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 /*
  39  * Architectures vary in how they handle caching for addresses
  40  * outside of main memory.
  41  *
  42  */
  43 static inline int uncached_access(struct file *file, unsigned long addr)
  44 {
  45 #if defined(CONFIG_IA64)
  46         /*
  47          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  48          */
  49         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  50 #elif defined(CONFIG_MIPS)
  51         {
  52                 extern int __uncached_access(struct file *file,
  53                                              unsigned long addr);
  54
  55                 return __uncached_access(file, addr);
  56         }
  57 #else
  58         /*
  59          * Accessing memory above the top the kernel knows about or through a file pointer
  60          * that was marked O_SYNC will be done non-cached.
  61          */
  62         if (file->f_flags & O_SYNC)
  63                 return 1;
  64         return addr >= __pa(high_memory);
  65 #endif
  66 }
  67
  68 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  69 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  70 {
  71         if (addr + count > __pa(high_memory))
  72                 return 0;
  73
  74         return 1;
  75 }
  76
  77 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  78 {
  79         return 1;
  80 }
  81 #endif
  82
  83 #ifdef CONFIG_STRICT_DEVMEM
  84 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  85 {
  86         u64 from = ((u64)pfn) << PAGE_SHIFT;
  87         u64 to = from + size;
  88         u64 cursor = from;
  89
  90         while (cursor < to) {
  91                 if (!devmem_is_allowed(pfn)) {
  92                         printk(KERN_INFO
  93                 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
  94                                 current->comm, from, to);
  95                         return 0;
  96                 }
  97                 cursor += PAGE_SIZE;
  98                 pfn++;
  99         }
 100         return 1;
 101 }
 102 #else
 103 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 104 {
 105         return 1;
 106 }
 107 #endif
 108
 109 void __attribute__((weak)) unxlate_dev_mem_ptr(unsigned long phys, void *addr)
 110 {
 111 }
 112
 113 /*
 114  * This funcion reads the *physical* memory. The f_pos points directly to the
 115  * memory location.
 116  */
 117 static ssize_t read_mem(struct file * file, char __user * buf,
 118                         size_t count, loff_t *ppos)
 119 {
 120         unsigned long p = *ppos;
 121         ssize_t read, sz;
 122         char *ptr;
 123
 124         if (!valid_phys_addr_range(p, count))
 125                 return -EFAULT;
 126         read = 0;
 127 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 128         /* we don't have page 0 mapped on sparc and m68k.. */
 129         if (p < PAGE_SIZE) {
 130                 sz = PAGE_SIZE - p;
 131                 if (sz > count)
 132                         sz = count;
 133                 if (sz > 0) {
 134                         if (clear_user(buf, sz))
 135                                 return -EFAULT;
 136                         buf += sz;
 137                         p += sz;
 138                         count -= sz;
 139                         read += sz;
 140                 }
 141         }
 142 #endif
 143
 144         while (count > 0) {
 145                 /*
 146                  * Handle first page in case it's not aligned
 147                  */
 148                 if (-p & (PAGE_SIZE - 1))
 149                         sz = -p & (PAGE_SIZE - 1);
 150                 else
 151                         sz = PAGE_SIZE;
 152
 153                 sz = min_t(unsigned long, sz, count);
 154
 155                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 156                         return -EPERM;
 157
 158                 /*
 159                  * On ia64 if a page has been mapped somewhere as
 160                  * uncached, then it must also be accessed uncached
 161                  * by the kernel or data corruption may occur
 162                  */
 163                 ptr = xlate_dev_mem_ptr(p);
 164                 if (!ptr)
 165                         return -EFAULT;
 166
 167                 if (copy_to_user(buf, ptr, sz)) {
 168                         unxlate_dev_mem_ptr(p, ptr);
 169                         return -EFAULT;
 170                 }
 171
 172                 unxlate_dev_mem_ptr(p, ptr);
 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                 unsigned long sz = PAGE_SIZE - p;
 201                 if (sz > count)
 202                         sz = count;
 203                 /* Hmm. Do something? */
 204                 buf += sz;
 205                 p += sz;
 206                 count -= sz;
 207                 written += sz;
 208         }
 209 #endif
 210
 211         while (count > 0) {
 212                 /*
 213                  * Handle first page in case it's not aligned
 214                  */
 215                 if (-p & (PAGE_SIZE - 1))
 216                         sz = -p & (PAGE_SIZE - 1);
