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/crash_dump.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/splice.h>
  28 #include <linux/pfn.h>
  29
  30 #include <asm/uaccess.h>
  31 #include <asm/io.h>
  32
  33 #ifdef CONFIG_IA64
  34 # include <linux/efi.h>
  35 #endif
  36
  37 static inline unsigned long size_inside_page(unsigned long start,
  38                                              unsigned long size)
  39 {
  40         unsigned long sz;
  41
  42         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  43
  44         return min(sz, size);
  45 }
  46
  47 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  48 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  49 {
  50         if (addr + count > __pa(high_memory))
  51                 return 0;
  52
  53         return 1;
  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         unsigned long 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         unsigned long 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, unsigned long 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         unsigned long 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 and VM_RESERVED */
 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_CRASH_DUMP
 360 /*
 361  * Read memory corresponding to the old kernel.
 362  */
 363 static ssize_t read_oldmem(struct file *file, char __user *buf,
 364                                 size_t count, loff_t *ppos)
 365 {
 366         unsigned long pfn, offset;
 367         size_t read = 0, csize;
 368         int rc = 0;
 369
 370         while (count) {
 371                 pfn = *ppos / PAGE_SIZE;
 372                 if (pfn > saved_max_pfn)
 373                         return read;
 374
 375                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 376                 if (count > PAGE_SIZE - offset)
 377                         csize = PAGE_SIZE - offset;
 378                 else
 379                         csize = count;
 380
 381                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 382                 if (rc < 0)
 383                         return rc;
 384                 buf += csize;
 385                 *ppos += csize;
 386                 read += csize;
 387                 count -= csize;
 388         }
 389         return read;
 390 }
 391 #endif
 392
 393 #ifdef CONFIG_DEVKMEM
 394 /*
 395  * This function reads the *virtual* memory as seen by the kernel.
 396  */
 397 static ssize_t read_kmem(struct file *file, char __user *buf,
 398                          size_t count, loff_t *ppos)
 399 {
 400         unsigned long p = *ppos;
 401         ssize_t low_count, read, sz;
 402         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 403         int err = 0;
 404
 405         read = 0;
 406         if (p < (unsigned long) high_memory) {
 407                 low_count = count;
 408                 if (count > (unsigned long)high_memory - p)
 409                         low_count = (unsigned long)high_memory - p;
 410
 411 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 412                 /* we don't have page 0 mapped on sparc and m68k.. */
 413                 if (p < PAGE_SIZE && low_count > 0) {
 414                         sz = size_inside_page(p, low_count);
 415                         if (clear_user(buf, sz))
 416                                 return -EFAULT;
 417                         buf += sz;
 418                         p += sz;
 419                         read += sz;
 420                         low_count -= sz;
 421                         count -= sz;
 422                 }
 423 #endif
 424                 while (low_count > 0) {
 425                         sz = size_inside_page(p, low_count);
 426
 427                         /*
 428                          * On ia64 if a page has been mapped somewhere as
 429                          * uncached, then it must also be accessed uncached
 430                          * by the kernel or data corruption may occur
 431                          */
 432                         kbuf = xlate_dev_kmem_ptr((char *)p);
 433
 434                         if (copy_to_user(buf, kbuf, sz))
 435                                 return -EFAULT;
 436                         buf += sz;
 437                         p += sz;
 438                         read += sz;
 439                         low_count -= sz;
 440                         count -= sz;
 441                 }
 442         }
 443
 444         if (count > 0) {
 445                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 446                 if (!kbuf)
 447                         return -ENOMEM;
 448                 while (count > 0) {
 449                         sz = size_inside_page(p, count);
 450                         if (!is_vmalloc_or_module_addr((void *)p)) {
 451                                 err = -ENXIO;
 452                                 break;
 453                         }
 454                         sz = vread(kbuf, (char *)p, sz);
 455                         if (!sz)
 456                                 break;
 457                         if (copy_to_user(buf, kbuf, sz)) {
 458                                 err = -EFAULT;
 459                                 break;
 460                         }
 461                         count -= sz;
 462                         buf += sz;
 463                         read += sz;
 464                         p += sz;
 465                 }
 466                 free_page((unsigned long)kbuf);
 467         }
 468         *ppos = p;
 469         return read ? read : err;
 470 }
 471
 472
 473 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 474                                 size_t count, loff_t *ppos)
 475 {
 476         ssize_t written, sz;
 477         unsigned long copied;
 478
 479         written = 0;
 480 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 481         /* we don't have page 0 mapped on sparc and m68k.. */
 482         if (p < PAGE_SIZE) {
 483                 sz = size_inside_page(p, count);
 484                 /* Hmm. Do something? */
 485                 buf += sz;
 486                 p += sz;
 487                 count -= sz;
 488                 written += sz;
 489         }
 490 #endif
 491
 492         while (count > 0) {
 493                 char *ptr;
 494
 495                 sz = size_inside_page(p, count);
 496
 497                 /*
 498                  * On ia64 if a page has been mapped somewhere as uncached, then
 499                  * it must also be accessed uncached by the kernel or data
 500                  * corruption may occur.
