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 mmapping 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/backing-dev.h>
  25 #include <linux/bootmem.h>
  26 #include <linux/splice.h>
  27 #include <linux/pfn.h>
  28 #include <linux/export.h>
  29 #include <linux/io.h>
  30 #include <linux/aio.h>
  31
  32 #include <asm/uaccess.h>
  33
  34 #ifdef CONFIG_IA64
  35 # include <linux/efi.h>
  36 #endif
  37
  38 #define DEVPORT_MINOR   4
  39
  40 static inline unsigned long size_inside_page(unsigned long start,
  41                                              unsigned long size)
  42 {
  43         unsigned long sz;
  44
  45         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  46
  47         return min(sz, size);
  48 }
  49
  50 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  51 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
  52 {
  53         return addr + count <= __pa(high_memory);
  54 }
  55
  56 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  57 {
  58         return 1;
  59 }
  60 #endif
  61
  62 #ifdef CONFIG_STRICT_DEVMEM
  63 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  64 {
  65         u64 from = ((u64)pfn) << PAGE_SHIFT;
  66         u64 to = from + size;
  67         u64 cursor = from;
  68
  69         while (cursor < to) {
  70                 if (!devmem_is_allowed(pfn)) {
  71                         printk(KERN_INFO
  72                 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
  73                                 current->comm, from, to);
  74                         return 0;
  75                 }
  76                 cursor += PAGE_SIZE;
  77                 pfn++;
  78         }
  79         return 1;
  80 }
  81 #else
  82 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  83 {
  84         return 1;
  85 }
  86 #endif
  87
  88 void __weak unxlate_dev_mem_ptr(unsigned long phys, void *addr)
  89 {
  90 }
  91
  92 /*
  93  * This funcion reads the *physical* memory. The f_pos points directly to the
  94  * memory location.
  95  */
  96 static ssize_t read_mem(struct file *file, char __user *buf,
  97                         size_t count, loff_t *ppos)
  98 {
  99         phys_addr_t p = *ppos;
 100         ssize_t read, sz;
 101         char *ptr;
 102
 103         if (p != *ppos)
 104                 return 0;
 105
 106         if (!valid_phys_addr_range(p, count))
 107                 return -EFAULT;
 108         read = 0;
 109 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 110         /* we don't have page 0 mapped on sparc and m68k.. */
 111         if (p < PAGE_SIZE) {
 112                 sz = size_inside_page(p, count);
 113                 if (sz > 0) {
 114                         if (clear_user(buf, sz))
 115                                 return -EFAULT;
 116                         buf += sz;
 117                         p += sz;
 118                         count -= sz;
 119                         read += sz;
 120                 }
 121         }
 122 #endif
 123
 124         while (count > 0) {
 125                 unsigned long remaining;
 126
 127                 sz = size_inside_page(p, count);
 128
 129                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 130                         return -EPERM;
 131
 132                 /*
 133                  * On ia64 if a page has been mapped somewhere as uncached, then
 134                  * it must also be accessed uncached by the kernel or data
 135                  * corruption may occur.
 136                  */
 137                 ptr = xlate_dev_mem_ptr(p);
 138                 if (!ptr)
 139                         return -EFAULT;
 140
 141                 remaining = copy_to_user(buf, ptr, sz);
 142                 unxlate_dev_mem_ptr(p, ptr);
 143                 if (remaining)
 144                         return -EFAULT;
 145
 146                 buf += sz;
 147                 p += sz;
 148                 count -= sz;
 149                 read += sz;
 150         }
 151
 152         *ppos += read;
 153         return read;
 154 }
 155
 156 static ssize_t write_mem(struct file *file, const char __user *buf,
 157                          size_t count, loff_t *ppos)
 158 {
 159         phys_addr_t p = *ppos;
 160         ssize_t written, sz;
 161         unsigned long copied;
 162         void *ptr;
 163
 164         if (p != *ppos)
 165                 return -EFBIG;
 166
 167         if (!valid_phys_addr_range(p, count))
 168                 return -EFAULT;
 169
 170         written = 0;
 171
 172 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 173         /* we don't have page 0 mapped on sparc and m68k.. */
 174         if (p < PAGE_SIZE) {
 175                 sz = size_inside_page(p, count);
 176                 /* Hmm. Do something? */
 177                 buf += sz;
 178                 p += sz;
 179                 count -= sz;
 180                 written += sz;
 181         }
 182 #endif
 183
 184         while (count > 0) {
 185                 sz = size_inside_page(p, count);
 186
 187                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 188                         return -EPERM;
 189
 190                 /*
 191                  * On ia64 if a page has been mapped somewhere as uncached, then
 192                  * it must also be accessed uncached by the kernel or data
 193                  * corruption may occur.