 217                 else
 218                         sz = PAGE_SIZE;
 219
 220                 sz = min_t(unsigned long, sz, count);
 221
 222                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 223                         return -EPERM;
 224
 225                 /*
 226                  * On ia64 if a page has been mapped somewhere as
 227                  * uncached, then it must also be accessed uncached
 228                  * by the kernel or data corruption may occur
 229                  */
 230                 ptr = xlate_dev_mem_ptr(p);
 231                 if (!ptr) {
 232                         if (written)
 233                                 break;
 234                         return -EFAULT;
 235                 }
 236
 237                 copied = copy_from_user(ptr, buf, sz);
 238                 if (copied) {
 239                         written += sz - copied;
 240                         unxlate_dev_mem_ptr(p, ptr);
 241                         if (written)
 242                                 break;
 243                         return -EFAULT;
 244                 }
 245
 246                 unxlate_dev_mem_ptr(p, ptr);
 247
 248                 buf += sz;
 249                 p += sz;
 250                 count -= sz;
 251                 written += sz;
 252         }
 253
 254         *ppos += written;
 255         return written;
 256 }
 257
 258 int __attribute__((weak)) phys_mem_access_prot_allowed(struct file *file,
 259         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 260 {
 261         return 1;
 262 }
 263
 264 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 265 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 266                                      unsigned long size, pgprot_t vma_prot)
 267 {
 268 #ifdef pgprot_noncached
 269         unsigned long offset = pfn << PAGE_SHIFT;
 270
 271         if (uncached_access(file, offset))
 272                 return pgprot_noncached(vma_prot);
 273 #endif
 274         return vma_prot;
 275 }
 276 #endif
 277
 278 #ifndef CONFIG_MMU
 279 static unsigned long get_unmapped_area_mem(struct file *file,
 280                                            unsigned long addr,
 281                                            unsigned long len,
 282                                            unsigned long pgoff,
 283                                            unsigned long flags)
 284 {
 285         if (!valid_mmap_phys_addr_range(pgoff, len))
 286                 return (unsigned long) -EINVAL;
 287         return pgoff << PAGE_SHIFT;
 288 }
 289
 290 /* can't do an in-place private mapping if there's no MMU */
 291 static inline int private_mapping_ok(struct vm_area_struct *vma)
 292 {
 293         return vma->vm_flags & VM_MAYSHARE;
 294 }
 295 #else
 296 #define get_unmapped_area_mem   NULL
 297
 298 static inline int private_mapping_ok(struct vm_area_struct *vma)
 299 {
 300         return 1;
 301 }
 302 #endif
 303
 304 static struct vm_operations_struct mmap_mem_ops = {
 305 #ifdef CONFIG_HAVE_IOREMAP_PROT
 306         .access = generic_access_phys
 307 #endif
 308 };
 309
 310 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 311 {
 312         size_t size = vma->vm_end - vma->vm_start;
 313
 314         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 315                 return -EINVAL;
 316
 317         if (!private_mapping_ok(vma))
 318                 return -ENOSYS;
 319
 320         if (!range_is_allowed(vma->vm_pgoff, size))
 321                 return -EPERM;
 322
 323         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 324                                                 &vma->vm_page_prot))
 325                 return -EINVAL;
 326
 327         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 328                                                  size,
 329                                                  vma->vm_page_prot);
 330
 331         vma->vm_ops = &mmap_mem_ops;
 332
 333         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 334         if (remap_pfn_range(vma,
 335                             vma->vm_start,
 336                             vma->vm_pgoff,
 337                             size,
 338                             vma->vm_page_prot)) {
 339                 return -EAGAIN;
 340         }
 341         return 0;
 342 }
 343
 344 #ifdef CONFIG_DEVKMEM
 345 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 346 {
 347         unsigned long pfn;
 348
 349         /* Turn a kernel-virtual address into a physical page frame */
 350         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 351
 352         /*
 353          * RED-PEN: on some architectures there is more mapped memory
 354          * than available in mem_map which pfn_valid checks
 355          * for. Perhaps should add a new macro here.
 356          *
 357          * RED-PEN: vmalloc is not supported right now.
 358          */
 359         if (!pfn_valid(pfn))
 360                 return -EIO;
 361
 362         vma->vm_pgoff = pfn;
 363         return mmap_mem(file, vma);
 364 }
 365 #endif
 366
 367 #ifdef CONFIG_CRASH_DUMP
 368 /*
 369  * Read memory corresponding to the old kernel.
