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 mmaping 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 #include <linux/smp_lock.h>
  30
  31 #include <asm/uaccess.h>
  32 #include <asm/io.h>
  33
  34 #ifdef CONFIG_IA64
  35 # include <linux/efi.h>
  36 #endif
  37
  38 static inline unsigned long size_inside_page(unsigned long start,
  39                                              unsigned long size)
  40 {
  41         unsigned long sz;
  42
  43         if (-start & (PAGE_SIZE - 1))
  44                 sz = -start & (PAGE_SIZE - 1);
  45         else
  46                 sz = PAGE_SIZE;
  47
  48         return min_t(unsigned long, sz, size);
  49 }
  50
  51 /*
  52  * Architectures vary in how they handle caching for addresses
  53  * outside of main memory.
  54  *
  55  */
  56 static inline int uncached_access(struct file *file, unsigned long addr)
  57 {
  58 #if defined(CONFIG_IA64)
  59         /*
  60          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  61          */
  62         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  63 #elif defined(CONFIG_MIPS)
  64         {
  65                 extern int __uncached_access(struct file *file,
  66                                              unsigned long addr);
  67
  68                 return __uncached_access(file, addr);
  69         }
  70 #else
  71         /*
  72          * Accessing memory above the top the kernel knows about or through a file pointer
  73          * that was marked O_SYNC will be done non-cached.
  74          */
  75         if (file->f_flags & O_SYNC)
  76                 return 1;
  77         return addr >= __pa(high_memory);
  78 #endif
  79 }
  80
  81 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  82 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  83 {
  84         if (addr + count > __pa(high_memory))
  85                 return 0;
  86
  87         return 1;
  88 }
  89
  90 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  91 {
  92         return 1;
  93 }
  94 #endif
  95
  96 #ifdef CONFIG_STRICT_DEVMEM
  97 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  98 {
  99         u64 from = ((u64)pfn) << PAGE_SHIFT;
 100         u64 to = from + size;
 101         u64 cursor = from;
 102
 103         while (cursor < to) {
 104                 if (!devmem_is_allowed(pfn)) {
 105                         printk(KERN_INFO
 106                 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
 107                                 current->comm, from, to);
 108                         return 0;
 109                 }
 110                 cursor += PAGE_SIZE;
 111                 pfn++;
 112         }
 113         return 1;
 114 }
 115 #else
 116 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 117 {
 118         return 1;
 119 }
 120 #endif
 121
 122 void __attribute__((weak)) unxlate_dev_mem_ptr(unsigned long phys, void *addr)
 123 {
 124 }
 125
 126 /*
 127  * This funcion reads the *physical* memory. The f_pos points directly to the
 128  * memory location.
 129  */
 130 static ssize_t read_mem(struct file * file, char __user * buf,
 131                         size_t count, loff_t *ppos)
 132 {
 133         unsigned long p = *ppos;
 134         ssize_t read, sz;
 135         char *ptr;
 136
 137         if (!valid_phys_addr_range(p, count))
 138                 return -EFAULT;
 139         read = 0;
 140 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 141         /* we don't have page 0 mapped on sparc and m68k.. */
 142         if (p < PAGE_SIZE) {
 143                 sz = PAGE_SIZE - p;
 144                 if (sz > count)
 145                         sz = count;
 146                 if (sz > 0) {
 147                         if (clear_user(buf, sz))
 148                                 return -EFAULT;
 149                         buf += sz;
 150                         p += sz;
 151                         count -= sz;
 152                         read += sz;
 153                 }
 154         }
 155 #endif
 156
 157         while (count > 0) {
 158                 /*
 159                  * Handle first page in case it's not aligned
 160                  */
 161                 if (-p & (PAGE_SIZE - 1))
 162                         sz = -p & (PAGE_SIZE - 1);
