drivers/char/mem.c

   1 /*
   2  *  linux/drivers/char/mem.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *
   6  *  Added devfs support.
   7  *    Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
   8  *  Shared /dev/zero mmaping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
   9  */
  10
  11 #include <linux/mm.h>
  12 #include <linux/miscdevice.h>
  13 #include <linux/slab.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/mman.h>
  16 #include <linux/random.h>
  17 #include <linux/init.h>
  18 #include <linux/raw.h>
  19 #include <linux/tty.h>
  20 #include <linux/capability.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/device.h>
  23 #include <linux/highmem.h>
  24 #include <linux/crash_dump.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/splice.h>
  28 #include <linux/pfn.h>
  29 #include <linux/smp_lock.h>
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/io.h>
  33
  34 #ifdef CONFIG_IA64
  35 # include <linux/efi.h>
  36 #endif
  37
  38 static inline unsigned long size_inside_page(unsigned long start,
  39                                              unsigned long size)
  40 {
  41         unsigned long sz;
  42
  43         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  44
  45         return min(sz, size);
  46 }
  47
  48 /*
  49  * Architectures vary in how they handle caching for addresses
  50  * outside of main memory.
  51  *
  52  */
  53 static inline int uncached_access(struct file *file, unsigned long addr)
  54 {
  55 #if defined(CONFIG_IA64)
  56         /*
  57          * On ia64, we ignore O_DSYNC because we cannot tolerate memory attribute aliases.
  58          */
  59         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  60 #elif defined(CONFIG_MIPS)
  61         {
  62                 extern int __uncached_access(struct file *file,
  63                                              unsigned long addr);
  64
  65                 return __uncached_access(file, addr);
  66         }
  67 #else
  68         /*
  69          * Accessing memory above the top the kernel knows about or through a file pointer
  70          * that was marked O_DSYNC will be done non-cached.
  71          */
  72         if (file->f_flags & O_DSYNC)
  73                 return 1;
  74         return addr >= __pa(high_memory);
  75 #endif
  76 }
  77
  78 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  79 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  80 {
  81         if (addr + count > __pa(high_memory))
  82                 return 0;
  83
  84         return 1;
  85 }
  86
  87 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  88 {
  89         return 1;
  90 }
  91 #endif
  92
  93 #ifdef CONFIG_STRICT_DEVMEM
  94 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  95 {
  96         u64 from = ((u64)pfn) << PAGE_SHIFT;
  97         u64 to = from + size;
  98         u64 cursor = from;
  99
 100         while (cursor < to) {
 101                 if (!devmem_is_allowed(pfn)) {
 102                         printk(KERN_INFO
 103                 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
 104                                 current->comm, from, to);
 105                         return 0;
 106                 }
 107                 cursor += PAGE_SIZE;
 108                 pfn++;
 109         }
 110         return 1;
 111 }
 112 #else
 113 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 114 {
 115         return 1;
 116 }
 117 #endif
 118
 119 void __attribute__((weak)) unxlate_dev_mem_ptr(unsigned long phys, void *addr)
 120 {
 121 }
 122
 123 /*
 124  * This funcion reads the *physical* memory. The f_pos points directly to the
 125  * memory location.
