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         if (-start & (PAGE_SIZE - 1))
  44                 sz = -start & (PAGE_SIZE - 1);
  45         else
  46                 sz = PAGE_SIZE;
  47
  48         return min_t(unsigned long, sz, size);
  49 }
  50
  51 /*
  52  * Architectures vary in how they handle caching for addresses
  53  * outside of main memory.
  54  *
  55  */
  56 static inline int uncached_access(struct file *file, unsigned long addr)
  57 {
  58 #if defined(CONFIG_IA64)
  59         /*
  60          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  61          */
  62         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  63 #elif defined(CONFIG_MIPS)
  64         {
  65                 extern int __uncached_access(struct file *file,
  66                                              unsigned long addr);
  67
  68                 return __uncached_access(file, addr);
  69         }
  70 #else
  71         /*
  72          * Accessing memory above the top the kernel knows about or through a file pointer
  73          * that was marked O_SYNC will be done non-cached.
  74          */
  75         if (file->f_flags & O_SYNC)
  76                 return 1;
  77         return addr >= __pa(high_memory);
  78 #endif
  79 }
  80
  81 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  82 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  83 {
  84         if (addr + count > __pa(high_memory))
  85                 return 0;
  86
  87         return 1;
  88 }
  89
  90 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  91 {
  92         return 1;
  93 }
  94 #endif
  95
  96 #ifdef CONFIG_STRICT_DEVMEM
  97 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  98 {
  99         u64 from = ((u64)pfn) << PAGE_SHIFT;
 100         u64 to = from + size;
 101         u64 cursor = from;
 102
 103         while (cursor < to) {
 104                 if (!devmem_is_allowed(pfn)) {
 105                         printk(KERN_INFO
 106                 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
 107                                 current->comm, from, to);
 108                         return 0;
 109                 }
 110                 cursor += PAGE_SIZE;
 111                 pfn++;
 112         }
 113         return 1;
 114 }
 115 #else
 116 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 117 {
 118         return 1;
 119 }
 120 #endif
 121
 122 void __attribute__((weak)) unxlate_dev_mem_ptr(unsigned long phys, void *addr)
 123 {
 124 }
 125
 126 /*
 127  * This funcion reads the *physical* memory. The f_pos points directly to the
 128  * memory location.
 129  */
 130 static ssize_t read_mem(struct file * file, char __user * buf,
 131                         size_t count, loff_t *ppos)
 132 {
 133         unsigned long p = *ppos;
 134         ssize_t read, sz;
 135         char *ptr;
 136
 137         if (!valid_phys_addr_range(p, count))
 138                 return -EFAULT;
 139         read = 0;
 140 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 141         /* we don't have page 0 mapped on sparc and m68k.. */
 142         if (p < PAGE_SIZE) {
 143                 sz = PAGE_SIZE - p;
 144                 if (sz > count)
 145                         sz = count;
 146                 if (sz > 0) {
 147                         if (clear_user(buf, sz))
 148                                 return -EFAULT;
 149                         buf += sz;
 150                         p += sz;
 151                         count -= sz;
 152                         read += sz;
 153                 }
 154         }
 155 #endif
 156
 157         while (count > 0) {
 158                 /*
 159                  * Handle first page in case it's not aligned
 160                  */
 161                 if (-p & (PAGE_SIZE - 1))
 162                         sz = -p & (PAGE_SIZE - 1);
 163                 else
 164                         sz = PAGE_SIZE;
 165
 166                 sz = min_t(unsigned long, sz, count);
 167
 168                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 169                         return -EPERM;
 170
 171                 /*
 172                  * On ia64 if a page has been mapped somewhere as
 173                  * uncached, then it must also be accessed uncached
 174                  * by the kernel or data corruption may occur
 175                  */
 176                 ptr = xlate_dev_mem_ptr(p);
 177                 if (!ptr)
 178                         return -EFAULT;
 179
 180                 if (copy_to_user(buf, ptr, sz)) {
 181                         unxlate_dev_mem_ptr(p, ptr);
 182                         return -EFAULT;
 183                 }
 184
 185                 unxlate_dev_mem_ptr(p, ptr);
 186
 187                 buf += sz;
 188                 p += sz;
 189                 count -= sz;
 190                 read += sz;
 191         }
 192
 193         *ppos += read;
 194         return read;
 195 }
 196
 197 static ssize_t write_mem(struct file * file, const char __user * buf,
 198                          size_t count, loff_t *ppos)
 199 {
 200         unsigned long p = *ppos;
 201         ssize_t written, sz;
 202         unsigned long copied;
 203         void *ptr;
 204
 205         if (!valid_phys_addr_range(p, count))
 206                 return -EFAULT;
 207
 208         written = 0;
 209
 210 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 211         /* we don't have page 0 mapped on sparc and m68k.. */