 194                  */
 195                 ptr = xlate_dev_mem_ptr(p);
 196                 if (!ptr) {
 197                         if (written)
 198                                 break;
 199                         return -EFAULT;
 200                 }
 201
 202                 copied = copy_from_user(ptr, buf, sz);
 203                 unxlate_dev_mem_ptr(p, ptr);
 204                 if (copied) {
 205                         written += sz - copied;
 206                         if (written)
 207                                 break;
 208                         return -EFAULT;
 209                 }
 210
 211                 buf += sz;
 212                 p += sz;
 213                 count -= sz;
 214                 written += sz;
 215         }
 216
 217         *ppos += written;
 218         return written;
 219 }
 220
 221 int __weak phys_mem_access_prot_allowed(struct file *file,
 222         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 223 {
 224         return 1;
 225 }
 226
 227 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 228
 229 /*
 230  * Architectures vary in how they handle caching for addresses
 231  * outside of main memory.
 232  *
 233  */
 234 #ifdef pgprot_noncached
 235 static int uncached_access(struct file *file, phys_addr_t addr)
 236 {
 237 #if defined(CONFIG_IA64)
 238         /*
 239          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 240          * attribute aliases.
 241          */
 242         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 243 #elif defined(CONFIG_MIPS)
 244         {
 245                 extern int __uncached_access(struct file *file,
 246                                              unsigned long addr);
 247
 248                 return __uncached_access(file, addr);
 249         }
 250 #else
 251         /*
 252          * Accessing memory above the top the kernel knows about or through a
 253          * file pointer
 254          * that was marked O_DSYNC will be done non-cached.
 255          */
 256         if (file->f_flags & O_DSYNC)
 257                 return 1;
 258         return addr >= __pa(high_memory);
 259 #endif
 260 }
 261 #endif
 262
 263 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 264                                      unsigned long size, pgprot_t vma_prot)
 265 {
 266 #ifdef pgprot_noncached
 267         phys_addr_t offset = pfn << PAGE_SHIFT;
 268
 269         if (uncached_access(file, offset))
 270                 return pgprot_noncached(vma_prot);
 271 #endif
 272         return vma_prot;
 273 }
 274 #endif
 275
 276 #ifndef CONFIG_MMU
 277 static unsigned long get_unmapped_area_mem(struct file *file,
 278                                            unsigned long addr,
 279                                            unsigned long len,
 280                                            unsigned long pgoff,
 281                                            unsigned long flags)
 282 {
 283         if (!valid_mmap_phys_addr_range(pgoff, len))
 284                 return (unsigned long) -EINVAL;
 285         return pgoff << PAGE_SHIFT;
 286 }
 287
 288 /* can't do an in-place private mapping if there's no MMU */
 289 static inline int private_mapping_ok(struct vm_area_struct *vma)
 290 {
 291         return vma->vm_flags & VM_MAYSHARE;
 292 }
 293 #else
 294 #define get_unmapped_area_mem   NULL
 295
 296 static inline int private_mapping_ok(struct vm_area_struct *vma)
 297 {
 298         return 1;
 299 }
 300 #endif
 301
 302 static const struct vm_operations_struct mmap_mem_ops = {
 303 #ifdef CONFIG_HAVE_IOREMAP_PROT
 304         .access = generic_access_phys
 305 #endif
 306 };
 307
 308 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 309 {
 310         size_t size = vma->vm_end - vma->vm_start;
 311
 312         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 313                 return -EINVAL;
 314
 315         if (!private_mapping_ok(vma))
 316                 return -ENOSYS;
 317
 318         if (!range_is_allowed(vma->vm_pgoff, size))
 319                 return -EPERM;
 320
 321         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 322                                                 &vma->vm_page_prot))
 323                 return -EINVAL;
 324
 325         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 326                                                  size,
 327                                                  vma->vm_page_prot);
 328
 329         vma->vm_ops = &mmap_mem_ops;
 330
 331         /* Remap-pfn-range will mark the range VM_IO */
 332         if (remap_pfn_range(vma,
 333                             vma->vm_start,
 334                             vma->vm_pgoff,
 335                             size,
 336                             vma->vm_page_prot)) {
 337                 return -EAGAIN;
 338         }
 339         return 0;
 340 }
 341
 342 #ifdef CONFIG_DEVKMEM
 343 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 344 {
 345         unsigned long pfn;
 346
 347         /* Turn a kernel-virtual address into a physical page frame */
 348         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 349
 350         /*
 351          * RED-PEN: on some architectures there is more mapped memory than
 352          * available in mem_map which pfn_valid checks for. Perhaps should add a
 353          * new macro here.