 370  */
 371 static ssize_t read_oldmem(struct file *file, char __user *buf,
 372                                 size_t count, loff_t *ppos)
 373 {
 374         unsigned long pfn, offset;
 375         size_t read = 0, csize;
 376         int rc = 0;
 377
 378         while (count) {
 379                 pfn = *ppos / PAGE_SIZE;
 380                 if (pfn > saved_max_pfn)
 381                         return read;
 382
 383                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 384                 if (count > PAGE_SIZE - offset)
 385                         csize = PAGE_SIZE - offset;
 386                 else
 387                         csize = count;
 388
 389                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 390                 if (rc < 0)
 391                         return rc;
 392                 buf += csize;
 393                 *ppos += csize;
 394                 read += csize;
 395                 count -= csize;
 396         }
 397         return read;
 398 }
 399 #endif
 400
 401 #ifdef CONFIG_DEVKMEM
 402 /*
 403  * This function reads the *virtual* memory as seen by the kernel.
 404  */
 405 static ssize_t read_kmem(struct file *file, char __user *buf,
 406                          size_t count, loff_t *ppos)
 407 {
 408         unsigned long p = *ppos;
 409         ssize_t low_count, read, sz;
 410         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 411
 412         read = 0;
 413         if (p < (unsigned long) high_memory) {
 414                 low_count = count;
 415                 if (count > (unsigned long) high_memory - p)
 416                         low_count = (unsigned long) high_memory - p;
 417
 418 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 419                 /* we don't have page 0 mapped on sparc and m68k.. */
 420                 if (p < PAGE_SIZE && low_count > 0) {
 421                         size_t tmp = PAGE_SIZE - p;
 422                         if (tmp > low_count) tmp = low_count;
 423                         if (clear_user(buf, tmp))
 424                                 return -EFAULT;
 425                         buf += tmp;
 426                         p += tmp;
 427                         read += tmp;
 428                         low_count -= tmp;
 429                         count -= tmp;
 430                 }
 431 #endif
 432                 while (low_count > 0) {
 433                         /*
 434                          * Handle first page in case it's not aligned
 435                          */
 436                         if (-p & (PAGE_SIZE - 1))
 437                                 sz = -p & (PAGE_SIZE - 1);
 438                         else
 439                                 sz = PAGE_SIZE;
 440
 441                         sz = min_t(unsigned long, sz, low_count);
 442
 443                         /*
 444                          * On ia64 if a page has been mapped somewhere as
 445                          * uncached, then it must also be accessed uncached
 446                          * by the kernel or data corruption may occur
 447                          */
 448                         kbuf = xlate_dev_kmem_ptr((char *)p);
 449
 450                         if (copy_to_user(buf, kbuf, sz))
 451                                 return -EFAULT;
 452                         buf += sz;
 453                         p += sz;
 454                         read += sz;
 455                         low_count -= sz;
 456                         count -= sz;
 457                 }
 458         }
 459
 460         if (count > 0) {
 461                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 462                 if (!kbuf)
 463                         return -ENOMEM;
 464                 while (count > 0) {
 465                         int len = count;
 466
 467                         if (len > PAGE_SIZE)
 468                                 len = PAGE_SIZE;
 469                         len = vread(kbuf, (char *)p, len);
 470                         if (!len)
 471                                 break;
 472                         if (copy_to_user(buf, kbuf, len)) {
 473                                 free_page((unsigned long)kbuf);
 474                                 return -EFAULT;
 475                         }
 476                         count -= len;
 477                         buf += len;
 478                         read += len;
 479                         p += len;
 480                 }
 481                 free_page((unsigned long)kbuf);
 482         }
 483         *ppos = p;
 484         return read;
 485 }
 486
 487
 488 static inline ssize_t
 489 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 490               size_t count, loff_t *ppos)
 491 {
 492         ssize_t written, sz;
 493         unsigned long copied;
 494
 495         written = 0;
 496 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 497         /* we don't have page 0 mapped on sparc and m68k.. */
 498         if (realp < PAGE_SIZE) {
 499                 unsigned long sz = PAGE_SIZE - realp;
 500                 if (sz > count)
 501                         sz = count;
 502                 /* Hmm. Do something? */
 503                 buf += sz;
 504                 p += sz;
 505                 realp += sz;
 506                 count -= sz;
 507                 written += sz;
 508         }
 509 #endif
 510
 511         while (count > 0) {
 512                 char *ptr;
 513                 /*
 514                  * Handle first page in case it's not aligned
 515                  */
 516                 if (-realp & (PAGE_SIZE - 1))
 517                         sz = -realp & (PAGE_SIZE - 1);
 518                 else
 519                         sz = PAGE_SIZE;
 520
 521                 sz = min_t(unsigned long, sz, count);
 522
 523                 /*
 524                  * On ia64 if a page has been mapped somewhere as
 525                  * uncached, then it must also be accessed uncached
 526                  * by the kernel or data corruption may occur
 527                  */
 528                 ptr = xlate_dev_kmem_ptr(p);
 529
 530                 copied = copy_from_user(ptr, buf, sz);
 531                 if (copied) {
 532                         written += sz - copied;
 533                         if (written)
 534                                 break;
 535                         return -EFAULT;
 536                 }
 537                 buf += sz;
 538                 p += sz;
 539                 realp += sz;
 540                 count -= sz;
 541                 written += sz;
 542         }
 543
 544         *ppos += written;
 545         return written;
 546 }
 547
 548
 549 /*
 550  * This function writes to the *virtual* memory as seen by the kernel.
