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