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 <[email protected]>
   8  *  Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <[email protected]>
   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/backing-dev.h>
  25 #include <linux/splice.h>
  26 #include <linux/pfn.h>
  27 #include <linux/export.h>
  28 #include <linux/io.h>
  29 #include <linux/aio.h>
  30
  31 #include <asm/uaccess.h>
  32
  33 #ifdef CONFIG_IA64
  34 # include <linux/efi.h>
  35 #endif
  36
  37 #define DEVPORT_MINOR   4
  38
  39 static inline unsigned long size_inside_page(unsigned long start,
  40                                              unsigned long size)
  41 {
  42         unsigned long sz;
  43
  44         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  45
  46         return min(sz, size);
  47 }
  48
  49 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  50 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
  51 {
  52         return addr + count <= __pa(high_memory);
  53 }
  54
  55 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  56 {
  57         return 1;
  58 }
  59 #endif
  60
  61 #ifdef CONFIG_STRICT_DEVMEM
  62 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  63 {
  64         u64 from = ((u64)pfn) << PAGE_SHIFT;
  65         u64 to = from + size;
  66         u64 cursor = from;
  67
  68         while (cursor < to) {
  69                 if (!devmem_is_allowed(pfn)) {
  70                         printk(KERN_INFO
  71                 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
  72                                 current->comm, from, to);
  73                         return 0;
  74                 }
  75                 cursor += PAGE_SIZE;
  76                 pfn++;
  77         }
  78         return 1;
  79 }
  80 #else
  81 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  82 {
  83         return 1;
  84 }
  85 #endif
  86
  87 void __weak unxlate_dev_mem_ptr(unsigned long phys, void *addr)
  88 {
  89 }
  90
  91 /*
  92  * This funcion reads the *physical* memory. The f_pos points directly to the
  93  * memory location.
  94  */
  95 static ssize_t read_mem(struct file *file, char __user *buf,
  96                         size_t count, loff_t *ppos)
  97 {
  98         phys_addr_t p = *ppos;
  99         ssize_t read, sz;
 100         char *ptr;
 101
 102         if (!valid_phys_addr_range(p, count))
 103                 return -EFAULT;
 104         read = 0;
 105 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 106         /* we don't have page 0 mapped on sparc and m68k.. */
 107         if (p < PAGE_SIZE) {
 108                 sz = size_inside_page(p, count);
 109                 if (sz > 0) {
 110                         if (clear_user(buf, sz))
 111                                 return -EFAULT;
 112                         buf += sz;
 113                         p += sz;
 114                         count -= sz;
 115                         read += sz;
 116                 }
 117         }
 118 #endif
 119
 120         while (count > 0) {
 121                 unsigned long remaining;
 122
 123                 sz = size_inside_page(p, count);
 124
 125                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 126                         return -EPERM;
 127
 128                 /*
 129                  * On ia64 if a page has been mapped somewhere as uncached, then
 130                  * it must also be accessed uncached by the kernel or data
 131                  * corruption may occur.
 132                  */
 133                 ptr = xlate_dev_mem_ptr(p);
 134                 if (!ptr)
 135                         return -EFAULT;
 136
 137                 remaining = copy_to_user(buf, ptr, sz);
 138                 unxlate_dev_mem_ptr(p, ptr);
 139                 if (remaining)
 140                         return -EFAULT;
 141
 142                 buf += sz;
 143                 p += sz;
 144                 count -= sz;
 145                 read += sz;
 146         }
 147
 148         *ppos += read;
 149         return read;
 150 }
 151
 152 static ssize_t write_mem(struct file *file, const char __user *buf,
 153                          size_t count, loff_t *ppos)
 154 {
 155         phys_addr_t p = *ppos;
 156         ssize_t written, sz;
 157         unsigned long copied;
 158         void *ptr;
 159
 160         if (!valid_phys_addr_range(p, count))
 161                 return -EFAULT;
 162
 163         written = 0;
 164
 165 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 166         /* we don't have page 0 mapped on sparc and m68k.. */
 167         if (p < PAGE_SIZE) {
 168                 sz = size_inside_page(p, count);
 169                 /* Hmm. Do something? */
 170                 buf += sz;
 171                 p += sz;
 172                 count -= sz;
 173                 written += sz;
 174         }
 175 #endif
 176
 177         while (count > 0) {
 178                 sz = size_inside_page(p, count);
 179
 180                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 181                         return -EPERM;
 182
 183                 /*
 184                  * On ia64 if a page has been mapped somewhere as uncached, then
 185                  * it must also be accessed uncached by the kernel or data
 186                  * corruption may occur.