 163                 else
 164                         sz = PAGE_SIZE;
 165
 166                 sz = min_t(unsigned long, sz, count);
 167
 168                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 169                         return -EPERM;
 170
 171                 /*
 172                  * On ia64 if a page has been mapped somewhere as
 173                  * uncached, then it must also be accessed uncached
 174                  * by the kernel or data corruption may occur
 175                  */
 176                 ptr = xlate_dev_mem_ptr(p);
 177                 if (!ptr)
 178                         return -EFAULT;
 179
 180                 if (copy_to_user(buf, ptr, sz)) {
 181                         unxlate_dev_mem_ptr(p, ptr);
 182                         return -EFAULT;
 183                 }
 184
 185                 unxlate_dev_mem_ptr(p, ptr);
 186
 187                 buf += sz;
 188                 p += sz;
 189                 count -= sz;
 190                 read += sz;
 191         }
 192
 193         *ppos += read;
 194         return read;
 195 }
 196
 197 static ssize_t write_mem(struct file * file, const char __user * buf,
 198                          size_t count, loff_t *ppos)
 199 {
 200         unsigned long p = *ppos;
 201         ssize_t written, sz;
 202         unsigned long copied;
 203         void *ptr;
 204
 205         if (!valid_phys_addr_range(p, count))
 206                 return -EFAULT;
 207
 208         written = 0;
 209
 210 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 211         /* we don't have page 0 mapped on sparc and m68k.. */
 212         if (p < PAGE_SIZE) {
 213                 unsigned long sz = PAGE_SIZE - p;
 214                 if (sz > count)
 215                         sz = count;
 216                 /* Hmm. Do something? */
 217                 buf += sz;
 218                 p += sz;
 219                 count -= sz;
 220                 written += sz;
 221         }
 222 #endif
 223
 224         while (count > 0) {
 225                 /*
 226                  * Handle first page in case it's not aligned
 227                  */
 228                 if (-p & (PAGE_SIZE - 1))
 229                         sz = -p & (PAGE_SIZE - 1);
 230                 else
 231                         sz = PAGE_SIZE;
 232
 233                 sz = min_t(unsigned long, sz, count);
 234
 235                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 236                         return -EPERM;
 237
 238                 /*
 239                  * On ia64 if a page has been mapped somewhere as
 240                  * uncached, then it must also be accessed uncached
 241                  * by the kernel or data corruption may occur
 242                  */
 243                 ptr = xlate_dev_mem_ptr(p);
 244                 if (!ptr) {
 245                         if (written)
 246                                 break;
 247                         return -EFAULT;
 248                 }
 249
 250                 copied = copy_from_user(ptr, buf, sz);
 251                 if (copied) {
 252                         written += sz - copied;
 253                         unxlate_dev_mem_ptr(p, ptr);
 254                         if (written)
 255                                 break;
 256                         return -EFAULT;
 257                 }
 258
 259                 unxlate_dev_mem_ptr(p, ptr);
 260
 261                 buf += sz;
 262                 p += sz;
 263                 count -= sz;
 264                 written += sz;
 265         }
 266
 267         *ppos += written;
 268         return written;
 269 }
 270
 271 int __attribute__((weak)) phys_mem_access_prot_allowed(struct file *file,
 272         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 273 {
 274         return 1;
 275 }
 276
 277 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 278 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 279                                      unsigned long size, pgprot_t vma_prot)
 280 {
 281 #ifdef pgprot_noncached
 282         unsigned long offset = pfn << PAGE_SHIFT;
 283
 284         if (uncached_access(file, offset))
 285                 return pgprot_noncached(vma_prot);
 286 #endif
 287         return vma_prot;
 288 }
 289 #endif
 290