 551  */
 552 static ssize_t write_kmem(struct file * file, const char __user * buf,
 553                           size_t count, loff_t *ppos)
 554 {
 555         unsigned long p = *ppos;
 556         ssize_t wrote = 0;
 557         ssize_t virtr = 0;
 558         ssize_t written;
 559         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 560
 561         if (p < (unsigned long) high_memory) {
 562
 563                 wrote = count;
 564                 if (count > (unsigned long) high_memory - p)
 565                         wrote = (unsigned long) high_memory - p;
 566
 567                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 568                 if (written != wrote)
 569                         return written;
 570                 wrote = written;
 571                 p += wrote;
 572                 buf += wrote;
 573                 count -= wrote;
 574         }
 575
 576         if (count > 0) {
 577                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 578                 if (!kbuf)
 579                         return wrote ? wrote : -ENOMEM;
 580                 while (count > 0) {
 581                         int len = count;
 582
 583                         if (len > PAGE_SIZE)
 584                                 len = PAGE_SIZE;
 585                         if (len) {
 586                                 written = copy_from_user(kbuf, buf, len);
 587                                 if (written) {
 588                                         if (wrote + virtr)
 589                                                 break;
 590                                         free_page((unsigned long)kbuf);
 591                                         return -EFAULT;
 592                                 }
 593                         }
 594                         len = vwrite(kbuf, (char *)p, len);
 595                         count -= len;
 596                         buf += len;
 597                         virtr += len;
 598                         p += len;
 599                 }
 600                 free_page((unsigned long)kbuf);
 601         }
 602
 603         *ppos = p;
 604         return virtr + wrote;
 605 }
 606 #endif
 607
 608 #ifdef CONFIG_DEVPORT
 609 static ssize_t read_port(struct file * file, char __user * buf,
 610                          size_t count, loff_t *ppos)
 611 {
 612         unsigned long i = *ppos;
 613         char __user *tmp = buf;
 614
 615         if (!access_ok(VERIFY_WRITE, buf, count))
 616                 return -EFAULT;
 617         while (count-- > 0 && i < 65536) {
 618                 if (__put_user(inb(i),tmp) < 0)
 619                         return -EFAULT;
 620                 i++;
 621                 tmp++;
 622         }
 623         *ppos = i;
 624         return tmp-buf;
 625 }
 626
 627 static ssize_t write_port(struct file * file, const char __user * buf,
 628                           size_t count, loff_t *ppos)
 629 {
 630         unsigned long i = *ppos;
 631         const char __user * tmp = buf;
 632
 633         if (!access_ok(VERIFY_READ,buf,count))
 634                 return -EFAULT;
 635         while (count-- > 0 && i < 65536) {
 636                 char c;
 637                 if (__get_user(c, tmp)) {
 638                         if (tmp > buf)
 639                                 break;
 640                         return -EFAULT;
 641                 }
 642                 outb(c,i);
 643                 i++;
 644                 tmp++;
 645         }
 646         *ppos = i;
 647         return tmp-buf;
 648 }
 649 #endif
 650
 651 static ssize_t read_null(struct file * file, char __user * buf,
 652                          size_t count, loff_t *ppos)
 653 {
 654         return 0;
 655 }
 656
 657 static ssize_t write_null(struct file * file, const char __user * buf,
 658                           size_t count, loff_t *ppos)
 659 {
 660         return count;
 661 }
 662
 663 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 664                         struct splice_desc *sd)
 665 {
 666         return sd->len;
 667 }
 668
 669 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 670                                  loff_t *ppos, size_t len, unsigned int flags)
 671 {
 672         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 673 }
 674
 675 static ssize_t read_zero(struct file * file, char __user * buf,
 676                          size_t count, loff_t *ppos)
 677 {
 678         size_t written;
 679
 680         if (!count)
 681                 return 0;
 682
 683         if (!access_ok(VERIFY_WRITE, buf, count))
 684                 return -EFAULT;
 685
 686         written = 0;
 687         while (count) {
 688                 unsigned long unwritten;
 689                 size_t chunk = count;
 690
 691                 if (chunk > PAGE_SIZE)
 692                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 693                 unwritten = clear_user(buf, chunk);