 501                  */
 502                 ptr = xlate_dev_kmem_ptr((char *)p);
 503
 504                 copied = copy_from_user(ptr, buf, sz);
 505                 if (copied) {
 506                         written += sz - copied;
 507                         if (written)
 508                                 break;
 509                         return -EFAULT;
 510                 }
 511                 buf += sz;
 512                 p += sz;
 513                 count -= sz;
 514                 written += sz;
 515         }
 516
 517         *ppos += written;
 518         return written;
 519 }
 520
 521 /*
 522  * This function writes to the *virtual* memory as seen by the kernel.
 523  */
 524 static ssize_t write_kmem(struct file *file, const char __user *buf,
 525                           size_t count, loff_t *ppos)
 526 {
 527         unsigned long p = *ppos;
 528         ssize_t wrote = 0;
 529         ssize_t virtr = 0;
 530         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 531         int err = 0;
 532
 533         if (p < (unsigned long) high_memory) {
 534                 unsigned long to_write = min_t(unsigned long, count,
 535                                                (unsigned long)high_memory - p);
 536                 wrote = do_write_kmem(p, buf, to_write, ppos);
 537                 if (wrote != to_write)
 538                         return wrote;
 539                 p += wrote;
 540                 buf += wrote;
 541                 count -= wrote;
 542         }
 543
 544         if (count > 0) {
 545                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 546                 if (!kbuf)
 547                         return wrote ? wrote : -ENOMEM;
 548                 while (count > 0) {
 549                         unsigned long sz = size_inside_page(p, count);
 550                         unsigned long n;
 551
 552                         if (!is_vmalloc_or_module_addr((void *)p)) {
 553                                 err = -ENXIO;
 554                                 break;
 555                         }
 556                         n = copy_from_user(kbuf, buf, sz);
 557                         if (n) {
 558                                 err = -EFAULT;
 559                                 break;
 560                         }
 561                         vwrite(kbuf, (char *)p, sz);
 562                         count -= sz;
 563                         buf += sz;
 564                         virtr += sz;
 565                         p += sz;
 566                 }
 567                 free_page((unsigned long)kbuf);
 568         }
 569
 570         *ppos = p;
 571         return virtr + wrote ? : err;
 572 }
 573 #endif
 574
 575 #ifdef CONFIG_DEVPORT
 576 static ssize_t read_port(struct file *file, char __user *buf,
 577                          size_t count, loff_t *ppos)
 578 {
 579         unsigned long i = *ppos;
 580         char __user *tmp = buf;
 581
 582         if (!access_ok(VERIFY_WRITE, buf, count))
 583                 return -EFAULT;
 584         while (count-- > 0 && i < 65536) {
 585                 if (__put_user(inb(i), tmp) < 0)
 586                         return -EFAULT;
 587                 i++;
 588                 tmp++;
 589         }
 590         *ppos = i;
 591         return tmp-buf;
 592 }
 593
 594 static ssize_t write_port(struct file *file, const char __user *buf,
 595                           size_t count, loff_t *ppos)
 596 {
 597         unsigned long i = *ppos;
 598         const char __user * tmp = buf;
 599
 600         if (!access_ok(VERIFY_READ, buf, count))
 601                 return -EFAULT;
 602         while (count-- > 0 && i < 65536) {
 603                 char c;
 604                 if (__get_user(c, tmp)) {
 605                         if (tmp > buf)
 606                                 break;
 607                         return -EFAULT;
 608                 }
 609                 outb(c, i);
 610                 i++;
 611                 tmp++;
 612         }
 613         *ppos = i;
 614         return tmp-buf;
 615 }
 616 #endif
 617
 618 static ssize_t read_null(struct file *file, char __user *buf,
 619                          size_t count, loff_t *ppos)
 620 {
 621         return 0;
 622 }
 623
 624 static ssize_t write_null(struct file *file, const char __user *buf,
 625                           size_t count, loff_t *ppos)
 626 {
 627         return count;
 628 }
 629
 630 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 631                         struct splice_desc *sd)
 632 {
 633         return sd->len;
 634 }
 635
