unsigned long offset = pfn << PAGE_SHIFT;
 264
 265         if (uncached_access(file, offset))
 266                 return pgprot_noncached(vma_prot);
 267 #endif
 268         return vma_prot;
 269 }
 270 #endif
 271
 272 #ifndef CONFIG_MMU
 273 static unsigned long get_unmapped_area_mem(struct file *file,
 274                                            unsigned long addr,
 275                                            unsigned long len,
 276                                            unsigned long pgoff,
 277                                            unsigned long flags)
 278 {
 279         if (!valid_mmap_phys_addr_range(pgoff, len))
 280                 return (unsigned long) -EINVAL;
 281         return pgoff << PAGE_SHIFT;
 282 }
 283
 284 /* can't do an in-place private mapping if there's no MMU */
 285 static inline int private_mapping_ok(struct vm_area_struct *vma)
 286 {
 287         return vma->vm_flags & VM_MAYSHARE;
 288 }
 289 #else
 290 #define get_unmapped_area_mem   NULL
 291
 292 static inline int private_mapping_ok(struct vm_area_struct *vma)
 293 {
 294         return 1;
 295 }
 296 #endif
 297
 298 static const struct vm_operations_struct mmap_mem_ops = {
 299 #ifdef CONFIG_HAVE_IOREMAP_PROT
 300         .access = generic_access_phys
 301 #endif
 302 };
 303
 304 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 305 {
 306         size_t size = vma->vm_end - vma->vm_start;
 307
 308         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 309                 return -EINVAL;
 310
 311         if (!private_mapping_ok(vma))
 312                 return -ENOSYS;
 313
 314         if (!range_is_allowed(vma->vm_pgoff, size))
 315                 return -EPERM;
 316
 317         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 318                                                 &vma->vm_page_prot))
 319                 return -EINVAL;
 320
 321         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 322                                                  size,
 323                                                  vma->vm_page_prot);
 324
 325         vma->vm_ops = &mmap_mem_ops;
 326
 327         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 328         if (remap_pfn_range(vma,
 329                             vma->vm_start,
 330                             vma->vm_pgoff,
 331                             size,
 332                             vma->vm_page_prot)) {
 333                 return -EAGAIN;
 334         }
 335         return 0;
 336 }
 337
 338 #ifdef CONFIG_DEVKMEM
 339 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 340 {
 341         unsigned long pfn;
 342
 343         /* Turn a kernel-virtual address into a physical page frame */
 344         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 345
 346         /*
 347          * RED-PEN: on some architectures there is more mapped memory
 348          * than available in mem_map which pfn_valid checks
 349          * for. Perhaps should add a new macro here.
 350          *
 351          * RED-PEN: vmalloc is not supported right now.
 352          */
 353         if (!pfn_valid(pfn))
 354                 return -EIO;
 355
 356         vma->vm_pgoff = pfn;
 357         return mmap_mem(file, vma);
 358 }
 359 #endif
 360
 361 #ifdef CONFIG_CRASH_DUMP
 362 /*
 363  * Read memory corresponding to the old kernel.
 364  */
 365 static ssize_t read_oldmem(struct file *file, char __user *buf,
 366                                 size_t count, loff_t *ppos)
 367 {
 368         unsigned long pfn, offset;
 369         size_t read = 0, csize;
 370         int rc = 0;
 371
 372         while (count) {
 373                 pfn = *ppos / PAGE_SIZE;
 374                 if (pfn > saved_max_pfn)
 375                         return read;
 376
 377                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 378                 if (count > PAGE_SIZE - offset)
 379                         csize = PAGE_SIZE - offset;
 380                 else
 381                         csize = count;
 382
 383                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 384                 if (rc < 0)
 385                         return rc;
 386                 buf += csize;
 387                 *ppos += csize;
 388                 read += csize;
 389                 count -= csize;
 390         }
 391         return read;
 392 }
 393 #endif
 394
 395 #ifdef CONFIG_DEVKMEM
 396 /*
 397  * This function reads the *virtual* memory as seen by the kernel.
