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_NONPROMISC_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 void __attribute__((weak))
 305 map_devmem(unsigned long pfn, unsigned long len, pgprot_t prot)
 306 {
 307         /* nothing. architectures can override. */
 308 }
 309
 310 void __attribute__((weak))
 311 unmap_devmem(unsigned long pfn, unsigned long len, pgprot_t prot)
 312 {
 313         /* nothing. architectures can override. */
 314 }
 315
 316 static void mmap_mem_open(struct vm_area_struct *vma)
 317 {
 318         map_devmem(vma->vm_pgoff,  vma->vm_end - vma->vm_start,
 319                         vma->vm_page_prot);
 320 }
 321
 322 static void mmap_mem_close(struct vm_area_struct *vma)
 323 {
 324         unmap_devmem(vma->vm_pgoff,  vma->vm_end - vma->vm_start,
 325                         vma->vm_page_prot);
 326 }
 327
 328 static struct vm_operations_struct mmap_mem_ops = {
 329         .open  = mmap_mem_open,
 330         .close = mmap_mem_close
 331 };
 332
 333 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 334 {
 335         size_t size = vma->vm_end - vma->vm_start;
 336
 337         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 338                 return -EINVAL;
 339
 340         if (!private_mapping_ok(vma))
 341                 return -ENOSYS;
 342
 343         if (!range_is_allowed(vma->vm_pgoff, size))
 344                 return -EPERM;
 345
 346         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 347                                                 &vma->vm_page_prot))
 348                 return -EINVAL;
 349
 350         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 351                                                  size,
 352                                                  vma->vm_page_prot);
 353
 354         vma->vm_ops = &mmap_mem_ops;
 355
 356         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 357         if (remap_pfn_range(vma,
 358                             vma->vm_start,
 359                             vma->vm_pgoff,
 360                             size,
 361                             vma->vm_page_prot)) {
 362                 unmap_devmem(vma->vm_pgoff, size, vma->vm_page_prot);
 363                 return -EAGAIN;
 364         }
 365         return 0;
 366 }
 367
 368 #ifdef CONFIG_DEVKMEM
 369 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 370 {
 371         unsigned long pfn;
 372
 373         /* Turn a kernel-virtual address into a physical page frame */
 374         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 375
 376         /*
 377          * RED-PEN: on some architectures there is more mapped memory
 378          * than available in mem_map which pfn_valid checks
 379          * for. Perhaps should add a new macro here.
 380          *
 381          * RED-PEN: vmalloc is not supported right now.
 382          */
 383         if (!pfn_valid(pfn))
 384                 return -EIO;
 385
 386         vma->vm_pgoff = pfn;
 387         return mmap_mem(file, vma);
 388 }
 389 #endif
 390
 391 #ifdef CONFIG_CRASH_DUMP
 392 /*
 393  * Read memory corresponding to the old kernel.
 394  */
 395 static ssize_t read_oldmem(struct file *file, char __user *buf,
 396                                 size_t count, loff_t *ppos)
 397 {
 398         unsigned long pfn, offset;
 399         size_t read = 0, csize;
 400         int rc = 0;
 401
 402         while (count) {
 403                 pfn = *ppos / PAGE_SIZE;
 404                 if (pfn > saved_max_pfn)
 405                         return read;
 406
 407                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 408                 if (count > PAGE_SIZE - offset)
 409                         csize = PAGE_SIZE - offset;
 410                 else
 411                         csize = count;
 412
 413                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 414                 if (rc < 0)
 415                         return rc;
 416                 buf += csize;
 417                 *ppos += csize;
 418                 read += csize;
 419                 count -= csize;
 420         }
 421         return read;
 422 }
 423 #endif
 424
 425 extern long vread(char *buf, char *addr, unsigned long count);
 426 extern long vwrite(char *buf, char *addr, unsigned long count);
 427
 428 #ifdef CONFIG_DEVKMEM
 429 /*
 430  * This function reads the *virtual* memory as seen by the kernel.