 636 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 637                                  loff_t *ppos, size_t len, unsigned int flags)
 638 {
 639         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 640 }
 641
 642 static ssize_t read_zero(struct file *file, char __user *buf,
 643                          size_t count, loff_t *ppos)
 644 {
 645         size_t written;
 646
 647         if (!count)
 648                 return 0;
 649
 650         if (!access_ok(VERIFY_WRITE, buf, count))
 651                 return -EFAULT;
 652
 653         written = 0;
 654         while (count) {
 655                 unsigned long unwritten;
 656                 size_t chunk = count;
 657
 658                 if (chunk > PAGE_SIZE)
 659                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 660                 unwritten = __clear_user(buf, chunk);
 661                 written += chunk - unwritten;
 662                 if (unwritten)
 663                         break;
 664                 if (signal_pending(current))
 665                         return written ? written : -ERESTARTSYS;
 666                 buf += chunk;
 667                 count -= chunk;
 668                 cond_resched();
 669         }
 670         return written ? written : -EFAULT;
 671 }
 672
 673 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 674 {
 675 #ifndef CONFIG_MMU
 676         return -ENOSYS;
 677 #endif
 678         if (vma->vm_flags & VM_SHARED)
 679                 return shmem_zero_setup(vma);
 680         return 0;
 681 }
 682
 683 static ssize_t write_full(struct file *file, const char __user *buf,
 684                           size_t count, loff_t *ppos)
 685 {
 686         return -ENOSPC;
 687 }
 688
 689 /*
 690  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 691  * can fopen() both devices with "a" now.  This was previously impossible.
 692  * -- SRB.
 693  */
 694 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 695 {
 696         return file->f_pos = 0;
 697 }
 698
 699 /*
 700  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 701  * check against negative addresses: they are ok. The return value is weird,
 702  * though, in that case (0).
 703  *
 704  * also note that seeking relative to the "end of file" isn't supported:
 705  * it has no meaning, so it returns -EINVAL.
 706  */
 707 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 708 {
 709         loff_t ret;
 710
 711         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 712         switch (orig) {
 713         case SEEK_CUR:
 714                 offset += file->f_pos;
 715         case SEEK_SET:
 716                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 717                 if ((unsigned long long)offset >= ~0xFFFULL) {
 718                         ret = -EOVERFLOW;
 719                         break;
 720                 }
 721                 file->f_pos = offset;
 722                 ret = file->f_pos;
 723                 force_successful_syscall_return();
 724                 break;
 725         default:
 726                 ret = -EINVAL;
 727         }
 728         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 729         return ret;
 730 }
 731
 732 static int open_port(struct inode * inode, struct file * filp)
 733 {
 734         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 735 }
 736
 737 #define zero_lseek      null_lseek
 738 #define full_lseek      null_lseek
 739 #define write_zero      write_null
 740 #define read_full       read_zero
 741 #define open_mem        open_port
 742 #define open_kmem       open_mem
 743 #define open_oldmem     open_mem
 744
 745 static const struct file_operations mem_fops = {
 746         .llseek         = memory_lseek,
 747         .read           = read_mem,
 748         .write          = write_mem,
 749         .mmap           = mmap_mem,
 750         .open           = open_mem,
 751         .get_unmapped_area = get_unmapped_area_mem,
 752 };
 753
 754 #ifdef CONFIG_DEVKMEM
 755 static const struct file_operations kmem_fops = {
 756         .llseek         = memory_lseek,
 757         .read           = read_kmem,
 758         .write          = write_kmem,
 759         .mmap           = mmap_kmem,
 760         .open           = open_kmem,
 761         .get_unmapped_area = get_unmapped_area_mem,
 762 };
 763 #endif
 764
 765 static const struct file_operations null_fops = {
 766         .llseek         = null_lseek,
 767         .read           = read_null,
 768         .write          = write_null,
 769         .splice_write   = splice_write_null,
 770 };
 771
 772 #ifdef CONFIG_DEVPORT