 291 #ifndef CONFIG_MMU
 292 static unsigned long get_unmapped_area_mem(struct file *file,
 293                                            unsigned long addr,
 294                                            unsigned long len,
 295                                            unsigned long pgoff,
 296                                            unsigned long flags)
 297 {
 298         if (!valid_mmap_phys_addr_range(pgoff, len))
 299                 return (unsigned long) -EINVAL;
 300         return pgoff << PAGE_SHIFT;
 301 }
 302
 303 /* can't do an in-place private mapping if there's no MMU */
 304 static inline int private_mapping_ok(struct vm_area_struct *vma)
 305 {
 306         return vma->vm_flags & VM_MAYSHARE;
 307 }
 308 #else
 309 #define get_unmapped_area_mem   NULL
 310
 311 static inline int private_mapping_ok(struct vm_area_struct *vma)
 312 {
 313         return 1;
 314 }
 315 #endif
 316
 317 static const struct vm_operations_struct mmap_mem_ops = {
 318 #ifdef CONFIG_HAVE_IOREMAP_PROT
 319         .access = generic_access_phys
 320 #endif
 321 };
 322
 323 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 324 {
 325         size_t size = vma->vm_end - vma->vm_start;
 326
 327         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 328                 return -EINVAL;
 329
 330         if (!private_mapping_ok(vma))
 331                 return -ENOSYS;
 332
 333         if (!range_is_allowed(vma->vm_pgoff, size))
 334                 return -EPERM;
 335
 336         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 337                                                 &vma->vm_page_prot))
 338                 return -EINVAL;
 339
 340         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 341                                                  size,
 342                                                  vma->vm_page_prot);
 343
 344         vma->vm_ops = &mmap_mem_ops;
 345
 346         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 347         if (remap_pfn_range(vma,
 348                             vma->vm_start,
 349                             vma->vm_pgoff,
 350                             size,
 351                             vma->vm_page_prot)) {
 352                 return -EAGAIN;
 353         }
 354         return 0;
 355 }
 356
 357 #ifdef CONFIG_DEVKMEM
 358 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 359 {
 360         unsigned long pfn;
 361
 362         /* Turn a kernel-virtual address into a physical page frame */
 363         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 364
 365         /*
 366          * RED-PEN: on some architectures there is more mapped memory
 367          * than available in mem_map which pfn_valid checks
 368          * for. Perhaps should add a new macro here.
 369          *
 370          * RED-PEN: vmalloc is not supported right now.
 371          */
 372         if (!pfn_valid(pfn))
 373                 return -EIO;
 374
 375         vma->vm_pgoff = pfn;
 376         return mmap_mem(file, vma);
 377 }
 378 #endif
 379
 380 #ifdef CONFIG_CRASH_DUMP
 381 /*
 382  * Read memory corresponding to the old kernel.
 383  */
 384 static ssize_t read_oldmem(struct file *file, char __user *buf,
 385                                 size_t count, loff_t *ppos)
 386 {
 387         unsigned long pfn, offset;
 388         size_t read = 0, csize;
 389         int rc = 0;
 390
 391         while (count) {
 392                 pfn = *ppos / PAGE_SIZE;
 393                 if (pfn > saved_max_pfn)
 394                         return read;
 395
 396                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 397                 if (count > PAGE_SIZE - offset)
 398                         csize = PAGE_SIZE - offset;
 399                 else
 400                         csize = count;
 401
 402                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 403                 if (rc < 0)
 404                         return rc;
 405                 buf += csize;
 406                 *ppos += csize;
 407                 read += csize;
 408                 count -= csize;
 409         }
 410         return read;
 411 }
 412 #endif
 413
 414 #ifdef CONFIG_DEVKMEM
 415 /*
 416  * This function reads the *virtual* memory as seen by the kernel.