 431  */
 432 static ssize_t read_kmem(struct file *file, char __user *buf,
 433                          size_t count, loff_t *ppos)
 434 {
 435         unsigned long p = *ppos;
 436         ssize_t low_count, read, sz;
 437         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 438
 439         read = 0;
 440         if (p < (unsigned long) high_memory) {
 441                 low_count = count;
 442                 if (count > (unsigned long) high_memory - p)
 443                         low_count = (unsigned long) high_memory - p;
 444
 445 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 446                 /* we don't have page 0 mapped on sparc and m68k.. */
 447                 if (p < PAGE_SIZE && low_count > 0) {
 448                         size_t tmp = PAGE_SIZE - p;
 449                         if (tmp > low_count) tmp = low_count;
 450                         if (clear_user(buf, tmp))
 451                                 return -EFAULT;
 452                         buf += tmp;
 453                         p += tmp;
 454                         read += tmp;
 455                         low_count -= tmp;
 456                         count -= tmp;
 457                 }
 458 #endif
 459                 while (low_count > 0) {
 460                         /*
 461                          * Handle first page in case it's not aligned
 462                          */
 463                         if (-p & (PAGE_SIZE - 1))
 464                                 sz = -p & (PAGE_SIZE - 1);
 465                         else
 466                                 sz = PAGE_SIZE;
 467
 468                         sz = min_t(unsigned long, sz, low_count);
 469
 470                         /*
 471                          * On ia64 if a page has been mapped somewhere as
 472                          * uncached, then it must also be accessed uncached
 473                          * by the kernel or data corruption may occur
 474                          */
 475                         kbuf = xlate_dev_kmem_ptr((char *)p);
 476
 477                         if (copy_to_user(buf, kbuf, sz))
 478                                 return -EFAULT;
 479                         buf += sz;
 480                         p += sz;
 481                         read += sz;
 482                         low_count -= sz;
 483                         count -= sz;
 484                 }
 485         }
 486
 487         if (count > 0) {
 488                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 489                 if (!kbuf)
 490                         return -ENOMEM;
 491                 while (count > 0) {
 492                         int len = count;
 493
 494                         if (len > PAGE_SIZE)
 495                                 len = PAGE_SIZE;
 496                         len = vread(kbuf, (char *)p, len);
 497                         if (!len)
 498                                 break;
 499                         if (copy_to_user(buf, kbuf, len)) {
 500                                 free_page((unsigned long)kbuf);
 501                                 return -EFAULT;
 502                         }
 503                         count -= len;
 504                         buf += len;
 505                         read += len;
 506                         p += len;
 507                 }
 508                 free_page((unsigned long)kbuf);
 509         }
 510         *ppos = p;
 511         return read;
 512 }
 513
 514
 515 static inline ssize_t
 516 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 517               size_t count, loff_t *ppos)
 518 {
 519         ssize_t written, sz;
 520         unsigned long copied;
 521
 522         written = 0;
 523 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 524         /* we don't have page 0 mapped on sparc and m68k.. */
 525         if (realp < PAGE_SIZE) {
 526                 unsigned long sz = PAGE_SIZE - realp;
 527                 if (sz > count)
 528                         sz = count;
 529                 /* Hmm. Do something? */
 530                 buf += sz;
 531                 p += sz;
 532                 realp += sz;
 533                 count -= sz;
 534                 written += sz;
 535         }
 536 #endif
 537
 538         while (count > 0) {
 539                 char *ptr;
 540                 /*
 541                  * Handle first page in case it's not aligned
 542                  */
 543                 if (-realp & (PAGE_SIZE - 1))
 544                         sz = -realp & (PAGE_SIZE - 1);
 545                 else
 546                         sz = PAGE_SIZE;
 547
 548                 sz = min_t(unsigned long, sz, count);
 549
 550                 /*
 551                  * On ia64 if a page has been mapped somewhere as
 552                  * uncached, then it must also be accessed uncached
 553                  * by the kernel or data corruption may occur
 554                  */
 555                 ptr = xlate_dev_kmem_ptr(p);
 556
 557                 copied = copy_from_user(ptr, buf, sz);
 558                 if (copied) {
 559                         written += sz - copied;
 560                         if (written)
 561                                 break;
 562                         return -EFAULT;
 563                 }
 564                 buf += sz;
 565                 p += sz;
 566                 realp += sz;
 567                 count -= sz;
 568                 written += sz;
 569         }
 570
 571         *ppos += written;
 572         return written;
 573 }
 574
 575
 576 /*
 577  * This function writes to the *virtual* memory as seen by the kernel.