 212         if (p < PAGE_SIZE) {
 213                 unsigned long sz = PAGE_SIZE - p;
 214                 if (sz > count)
 215                         sz = count;
 216                 /* Hmm. Do something? */
 217                 buf += sz;
 218                 p += sz;
 219                 count -= sz;
 220                 written += sz;
 221         }
 222 #endif
 223
 224         while (count > 0) {
 225                 /*
 226                  * Handle first page in case it's not aligned
 227                  */
 228                 if (-p & (PAGE_SIZE - 1))
 229                         sz = -p & (PAGE_SIZE - 1);
 230                 else
 231                         sz = PAGE_SIZE;
 232
 233                 sz = min_t(unsigned long, sz, count);
 234
 235                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 236                         return -EPERM;
 237
 238                 /*
 239                  * On ia64 if a page has been mapped somewhere as
 240                  * uncached, then it must also be accessed uncached
 241                  * by the kernel or data corruption may occur
 242                  */
 243                 ptr = xlate_dev_mem_ptr(p);
 244                 if (!ptr) {
 245                         if (written)
 246                                 break;
 247                         return -EFAULT;
 248                 }
 249
 250                 copied = copy_from_user(ptr, buf, sz);
 251                 if (copied) {
 252                         written += sz - copied;
 253                         unxlate_dev_mem_ptr(p, ptr);
 254                         if (written)
 255                                 break;
 256                         return -EFAULT;
 257                 }
 258
 259                 unxlate_dev_mem_ptr(p, ptr);
 260
 261                 buf += sz;
 262                 p += sz;
 263                 count -= sz;
 264                 written += sz;
 265         }
 266
 267         *ppos += written;
 268         return written;
 269 }
 270
 271 int __attribute__((weak)) phys_mem_access_prot_allowed(struct file *file,
 272         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 273 {
 274         return 1;
 275 }
 276
 277 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 278 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 279                                      unsigned long size, pgprot_t vma_prot)
 280 {
 281 #ifdef pgprot_noncached
 282         unsigned long offset = pfn << PAGE_SHIFT;
 283
 284         if (uncached_access(file, offset))
 285                 return pgprot_noncached(vma_prot);
 286 #endif
 287         return vma_prot;
 288 }
 289 #endif
 290
 291 #ifndef CONFIG_MMU
 292 static unsigned long get_unmapped_area_mem(struct file *file,
 293                                            unsigned long addr,
 294                                            unsigned long len,
 295                                            unsigned long pgoff,
 296                                            unsigned long flags)
 297 {
 298         if (!valid_mmap_phys_addr_range(pgoff, len))
 299                 return (unsigned long) -EINVAL;
 300         return pgoff << PAGE_SHIFT;
 301 }
 302
 303 /* can't do an in-place private mapping if there's no MMU */
 304 static inline int private_mapping_ok(struct vm_area_struct *vma)
 305 {
 306         return vma->vm_flags & VM_MAYSHARE;
 307 }
 308 #else
 309 #define get_unmapped_area_mem   NULL
 310
 311 static inline int private_mapping_ok(struct vm_area_struct *vma)
 312 {
 313         return 1;
 314 }
 315 #endif
 316
 317 static const struct vm_operations_struct mmap_mem_ops = {
 318 #ifdef CONFIG_HAVE_IOREMAP_PROT
 319         .access = generic_access_phys
 320 #endif
 321 };
 322
 323 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 324 {
 325         size_t size = vma->vm_end - vma->vm_start;
 326
 327         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 328                 return -EINVAL;
 329
 330         if (!private_mapping_ok(vma))
 331                 return -ENOSYS;
 332
 333         if (!range_is_allowed(vma->vm_pgoff, size))
 334                 return -EPERM;
 335
 336         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 337                                                 &vma->vm_page_prot))
 338                 return -EINVAL;
 339
 340         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 341                                                  size,
 342                                                  vma->vm_page_prot);
 343
 344         vma->vm_ops = &mmap_mem_ops;
 345
 346         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 347         if (remap_pfn_range(vma,
 348                             vma->vm_start,
 349                             vma->vm_pgoff,
 350                             size,
 351                             vma->vm_page_prot)) {
 352                 return -EAGAIN;
 353         }
 354         return 0;
 355 }
 356
 357 #ifdef CONFIG_DEVKMEM
 358 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 359 {
 360         unsigned long pfn;
 361
 362         /* Turn a kernel-virtual address into a physical page frame */
 363         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 364
 365         /*
 366          * RED-PEN: on some architectures there is more mapped memory
 367          * than available in mem_map which pfn_valid checks
 368          * for. Perhaps should add a new macro here.