 187                  */
 188                 ptr = xlate_dev_mem_ptr(p);
 189                 if (!ptr) {
 190                         if (written)
 191                                 break;
 192                         return -EFAULT;
 193                 }
 194
 195                 copied = copy_from_user(ptr, buf, sz);
 196                 unxlate_dev_mem_ptr(p, ptr);
 197                 if (copied) {
 198                         written += sz - copied;
 199                         if (written)
 200                                 break;
 201                         return -EFAULT;
 202                 }
 203
 204                 buf += sz;
 205                 p += sz;
 206                 count -= sz;
 207                 written += sz;
 208         }
 209
 210         *ppos += written;
 211         return written;
 212 }
 213
 214 int __weak phys_mem_access_prot_allowed(struct file *file,
 215         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 216 {
 217         return 1;
 218 }
 219
 220 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 221
 222 /*
 223  * Architectures vary in how they handle caching for addresses
 224  * outside of main memory.
 225  *
 226  */
 227 #ifdef pgprot_noncached
 228 static int uncached_access(struct file *file, phys_addr_t addr)
 229 {
 230 #if defined(CONFIG_IA64)
 231         /*
 232          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 233          * attribute aliases.
 234          */
 235         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 236 #elif defined(CONFIG_MIPS)
 237         {
 238                 extern int __uncached_access(struct file *file,
 239                                              unsigned long addr);
 240
 241                 return __uncached_access(file, addr);
 242         }
 243 #else
 244         /*
 245          * Accessing memory above the top the kernel knows about or through a
 246          * file pointer
 247          * that was marked O_DSYNC will be done non-cached.
 248          */
 249         if (file->f_flags & O_DSYNC)
 250                 return 1;
 251         return addr >= __pa(high_memory);
 252 #endif
 253 }
 254 #endif
 255
 256 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 257                                      unsigned long size, pgprot_t vma_prot)
 258 {
 259 #ifdef pgprot_noncached
 260         phys_addr_t offset = pfn << PAGE_SHIFT;
 261
 262         if (uncached_access(file, offset))
 263                 return pgprot_noncached(vma_prot);
 264 #endif
 265         return vma_prot;
 266 }
 267 #endif
 268
 269 #ifndef CONFIG_MMU
 270 static unsigned long get_unmapped_area_mem(struct file *file,
 271                                            unsigned long addr,
 272                                            unsigned long len,
 273                                            unsigned long pgoff,
 274                                            unsigned long flags)
 275 {
 276         if (!valid_mmap_phys_addr_range(pgoff, len))
 277                 return (unsigned long) -EINVAL;
 278         return pgoff << PAGE_SHIFT;
 279 }
 280
 281 /* can't do an in-place private mapping if there's no MMU */
 282 static inline int private_mapping_ok(struct vm_area_struct *vma)
 283 {
 284         return vma->vm_flags & VM_MAYSHARE;
 285 }
 286 #else
 287 #define get_unmapped_area_mem   NULL
 288
 289 static inline int private_mapping_ok(struct vm_area_struct *vma)
 290 {
 291         return 1;
 292 }
 293 #endif
 294
 295 static const struct vm_operations_struct mmap_mem_ops = {
 296 #ifdef CONFIG_HAVE_IOREMAP_PROT
 297         .access = generic_access_phys
 298 #endif
 299 };
 300
 301 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 302 {
 303         size_t size = vma->vm_end - vma->vm_start;
 304
 305         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 306                 return -EINVAL;
 307
 308         if (!private_mapping_ok(vma))
 309                 return -ENOSYS;
 310
 311         if (!range_is_allowed(vma->vm_pgoff, size))
 312                 return -EPERM;
 313
 314         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 315                                                 &vma->vm_page_prot))
 316                 return -EINVAL;
 317
 318         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 319                                                  size,
 320                                                  vma->vm_page_prot);
 321
 322         vma->vm_ops = &mmap_mem_ops;
 323
 324         /* Remap-pfn-range will mark the range VM_IO */
 325         if (remap_pfn_range(vma,
 326                             vma->vm_start,
 327                             vma->vm_pgoff,
 328                             size,
 329                             vma->vm_page_prot)) {
 330                 return -EAGAIN;
 331         }
 332         return 0;
 333 }
 334
 335 #ifdef CONFIG_DEVKMEM
 336 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 337 {
 338         unsigned long pfn;
 339
 340         /* Turn a kernel-virtual address into a physical page frame */
 341         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 342
 343         /*
 344          * RED-PEN: on some architectures there is more mapped memory than
 345          * available in mem_map which pfn_valid checks for. Perhaps should add a
 346          * new macro here.