 773 static const struct file_operations port_fops = {
 774         .llseek         = memory_lseek,
 775         .read           = read_port,
 776         .write          = write_port,
 777         .open           = open_port,
 778 };
 779 #endif
 780
 781 static const struct file_operations zero_fops = {
 782         .llseek         = zero_lseek,
 783         .read           = read_zero,
 784         .write          = 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 #ifdef CONFIG_CRASH_DUMP
 805 static const struct file_operations oldmem_fops = {
 806         .read   = read_oldmem,
 807         .open   = open_oldmem,
 808         .llseek = default_llseek,
 809 };
 810 #endif
 811
 812 static ssize_t kmsg_write(struct file *file, const char __user *buf,
 813                           size_t count, loff_t *ppos)
 814 {
 815         char *tmp;
 816         ssize_t ret;
 817
 818         tmp = kmalloc(count + 1, GFP_KERNEL);
 819         if (tmp == NULL)
 820                 return -ENOMEM;
 821         ret = -EFAULT;
 822         if (!copy_from_user(tmp, buf, count)) {
 823                 tmp[count] = 0;
 824                 ret = printk("%s", tmp);
 825                 if (ret > count)
 826                         /* printk can add a prefix */
 827                         ret = count;
 828         }
 829         kfree(tmp);
 830         return ret;
 831 }
 832
 833 static const struct file_operations kmsg_fops = {
 834         .write = kmsg_write,
 835         .llseek = noop_llseek,
 836 };
 837
 838 static const struct memdev {
 839         const char *name;
 840         mode_t mode;
 841         const struct file_operations *fops;
 842         struct backing_dev_info *dev_info;
 843 } devlist[] = {
 844          [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
 845 #ifdef CONFIG_DEVKMEM
 846          [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
 847 #endif
 848          [3] = { "null", 0666, &null_fops, NULL },
 849 #ifdef CONFIG_DEVPORT
 850          [4] = { "port", 0, &port_fops, NULL },
 851 #endif
 852          [5] = { "zero", 0666, &zero_fops, &zero_bdi },
 853          [7] = { "full", 0666, &full_fops, NULL },
 854          [8] = { "random", 0666, &random_fops, NULL },
 855          [9] = { "urandom", 0666, &urandom_fops, NULL },
 856         [11] = { "kmsg", 0, &kmsg_fops, NULL },
 857 #ifdef CONFIG_CRASH_DUMP
 858         [12] = { "oldmem", 0, &oldmem_fops, NULL },
 859 #endif
 860 };
 861
 862 static int memory_open(struct inode *inode, struct file *filp)
 863 {
 864         int minor;
 865         const struct memdev *dev;
 866
 867         minor = iminor(inode);
 868         if (minor >= ARRAY_SIZE(devlist))
 869                 return -ENXIO;
 870
 871         dev = &devlist[minor];
 872         if (!dev->fops)
 873                 return -ENXIO;
 874
 875         filp->f_op = dev->fops;
 876         if (dev->dev_info)
 877                 filp->f_mapping->backing_dev_info = dev->dev_info;
 878
 879         /* Is /dev/mem or /dev/kmem ? */
 880         if (dev->dev_info == &directly_mappable_cdev_bdi)
 881                 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
 882
 883         if (dev->fops->open)
 884                 return dev->fops->open(inode, filp);
 885
 886         return 0;
 887 }
 888
 889 static const struct file_operations memory_fops = {
 890         .open = memory_open,
 891         .llseek = noop_llseek,
 892 };
 893
 894 static char *mem_devnode(struct device *dev, mode_t *mode)
 895 {
 896         if (mode && devlist[MINOR(dev->devt)].mode)
 897                 *mode = devlist[MINOR(dev->devt)].mode;
 898         return NULL;
 899 }
 900
 901 static struct class *mem_class;
 902
 903 static int __init chr_dev_init(void)
 904 {
 905         int minor;
 906         int err;
 907
 908         err = bdi_init(&zero_bdi);
 909         if (err)
 910                 return err;
 911
 912         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 913                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 914
 915         mem_class = class_create(THIS_MODULE, "mem");
 916         if (IS_ERR(mem_class))
 917                 return PTR_ERR(mem_class);
 918
 919         mem_class->devnode = mem_devnode;
 920         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 921                 if (!devlist[minor].name)
 922                         continue;
 923                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 924                               NULL, devlist[minor].name);
 925         }
 926
 927         return tty_init();
 928 }
 929
 930 fs_initcall(chr_dev_init);