 694                 written += chunk - unwritten;
 695                 if (unwritten)
 696                         break;
 697                 /* Consider changing this to just 'signal_pending()' with lots of testing */
 698                 if (fatal_signal_pending(current))
 699                         return written ? written : -EINTR;
 700                 buf += chunk;
 701                 count -= chunk;
 702                 cond_resched();
 703         }
 704         return written ? written : -EFAULT;
 705 }
 706
 707 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 708 {
 709 #ifndef CONFIG_MMU
 710         return -ENOSYS;
 711 #endif
 712         if (vma->vm_flags & VM_SHARED)
 713                 return shmem_zero_setup(vma);
 714         return 0;
 715 }
 716
 717 static ssize_t write_full(struct file * file, const char __user * buf,
 718                           size_t count, loff_t *ppos)
 719 {
 720         return -ENOSPC;
 721 }
 722
 723 /*
 724  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 725  * can fopen() both devices with "a" now.  This was previously impossible.
 726  * -- SRB.
 727  */
 728
 729 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 730 {
 731         return file->f_pos = 0;
 732 }
 733
 734 /*
 735  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 736  * check against negative addresses: they are ok. The return value is weird,
 737  * though, in that case (0).
 738  *
 739  * also note that seeking relative to the "end of file" isn't supported:
 740  * it has no meaning, so it returns -EINVAL.
 741  */
 742 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 743 {
 744         loff_t ret;
 745
 746         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 747         switch (orig) {
 748                 case 0:
 749                         file->f_pos = offset;
 750                         ret = file->f_pos;
 751                         force_successful_syscall_return();
 752                         break;
 753                 case 1:
 754                         file->f_pos += offset;
 755                         ret = file->f_pos;
 756                         force_successful_syscall_return();
 757                         break;
 758                 default:
 759                         ret = -EINVAL;
 760         }
 761         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 762         return ret;
 763 }
 764
 765 static int open_port(struct inode * inode, struct file * filp)
 766 {
 767         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 768 }
 769
 770 #define zero_lseek      null_lseek
 771 #define full_lseek      null_lseek
 772 #define write_zero      write_null
 773 #define read_full       read_zero
 774 #define open_mem        open_port
 775 #define open_kmem       open_mem
 776 #define open_oldmem     open_mem
 777
 778 static const struct file_operations mem_fops = {
 779         .llseek         = memory_lseek,
 780         .read           = read_mem,
 781         .write          = write_mem,
 782         .mmap           = mmap_mem,
 783         .open           = open_mem,
 784         .get_unmapped_area = get_unmapped_area_mem,
 785 };
 786
 787 #ifdef CONFIG_DEVKMEM
 788 static const struct file_operations kmem_fops = {
 789         .llseek         = memory_lseek,
 790         .read           = read_kmem,
 791         .write          = write_kmem,
 792         .mmap           = mmap_kmem,
 793         .open           = open_kmem,
 794         .get_unmapped_area = get_unmapped_area_mem,
 795 };
 796 #endif
 797
 798 static const struct file_operations null_fops = {
 799         .llseek         = null_lseek,
 800         .read           = read_null,
 801         .write          = write_null,
 802         .splice_write   = splice_write_null,
 803 };
 804
 805 #ifdef CONFIG_DEVPORT
 806 static const struct file_operations port_fops = {
 807         .llseek         = memory_lseek,
 808         .read           = read_port,
 809         .write          = write_port,
 810         .open           = open_port,
 811 };
 812 #endif
 813
 814 static const struct file_operations zero_fops = {
 815         .llseek         = zero_lseek,
 816         .read           = read_zero,
 817         .write          = write_zero,
 818         .mmap           = mmap_zero,
 819 };
 820
 821 /*
 822  * capabilities for /dev/zero
 823  * - permits private mappings, "copies" are taken of the source of zeros
 824  */
 825 static struct backing_dev_info zero_bdi = {
 826         .capabilities   = BDI_CAP_MAP_COPY,
 827 };
 828
 829 static const struct file_operations full_fops = {
 830         .llseek         = full_lseek,
 831         .read           = read_full,
 832         .write          = write_full,
 833 };
 834
 835 #ifdef CONFIG_CRASH_DUMP
 836 static const struct file_operations oldmem_fops = {
 837         .read   = read_oldmem,