 398  */
 399 static ssize_t read_kmem(struct file *file, char __user *buf,
 400                          size_t count, loff_t *ppos)
 401 {
 402         unsigned long p = *ppos;
 403         ssize_t low_count, read, sz;
 404         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 405
 406         read = 0;
 407         if (p < (unsigned long) high_memory) {
 408                 low_count = count;
 409                 if (count > (unsigned long) high_memory - p)
 410                         low_count = (unsigned long) high_memory - p;
 411
 412 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 413                 /* we don't have page 0 mapped on sparc and m68k.. */
 414                 if (p < PAGE_SIZE && low_count > 0) {
 415                         size_t tmp = PAGE_SIZE - p;
 416                         if (tmp > low_count) tmp = low_count;
 417                         if (clear_user(buf, tmp))
 418                                 return -EFAULT;
 419                         buf += tmp;
 420                         p += tmp;
 421                         read += tmp;
 422                         low_count -= tmp;
 423                         count -= tmp;
 424                 }
 425 #endif
 426                 while (low_count > 0) {
 427                         sz = size_inside_page(p, low_count);
 428
 429                         /*
 430                          * On ia64 if a page has been mapped somewhere as
 431                          * uncached, then it must also be accessed uncached
 432                          * by the kernel or data corruption may occur
 433                          */
 434                         kbuf = xlate_dev_kmem_ptr((char *)p);
 435
 436                         if (copy_to_user(buf, kbuf, sz))
 437                                 return -EFAULT;
 438                         buf += sz;
 439                         p += sz;
 440                         read += sz;
 441                         low_count -= sz;
 442                         count -= sz;
 443                 }
 444         }
 445
 446         if (count > 0) {
 447                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 448                 if (!kbuf)
 449                         return -ENOMEM;
 450                 while (count > 0) {
 451                         int len = size_inside_page(p, count);
 452
 453                         len = vread(kbuf, (char *)p, len);
 454                         if (!len)
 455                                 break;
 456                         if (copy_to_user(buf, kbuf, len)) {
 457                                 free_page((unsigned long)kbuf);
 458                                 return -EFAULT;
 459                         }
 460                         count -= len;
 461                         buf += len;
 462                         read += len;
 463                         p += len;
 464                 }
 465                 free_page((unsigned long)kbuf);
 466         }
 467         *ppos = p;
 468         return read;
 469 }
 470
 471
 472 static inline ssize_t
 473 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 474               size_t count, loff_t *ppos)
 475 {
 476         ssize_t written, sz;
 477         unsigned long copied;
 478
 479         written = 0;
 480 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 481         /* we don't have page 0 mapped on sparc and m68k.. */
 482         if (realp < PAGE_SIZE) {
 483                 unsigned long sz = PAGE_SIZE - realp;
 484                 if (sz > count)
 485                         sz = count;
 486                 /* Hmm. Do something? */
 487                 buf += sz;
 488                 p += sz;
 489                 realp += sz;
 490                 count -= sz;
 491                 written += sz;
 492         }
 493 #endif
 494
 495         while (count > 0) {
 496                 char *ptr;
 497
 498                 sz = size_inside_page(realp, count);
 499
 500                 /*
 501                  * On ia64 if a page has been mapped somewhere as
 502                  * uncached, then it must also be accessed uncached
 503                  * by the kernel or data corruption may occur
 504                  */
 505                 ptr = xlate_dev_kmem_ptr(p);
 506
 507                 copied = copy_from_user(ptr, buf, sz);
 508                 if (copied) {
 509                         written += sz - copied;
 510                         if (written)
 511                                 break;
 512                         return -EFAULT;
 513                 }
 514                 buf += sz;
 515                 p += sz;
 516                 realp += sz;
 517                 count -= sz;
 518                 written += sz;
 519         }
 520
 521         *ppos += written;
 522         return written;
 523 }
 524
 525
 526 /*
 527  * This function writes to the *virtual* memory as seen by the kernel.