 417  */
 418 static ssize_t read_kmem(struct file *file, char __user *buf,
 419                          size_t count, loff_t *ppos)
 420 {
 421         unsigned long p = *ppos;
 422         ssize_t low_count, read, sz;
 423         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 424         int err = 0;
 425
 426         read = 0;
 427         if (p < (unsigned long) high_memory) {
 428                 low_count = count;
 429                 if (count > (unsigned long) high_memory - p)
 430                         low_count = (unsigned long) high_memory - p;
 431
 432 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 433                 /* we don't have page 0 mapped on sparc and m68k.. */
 434                 if (p < PAGE_SIZE && low_count > 0) {
 435                         size_t tmp = PAGE_SIZE - p;
 436                         if (tmp > low_count) tmp = low_count;
 437                         if (clear_user(buf, tmp))
 438                                 return -EFAULT;
 439                         buf += tmp;
 440                         p += tmp;
 441                         read += tmp;
 442                         low_count -= tmp;
 443                         count -= tmp;
 444                 }
 445 #endif
 446                 while (low_count > 0) {
 447                         sz = size_inside_page(p, low_count);
 448
 449                         /*
 450                          * On ia64 if a page has been mapped somewhere as
 451                          * uncached, then it must also be accessed uncached
 452                          * by the kernel or data corruption may occur
 453                          */
 454                         kbuf = xlate_dev_kmem_ptr((char *)p);
 455
 456                         if (copy_to_user(buf, kbuf, sz))
 457                                 return -EFAULT;
 458                         buf += sz;
 459                         p += sz;
 460                         read += sz;
 461                         low_count -= sz;
 462                         count -= sz;
 463                 }
 464         }
 465
 466         if (count > 0) {
 467                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 468                 if (!kbuf)
 469                         return -ENOMEM;
 470                 while (count > 0) {
 471                         int len = size_inside_page(p, count);
 472
 473                         if (!is_vmalloc_or_module_addr((void *)p)) {
 474                                 err = -ENXIO;
 475                                 break;
 476                         }
 477                         len = vread(kbuf, (char *)p, len);
 478                         if (!len)
 479                                 break;
 480                         if (copy_to_user(buf, kbuf, len)) {
 481                                 err = -EFAULT;
 482                                 break;
 483                         }
 484                         count -= len;
 485                         buf += len;
 486                         read += len;
 487                         p += len;
 488                 }
 489                 free_page((unsigned long)kbuf);
 490         }
 491         *ppos = p;
 492         return read ? read : err;
 493 }
 494
 495
 496 static inline ssize_t
 497 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 498               size_t count, loff_t *ppos)
 499 {
 500         ssize_t written, sz;
 501         unsigned long copied;
 502
 503         written = 0;
 504 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 505         /* we don't have page 0 mapped on sparc and m68k.. */
 506         if (realp < PAGE_SIZE) {
 507                 unsigned long sz = PAGE_SIZE - realp;
 508                 if (sz > count)
 509                         sz = count;
 510                 /* Hmm. Do something? */
 511                 buf += sz;
 512                 p += sz;
 513                 realp += sz;
 514                 count -= sz;
 515                 written += sz;
 516         }
 517 #endif
 518
 519         while (count > 0) {
 520                 char *ptr;
 521
 522                 sz = size_inside_page(realp, count);
 523
 524                 /*
 525                  * On ia64 if a page has been mapped somewhere as
 526                  * uncached, then it must also be accessed uncached
 527                  * by the kernel or data corruption may occur
 528                  */
 529                 ptr = xlate_dev_kmem_ptr(p);
 530
 531                 copied = copy_from_user(ptr, buf, sz);
 532                 if (copied) {
 533                         written += sz - copied;
 534                         if (written)
 535                                 break;
 536                         return -EFAULT;
 537                 }
 538                 buf += sz;
 539                 p += sz;
 540                 realp += sz;
 541                 count -= sz;
 542                 written += sz;
 543         }
 544
 545         *ppos += written;
 546         return written;
 547 }
 548
 549
 550 /*
 551  * This function writes to the *virtual* memory as seen by the kernel.