 578  */
 579 static ssize_t write_kmem(struct file * file, const char __user * buf,
 580                           size_t count, loff_t *ppos)
 581 {
 582         unsigned long p = *ppos;
 583         ssize_t wrote = 0;
 584         ssize_t virtr = 0;
 585         ssize_t written;
 586         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 587
 588         if (p < (unsigned long) high_memory) {
 589
 590                 wrote = count;
 591                 if (count > (unsigned long) high_memory - p)
 592                         wrote = (unsigned long) high_memory - p;
 593
 594                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 595                 if (written != wrote)
 596                         return written;
 597                 wrote = written;
 598                 p += wrote;
 599                 buf += wrote;
 600                 count -= wrote;
 601         }
 602
 603         if (count > 0) {
 604                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 605                 if (!kbuf)
 606                         return wrote ? wrote : -ENOMEM;
 607                 while (count > 0) {
 608                         int len = count;
 609
 610                         if (len > PAGE_SIZE)
 611                                 len = PAGE_SIZE;
 612                         if (len) {
 613                                 written = copy_from_user(kbuf, buf, len);
 614                                 if (written) {
 615                                         if (wrote + virtr)
 616                                                 break;
 617                                         free_page((unsigned long)kbuf);
 618                                         return -EFAULT;
 619                                 }
 620                         }
 621                         len = vwrite(kbuf, (char *)p, len);
 622                         count -= len;
 623                         buf += len;
 624                         virtr += len;
 625                         p += len;
 626                 }
 627                 free_page((unsigned long)kbuf);
 628         }
 629
 630         *ppos = p;
 631         return virtr + wrote;
 632 }
 633 #endif
 634
 635 #ifdef CONFIG_DEVPORT
 636 static ssize_t read_port(struct file * file, char __user * buf,
 637                          size_t count, loff_t *ppos)
 638 {
 639         unsigned long i = *ppos;
 640         char __user *tmp = buf;
 641
 642         if (!access_ok(VERIFY_WRITE, buf, count))
 643                 return -EFAULT;
 644         while (count-- > 0 && i < 65536) {
 645                 if (__put_user(inb(i),tmp) < 0)
 646                         return -EFAULT;
 647                 i++;
 648                 tmp++;
 649         }
 650         *ppos = i;
 651         return tmp-buf;
 652 }
 653
 654 static ssize_t write_port(struct file * file, const char __user * buf,
 655                           size_t count, loff_t *ppos)
 656 {
 657         unsigned long i = *ppos;
 658         const char __user * tmp = buf;
 659
 660         if (!access_ok(VERIFY_READ,buf,count))
 661                 return -EFAULT;
 662         while (count-- > 0 && i < 65536) {
 663                 char c;
 664                 if (__get_user(c, tmp)) {
 665                         if (tmp > buf)
 666                                 break;
 667                         return -EFAULT;
 668                 }
 669                 outb(c,i);
 670                 i++;
 671                 tmp++;
 672         }
 673         *ppos = i;
 674         return tmp-buf;
 675 }
 676 #endif
 677
 678 static ssize_t read_null(struct file * file, char __user * buf,
 679                          size_t count, loff_t *ppos)
 680 {
 681         return 0;
 682 }
 683
 684 static ssize_t write_null(struct file * file, const char __user * buf,
 685                           size_t count, loff_t *ppos)
 686 {
 687         return count;
 688 }
 689
 690 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 691                         struct splice_desc *sd)
 692 {
 693         return sd->len;
 694 }
 695
 696 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 697                                  loff_t *ppos, size_t len, unsigned int flags)
 698 {
 699         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 700 }
 701
 702 static ssize_t read_zero(struct file * file, char __user * buf,
 703                          size_t count, loff_t *ppos)
 704 {
 705         size_t written;
 706
 707         if (!count)
 708                 return 0;
 709
 710         if (!access_ok(VERIFY_WRITE, buf, count))
 711                 return -EFAULT;
 712
 713         written = 0;
 714         while (count) {
 715                 unsigned long unwritten;
 716                 size_t chunk = count;
 717
 718                 if (chunk > PAGE_SIZE)
 719                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 720                 unwritten = clear_user(buf, chunk);
 721                 written += chunk - unwritten;
 722                 if (unwritten)
 723                         break;
 724                 buf += chunk;
 725                 count -= chunk;
 726                 cond_resched();
 727         }
 728         return written ? written : -EFAULT;
 729 }
 730
 731 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 732 {
 733 #ifndef CONFIG_MMU
 734         return -ENOSYS;
 735 #endif
 736         if (vma->vm_flags & VM_SHARED)
 737                 return shmem_zero_setup(vma);
 738         return 0;
 739 }
 740
 741 static ssize_t write_full(struct file * file, const char __user * buf,
 742                           size_t count, loff_t *ppos)
 743 {
 744         return -ENOSPC;
 745 }
 746
 747 /*
 748  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 749  * can fopen() both devices with "a" now.  This was previously impossible.