 126  */
 127 static ssize_t read_mem(struct file * file, char __user * buf,
 128                         size_t count, loff_t *ppos)
 129 {
 130         unsigned long p = *ppos;
 131         ssize_t read, sz;
 132         char *ptr;
 133
 134         if (!valid_phys_addr_range(p, count))
 135                 return -EFAULT;
 136         read = 0;
 137 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 138         /* we don't have page 0 mapped on sparc and m68k.. */
 139         if (p < PAGE_SIZE) {
 140                 sz = size_inside_page(p, count);
 141                 if (sz > 0) {
 142                         if (clear_user(buf, sz))
 143                                 return -EFAULT;
 144                         buf += sz;
 145                         p += sz;
 146                         count -= sz;
 147                         read += sz;
 148                 }
 149         }
 150 #endif
 151
 152         while (count > 0) {
 153                 unsigned long remaining;
 154
 155                 sz = size_inside_page(p, count);
 156
 157                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 158                         return -EPERM;
 159
 160                 /*
 161                  * On ia64 if a page has been mapped somewhere as
 162                  * uncached, then it must also be accessed uncached
 163                  * by the kernel or data corruption may occur
 164                  */
 165                 ptr = xlate_dev_mem_ptr(p);
 166                 if (!ptr)
 167                         return -EFAULT;
 168
 169                 remaining = copy_to_user(buf, ptr, sz);
 170                 unxlate_dev_mem_ptr(p, ptr);
 171                 if (remaining)
 172                         return -EFAULT;
 173
 174                 buf += sz;
 175                 p += sz;
 176                 count -= sz;
 177                 read += sz;
 178         }
 179
 180         *ppos += read;
 181         return read;
 182 }
 183
 184 static ssize_t write_mem(struct file * file, const char __user * buf,
 185                          size_t count, loff_t *ppos)
 186 {
 187         unsigned long p = *ppos;
 188         ssize_t written, sz;
 189         unsigned long copied;
 190         void *ptr;
 191
 192         if (!valid_phys_addr_range(p, count))
 193                 return -EFAULT;
 194
 195         written = 0;
 196
 197 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 198         /* we don't have page 0 mapped on sparc and m68k.. */
 199         if (p < PAGE_SIZE) {
 200                 sz = size_inside_page(p, count);
 201                 /* Hmm. Do something? */
 202                 buf += sz;
 203                 p += sz;
 204                 count -= sz;
 205                 written += sz;
 206         }
 207 #endif
 208
 209         while (count > 0) {
 210                 sz = size_inside_page(p, count);
 211
 212                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 213                         return -EPERM;
 214
 215                 /*
 216                  * On ia64 if a page has been mapped somewhere as
 217                  * uncached, then it must also be accessed uncached
 218                  * by the kernel or data corruption may occur
 219                  */
 220                 ptr = xlate_dev_mem_ptr(p);
 221                 if (!ptr) {
 222                         if (written)
 223                                 break;
 224                         return -EFAULT;
 225                 }
 226
 227                 copied = copy_from_user(ptr, buf, sz);
 228                 unxlate_dev_mem_ptr(p, ptr);
 229                 if (copied) {
 230                         written += sz - copied;
 231                         if (written)
 232                                 break;
 233                         return -EFAULT;
 234                 }
 235
 236                 buf += sz;
 237                 p += sz;
 238                 count -= sz;
 239                 written += sz;
 240         }
 241
 242         *ppos += written;
 243         return written;
 244 }
 245
 246 int __attribute__((weak)) phys_mem_access_prot_allowed(struct file *file,
 247         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 248 {
 249         return 1;
 250 }
 251
 252 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 253 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 254                                      unsigned long size, pgprot_t vma_prot)
 255 {
 256 #ifdef pgprot_noncached
 257         unsigned long offset = pfn << PAGE_SHIFT;
 258
 259         if (uncached_access(file, offset))
 260                 return pgprot_noncached(vma_prot);
 261 #endif
 262         return vma_prot;
 263 }
 264 #endif
 265
 266 #ifndef CONFIG_MMU
 267 static unsigned long get_unmapped_area_mem(struct file *file,
 268                                            unsigned long addr,
 269                                            unsigned long len,
 270                                            unsigned long pgoff,
 271                                            unsigned long flags)
 272 {
 273         if (!valid_mmap_phys_addr_range(pgoff, len))
 274                 return (unsigned long) -EINVAL;
 275         return pgoff << PAGE_SHIFT;
 276 }
 277
 278 /* can't do an in-place private mapping if there's no MMU */
 279 static inline int private_mapping_ok(struct vm_area_struct *vma)
 280 {
 281         return vma->vm_flags & VM_MAYSHARE;
 282 }
 283 #else
 284 #define get_unmapped_area_mem   NULL
 285
 286 static inline int private_mapping_ok(struct vm_area_struct *vma)
 287 {
 288         return 1;
 289 }
 290 #endif
 291
 292 static const struct vm_operations_struct mmap_mem_ops = {
 293 #ifdef CONFIG_HAVE_IOREMAP_PROT
 294         .access = generic_access_phys
 295 #endif
 296 };
 297
 298 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 299 {
 300         size_t size = vma->vm_end - vma->vm_start;
 301
 302         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 303                 return -EINVAL;
 304
 305         if (!private_mapping_ok(vma))
 306                 return -ENOSYS;
 307
 308         if (!range_is_allowed(vma->vm_pgoff, size))
 309                 return -EPERM;
 310
 311         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 312                                                 &vma->vm_page_prot))
 313                 return -EINVAL;
 314
 315         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 316                                                  size,
 317                                                  vma->vm_page_prot);
 318
 319         vma->vm_ops = &mmap_mem_ops;
 320
 321         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 322         if (remap_pfn_range(vma,
 323                             vma->vm_start,
 324                             vma->vm_pgoff,
 325                             size,
 326                             vma->vm_page_prot)) {
 327                 return -EAGAIN;
 328         }
 329         return 0;
 330 }
 331
 332 #ifdef CONFIG_DEVKMEM
 333 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 334 {
 335         unsigned long pfn;
 336
 337         /* Turn a kernel-virtual address into a physical page frame */
 338         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 339
 340         /*
 341          * RED-PEN: on some architectures there is more mapped memory
 342          * than available in mem_map which pfn_valid checks
 343          * for. Perhaps should add a new macro here.