 369          *
 370          * RED-PEN: vmalloc is not supported right now.
 371          */
 372         if (!pfn_valid(pfn))
 373                 return -EIO;
 374
 375         vma->vm_pgoff = pfn;
 376         return mmap_mem(file, vma);
 377 }
 378 #endif
 379
 380 #ifdef CONFIG_CRASH_DUMP
 381 /*
 382  * Read memory corresponding to the old kernel.
 383  */
 384 static ssize_t read_oldmem(struct file *file, char __user *buf,
 385                                 size_t count, loff_t *ppos)
 386 {
 387         unsigned long pfn, offset;
 388         size_t read = 0, csize;
 389         int rc = 0;
 390
 391         while (count) {
 392                 pfn = *ppos / PAGE_SIZE;
 393                 if (pfn > saved_max_pfn)
 394                         return read;
 395
 396                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 397                 if (count > PAGE_SIZE - offset)
 398                         csize = PAGE_SIZE - offset;
 399                 else
 400                         csize = count;
 401
 402                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 403                 if (rc < 0)
 404                         return rc;
 405                 buf += csize;
 406                 *ppos += csize;
 407                 read += csize;
 408                 count -= csize;
 409         }
 410         return read;
 411 }
 412 #endif
 413
 414 #ifdef CONFIG_DEVKMEM
 415 /*
 416  * This function reads the *virtual* memory as seen by the kernel.
 417  */
 418 static ssize_t read_kmem(struct file *file, char __user *buf,
 419                          size_t count, loff_t *ppos)
 420 {
 421         unsigned long p = *ppos;
 422         ssize_t low_count, read, sz;
 423         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 424
 425         read = 0;
 426         if (p < (unsigned long) high_memory) {
 427                 low_count = count;
 428                 if (count > (unsigned long) high_memory - p)
 429                         low_count = (unsigned long) high_memory - p;
 430
 431 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 432                 /* we don't have page 0 mapped on sparc and m68k.. */
 433                 if (p < PAGE_SIZE && low_count > 0) {
 434                         size_t tmp = PAGE_SIZE - p;
 435                         if (tmp > low_count) tmp = low_count;
 436                         if (clear_user(buf, tmp))
 437                                 return -EFAULT;
 438                         buf += tmp;
 439                         p += tmp;
 440                         read += tmp;
 441                         low_count -= tmp;
 442                         count -= tmp;
 443                 }
 444 #endif
 445                 while (low_count > 0) {
 446                         sz = size_inside_page(p, low_count);
 447
 448                         /*
 449                          * On ia64 if a page has been mapped somewhere as
 450                          * uncached, then it must also be accessed uncached
 451                          * by the kernel or data corruption may occur
 452                          */
 453                         kbuf = xlate_dev_kmem_ptr((char *)p);
 454
 455                         if (copy_to_user(buf, kbuf, sz))
 456                                 return -EFAULT;
 457                         buf += sz;
 458                         p += sz;
 459                         read += sz;
 460                         low_count -= sz;
 461                         count -= sz;
 462                 }
 463         }
 464
 465         if (count > 0) {
 466                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 467                 if (!kbuf)
 468                         return -ENOMEM;
 469                 while (count > 0) {
 470                         int len = size_inside_page(p, count);
 471
 472                         len = vread(kbuf, (char *)p, len);
 473                         if (!len)
 474                                 break;
 475                         if (copy_to_user(buf, kbuf, len)) {
 476                                 free_page((unsigned long)kbuf);
 477                                 return -EFAULT;
 478                         }
 479                         count -= len;
 480                         buf += len;
 481                         read += len;
 482                         p += len;
 483                 }
 484                 free_page((unsigned long)kbuf);
 485         }
 486         *ppos = p;
 487         return read;
 488 }
 489
 490
 491 static inline ssize_t
 492 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 493               size_t count, loff_t *ppos)
 494 {
 495         ssize_t written, sz;
 496         unsigned long copied;
 497
 498         written = 0;
 499 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 500         /* we don't have page 0 mapped on sparc and m68k.. */
 501         if (realp < PAGE_SIZE) {
 502                 unsigned long sz = PAGE_SIZE - realp;