 838         .open   = open_oldmem,
 839 };
 840 #endif
 841
 842 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 843                           size_t count, loff_t *ppos)
 844 {
 845         char *tmp;
 846         ssize_t ret;
 847
 848         tmp = kmalloc(count + 1, GFP_KERNEL);
 849         if (tmp == NULL)
 850                 return -ENOMEM;
 851         ret = -EFAULT;
 852         if (!copy_from_user(tmp, buf, count)) {
 853                 tmp[count] = 0;
 854                 ret = printk("%s", tmp);
 855                 if (ret > count)
 856                         /* printk can add a prefix */
 857                         ret = count;
 858         }
 859         kfree(tmp);
 860         return ret;
 861 }
 862
 863 static const struct file_operations kmsg_fops = {
 864         .write =        kmsg_write,
 865 };
 866
 867 static int memory_open(struct inode * inode, struct file * filp)
 868 {
 869         int ret = 0;
 870
 871         lock_kernel();
 872         switch (iminor(inode)) {
 873                 case 1:
 874                         filp->f_op = &mem_fops;
 875                         filp->f_mapping->backing_dev_info =
 876                                 &directly_mappable_cdev_bdi;
 877                         break;
 878 #ifdef CONFIG_DEVKMEM
 879                 case 2:
 880                         filp->f_op = &kmem_fops;
 881                         filp->f_mapping->backing_dev_info =
 882                                 &directly_mappable_cdev_bdi;
 883                         break;
 884 #endif
 885                 case 3:
 886                         filp->f_op = &null_fops;
 887                         break;
 888 #ifdef CONFIG_DEVPORT
 889                 case 4:
 890                         filp->f_op = &port_fops;
 891                         break;
 892 #endif
 893                 case 5:
 894                         filp->f_mapping->backing_dev_info = &zero_bdi;
 895                         filp->f_op = &zero_fops;
 896                         break;
 897                 case 7:
 898                         filp->f_op = &full_fops;
 899                         break;
 900                 case 8:
 901                         filp->f_op = &random_fops;
 902                         break;
 903                 case 9:
 904                         filp->f_op = &urandom_fops;
 905                         break;
 906                 case 11:
 907                         filp->f_op = &kmsg_fops;
 908                         break;
 909 #ifdef CONFIG_CRASH_DUMP
 910                 case 12:
 911                         filp->f_op = &oldmem_fops;
 912                         break;
 913 #endif
 914                 default:
 915                         unlock_kernel();
 916                         return -ENXIO;
 917         }
 918         if (filp->f_op && filp->f_op->open)
 919                 ret = filp->f_op->open(inode,filp);
 920         unlock_kernel();
 921         return ret;
 922 }
 923
 924 static const struct file_operations memory_fops = {
 925         .open           = memory_open,  /* just a selector for the real open */
 926 };
 927
 928 static const struct {
 929         unsigned int            minor;
 930         char                    *name;
 931         umode_t                 mode;
 932         const struct file_operations    *fops;
 933 } devlist[] = { /* list of minor devices */
 934         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 935 #ifdef CONFIG_DEVKMEM
 936         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 937 #endif
 938         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 939 #ifdef CONFIG_DEVPORT
 940         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 941 #endif
 942         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 943         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 944         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 945         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 946         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 947 #ifdef CONFIG_CRASH_DUMP
 948         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 949 #endif
 950 };
 951
 952 static struct class *mem_class;
 953
 954 static int __init chr_dev_init(void)
 955 {
 956         int i;
 957         int err;
 958
 959         err = bdi_init(&zero_bdi);
 960         if (err)
 961                 return err;
 962
 963         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 964                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 965
 966         mem_class = class_create(THIS_MODULE, "mem");
 967         for (i = 0; i < ARRAY_SIZE(devlist); i++)
 968                 device_create(mem_class, NULL,
 969                               MKDEV(MEM_MAJOR, devlist[i].minor), NULL,
 970                               devlist[i].name);
 971
 972         return 0;
 973 }
 974
 975 fs_initcall(chr_dev_init);