 347          *
 348          * RED-PEN: vmalloc is not supported right now.
 349          */
 350         if (!pfn_valid(pfn))
 351                 return -EIO;
 352
 353         vma->vm_pgoff = pfn;
 354         return mmap_mem(file, vma);
 355 }
 356 #endif
 357
 358 #ifdef CONFIG_DEVKMEM
 359 /*
 360  * This function reads the *virtual* memory as seen by the kernel.
 361  */
 362 static ssize_t read_kmem(struct file *file, char __user *buf,
 363                          size_t count, loff_t *ppos)
 364 {
 365         unsigned long p = *ppos;
 366         ssize_t low_count, read, sz;
 367         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 368         int err = 0;
 369
 370         read = 0;
 371         if (p < (unsigned long) high_memory) {
 372                 low_count = count;
 373                 if (count > (unsigned long)high_memory - p)
 374                         low_count = (unsigned long)high_memory - p;
 375
 376 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 377                 /* we don't have page 0 mapped on sparc and m68k.. */
 378                 if (p < PAGE_SIZE && low_count > 0) {
 379                         sz = size_inside_page(p, low_count);
 380                         if (clear_user(buf, sz))
 381                                 return -EFAULT;
 382                         buf += sz;
 383                         p += sz;
 384                         read += sz;
 385                         low_count -= sz;
 386                         count -= sz;
 387                 }
 388 #endif
 389                 while (low_count > 0) {
 390                         sz = size_inside_page(p, low_count);
 391
 392                         /*
 393                          * On ia64 if a page has been mapped somewhere as
 394                          * uncached, then it must also be accessed uncached
 395                          * by the kernel or data corruption may occur
 396                          */
 397                         kbuf = xlate_dev_kmem_ptr((char *)p);
 398
 399                         if (copy_to_user(buf, kbuf, sz))
 400                                 return -EFAULT;
 401                         buf += sz;
 402                         p += sz;
 403                         read += sz;
 404                         low_count -= sz;
 405                         count -= sz;
 406                 }
 407         }
 408
 409         if (count > 0) {
 410                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 411                 if (!kbuf)
 412                         return -ENOMEM;
 413                 while (count > 0) {
 414                         sz = size_inside_page(p, count);
 415                         if (!is_vmalloc_or_module_addr((void *)p)) {
 416                                 err = -ENXIO;
 417                                 break;
 418                         }
 419                         sz = vread(kbuf, (char *)p, sz);
 420                         if (!sz)
 421                                 break;
 422                         if (copy_to_user(buf, kbuf, sz)) {
 423                                 err = -EFAULT;
 424                                 break;
 425                         }
 426                         count -= sz;
 427                         buf += sz;
 428                         read += sz;
 429                         p += sz;
 430                 }
 431                 free_page((unsigned long)kbuf);
 432         }
 433         *ppos = p;
 434         return read ? read : err;
 435 }
 436
 437
 438 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 439                                 size_t count, loff_t *ppos)
 440 {
 441         ssize_t written, sz;
 442         unsigned long copied;
 443
 444         written = 0;
 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) {
 448                 sz = size_inside_page(p, count);
 449                 /* Hmm. Do something? */
 450                 buf += sz;
 451                 p += sz;
 452                 count -= sz;
 453                 written += sz;
 454         }
 455 #endif
 456
 457         while (count > 0) {
 458                 char *ptr;
 459
 460                 sz = size_inside_page(p, count);
 461
 462                 /*
 463                  * On ia64 if a page has been mapped somewhere as uncached, then
 464                  * it must also be accessed uncached by the kernel or data
 465                  * corruption may occur.