 503                 if (sz > count)
 504                         sz = count;
 505                 /* Hmm. Do something? */
 506                 buf += sz;
 507                 p += sz;
 508                 realp += sz;
 509                 count -= sz;
 510                 written += sz;
 511         }
 512 #endif
 513
 514         while (count > 0) {
 515                 char *ptr;
 516
 517                 sz = size_inside_page(realp, count);
 518
 519                 /*
 520                  * On ia64 if a page has been mapped somewhere as
 521                  * uncached, then it must also be accessed uncached
 522                  * by the kernel or data corruption may occur
 523                  */
 524                 ptr = xlate_dev_kmem_ptr(p);
 525
 526                 copied = copy_from_user(ptr, buf, sz);
 527                 if (copied) {
 528                         written += sz - copied;
 529                         if (written)
 530                                 break;
 531                         return -EFAULT;
 532                 }
 533                 buf += sz;
 534                 p += sz;
 535                 realp += sz;
 536                 count -= sz;
 537                 written += sz;
 538         }
 539
 540         *ppos += written;
 541         return written;
 542 }
 543
 544
 545 /*
 546  * This function writes to the *virtual* memory as seen by the kernel.
 547  */
 548 static ssize_t write_kmem(struct file * file, const char __user * buf,
 549                           size_t count, loff_t *ppos)
 550 {
 551         unsigned long p = *ppos;
 552         ssize_t wrote = 0;
 553         ssize_t virtr = 0;
 554         ssize_t written;
 555         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 556
 557         if (p < (unsigned long) high_memory) {
 558
 559                 wrote = count;
 560                 if (count > (unsigned long) high_memory - p)
 561                         wrote = (unsigned long) high_memory - p;
 562
 563                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 564                 if (written != wrote)
 565                         return written;
 566                 wrote = written;
 567                 p += wrote;
 568                 buf += wrote;
 569                 count -= wrote;
 570         }
 571
 572         if (count > 0) {
 573                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 574                 if (!kbuf)
 575                         return wrote ? wrote : -ENOMEM;
 576                 while (count > 0) {
 577                         int len = size_inside_page(p, count);
 578
 579                         if (len) {
 580                                 written = copy_from_user(kbuf, buf, len);
 581                                 if (written) {
 582                                         if (wrote + virtr)
 583                                                 break;
 584                                         free_page((unsigned long)kbuf);
 585                                         return -EFAULT;
 586                                 }
 587                         }
 588                         len = vwrite(kbuf, (char *)p, len);
 589                         count -= len;
 590                         buf += len;
 591                         virtr += len;
 592                         p += len;
 593                 }
 594                 free_page((unsigned long)kbuf);
 595         }
 596
 597         *ppos = p;
 598         return virtr + wrote;
 599 }
 600 #endif
 601
 602 #ifdef CONFIG_DEVPORT
 603 static ssize_t read_port(struct file * file, char __user * buf,
 604                          size_t count, loff_t *ppos)
 605 {
 606         unsigned long i = *ppos;
 607         char __user *tmp = buf;
 608
 609         if (!access_ok(VERIFY_WRITE, buf, count))
 610                 return -EFAULT;
 611         while (count-- > 0 && i < 65536) {
 612                 if (__put_user(inb(i),tmp) < 0)
 613                         return -EFAULT;
 614                 i++;
 615                 tmp++;
 616         }
 617         *ppos = i;
 618         return tmp-buf;
 619 }
 620
 621 static ssize_t write_port(struct file * file, const char __user * buf,
 622                           size_t count, loff_t *ppos)
 623 {
 624         unsigned long i = *ppos;
 625         const char __user * tmp = buf;
 626
 627         if (!access_ok(VERIFY_READ,buf,count))
 628                 return -EFAULT;
 629         while (count-- > 0 && i < 65536) {
 630                 char c;
 631                 if (__get_user(c, tmp)) {
 632                         if (tmp > buf)
 633                                 break;
 634                         return -EFAULT;
 635                 }
 636                 outb(c,i);
 637                 i++;
 638                 tmp++;
 639         }
 640         *ppos = i;
 641         return tmp-buf;
 642 }
 643 #endif
 644
