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