 528  */
 529 static ssize_t write_kmem(struct file * file, const char __user * buf,
 530                           size_t count, loff_t *ppos)
 531 {
 532         unsigned long p = *ppos;
 533         ssize_t wrote = 0;
 534         ssize_t virtr = 0;
 535         ssize_t written;
 536         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 537
 538         if (p < (unsigned long) high_memory) {
 539
 540                 wrote = count;
 541                 if (count > (unsigned long) high_memory - p)
 542                         wrote = (unsigned long) high_memory - p;
 543
 544                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 545                 if (written != wrote)
 546                         return written;
 547                 wrote = written;
 548                 p += wrote;
 549                 buf += wrote;
 550                 count -= wrote;
 551         }
 552
 553         if (count > 0) {
 554                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 555                 if (!kbuf)
 556                         return wrote ? wrote : -ENOMEM;
 557                 while (count > 0) {
 558                         int len = size_inside_page(p, count);
 559
 560                         written = copy_from_user(kbuf, buf, len);
 561                         if (written) {
 562                                 if (wrote + virtr)
 563                                         break;
 564                                 free_page((unsigned long)kbuf);
 565                                 return -EFAULT;
 566                         }
 567                         len = vwrite(kbuf, (char *)p, len);
 568                         count -= len;
 569                         buf += len;
 570                         virtr += len;
 571                         p += len;
 572                 }
 573                 free_page((unsigned long)kbuf);
 574         }
 575
 576         *ppos = p;
 577         return virtr + wrote;
 578 }
 579 #endif
 580
 581 #ifdef CONFIG_DEVPORT
 582 static ssize_t read_port(struct file * file, char __user * buf,
 583                          size_t count, loff_t *ppos)
 584 {
 585         unsigned long i = *ppos;
 586         char __user *tmp = buf;
 587
 588         if (!access_ok(VERIFY_WRITE, buf, count))
 589                 return -EFAULT;
 590         while (count-- > 0 && i < 65536) {
 591                 if (__put_user(inb(i),tmp) < 0)
 592                         return -EFAULT;
 593                 i++;
 594                 tmp++;
 595         }
 596         *ppos = i;
 597         return tmp-buf;
 598 }
 599
 600 static ssize_t write_port(struct file * file, const char __user * buf,
 601                           size_t count, loff_t *ppos)
 602 {
 603         unsigned long i = *ppos;
 604         const char __user * tmp = buf;
 605
 606         if (!access_ok(VERIFY_READ,buf,count))
 607                 return -EFAULT;
 608         while (count-- > 0 && i < 65536) {
 609                 char c;
 610                 if (__get_user(c, tmp)) {
 611                         if (tmp > buf)
 612                                 break;
 613                         return -EFAULT;
 614                 }
 615                 outb(c,i);
 616                 i++;
 617                 tmp++;
 618         }
 619         *ppos = i;
 620         return tmp-buf;
 621 }
 622 #endif
 623
 624 static ssize_t read_null(struct file * file, char __user * buf,
 625                          size_t count, loff_t *ppos)
 626 {
 627         return 0;
 628 }
 629
 630 static ssize_t write_null(struct file * file, const char __user * buf,
 631                           size_t count, loff_t *ppos)
 632 {
 633         return count;
 634 }
 635
 636 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 637                         struct splice_desc *sd)
 638 {
 639         return sd->len;
 640 }
 641
 642 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 643                                  loff_t *ppos, size_t len, unsigned int flags)
 644 {
 645         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 646 }
 647
 648 static ssize_t read_zero(struct file * file, char __user * buf,
 649                          size_t count, loff_t *ppos)
 650 {
 651         size_t written;
 652
 653         if (!count)
 654                 return 0;
 655
 656         if (!access_ok(VERIFY_WRITE, buf, count))
 657                 return -EFAULT;
 658
 659         written = 0;
 660         while (count) {
 661                 unsigned long unwritten;
 662                 size_t chunk = count;
 663
 664                 if (chunk > PAGE_SIZE)
 665                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 666                 unwritten = __clear_user(buf, chunk);
 667                 written += chunk - unwritten;
 668                 if (unwritten)
 669                         break;
 670                 if (signal_pending(current))
 671                         return written ? written : -ERESTARTSYS;
 672                 buf += chunk;
 673                 count -= chunk;
 674                 cond_resched();
 675         }
 676         return written ? written : -EFAULT;
 677 }
 678
 679 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 680 {
 681 #ifndef CONFIG_MMU
 682         return -ENOSYS;
 683 #endif
 684         if (vma->vm_flags & VM_SHARED)
 685                 return shmem_zero_setup(vma);
 686         return 0;
 687 }
 688
 689 static ssize_t write_full(struct file * file, const char __user * buf,
 690                           size_t count, loff_t *ppos)
 691 {
 692         return -ENOSPC;
 693 }
 694
 695 /*
 696  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 697  * can fopen() both devices with "a" now.  This was previously impossible.