 466                  */
 467                 ptr = xlate_dev_kmem_ptr((char *)p);
 468
 469                 copied = copy_from_user(ptr, buf, sz);
 470                 if (copied) {
 471                         written += sz - copied;
 472                         if (written)
 473                                 break;
 474                         return -EFAULT;
 475                 }
 476                 buf += sz;
 477                 p += sz;
 478                 count -= sz;
 479                 written += sz;
 480         }
 481
 482         *ppos += written;
 483         return written;
 484 }
 485
 486 /*
 487  * This function writes to the *virtual* memory as seen by the kernel.
 488  */
 489 static ssize_t write_kmem(struct file *file, const char __user *buf,
 490                           size_t count, loff_t *ppos)
 491 {
 492         unsigned long p = *ppos;
 493         ssize_t wrote = 0;
 494         ssize_t virtr = 0;
 495         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 496         int err = 0;
 497
 498         if (p < (unsigned long) high_memory) {
 499                 unsigned long to_write = min_t(unsigned long, count,
 500                                                (unsigned long)high_memory - p);
 501                 wrote = do_write_kmem(p, buf, to_write, ppos);
 502                 if (wrote != to_write)
 503                         return wrote;
 504                 p += wrote;
 505                 buf += wrote;
 506                 count -= wrote;
 507         }
 508
 509         if (count > 0) {
 510                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 511                 if (!kbuf)
 512                         return wrote ? wrote : -ENOMEM;
 513                 while (count > 0) {
 514                         unsigned long sz = size_inside_page(p, count);
 515                         unsigned long n;
 516
 517                         if (!is_vmalloc_or_module_addr((void *)p)) {
 518                                 err = -ENXIO;
 519                                 break;
 520                         }
 521                         n = copy_from_user(kbuf, buf, sz);
 522                         if (n) {
 523                                 err = -EFAULT;
 524                                 break;
 525                         }
 526                         vwrite(kbuf, (char *)p, sz);
 527                         count -= sz;
 528                         buf += sz;
 529                         virtr += sz;
 530                         p += sz;
 531                 }
 532                 free_page((unsigned long)kbuf);
 533         }
 534
 535         *ppos = p;
 536         return virtr + wrote ? : err;
 537 }
 538 #endif
 539
 540 #ifdef CONFIG_DEVPORT
 541 static ssize_t read_port(struct file *file, char __user *buf,
 542                          size_t count, loff_t *ppos)
 543 {
 544         unsigned long i = *ppos;
 545         char __user *tmp = buf;
 546
 547         if (!access_ok(VERIFY_WRITE, buf, count))
 548                 return -EFAULT;
 549         while (count-- > 0 && i < 65536) {
 550                 if (__put_user(inb(i), tmp) < 0)
 551                         return -EFAULT;
 552                 i++;
 553                 tmp++;
 554         }
 555         *ppos = i;
 556         return tmp-buf;
 557 }
 558
 559 static ssize_t write_port(struct file *file, const char __user *buf,
 560                           size_t count, loff_t *ppos)
 561 {
 562         unsigned long i = *ppos;
 563         const char __user *tmp = buf;
 564
 565         if (!access_ok(VERIFY_READ, buf, count))
 566                 return -EFAULT;
 567         while (count-- > 0 && i < 65536) {
 568                 char c;
 569                 if (__get_user(c, tmp)) {
 570                         if (tmp > buf)
 571                                 break;
 572                         return -EFAULT;
 573                 }
 574                 outb(c, i);
 575                 i++;
 576                 tmp++;
 577         }
 578         *ppos = i;
 579         return tmp-buf;
 580 }
 581 #endif
 582
 583 static ssize_t read_null(struct file *file, char __user *buf,
 584                          size_t count, loff_t *ppos)
 585 {
 586         return 0;
 587 }
 588
 589 static ssize_t write_null(struct file *file, const char __user *buf,
 590                           size_t count, loff_t *ppos)
 591 {
 592         return count;
 593 }
 594
 595 static ssize_t aio_read_null(struct kiocb *iocb, const struct iovec *iov,
 596                              unsigned long nr_segs, loff_t pos)
 597 {
 598         return 0;
 599 }
 600
 601 static ssize_t aio_write_null(struct kiocb *iocb, const struct iovec *iov,
 602                               unsigned long nr_segs, loff_t pos)
 603 {
 604         return iov_length(iov, nr_segs);
 605 }
 606