 344          *
 345          * RED-PEN: vmalloc is not supported right now.
 346          */
 347         if (!pfn_valid(pfn))
 348                 return -EIO;
 349
 350         vma->vm_pgoff = pfn;
 351         return mmap_mem(file, vma);
 352 }
 353 #endif
 354
 355 #ifdef CONFIG_CRASH_DUMP
 356 /*
 357  * Read memory corresponding to the old kernel.
 358  */
 359 static ssize_t read_oldmem(struct file *file, char __user *buf,
 360                                 size_t count, loff_t *ppos)
 361 {
 362         unsigned long pfn, offset;
 363         size_t read = 0, csize;
 364         int rc = 0;
 365
 366         while (count) {
 367                 pfn = *ppos / PAGE_SIZE;
 368                 if (pfn > saved_max_pfn)
 369                         return read;
 370
 371                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 372                 if (count > PAGE_SIZE - offset)
 373                         csize = PAGE_SIZE - offset;
 374                 else
 375                         csize = count;
 376
 377                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 378                 if (rc < 0)
 379                         return rc;
 380                 buf += csize;
 381                 *ppos += csize;
 382                 read += csize;
 383                 count -= csize;
 384         }
 385         return read;
 386 }
 387 #endif
 388
 389 #ifdef CONFIG_DEVKMEM
 390 /*
 391  * This function reads the *virtual* memory as seen by the kernel.
 392  */
 393 static ssize_t read_kmem(struct file *file, char __user *buf,
 394                          size_t count, loff_t *ppos)
 395 {
 396         unsigned long p = *ppos;
 397         ssize_t low_count, read, sz;
 398         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 399
 400         read = 0;
 401         if (p < (unsigned long) high_memory) {
 402                 low_count = count;
 403                 if (count > (unsigned long) high_memory - p)
 404                         low_count = (unsigned long) high_memory - p;
 405
 406 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 407                 /* we don't have page 0 mapped on sparc and m68k.. */
 408                 if (p < PAGE_SIZE && low_count > 0) {
 409                         sz = size_inside_page(p, low_count);
 410                         if (clear_user(buf, sz))
 411                                 return -EFAULT;
 412                         buf += sz;
 413                         p += sz;
 414                         read += sz;
 415                         low_count -= sz;
 416                         count -= sz;
 417                 }
 418 #endif
 419                 while (low_count > 0) {
 420                         sz = size_inside_page(p, low_count);
 421
 422                         /*
 423                          * On ia64 if a page has been mapped somewhere as
 424                          * uncached, then it must also be accessed uncached
 425                          * by the kernel or data corruption may occur
 426                          */
 427                         kbuf = xlate_dev_kmem_ptr((char *)p);
 428
 429                         if (copy_to_user(buf, kbuf, sz))
 430                                 return -EFAULT;
 431                         buf += sz;
 432                         p += sz;
 433                         read += sz;
 434                         low_count -= sz;
 435                         count -= sz;
 436                 }
 437         }
 438
 439         if (count > 0) {
 440                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 441                 if (!kbuf)
 442                         return -ENOMEM;
 443                 while (count > 0) {
 444                         sz = size_inside_page(p, count);
 445                         sz = vread(kbuf, (char *)p, sz);
 446                         if (!sz)
 447                                 break;
 448                         if (copy_to_user(buf, kbuf, sz)) {
 449                                 free_page((unsigned long)kbuf);
 450                                 return -EFAULT;
 451                         }
 452                         count -= sz;
 453                         buf += sz;
 454                         read += sz;
 455                         p += sz;
 456                 }
 457                 free_page((unsigned long)kbuf);
 458         }
 459         *ppos = p;
 460         return read;
 461 }
 462
 463
 464 static inline ssize_t
 465 do_write_kmem(unsigned long p, const char __user *buf,
 466               size_t count, loff_t *ppos)
 467 {
 468         ssize_t written, sz;
 469         unsigned long copied;
 470
 471         written = 0;
 472 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 473         /* we don't have page 0 mapped on sparc and m68k.. */
 474         if (p < PAGE_SIZE) {