 698  * -- SRB.
 699  */
 700
 701 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 702 {
 703         return file->f_pos = 0;
 704 }
 705
 706 /*
 707  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 708  * check against negative addresses: they are ok. The return value is weird,
 709  * though, in that case (0).
 710  *
 711  * also note that seeking relative to the "end of file" isn't supported:
 712  * it has no meaning, so it returns -EINVAL.
 713  */
 714 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 715 {
 716         loff_t ret;
 717
 718         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 719         switch (orig) {
 720                 case 0:
 721                         file->f_pos = offset;
 722                         ret = file->f_pos;
 723                         force_successful_syscall_return();
 724                         break;
 725                 case 1:
 726                         file->f_pos += offset;
 727                         ret = file->f_pos;
 728                         force_successful_syscall_return();
 729                         break;
 730                 default:
 731                         ret = -EINVAL;
 732         }
 733         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 734         return ret;
 735 }
 736
 737 static int open_port(struct inode * inode, struct file * filp)
 738 {
 739         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 740 }
 741
 742 #define zero_lseek      null_lseek
 743 #define full_lseek      null_lseek
 744 #define write_zero      write_null
 745 #define read_full       read_zero
 746 #define open_mem        open_port
 747 #define open_kmem       open_mem
 748 #define open_oldmem     open_mem
 749
 750 static const struct file_operations mem_fops = {
 751         .llseek         = memory_lseek,
 752         .read           = read_mem,
 753         .write          = write_mem,
 754         .mmap           = mmap_mem,
 755         .open           = open_mem,
 756         .get_unmapped_area = get_unmapped_area_mem,
 757 };
 758
 759 #ifdef CONFIG_DEVKMEM
 760 static const struct file_operations kmem_fops = {
 761         .llseek         = memory_lseek,
 762         .read           = read_kmem,
 763         .write          = write_kmem,
 764         .mmap           = mmap_kmem,
 765         .open           = open_kmem,
 766         .get_unmapped_area = get_unmapped_area_mem,
 767 };
 768 #endif
 769
 770 static const struct file_operations null_fops = {
 771         .llseek         = null_lseek,
 772         .read           = read_null,
 773         .write          = write_null,
 774         .splice_write   = splice_write_null,
 775 };
 776
 777 #ifdef CONFIG_DEVPORT
 778 static const struct file_operations port_fops = {
 779         .llseek         = memory_lseek,
 780         .read           = read_port,
 781         .write          = write_port,
 782         .open           = open_port,
 783 };
 784 #endif
 785
 786 static const struct file_operations zero_fops = {
 787         .llseek         = zero_lseek,
 788         .read           = read_zero,
 789         .write          = write_zero,
 790         .mmap           = mmap_zero,
 791 };
 792
 793 /*
 794  * capabilities for /dev/zero
 795  * - permits private mappings, "copies" are taken of the source of zeros
 796  */
 797 static struct backing_dev_info zero_bdi = {
 798         .name           = "char/mem",
 799         .capabilities   = BDI_CAP_MAP_COPY,
 800 };
 801
 802 static const struct file_operations full_fops = {
 803         .llseek         = full_lseek,
 804         .read           = read_full,
 805         .write          = write_full,
 806 };
 807
 808 #ifdef CONFIG_CRASH_DUMP