 552  */
 553 static ssize_t write_kmem(struct file * file, const char __user * buf,
 554                           size_t count, loff_t *ppos)
 555 {
 556         unsigned long p = *ppos;
 557         ssize_t wrote = 0;
 558         ssize_t virtr = 0;
 559         ssize_t written;
 560         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 561         int err = 0;
 562
 563         if (p < (unsigned long) high_memory) {
 564
 565                 wrote = count;
 566                 if (count > (unsigned long) high_memory - p)
 567                         wrote = (unsigned long) high_memory - p;
 568
 569                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 570                 if (written != wrote)
 571                         return written;
 572                 wrote = written;
 573                 p += wrote;
 574                 buf += wrote;
 575                 count -= wrote;
 576         }
 577
 578         if (count > 0) {
 579                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 580                 if (!kbuf)
 581                         return wrote ? wrote : -ENOMEM;
 582                 while (count > 0) {
 583                         int len = size_inside_page(p, count);
 584
 585                         if (!is_vmalloc_or_module_addr((void *)p)) {
 586                                 err = -ENXIO;
 587                                 break;
 588                         }
 589                         if (len) {
 590                                 written = copy_from_user(kbuf, buf, len);
 591                                 if (written) {
 592                                         err = -EFAULT;
 593                                         break;
 594                                 }
 595                         }
 596                         vwrite(kbuf, (char *)p, len);
 597                         count -= len;
 598                         buf += len;
 599                         virtr += len;
 600                         p += len;
 601                 }
 602                 free_page((unsigned long)kbuf);
 603         }
 604
 605         *ppos = p;
 606         return virtr + wrote ? : err;
 607 }
 608 #endif
 609
 610 #ifdef CONFIG_DEVPORT
 611 static ssize_t read_port(struct file * file, char __user * buf,
 612                          size_t count, loff_t *ppos)
 613 {
 614         unsigned long i = *ppos;
 615         char __user *tmp = buf;
 616
 617         if (!access_ok(VERIFY_WRITE, buf, count))
 618                 return -EFAULT;
 619         while (count-- > 0 && i < 65536) {
 620                 if (__put_user(inb(i),tmp) < 0)
 621                         return -EFAULT;
 622                 i++;
 623                 tmp++;
 624         }
 625         *ppos = i;
 626         return tmp-buf;
 627 }
 628
 629 static ssize_t write_port(struct file * file, const char __user * buf,
 630                           size_t count, loff_t *ppos)
 631 {
 632         unsigned long i = *ppos;
 633         const char __user * tmp = buf;
 634
 635         if (!access_ok(VERIFY_READ,buf,count))
 636                 return -EFAULT;
 637         while (count-- > 0 && i < 65536) {
 638                 char c;
 639                 if (__get_user(c, tmp)) {
 640                         if (tmp > buf)
 641                                 break;
 642                         return -EFAULT;
 643                 }
 644                 outb(c,i);
 645                 i++;
 646                 tmp++;
 647         }
 648         *ppos = i;
 649         return tmp-buf;
 650 }
 651 #endif
 652
 653 static ssize_t read_null(struct file * file, char __user * buf,
 654                          size_t count, loff_t *ppos)
 655 {
 656         return 0;
 657 }
 658
 659 static ssize_t write_null(struct file * file, const char __user * buf,
 660                           size_t count, loff_t *ppos)
 661 {
 662         return count;
 663 }
 664
 665 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 666                         struct splice_desc *sd)
 667 {
 668         return sd->len;
 669 }
 670
 671 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 672                                  loff_t *ppos, size_t len, unsigned int flags)
 673 {
 674         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 675 }
 676
 677 static ssize_t read_zero(struct file * file, char __user * buf,
 678                          size_t count, loff_t *ppos)
 679 {
 680         size_t written;
 681
 682         if (!count)
 683                 return 0;
 684
 685         if (!access_ok(VERIFY_WRITE, buf, count))
 686                 return -EFAULT;
 687
 688         written = 0;
 689         while (count) {
 690                 unsigned long unwritten;
 691                 size_t chunk = count;
 692
 693                 if (chunk > PAGE_SIZE)
 694                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 695                 unwritten = __clear_user(buf, chunk);
 696                 written += chunk - unwritten;
 697                 if (unwritten)
 698                         break;
 699                 if (signal_pending(current))
 700                         return written ? written : -ERESTARTSYS;
 701                 buf += chunk;
 702                 count -= chunk;
 703                 cond_resched();
 704         }
 705         return written ? written : -EFAULT;
 706 }
 707
 708 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 709 {
 710 #ifndef CONFIG_MMU
 711         return -ENOSYS;
 712 #endif
 713         if (vma->vm_flags & VM_SHARED)
 714                 return shmem_zero_setup(vma);
 715         return 0;
 716 }
 717
 718 static ssize_t write_full(struct file * file, const char __user * buf,
 719                           size_t count, loff_t *ppos)
 720 {
 721         return -ENOSPC;
 722 }
 723
 724 /*
 725  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 726  * can fopen() both devices with "a" now.  This was previously impossible.