 475                 sz = size_inside_page(p, count);
 476                 /* Hmm. Do something? */
 477                 buf += sz;
 478                 p += sz;
 479                 count -= sz;
 480                 written += sz;
 481         }
 482 #endif
 483
 484         while (count > 0) {
 485                 char *ptr;
 486
 487                 sz = size_inside_page(p, count);
 488
 489                 /*
 490                  * On ia64 if a page has been mapped somewhere as
 491                  * uncached, then it must also be accessed uncached
 492                  * by the kernel or data corruption may occur
 493                  */
 494                 ptr = xlate_dev_kmem_ptr((char *)p);
 495
 496                 copied = copy_from_user(ptr, buf, sz);
 497                 if (copied) {
 498                         written += sz - copied;
 499                         if (written)
 500                                 break;
 501                         return -EFAULT;
 502                 }
 503                 buf += sz;
 504                 p += sz;
 505                 count -= sz;
 506                 written += sz;
 507         }
 508
 509         *ppos += written;
 510         return written;
 511 }
 512
 513
 514 /*
 515  * This function writes to the *virtual* memory as seen by the kernel.
 516  */
 517 static ssize_t write_kmem(struct file * file, const char __user * buf,
 518                           size_t count, loff_t *ppos)
 519 {
 520         unsigned long p = *ppos;
 521         ssize_t wrote = 0;
 522         ssize_t virtr = 0;
 523         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 524
 525         if (p < (unsigned long) high_memory) {
 526                 unsigned long to_write = min_t(unsigned long, count,
 527                                                (unsigned long)high_memory - p);
 528                 wrote = do_write_kmem(p, buf, to_write, ppos);
 529                 if (wrote != to_write)
 530                         return wrote;
 531                 p += wrote;
 532                 buf += wrote;
 533                 count -= wrote;
 534         }
 535
 536         if (count > 0) {
 537                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 538                 if (!kbuf)
 539                         return wrote ? wrote : -ENOMEM;
 540                 while (count > 0) {
 541                         unsigned long sz = size_inside_page(p, count);
 542                         unsigned long n;
 543
 544                         n = copy_from_user(kbuf, buf, sz);
 545                         if (n) {
 546                                 if (wrote + virtr)
 547                                         break;
 548                                 free_page((unsigned long)kbuf);
 549                                 return -EFAULT;
 550                         }
 551                         sz = vwrite(kbuf, (char *)p, sz);
 552                         count -= sz;
 553                         buf += sz;
 554                         virtr += sz;
 555                         p += sz;
 556                 }
 557                 free_page((unsigned long)kbuf);
 558         }
 559
 560         *ppos = p;
 561         return virtr + wrote;
 562 }
 563 #endif
 564
 565 #ifdef CONFIG_DEVPORT
 566 static ssize_t read_port(struct file * file, char __user * buf,
 567                          size_t count, loff_t *ppos)
 568 {
 569         unsigned long i = *ppos;
 570         char __user *tmp = buf;
 571
 572         if (!access_ok(VERIFY_WRITE, buf, count))
 573                 return -EFAULT;
 574         while (count-- > 0 && i < 65536) {
 575                 if (__put_user(inb(i),tmp) < 0)
 576                         return -EFAULT;
 577                 i++;
 578                 tmp++;
 579         }
 580         *ppos = i;
 581         return tmp-buf;
 582 }
 583
 584 static ssize_t write_port(struct file * file, const char __user * buf,
 585                           size_t count, loff_t *ppos)
 586 {
 587         unsigned long i = *ppos;
 588         const char __user * tmp = buf;
 589
 590         if (!access_ok(VERIFY_READ,buf,count))
 591                 return -EFAULT;
 592         while (count-- > 0 && i < 65536) {
 593                 char c;
 594                 if (__get_user(c, tmp)) {
 595                         if (tmp > buf)
 596                                 break;
 597                         return -EFAULT;
 598                 }
 599                 outb(c,i);
 600                 i++;
 601                 tmp++;
 602         }
 603         *ppos = i;
 604         return tmp-buf;
 605 }
 606 #endif
 607
 608 static ssize_t read_null(struct file * file, char __user * buf,
 609                          size_t count, loff_t *ppos)
 610 {
 611         return 0;
 612 }
 613
 614 static ssize_t write_null(struct file * file, const char __user * buf,