 809 static const struct file_operations oldmem_fops = {
 810         .read   = read_oldmem,
 811         .open   = open_oldmem,
 812 };
 813 #endif
 814
 815 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 816                           size_t count, loff_t *ppos)
 817 {
 818         char *tmp;
 819         ssize_t ret;
 820
 821         tmp = kmalloc(count + 1, GFP_KERNEL);
 822         if (tmp == NULL)
 823                 return -ENOMEM;
 824         ret = -EFAULT;
 825         if (!copy_from_user(tmp, buf, count)) {
 826                 tmp[count] = 0;
 827                 ret = printk("%s", tmp);
 828                 if (ret > count)
 829                         /* printk can add a prefix */
 830                         ret = count;
 831         }
 832         kfree(tmp);
 833         return ret;
 834 }
 835
 836 static const struct file_operations kmsg_fops = {
 837         .write =        kmsg_write,
 838 };
 839
 840 static const struct memdev {
 841         const char *name;
 842         mode_t mode;
 843         const struct file_operations *fops;
 844         struct backing_dev_info *dev_info;
 845 } devlist[] = {
 846          [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
 847 #ifdef CONFIG_DEVKMEM
 848          [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
 849 #endif
 850          [3] = { "null", 0666, &null_fops, NULL },
 851 #ifdef CONFIG_DEVPORT
 852          [4] = { "port", 0, &port_fops, NULL },
 853 #endif
 854          [5] = { "zero", 0666, &zero_fops, &zero_bdi },
 855          [7] = { "full", 0666, &full_fops, NULL },
 856          [8] = { "random", 0666, &random_fops, NULL },
 857          [9] = { "urandom", 0666, &urandom_fops, NULL },
 858         [11] = { "kmsg", 0, &kmsg_fops, NULL },
 859 #ifdef CONFIG_CRASH_DUMP
 860         [12] = { "oldmem", 0, &oldmem_fops, NULL },
 861 #endif
 862 };
 863
 864 static int memory_open(struct inode *inode, struct file *filp)
 865 {
 866         int minor;
 867         const struct memdev *dev;
 868
 869         minor = iminor(inode);
 870         if (minor >= ARRAY_SIZE(devlist))
 871                 return -ENXIO;
 872
 873         dev = &devlist[minor];
 874         if (!dev->fops)
 875                 return -ENXIO;
 876
 877         filp->f_op = dev->fops;
 878         if (dev->dev_info)
 879                 filp->f_mapping->backing_dev_info = dev->dev_info;
 880
 881         if (dev->fops->open)
 882                 return dev->fops->open(inode, filp);
 883
 884         return 0;
 885 }
 886
 887 static const struct file_operations memory_fops = {
 888         .open           = memory_open,
 889 };
 890
 891 static char *mem_devnode(struct device *dev, mode_t *mode)
 892 {
 893         if (mode && devlist[MINOR(dev->devt)].mode)
 894                 *mode = devlist[MINOR(dev->devt)].mode;
 895         return NULL;
 896 }
 897
 898 static struct class *mem_class;
 899
 900 static int __init chr_dev_init(void)
 901 {
 902         int minor;
 903         int err;
 904
 905         err = bdi_init(&zero_bdi);
 906         if (err)
 907                 return err;
 908
 909         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 910                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 911
 912         mem_class = class_create(THIS_MODULE, "mem");
 913         mem_class->devnode = mem_devnode;
 914         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 915                 if (!devlist[minor].name)
 916                         continue;
 917                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 918                               NULL, devlist[minor].name);
 919         }
 920
 921         return 0;
 922 }
 923
 924 fs_initcall(chr_dev_init);