 645 static ssize_t read_null(struct file * file, char __user * buf,
 646                          size_t count, loff_t *ppos)
 647 {
 648         return 0;
 649 }
 650
 651 static ssize_t write_null(struct file * file, const char __user * buf,
 652                           size_t count, loff_t *ppos)
 653 {
 654         return count;
 655 }
 656
 657 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 658                         struct splice_desc *sd)
 659 {
 660         return sd->len;
 661 }
 662
 663 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 664                                  loff_t *ppos, size_t len, unsigned int flags)
 665 {
 666         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 667 }
 668
 669 static ssize_t read_zero(struct file * file, char __user * buf,
 670                          size_t count, loff_t *ppos)
 671 {
 672         size_t written;
 673
 674         if (!count)
 675                 return 0;
 676
 677         if (!access_ok(VERIFY_WRITE, buf, count))
 678                 return -EFAULT;
 679
 680         written = 0;
 681         while (count) {
 682                 unsigned long unwritten;
 683                 size_t chunk = count;
 684
 685                 if (chunk > PAGE_SIZE)
 686                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 687                 unwritten = __clear_user(buf, chunk);
 688                 written += chunk - unwritten;
 689                 if (unwritten)
 690                         break;
 691                 if (signal_pending(current))
 692                         return written ? written : -ERESTARTSYS;
 693                 buf += chunk;
 694                 count -= chunk;
 695                 cond_resched();
 696         }
 697         return written ? written : -EFAULT;
 698 }
 699
 700 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 701 {
 702 #ifndef CONFIG_MMU
 703         return -ENOSYS;
 704 #endif
 705         if (vma->vm_flags & VM_SHARED)
 706                 return shmem_zero_setup(vma);
 707         return 0;
 708 }
 709
 710 static ssize_t write_full(struct file * file, const char __user * buf,
 711                           size_t count, loff_t *ppos)
 712 {
 713         return -ENOSPC;
 714 }
 715
 716 /*
 717  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 718  * can fopen() both devices with "a" now.  This was previously impossible.
 719  * -- SRB.
 720  */
 721
 722 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 723 {
 724         return file->f_pos = 0;
 725 }
 726
 727 /*
 728  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 729  * check against negative addresses: they are ok. The return value is weird,
 730  * though, in that case (0).
 731  *
 732  * also note that seeking relative to the "end of file" isn't supported:
 733  * it has no meaning, so it returns -EINVAL.
 734  */
 735 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 736 {
 737         loff_t ret;
 738
 739         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 740         switch (orig) {
 741                 case 0:
 742                         file->f_pos = offset;
 743                         ret = file->f_pos;
 744                         force_successful_syscall_return();
 745                         break;
 746                 case 1:
 747                         file->f_pos += offset;
 748                         ret = file->f_pos;
 749                         force_successful_syscall_return();
 750                         break;
 751                 default:
 752                         ret = -EINVAL;
 753         }
 754         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 755         return ret;
 756 }
 757
 758 static int open_port(struct inode * inode, struct file * filp)
 759 {
 760         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 761 }
 762
 763 #define zero_lseek      null_lseek
 764 #define full_lseek      null_lseek
 765 #define write_zero      write_null
 766 #define read_full       read_zero
 767 #define open_mem        open_port
 768 #define open_kmem       open_mem
 769 #define open_oldmem     open_mem
 770
 771 static const struct file_operations mem_fops = {
 772         .llseek         = memory_lseek,
 773         .read           = read_mem,
 774         .write          = write_mem,
 775         .mmap           = mmap_mem,
 776         .open           = open_mem,
 777         .get_unmapped_area = get_unmapped_area_mem,
 778 };
 779
 780 #ifdef CONFIG_DEVKMEM
 781 static const struct file_operations kmem_fops = {
 782         .llseek         = memory_lseek,
 783         .read           = read_kmem,
 784         .write          = write_kmem,
 785         .mmap           = mmap_kmem,
 786         .open           = open_kmem,
 787         .get_unmapped_area = get_unmapped_area_mem,
 788 };
 789 #endif
 790
 791 static const struct file_operations null_fops = {
 792         .llseek         = null_lseek,
 793         .read           = read_null,