 615                           size_t count, loff_t *ppos)
 616 {
 617         return count;
 618 }
 619
 620 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 621                         struct splice_desc *sd)
 622 {
 623         return sd->len;
 624 }
 625
 626 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 627                                  loff_t *ppos, size_t len, unsigned int flags)
 628 {
 629         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 630 }
 631
 632 static ssize_t read_zero(struct file * file, char __user * buf,
 633                          size_t count, loff_t *ppos)
 634 {
 635         size_t written;
 636
 637         if (!count)
 638                 return 0;
 639
 640         if (!access_ok(VERIFY_WRITE, buf, count))
 641                 return -EFAULT;
 642
 643         written = 0;
 644         while (count) {
 645                 unsigned long unwritten;
 646                 size_t chunk = count;
 647
 648                 if (chunk > PAGE_SIZE)
 649                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 650                 unwritten = __clear_user(buf, chunk);
 651                 written += chunk - unwritten;
 652                 if (unwritten)
 653                         break;
 654                 if (signal_pending(current))
 655                         return written ? written : -ERESTARTSYS;
 656                 buf += chunk;
 657                 count -= chunk;
 658                 cond_resched();
 659         }
 660         return written ? written : -EFAULT;
 661 }
 662
 663 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 664 {
 665 #ifndef CONFIG_MMU
 666         return -ENOSYS;
 667 #endif
 668         if (vma->vm_flags & VM_SHARED)
 669                 return shmem_zero_setup(vma);
 670         return 0;
 671 }
 672
 673 static ssize_t write_full(struct file * file, const char __user * buf,
 674                           size_t count, loff_t *ppos)
 675 {
 676         return -ENOSPC;
 677 }
 678
 679 /*
 680  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 681  * can fopen() both devices with "a" now.  This was previously impossible.
 682  * -- SRB.
 683  */
 684
 685 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 686 {
 687         return file->f_pos = 0;
 688 }
 689
 690 /*
 691  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 692  * check against negative addresses: they are ok. The return value is weird,
 693  * though, in that case (0).
 694  *
 695  * also note that seeking relative to the "end of file" isn't supported:
 696  * it has no meaning, so it returns -EINVAL.
 697  */
 698 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 699 {
 700         loff_t ret;
 701
 702         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 703         switch (orig) {
 704                 case 0:
 705                         file->f_pos = offset;
 706                         ret = file->f_pos;
 707                         force_successful_syscall_return();
 708                         break;
 709                 case 1:
 710                         file->f_pos += offset;
 711                         ret = file->f_pos;
 712                         force_successful_syscall_return();
 713                         break;
 714                 default:
 715                         ret = -EINVAL;
 716         }
 717         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 718         return ret;
 719 }
 720
 721 static int open_port(struct inode * inode, struct file * filp)
 722 {
 723         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 724 }
 725
 726 #define zero_lseek      null_lseek
 727 #define full_lseek      null_lseek
 728 #define write_zero      write_null
 729 #define read_full       read_zero
 730 #define open_mem        open_port
 731 #define open_kmem       open_mem
 732 #define open_oldmem     open_mem
 733
 734 static const struct file_operations mem_fops = {
 735         .llseek         = memory_lseek,
 736         .read           = read_mem,
 737         .write          = write_mem,
 738         .mmap           = mmap_mem,
 739         .open           = open_mem,
 740         .get_unmapped_area = get_unmapped_area_mem,
 741 };
 742
 743 #ifdef CONFIG_DEVKMEM
 744 static const struct file_operations kmem_fops = {
 745         .llseek         = memory_lseek,
 746         .read           = read_kmem,
 747         .write          = write_kmem,
 748         .mmap           = mmap_kmem,
 749         .open           = open_kmem,
 750         .get_unmapped_area = get_unmapped_area_mem,
 751 };
 752 #endif
 753
 754 static const struct file_operations null_fops = {
 755         .llseek         = null_lseek,
 756         .read           = read_null,
 757         .write          = write_null,
 758         .splice_write   = splice_write_null,
 759 };
 760