 750  * -- SRB.
 751  */
 752
 753 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 754 {
 755         return file->f_pos = 0;
 756 }
 757
 758 /*
 759  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 760  * check against negative addresses: they are ok. The return value is weird,
 761  * though, in that case (0).
 762  *
 763  * also note that seeking relative to the "end of file" isn't supported:
 764  * it has no meaning, so it returns -EINVAL.
 765  */
 766 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 767 {
 768         loff_t ret;
 769
 770         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 771         switch (orig) {
 772                 case 0:
 773                         file->f_pos = offset;
 774                         ret = file->f_pos;
 775                         force_successful_syscall_return();
 776                         break;
 777                 case 1:
 778                         file->f_pos += offset;
 779                         ret = file->f_pos;
 780                         force_successful_syscall_return();
 781                         break;
 782                 default:
 783                         ret = -EINVAL;
 784         }
 785         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 786         return ret;
 787 }
 788
 789 static int open_port(struct inode * inode, struct file * filp)
 790 {
 791         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 792 }
 793
 794 #define zero_lseek      null_lseek
 795 #define full_lseek      null_lseek
 796 #define write_zero      write_null
 797 #define read_full       read_zero
 798 #define open_mem        open_port
 799 #define open_kmem       open_mem
 800 #define open_oldmem     open_mem
 801
 802 static const struct file_operations mem_fops = {
 803         .llseek         = memory_lseek,
 804         .read           = read_mem,
 805         .write          = write_mem,
 806         .mmap           = mmap_mem,
 807         .open           = open_mem,
 808         .get_unmapped_area = get_unmapped_area_mem,
 809 };
 810
 811 #ifdef CONFIG_DEVKMEM
 812 static const struct file_operations kmem_fops = {
 813         .llseek         = memory_lseek,
 814         .read           = read_kmem,
 815         .write          = write_kmem,
 816         .mmap           = mmap_kmem,
 817         .open           = open_kmem,
 818         .get_unmapped_area = get_unmapped_area_mem,
 819 };
 820 #endif
 821
 822 static const struct file_operations null_fops = {
 823         .llseek         = null_lseek,
 824         .read           = read_null,
 825         .write          = write_null,
 826         .splice_write   = splice_write_null,
 827 };
 828
 829 #ifdef CONFIG_DEVPORT
 830 static const struct file_operations port_fops = {
 831         .llseek         = memory_lseek,
 832         .read           = read_port,
 833         .write          = write_port,
 834         .open           = open_port,
 835 };
 836 #endif
 837
 838 static const struct file_operations zero_fops = {
 839         .llseek         = zero_lseek,
 840         .read           = read_zero,
 841         .write          = write_zero,
 842         .mmap           = mmap_zero,
 843 };
 844
 845 /*
 846  * capabilities for /dev/zero
 847  * - permits private mappings, "copies" are taken of the source of zeros
 848  */
 849 static struct backing_dev_info zero_bdi = {
 850         .capabilities   = BDI_CAP_MAP_COPY,
 851 };
 852
 853 static const struct file_operations full_fops = {
 854         .llseek         = full_lseek,
 855         .read           = read_full,
 856         .write          = write_full,
 857 };
 858
 859 #ifdef CONFIG_CRASH_DUMP
 860 static const struct file_operations oldmem_fops = {
 861         .read   = read_oldmem,
 862         .open   = open_oldmem,
 863 };
 864 #endif
 865
 866 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 867                           size_t count, loff_t *ppos)
 868 {
 869         char *tmp;
 870         ssize_t ret;
 871
 872         tmp = kmalloc(count + 1, GFP_KERNEL);
 873         if (tmp == NULL)
 874                 return -ENOMEM;
 875         ret = -EFAULT;
 876         if (!copy_from_user(tmp, buf, count)) {
 877                 tmp[count] = 0;
 878                 ret = printk("%s", tmp);
 879                 if (ret > count)