 354          *
 355          * RED-PEN: vmalloc is not supported right now.
 356          */
 357         if (!pfn_valid(pfn))
 358                 return -EIO;
 359
 360         vma->vm_pgoff = pfn;
 361         return mmap_mem(file, vma);
 362 }
 363 #endif
 364
 365 #ifdef CONFIG_DEVKMEM
 366 /*
 367  * This function reads the *virtual* memory as seen by the kernel.
 368  */
 369 static ssize_t read_kmem(struct file *file, char __user *buf,
 370                          size_t count, loff_t *ppos)
 371 {
 372         unsigned long p = *ppos;
 373         ssize_t low_count, read, sz;
 374         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 375         int err = 0;
 376
 377         read = 0;
 378         if (p < (unsigned long) high_memory) {
 379                 low_count = count;
 380                 if (count > (unsigned long)high_memory - p)
 381                         low_count = (unsigned long)high_memory - p;
 382
 383 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 384                 /* we don't have page 0 mapped on sparc and m68k.. */
 385                 if (p < PAGE_SIZE && low_count > 0) {
 386                         sz = size_inside_page(p, low_count);
 387                         if (clear_user(buf, sz))
 388                                 return -EFAULT;
 389                         buf += sz;
 390                         p += sz;
 391                         read += sz;
 392                         low_count -= sz;
 393                         count -= sz;
 394                 }
 395 #endif
 396                 while (low_count > 0) {
 397                         sz = size_inside_page(p, low_count);
 398
 399                         /*
 400                          * On ia64 if a page has been mapped somewhere as
 401                          * uncached, then it must also be accessed uncached
 402                          * by the kernel or data corruption may occur
 403                          */
 404                         kbuf = xlate_dev_kmem_ptr((char *)p);
 405
 406                         if (copy_to_user(buf, kbuf, sz))
 407                                 return -EFAULT;
 408                         buf += sz;
 409                         p += sz;
 410                         read += sz;
 411                         low_count -= sz;
 412                         count -= sz;
 413                 }
 414         }
 415
 416         if (count > 0) {
 417                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 418                 if (!kbuf)
 419                         return -ENOMEM;
 420                 while (count > 0) {
 421                         sz = size_inside_page(p, count);
 422                         if (!is_vmalloc_or_module_addr((void *)p)) {
 423                                 err = -ENXIO;
 424                                 break;
 425                         }
 426                         sz = vread(kbuf, (char *)p, sz);
 427                         if (!sz)
 428                                 break;
 429                         if (copy_to_user(buf, kbuf, sz)) {
 430                                 err = -EFAULT;
 431                                 break;
 432                         }
 433                         count -= sz;
 434                         buf += sz;
 435                         read += sz;
 436                         p += sz;
 437                 }
 438                 free_page((unsigned long)kbuf);
 439         }
 440         *ppos = p;
 441         return read ? read : err;
 442 }
 443
 444
 445 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 446                                 size_t count, loff_t *ppos)
 447 {
 448         ssize_t written, sz;
 449         unsigned long copied;
 450
 451         written = 0;
 452 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 453         /* we don't have page 0 mapped on sparc and m68k.. */
 454         if (p < PAGE_SIZE) {
 455                 sz = size_inside_page(p, count);
 456                 /* Hmm. Do something? */
 457                 buf += sz;
 458                 p += sz;
 459                 count -= sz;
 460                 written += sz;
 461         }
 462 #endif
 463
 464         while (count > 0) {
 465                 char *ptr;
 466
 467                 sz = size_inside_page(p, count);
 468
 469                 /*
 470                  * On ia64 if a page has been mapped somewhere as uncached, then
 471                  * it must also be accessed uncached by the kernel or data
 472                  * corruption may occur.