 761 #ifdef CONFIG_DEVPORT
 762 static const struct file_operations port_fops = {
 763         .llseek         = memory_lseek,
 764         .read           = read_port,
 765         .write          = write_port,
 766         .open           = open_port,
 767 };
 768 #endif
 769
 770 static const struct file_operations zero_fops = {
 771         .llseek         = zero_lseek,
 772         .read           = read_zero,
 773         .write          = write_zero,
 774         .mmap           = mmap_zero,
 775 };
 776
 777 /*
 778  * capabilities for /dev/zero
 779  * - permits private mappings, "copies" are taken of the source of zeros
 780  */
 781 static struct backing_dev_info zero_bdi = {
 782         .name           = "char/mem",
 783         .capabilities   = BDI_CAP_MAP_COPY,
 784 };
 785
 786 static const struct file_operations full_fops = {
 787         .llseek         = full_lseek,
 788         .read           = read_full,
 789         .write          = write_full,
 790 };
 791
 792 #ifdef CONFIG_CRASH_DUMP
 793 static const struct file_operations oldmem_fops = {
 794         .read   = read_oldmem,
 795         .open   = open_oldmem,
 796 };
 797 #endif
 798
 799 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 800                           size_t count, loff_t *ppos)
 801 {
 802         char *tmp;
 803         ssize_t ret;
 804
 805         tmp = kmalloc(count + 1, GFP_KERNEL);
 806         if (tmp == NULL)
 807                 return -ENOMEM;
 808         ret = -EFAULT;
 809         if (!copy_from_user(tmp, buf, count)) {
 810                 tmp[count] = 0;
 811                 ret = printk("%s", tmp);
 812                 if (ret > count)
 813                         /* printk can add a prefix */
 814                         ret = count;
 815         }
 816         kfree(tmp);
 817         return ret;
 818 }
 819
 820 static const struct file_operations kmsg_fops = {
 821         .write =        kmsg_write,
 822 };
 823
 824 static const struct memdev {
 825         const char *name;
 826         mode_t mode;
 827         const struct file_operations *fops;
 828         struct backing_dev_info *dev_info;
 829 } devlist[] = {
 830          [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
 831 #ifdef CONFIG_DEVKMEM
 832          [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
 833 #endif
 834          [3] = { "null", 0666, &null_fops, NULL },
 835 #ifdef CONFIG_DEVPORT
 836          [4] = { "port", 0, &port_fops, NULL },
 837 #endif
 838          [5] = { "zero", 0666, &zero_fops, &zero_bdi },
 839          [7] = { "full", 0666, &full_fops, NULL },
 840          [8] = { "random", 0666, &random_fops, NULL },
 841          [9] = { "urandom", 0666, &urandom_fops, NULL },
 842         [11] = { "kmsg", 0, &kmsg_fops, NULL },
 843 #ifdef CONFIG_CRASH_DUMP
 844         [12] = { "oldmem", 0, &oldmem_fops, NULL },
 845 #endif
 846 };
 847
 848 static int memory_open(struct inode *inode, struct file *filp)
 849 {
 850         int minor;
 851         const struct memdev *dev;
 852         int ret = -ENXIO;
 853
 854         lock_kernel();
 855
 856         minor = iminor(inode);
 857         if (minor >= ARRAY_SIZE(devlist))
 858                 goto out;
 859
 860         dev = &devlist[minor];
 861         if (!dev->fops)
 862                 goto out;
 863
 864         filp->f_op = dev->fops;
 865         if (dev->dev_info)
 866                 filp->f_mapping->backing_dev_info = dev->dev_info;
 867
 868         if (dev->fops->open)
 869                 ret = dev->fops->open(inode, filp);
 870         else
 871                 ret = 0;
 872 out:
 873         unlock_kernel();
 874         return ret;
 875 }
 876
 877 static const struct file_operations memory_fops = {
 878         .open           = memory_open,
 879 };
 880
 881 static char *mem_devnode(struct device *dev, mode_t *mode)
 882 {
 883         if (mode && devlist[MINOR(dev->devt)].mode)
 884                 *mode = devlist[MINOR(dev->devt)].mode;
 885         return NULL;
 886 }
 887
 888 static struct class *mem_class;
 889
 890 static int __init chr_dev_init(void)
 891 {
 892         int minor;
 893         int err;
 894
 895         err = bdi_init(&zero_bdi);
 896         if (err)
 897                 return err;
 898
 899         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 900                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 901
 902         mem_class = class_create(THIS_MODULE, "mem");
 903         mem_class->devnode = mem_devnode;
 904         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 905                 if (!devlist[minor].name)
 906                         continue;
 907                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 908                               NULL, devlist[minor].name);
 909         }
 910
 911         return 0;
 912 }
 913
 914 fs_initcall(chr_dev_init);