 607 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 608                         struct splice_desc *sd)
 609 {
 610         return sd->len;
 611 }
 612
 613 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 614                                  loff_t *ppos, size_t len, unsigned int flags)
 615 {
 616         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 617 }
 618
 619 static ssize_t read_zero(struct file *file, char __user *buf,
 620                          size_t count, loff_t *ppos)
 621 {
 622         size_t written;
 623
 624         if (!count)
 625                 return 0;
 626
 627         if (!access_ok(VERIFY_WRITE, buf, count))
 628                 return -EFAULT;
 629
 630         written = 0;
 631         while (count) {
 632                 unsigned long unwritten;
 633                 size_t chunk = count;
 634
 635                 if (chunk > PAGE_SIZE)
 636                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 637                 unwritten = __clear_user(buf, chunk);
 638                 written += chunk - unwritten;
 639                 if (unwritten)
 640                         break;
 641                 if (signal_pending(current))
 642                         return written ? written : -ERESTARTSYS;
 643                 buf += chunk;
 644                 count -= chunk;
 645                 cond_resched();
 646         }
 647         return written ? written : -EFAULT;
 648 }
 649
 650 static ssize_t aio_read_zero(struct kiocb *iocb, const struct iovec *iov,
 651                              unsigned long nr_segs, loff_t pos)
 652 {
 653         size_t written = 0;
 654         unsigned long i;
 655         ssize_t ret;
 656
 657         for (i = 0; i < nr_segs; i++) {
 658                 ret = read_zero(iocb->ki_filp, iov[i].iov_base, iov[i].iov_len,
 659                                 &pos);
 660                 if (ret < 0)
 661                         break;
 662                 written += ret;
 663         }
 664
 665         return written ? written : -EFAULT;
 666 }
 667
 668 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 669 {
 670 #ifndef CONFIG_MMU
 671         return -ENOSYS;
 672 #endif
 673         if (vma->vm_flags & VM_SHARED)
 674                 return shmem_zero_setup(vma);
 675         return 0;
 676 }
 677
 678 static ssize_t write_full(struct file *file, const char __user *buf,
 679                           size_t count, loff_t *ppos)
 680 {
 681         return -ENOSPC;
 682 }
 683
 684 /*
 685  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 686  * can fopen() both devices with "a" now.  This was previously impossible.
 687  * -- SRB.
 688  */
 689 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 690 {
 691         return file->f_pos = 0;
 692 }
 693
 694 /*
 695  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 696  * check against negative addresses: they are ok. The return value is weird,
 697  * though, in that case (0).
 698  *
 699  * also note that seeking relative to the "end of file" isn't supported:
 700  * it has no meaning, so it returns -EINVAL.
 701  */
 702 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 703 {
 704         loff_t ret;
 705
 706         mutex_lock(&file_inode(file)->i_mutex);
 707         switch (orig) {
 708         case SEEK_CUR:
 709                 offset += file->f_pos;
 710         case SEEK_SET:
 711                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 712                 if (IS_ERR_VALUE((unsigned long long)offset)) {
 713                         ret = -EOVERFLOW;
 714                         break;
 715                 }
 716                 file->f_pos = offset;
 717                 ret = file->f_pos;
 718                 force_successful_syscall_return();
 719                 break;
 720         default:
 721                 ret = -EINVAL;
 722         }
 723         mutex_unlock(&file_inode(file)->i_mutex);
 724         return ret;
 725 }
 726
 727 static int open_port(struct inode *inode, struct file *filp)
 728 {
 729         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 730 }
 731
 732 #define zero_lseek      null_lseek
 733 #define full_lseek      null_lseek
 734 #define write_zero      write_null
 735 #define read_full       read_zero
 736 #define aio_write_zero  aio_write_null
 737 #define open_mem        open_port
 738 #define open_kmem       open_mem
 739
 740 static const struct file_operations mem_fops = {
 741         .llseek         = memory_lseek,
 742         .read           = read_mem,
 743         .write          = write_mem,
 744         .mmap           = mmap_mem,
 745         .open           = open_mem,
 746         .get_unmapped_area = get_unmapped_area_mem,
 747 };
 748
 749 #ifdef CONFIG_DEVKMEM
 750 static const struct file_operations kmem_fops = {
 751         .llseek         = memory_lseek,
 752         .read           = read_kmem,
 753         .write          = write_kmem,