 727  * -- SRB.
 728  */
 729
 730 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 731 {
 732         return file->f_pos = 0;
 733 }
 734
 735 /*
 736  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 737  * check against negative addresses: they are ok. The return value is weird,
 738  * though, in that case (0).
 739  *
 740  * also note that seeking relative to the "end of file" isn't supported:
 741  * it has no meaning, so it returns -EINVAL.
 742  */
 743 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 744 {
 745         loff_t ret;
 746
 747         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 748         switch (orig) {
 749                 case 0:
 750                         file->f_pos = offset;
 751                         ret = file->f_pos;
 752                         force_successful_syscall_return();
 753                         break;
 754                 case 1:
 755                         file->f_pos += offset;
 756                         ret = file->f_pos;
 757                         force_successful_syscall_return();
 758                         break;
 759                 default:
 760                         ret = -EINVAL;
 761         }
 762         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 763         return ret;
 764 }
 765
 766 static int open_port(struct inode * inode, struct file * filp)
 767 {
 768         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 769 }
 770
 771 #define zero_lseek      null_lseek
 772 #define full_lseek      null_lseek
 773 #define write_zero      write_null
 774 #define read_full       read_zero
 775 #define open_mem        open_port
 776 #define open_kmem       open_mem
 777 #define open_oldmem     open_mem
 778
 779 static const struct file_operations mem_fops = {
 780         .llseek         = memory_lseek,
 781         .read           = read_mem,
 782         .write          = write_mem,
 783         .mmap           = mmap_mem,
 784         .open           = open_mem,
 785         .get_unmapped_area = get_unmapped_area_mem,
 786 };
 787
 788 #ifdef CONFIG_DEVKMEM
 789 static const struct file_operations kmem_fops = {
 790         .llseek         = memory_lseek,
 791         .read           = read_kmem,
 792         .write          = write_kmem,
 793         .mmap           = mmap_kmem,
 794         .open           = open_kmem,
 795         .get_unmapped_area = get_unmapped_area_mem,
 796 };
 797 #endif
 798
 799 static const struct file_operations null_fops = {
 800         .llseek         = null_lseek,
 801         .read           = read_null,
 802         .write          = write_null,
 803         .splice_write   = splice_write_null,
 804 };
 805
 806 #ifdef CONFIG_DEVPORT
 807 static const struct file_operations port_fops = {
 808         .llseek         = memory_lseek,
 809         .read           = read_port,
 810         .write          = write_port,
 811         .open           = open_port,
 812 };
 813 #endif
 814
 815 static const struct file_operations zero_fops = {
 816         .llseek         = zero_lseek,
 817         .read           = read_zero,
 818         .write          = write_zero,
 819         .mmap           = mmap_zero,
 820 };
 821
 822 /*
 823  * capabilities for /dev/zero
 824  * - permits private mappings, "copies" are taken of the source of zeros
 825  * - no writeback happens
 826  */
 827 static struct backing_dev_info zero_bdi = {
 828         .name           = "char/mem",
 829         .capabilities   = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
 830 };
 831
 832 static const struct file_operations full_fops = {
 833         .llseek         = full_lseek,
 834         .read           = read_full,
 835         .write          = write_full,
 836 };
 837
 838 #ifdef CONFIG_CRASH_DUMP
 839 static const struct file_operations oldmem_fops = {