 880                         /* printk can add a prefix */
 881                         ret = count;
 882         }
 883         kfree(tmp);
 884         return ret;
 885 }
 886
 887 static const struct file_operations kmsg_fops = {
 888         .write =        kmsg_write,
 889 };
 890
 891 static int memory_open(struct inode * inode, struct file * filp)
 892 {
 893         int ret = 0;
 894
 895         lock_kernel();
 896         switch (iminor(inode)) {
 897                 case 1:
 898                         filp->f_op = &mem_fops;
 899                         filp->f_mapping->backing_dev_info =
 900                                 &directly_mappable_cdev_bdi;
 901                         break;
 902 #ifdef CONFIG_DEVKMEM
 903                 case 2:
 904                         filp->f_op = &kmem_fops;
 905                         filp->f_mapping->backing_dev_info =
 906                                 &directly_mappable_cdev_bdi;
 907                         break;
 908 #endif
 909                 case 3:
 910                         filp->f_op = &null_fops;
 911                         break;
 912 #ifdef CONFIG_DEVPORT
 913                 case 4:
 914                         filp->f_op = &port_fops;
 915                         break;
 916 #endif
 917                 case 5:
 918                         filp->f_mapping->backing_dev_info = &zero_bdi;
 919                         filp->f_op = &zero_fops;
 920                         break;
 921                 case 7:
 922                         filp->f_op = &full_fops;
 923                         break;
 924                 case 8:
 925                         filp->f_op = &random_fops;
 926                         break;
 927                 case 9:
 928                         filp->f_op = &urandom_fops;
 929                         break;
 930                 case 11:
 931                         filp->f_op = &kmsg_fops;
 932                         break;
 933 #ifdef CONFIG_CRASH_DUMP
 934                 case 12:
 935                         filp->f_op = &oldmem_fops;
 936                         break;
 937 #endif
 938                 default:
 939                         unlock_kernel();
 940                         return -ENXIO;
 941         }
 942         if (filp->f_op && filp->f_op->open)
 943                 ret = filp->f_op->open(inode,filp);
 944         unlock_kernel();
 945         return ret;
 946 }
 947
 948 static const struct file_operations memory_fops = {
 949         .open           = memory_open,  /* just a selector for the real open */
 950 };
 951
 952 static const struct {
 953         unsigned int            minor;
 954         char                    *name;
 955         umode_t                 mode;
 956         const struct file_operations    *fops;
 957 } devlist[] = { /* list of minor devices */
 958         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 959 #ifdef CONFIG_DEVKMEM
 960         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 961 #endif
 962         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 963 #ifdef CONFIG_DEVPORT
 964         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 965 #endif
 966         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 967         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 968         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 969         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 970         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 971 #ifdef CONFIG_CRASH_DUMP
 972         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 973 #endif
 974 };
 975
 976 static struct class *mem_class;
 977
 978 static int __init chr_dev_init(void)
 979 {
 980         int i;
 981         int err;
 982
 983         err = bdi_init(&zero_bdi);
 984         if (err)
 985                 return err;
 986
 987         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 988                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 989
 990         mem_class = class_create(THIS_MODULE, "mem");
 991         for (i = 0; i < ARRAY_SIZE(devlist); i++)
 992                 device_create(mem_class, NULL,
 993                               MKDEV(MEM_MAJOR, devlist[i].minor),
 994                               devlist[i].name);
 995
 996         return 0;
 997 }
 998
 999 fs_initcall(chr_dev_init);