 473                  */
 474                 ptr = xlate_dev_kmem_ptr((char *)p);
 475
 476                 copied = copy_from_user(ptr, buf, sz);
 477                 if (copied) {
 478                         written += sz - copied;
 479                         if (written)
 480                                 break;
 481                         return -EFAULT;
 482                 }
 483                 buf += sz;
 484                 p += sz;
 485                 count -= sz;
 486                 written += sz;
 487         }
 488
 489         *ppos += written;
 490         return written;
 491 }
 492
 493 /*
 494  * This function writes to the *virtual* memory as seen by the kernel.
 495  */
 496 static ssize_t write_kmem(struct file *file, const char __user *buf,
 497                           size_t count, loff_t *ppos)
 498 {
 499         unsigned long p = *ppos;
 500         ssize_t wrote = 0;
 501         ssize_t virtr = 0;
 502         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 503         int err = 0;
 504
 505         if (p < (unsigned long) high_memory) {
 506                 unsigned long to_write = min_t(unsigned long, count,
 507                                                (unsigned long)high_memory - p);
 508                 wrote = do_write_kmem(p, buf, to_write, ppos);
 509                 if (wrote != to_write)
 510                         return wrote;
 511                 p += wrote;
 512                 buf += wrote;
 513                 count -= wrote;
 514         }
 515
 516         if (count > 0) {
 517                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 518                 if (!kbuf)
 519                         return wrote ? wrote : -ENOMEM;
 520                 while (count > 0) {
 521                         unsigned long sz = size_inside_page(p, count);
 522                         unsigned long n;
 523
 524                         if (!is_vmalloc_or_module_addr((void *)p)) {
 525                                 err = -ENXIO;
 526                                 break;
 527                         }
 528                         n = copy_from_user(kbuf, buf, sz);
 529                         if (n) {
 530                                 err = -EFAULT;
 531                                 break;
 532                         }
 533                         vwrite(kbuf, (char *)p, sz);
 534                         count -= sz;
 535                         buf += sz;
 536                         virtr += sz;
 537                         p += sz;
 538                 }
 539                 free_page((unsigned long)kbuf);
 540         }
 541
 542         *ppos = p;
 543         return virtr + wrote ? : err;
 544 }
 545 #endif
 546
 547 #ifdef CONFIG_DEVPORT
 548 static ssize_t read_port(struct file *file, char __user *buf,
 549                          size_t count, loff_t *ppos)
 550 {
 551         unsigned long i = *ppos;
 552         char __user *tmp = buf;
 553
 554         if (!access_ok(VERIFY_WRITE, buf, count))
 555                 return -EFAULT;
 556         while (count-- > 0 && i < 65536) {
 557                 if (__put_user(inb(i), tmp) < 0)
 558                         return -EFAULT;
 559                 i++;
 560                 tmp++;
 561         }
 562         *ppos = i;
 563         return tmp-buf;
 564 }
 565
 566 static ssize_t write_port(struct file *file, const char __user *buf,
 567                           size_t count, loff_t *ppos)
 568 {
 569         unsigned long i = *ppos;
 570         const char __user *tmp = buf;
 571
 572         if (!access_ok(VERIFY_READ, buf, count))
 573                 return -EFAULT;
 574         while (count-- > 0 && i < 65536) {
 575                 char c;
 576                 if (__get_user(c, tmp)) {
 577                         if (tmp > buf)
 578                                 break;
 579                         return -EFAULT;
 580                 }
 581                 outb(c, i);
 582                 i++;
 583                 tmp++;
 584         }
 585         *ppos = i;
 586         return tmp-buf;
 587 }
 588 #endif
 589
 590 static ssize_t read_null(struct file *file, char __user *buf,
 591                          size_t count, loff_t *ppos)
 592 {
 593         return 0;
 594 }
 595
 596 static ssize_t write_null(struct file *file, const char __user *buf,
 597                           size_t count, loff_t *ppos)
 598 {
 599         return count;
 600 }
 601
 602 static ssize_t aio_read_null(struct kiocb *iocb, const struct iovec *iov,
 603                              unsigned long nr_segs, loff_t pos)
 604 {
 605         return 0;
 606 }
 607
 608 static ssize_t aio_write_null(struct kiocb *iocb, const struct iovec *iov,
 609                               unsigned long nr_segs, loff_t pos)
 610 {
 611         return iov_length(iov, nr_segs);
 612 }
 613