 794         .write          = write_null,
 795         .splice_write   = splice_write_null,
 796 };
 797
 798 #ifdef CONFIG_DEVPORT
 799 static const struct file_operations port_fops = {
 800         .llseek         = memory_lseek,
 801         .read           = read_port,
 802         .write          = write_port,
 803         .open           = open_port,
 804 };
 805 #endif
 806
 807 static const struct file_operations zero_fops = {
 808         .llseek         = zero_lseek,
 809         .read           = read_zero,
 810         .write          = write_zero,
 811         .mmap           = mmap_zero,
 812 };
 813
 814 /*
 815  * capabilities for /dev/zero
 816  * - permits private mappings, "copies" are taken of the source of zeros
 817  */
 818 static struct backing_dev_info zero_bdi = {
 819         .name           = "char/mem",
 820         .capabilities   = BDI_CAP_MAP_COPY,
 821 };
 822
 823 static const struct file_operations full_fops = {
 824         .llseek         = full_lseek,
 825         .read           = read_full,
 826         .write          = write_full,
 827 };
 828
 829 #ifdef CONFIG_CRASH_DUMP
 830 static const struct file_operations oldmem_fops = {
 831         .read   = read_oldmem,
 832         .open   = open_oldmem,
 833 };
 834 #endif
 835
 836 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 837                           size_t count, loff_t *ppos)
 838 {
 839         char *tmp;
 840         ssize_t ret;
 841
 842         tmp = kmalloc(count + 1, GFP_KERNEL);
 843         if (tmp == NULL)
 844                 return -ENOMEM;
 845         ret = -EFAULT;
 846         if (!copy_from_user(tmp, buf, count)) {
 847                 tmp[count] = 0;
 848                 ret = printk("%s", tmp);
 849                 if (ret > count)
 850                         /* printk can add a prefix */
 851                         ret = count;
 852         }
 853         kfree(tmp);
 854         return ret;
 855 }
 856
 857 static const struct file_operations kmsg_fops = {
 858         .write =        kmsg_write,
 859 };
 860
 861 static const struct memdev {
 862         const char *name;
 863         mode_t mode;
 864         const struct file_operations *fops;
 865         struct backing_dev_info *dev_info;
 866 } devlist[] = {
 867          [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
 868 #ifdef CONFIG_DEVKMEM
 869          [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
 870 #endif
 871          [3] = { "null", 0666, &null_fops, NULL },
 872 #ifdef CONFIG_DEVPORT
 873          [4] = { "port", 0, &port_fops, NULL },
 874 #endif
 875          [5] = { "zero", 0666, &zero_fops, &zero_bdi },
 876          [7] = { "full", 0666, &full_fops, NULL },
 877          [8] = { "random", 0666, &random_fops, NULL },
 878          [9] = { "urandom", 0666, &urandom_fops, NULL },
 879         [11] = { "kmsg", 0, &kmsg_fops, NULL },
 880 #ifdef CONFIG_CRASH_DUMP
 881         [12] = { "oldmem", 0, &oldmem_fops, NULL },
 882 #endif
 883 };
 884
 885 static int memory_open(struct inode *inode, struct file *filp)
 886 {
 887         int minor;
 888         const struct memdev *dev;
 889         int ret = -ENXIO;
 890
 891         lock_kernel();
 892
 893         minor = iminor(inode);
 894         if (minor >= ARRAY_SIZE(devlist))
 895                 goto out;
 896
 897         dev = &devlist[minor];
 898         if (!dev->fops)
 899                 goto out;
 900
 901         filp->f_op = dev->fops;
 902         if (dev->dev_info)
 903                 filp->f_mapping->backing_dev_info = dev->dev_info;
 904
 905         if (dev->fops->open)
 906                 ret = dev->fops->open(inode, filp);
 907         else
 908                 ret = 0;
 909 out:
 910         unlock_kernel();
 911         return ret;
 912 }
 913
 914 static const struct file_operations memory_fops = {
 915         .open           = memory_open,
 916 };
 917
 918 static char *mem_devnode(struct device *dev, mode_t *mode)
 919 {
 920         if (mode && devlist[MINOR(dev->devt)].mode)
 921                 *mode = devlist[MINOR(dev->devt)].mode;
 922         return NULL;
 923 }
 924
 925 static struct class *mem_class;
 926
 927 static int __init chr_dev_init(void)
 928 {
 929         int minor;
 930         int err;
 931
 932         err = bdi_init(&zero_bdi);
 933         if (err)
 934                 return err;
 935
 936         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 937                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 938
 939         mem_class = class_create(THIS_MODULE, "mem");
 940         mem_class->devnode = mem_devnode;
 941         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 942                 if (!devlist[minor].name)
 943                         continue;
 944                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 945                               NULL, devlist[minor].name);
 946         }
 947
 948         return 0;
 949 }
 950
 951 fs_initcall(chr_dev_init);