 840         .read   = read_oldmem,
 841         .open   = open_oldmem,
 842 };
 843 #endif
 844
 845 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 846                           size_t count, loff_t *ppos)
 847 {
 848         char *tmp;
 849         ssize_t ret;
 850
 851         tmp = kmalloc(count + 1, GFP_KERNEL);
 852         if (tmp == NULL)
 853                 return -ENOMEM;
 854         ret = -EFAULT;
 855         if (!copy_from_user(tmp, buf, count)) {
 856                 tmp[count] = 0;
 857                 ret = printk("%s", tmp);
 858                 if (ret > count)
 859                         /* printk can add a prefix */
 860                         ret = count;
 861         }
 862         kfree(tmp);
 863         return ret;
 864 }
 865
 866 static const struct file_operations kmsg_fops = {
 867         .write =        kmsg_write,
 868 };
 869
 870 static const struct memdev {
 871         const char *name;
 872         mode_t mode;
 873         const struct file_operations *fops;
 874         struct backing_dev_info *dev_info;
 875 } devlist[] = {
 876          [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
 877 #ifdef CONFIG_DEVKMEM
 878          [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
 879 #endif
 880          [3] = { "null", 0666, &null_fops, NULL },
 881 #ifdef CONFIG_DEVPORT
 882          [4] = { "port", 0, &port_fops, NULL },
 883 #endif
 884          [5] = { "zero", 0666, &zero_fops, &zero_bdi },
 885          [7] = { "full", 0666, &full_fops, NULL },
 886          [8] = { "random", 0666, &random_fops, NULL },
 887          [9] = { "urandom", 0666, &urandom_fops, NULL },
 888         [11] = { "kmsg", 0, &kmsg_fops, NULL },
 889 #ifdef CONFIG_CRASH_DUMP
 890         [12] = { "oldmem", 0, &oldmem_fops, NULL },
 891 #endif
 892 };
 893
 894 static int memory_open(struct inode *inode, struct file *filp)
 895 {
 896         int minor;
 897         const struct memdev *dev;
 898         int ret = -ENXIO;
 899
 900         lock_kernel();
 901
 902         minor = iminor(inode);
 903         if (minor >= ARRAY_SIZE(devlist))
 904                 goto out;
 905
 906         dev = &devlist[minor];
 907         if (!dev->fops)
 908                 goto out;
 909
 910         filp->f_op = dev->fops;
 911         if (dev->dev_info)
 912                 filp->f_mapping->backing_dev_info = dev->dev_info;
 913
 914         if (dev->fops->open)
 915                 ret = dev->fops->open(inode, filp);
 916         else
 917                 ret = 0;
 918 out:
 919         unlock_kernel();
 920         return ret;
 921 }
 922
 923 static const struct file_operations memory_fops = {
 924         .open           = memory_open,
 925 };
 926
 927 static char *mem_devnode(struct device *dev, mode_t *mode)
 928 {
 929         if (mode && devlist[MINOR(dev->devt)].mode)
 930                 *mode = devlist[MINOR(dev->devt)].mode;
 931         return NULL;
 932 }
 933
 934 static struct class *mem_class;
 935
 936 static int __init chr_dev_init(void)
 937 {
 938         int minor;
 939         int err;
 940
 941         err = bdi_init(&zero_bdi);
 942         if (err)
 943                 return err;
 944
 945         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 946                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 947
 948         mem_class = class_create(THIS_MODULE, "mem");
 949         mem_class->devnode = mem_devnode;
 950         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 951                 if (!devlist[minor].name)
 952                         continue;
 953                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 954                               NULL, devlist[minor].name);
 955         }
 956
 957         return 0;
 958 }
 959
 960 fs_initcall(chr_dev_init);