 614 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 615                         struct splice_desc *sd)
 616 {
 617         return sd->len;
 618 }
 619
 620 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 621                                  loff_t *ppos, size_t len, unsigned int flags)
 622 {
 623         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 624 }
 625
 626 static ssize_t read_zero(struct file *file, char __user *buf,
 627                          size_t count, loff_t *ppos)
 628 {
 629         size_t written;
 630
 631         if (!count)
 632                 return 0;
 633
 634         if (!access_ok(VERIFY_WRITE, buf, count))
 635                 return -EFAULT;
 636
 637         written = 0;
 638         while (count) {
 639                 unsigned long unwritten;
 640                 size_t chunk = count;
 641
 642                 if (chunk > PAGE_SIZE)
 643                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 644                 unwritten = __clear_user(buf, chunk);
 645                 written += chunk - unwritten;
 646                 if (unwritten)
 647                         break;
 648                 if (signal_pending(current))
 649                         return written ? written : -ERESTARTSYS;
 650                 buf += chunk;
 651                 count -= chunk;
 652                 cond_resched();
 653         }
 654         return written ? written : -EFAULT;
 655 }
 656
 657 static ssize_t aio_read_zero(struct kiocb *iocb, const struct iovec *iov,
 658                              unsigned long nr_segs, loff_t pos)
 659 {
 660         size_t written = 0;
 661         unsigned long i;
 662         ssize_t ret;
 663
 664         for (i = 0; i < nr_segs; i++) {
 665                 ret = read_zero(iocb->ki_filp, iov[i].iov_base, iov[i].iov_len,
 666                                 &pos);
 667                 if (ret < 0)
 668                         break;
 669                 written += ret;
 670         }
 671
 672         return written ? written : -EFAULT;
 673 }
 674
 675 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 676 {
 677 #ifndef CONFIG_MMU
 678         return -ENOSYS;
 679 #endif
 680         if (vma->vm_flags & VM_SHARED)
 681                 return shmem_zero_setup(vma);
 682         return 0;
 683 }
 684
 685 static ssize_t write_full(struct file *file, const char __user *buf,
 686                           size_t count, loff_t *ppos)
 687 {
 688         return -ENOSPC;
 689 }
 690
 691 /*
 692  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 693  * can fopen() both devices with "a" now.  This was previously impossible.
 694  * -- SRB.
 695  */
 696 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 697 {
 698         return file->f_pos = 0;
 699 }
 700
 701 /*
 702  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 703  * check against negative addresses: they are ok. The return value is weird,
 704  * though, in that case (0).
 705  *
 706  * also note that seeking relative to the "end of file" isn't supported:
 707  * it has no meaning, so it returns -EINVAL.
 708  */
 709 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 710 {
 711         loff_t ret;
 712
 713         mutex_lock(&file_inode(file)->i_mutex);
 714         switch (orig) {
 715         case SEEK_CUR:
 716                 offset += file->f_pos;
 717         case SEEK_SET:
 718                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 719                 if (IS_ERR_VALUE((unsigned long long)offset)) {
 720                         ret = -EOVERFLOW;
 721                         break;
 722                 }
 723                 file->f_pos = offset;
 724                 ret = file->f_pos;
 725                 force_successful_syscall_return();
 726                 break;
 727         default:
 728                 ret = -EINVAL;
 729         }
 730         mutex_unlock(&file_inode(file)->i_mutex);
 731         return ret;
 732 }
 733
 734 static int open_port(struct inode *inode, struct file *filp)
 735 {
 736         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 737 }
 738
 739 #define zero_lseek      null_lseek
 740 #define full_lseek      null_lseek
 741 #define write_zero      write_null
 742 #define read_full       read_zero
 743 #define aio_write_zero  aio_write_null
 744 #define open_mem        open_port
 745 #define open_kmem       open_mem
 746
 747 static const struct file_operations mem_fops = {
 748         .llseek         = memory_lseek,
 749         .read           = read_mem,
 750         .write          = write_mem,
 751         .mmap           = mmap_mem,
 752         .open           = open_mem,
 753         .get_unmapped_area = get_unmapped_area_mem,
 754 };
 755
 756 #ifdef CONFIG_DEVKMEM
 757 static const struct file_operations kmem_fops = {
 758         .llseek         = memory_lseek,
 759         .read           = read_kmem,
 760         .write          = write_kmem,