 754         .mmap           = mmap_kmem,
 755         .open           = open_kmem,
 756         .get_unmapped_area = get_unmapped_area_mem,
 757 };
 758 #endif
 759
 760 static const struct file_operations null_fops = {
 761         .llseek         = null_lseek,
 762         .read           = read_null,
 763         .write          = write_null,
 764         .aio_read       = aio_read_null,
 765         .aio_write      = aio_write_null,
 766         .splice_write   = splice_write_null,
 767 };
 768
 769 #ifdef CONFIG_DEVPORT
 770 static const struct file_operations port_fops = {
 771         .llseek         = memory_lseek,
 772         .read           = read_port,
 773         .write          = write_port,
 774         .open           = open_port,
 775 };
 776 #endif
 777
 778 static const struct file_operations zero_fops = {
 779         .llseek         = zero_lseek,
 780         .read           = read_zero,
 781         .write          = write_zero,
 782         .aio_read       = aio_read_zero,
 783         .aio_write      = aio_write_zero,
 784         .mmap           = mmap_zero,
 785 };
 786
 787 /*
 788  * capabilities for /dev/zero
 789  * - permits private mappings, "copies" are taken of the source of zeros
 790  * - no writeback happens
 791  */
 792 static struct backing_dev_info zero_bdi = {
 793         .name           = "char/mem",
 794         .capabilities   = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
 795 };
 796
 797 static const struct file_operations full_fops = {
 798         .llseek         = full_lseek,
 799         .read           = read_full,
 800         .write          = write_full,
 801 };
 802
 803 static const struct memdev {
 804         const char *name;
 805         umode_t mode;
 806         const struct file_operations *fops;
 807         struct backing_dev_info *dev_info;
 808 } devlist[] = {
 809          [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
 810 #ifdef CONFIG_DEVKMEM
 811          [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
 812 #endif
 813          [3] = { "null", 0666, &null_fops, NULL },
 814 #ifdef CONFIG_DEVPORT
 815          [4] = { "port", 0, &port_fops, NULL },
 816 #endif
 817          [5] = { "zero", 0666, &zero_fops, &zero_bdi },
 818          [7] = { "full", 0666, &full_fops, NULL },
 819          [8] = { "random", 0666, &random_fops, NULL },
 820          [9] = { "urandom", 0666, &urandom_fops, NULL },
 821 #ifdef CONFIG_PRINTK
 822         [11] = { "kmsg", 0644, &kmsg_fops, NULL },
 823 #endif
 824 };
 825
 826 static int memory_open(struct inode *inode, struct file *filp)
 827 {
 828         int minor;
 829         const struct memdev *dev;
 830
 831         minor = iminor(inode);
 832         if (minor >= ARRAY_SIZE(devlist))
 833                 return -ENXIO;
 834
 835         dev = &devlist[minor];
 836         if (!dev->fops)
 837                 return -ENXIO;
 838
 839         filp->f_op = dev->fops;
 840         if (dev->dev_info)
 841                 filp->f_mapping->backing_dev_info = dev->dev_info;
 842
 843         /* Is /dev/mem or /dev/kmem ? */
 844         if (dev->dev_info == &directly_mappable_cdev_bdi)
 845                 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
 846
 847         if (dev->fops->open)
 848                 return dev->fops->open(inode, filp);
 849
 850         return 0;
 851 }
 852
 853 static const struct file_operations memory_fops = {
 854         .open = memory_open,
 855         .llseek = noop_llseek,
 856 };
 857
 858 static char *mem_devnode(struct device *dev, umode_t *mode)
 859 {
 860         if (mode && devlist[MINOR(dev->devt)].mode)
 861                 *mode = devlist[MINOR(dev->devt)].mode;
 862         return NULL;
 863 }
 864
 865 static struct class *mem_class;
 866
 867 static int __init chr_dev_init(void)
 868 {
 869         int minor;
 870         int err;
 871
 872         err = bdi_init(&zero_bdi);
 873         if (err)
 874                 return err;
 875
 876         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 877                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 878
 879         mem_class = class_create(THIS_MODULE, "mem");
 880         if (IS_ERR(mem_class))
 881                 return PTR_ERR(mem_class);
 882
 883         mem_class->devnode = mem_devnode;
 884         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 885                 if (!devlist[minor].name)
 886                         continue;
 887
 888                 /*
 889                  * Create /dev/port?
 890                  */
 891                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 892                         continue;
 893
 894                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 895                               NULL, devlist[minor].name);
 896         }
 897
 898         return tty_init();
 899 }
 900
 901 fs_initcall(chr_dev_init);