 761         .mmap           = mmap_kmem,
 762         .open           = open_kmem,
 763         .get_unmapped_area = get_unmapped_area_mem,
 764 };
 765 #endif
 766
 767 static const struct file_operations null_fops = {
 768         .llseek         = null_lseek,
 769         .read           = read_null,
 770         .write          = write_null,
 771         .aio_read       = aio_read_null,
 772         .aio_write      = aio_write_null,
 773         .splice_write   = splice_write_null,
 774 };
 775
 776 #ifdef CONFIG_DEVPORT
 777 static const struct file_operations port_fops = {
 778         .llseek         = memory_lseek,
 779         .read           = read_port,
 780         .write          = write_port,
 781         .open           = open_port,
 782 };
 783 #endif
 784
 785 static const struct file_operations zero_fops = {
 786         .llseek         = zero_lseek,
 787         .read           = read_zero,
 788         .write          = write_zero,
 789         .aio_read       = aio_read_zero,
 790         .aio_write      = aio_write_zero,
 791         .mmap           = mmap_zero,
 792 };
 793
 794 /*
 795  * capabilities for /dev/zero
 796  * - permits private mappings, "copies" are taken of the source of zeros
 797  * - no writeback happens
 798  */
 799 static struct backing_dev_info zero_bdi = {
 800         .name           = "char/mem",
 801         .capabilities   = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
 802 };
 803
 804 static const struct file_operations full_fops = {
 805         .llseek         = full_lseek,
 806         .read           = read_full,
 807         .write          = write_full,
 808 };
 809
 810 static const struct memdev {
 811         const char *name;
 812         umode_t mode;
 813         const struct file_operations *fops;
 814         struct backing_dev_info *dev_info;
 815 } devlist[] = {
 816          [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
 817 #ifdef CONFIG_DEVKMEM
 818          [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
 819 #endif
 820          [3] = { "null", 0666, &null_fops, NULL },
 821 #ifdef CONFIG_DEVPORT
 822          [4] = { "port", 0, &port_fops, NULL },
 823 #endif
 824          [5] = { "zero", 0666, &zero_fops, &zero_bdi },
 825          [7] = { "full", 0666, &full_fops, NULL },
 826          [8] = { "random", 0666, &random_fops, NULL },
 827          [9] = { "urandom", 0666, &urandom_fops, NULL },
 828 #ifdef CONFIG_PRINTK
 829         [11] = { "kmsg", 0644, &kmsg_fops, NULL },
 830 #endif
 831 };
 832
 833 static int memory_open(struct inode *inode, struct file *filp)
 834 {
 835         int minor;
 836         const struct memdev *dev;
 837
 838         minor = iminor(inode);
 839         if (minor >= ARRAY_SIZE(devlist))
 840                 return -ENXIO;
 841
 842         dev = &devlist[minor];
 843         if (!dev->fops)
 844                 return -ENXIO;
 845
 846         filp->f_op = dev->fops;
 847         if (dev->dev_info)
 848                 filp->f_mapping->backing_dev_info = dev->dev_info;
 849
 850         /* Is /dev/mem or /dev/kmem ? */
 851         if (dev->dev_info == &directly_mappable_cdev_bdi)
 852                 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
 853
 854         if (dev->fops->open)
 855                 return dev->fops->open(inode, filp);
 856
 857         return 0;
 858 }
 859
 860 static const struct file_operations memory_fops = {
 861         .open = memory_open,
 862         .llseek = noop_llseek,
 863 };
 864
 865 static char *mem_devnode(struct device *dev, umode_t *mode)
 866 {
 867         if (mode && devlist[MINOR(dev->devt)].mode)
 868                 *mode = devlist[MINOR(dev->devt)].mode;
 869         return NULL;
 870 }
 871
 872 static struct class *mem_class;
 873
 874 static int __init chr_dev_init(void)
 875 {
 876         int minor;
 877         int err;
 878
 879         err = bdi_init(&zero_bdi);
 880         if (err)
 881                 return err;
 882
 883         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 884                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 885
 886         mem_class = class_create(THIS_MODULE, "mem");
 887         if (IS_ERR(mem_class))
 888                 return PTR_ERR(mem_class);
 889
 890         mem_class->devnode = mem_devnode;
 891         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 892                 if (!devlist[minor].name)
 893                         continue;
 894
 895                 /*
 896                  * Create /dev/port?
 897                  */
 898                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 899                         continue;
 900
 901                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 902                               NULL, devlist[minor].name);
 903         }
 904
 905         return tty_init();
 906 }
 907
 908 fs_initcall(chr_dev_init);