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 #include <linux/export.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         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  44
  45         return min(sz, size);
  46 }
  47
  48 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  49 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  50 {
  51         return addr + count <= __pa(high_memory);
  52 }
  53
  54 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  55 {
  56         return 1;
  57 }
  58 #endif
  59
  60 #ifdef CONFIG_STRICT_DEVMEM
  61 static inline int page_is_allowed(unsigned long pfn)
  62 {
  63         return devmem_is_allowed(pfn);
  64 }
  65 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  66 {
  67         u64 from = ((u64)pfn) << PAGE_SHIFT;
  68         u64 to = from + size;
  69         u64 cursor = from;
  70
  71         while (cursor < to) {
  72                 if (!devmem_is_allowed(pfn)) {
  73                         printk(KERN_INFO
  74                 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
  75                                 current->comm, from, to);
  76                         return 0;
  77                 }
  78                 cursor += PAGE_SIZE;
  79                 pfn++;
  80         }
  81         return 1;
  82 }
  83 #else
  84 static inline int page_is_allowed(unsigned long pfn)
  85 {
  86         return 1;
  87 }
  88 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  89 {
  90         return 1;
  91 }
  92 #endif
  93
  94 void __weak unxlate_dev_mem_ptr(unsigned long phys, void *addr)
  95 {
  96 }
  97
  98 /*
  99  * This funcion reads the *physical* memory. The f_pos points directly to the
 100  * memory location.
 101  */
 102 static ssize_t read_mem(struct file *file, char __user *buf,
 103                         size_t count, loff_t *ppos)
 104 {
 105         unsigned long p = *ppos;
 106         ssize_t read, sz;
 107         char *ptr;
 108
 109         if (!valid_phys_addr_range(p, count))
 110                 return -EFAULT;
 111         read = 0;
 112 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 113         /* we don't have page 0 mapped on sparc and m68k.. */
 114         if (p < PAGE_SIZE) {
 115                 sz = size_inside_page(p, count);
 116                 if (sz > 0) {
 117                         if (clear_user(buf, sz))
 118                                 return -EFAULT;
 119                         buf += sz;
 120                         p += sz;
 121                         count -= sz;
 122                         read += sz;
 123                 }
 124         }
 125 #endif
 126
 127         while (count > 0) {
 128                 unsigned long remaining;
 129                 int allowed;
 130
 131                 sz = size_inside_page(p, count);
 132
 133                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 134                 if (!allowed)
 135                         return -EPERM;
 136                 if (allowed == 2) {
 137                         /* Show zeros for restricted memory. */
 138                         remaining = clear_user(buf, sz);
 139                 } else {
 140                         /*
 141                          * On ia64 if a page has been mapped somewhere as
 142                          * uncached, then it must also be accessed uncached
 143                          * by the kernel or data corruption may occur.
 144                          */
 145                         ptr = xlate_dev_mem_ptr(p);
 146                         if (!ptr)
 147                                 return -EFAULT;
 148
 149                         remaining = copy_to_user(buf, ptr, sz);
 150
 151                         unxlate_dev_mem_ptr(p, ptr);
 152                 }
 153
 154                 if (remaining)
 155                         return -EFAULT;
 156
 157                 buf += sz;
 158                 p += sz;
 159                 count -= sz;
 160                 read += sz;
 161         }
 162
 163         *ppos += read;
 164         return read;
 165 }
 166
 167 static ssize_t write_mem(struct file *file, const char __user *buf,
 168                          size_t count, loff_t *ppos)
 169 {
 170         unsigned long p = *ppos;
 171         ssize_t written, sz;
 172         unsigned long copied;
 173         void *ptr;
 174
 175         if (!valid_phys_addr_range(p, count))
 176                 return -EFAULT;
 177
 178         written = 0;
 179
 180 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 181         /* we don't have page 0 mapped on sparc and m68k.. */
 182         if (p < PAGE_SIZE) {
 183                 sz = size_inside_page(p, count);
 184                 /* Hmm. Do something? */
 185                 buf += sz;
 186                 p += sz;
 187                 count -= sz;
 188                 written += sz;
 189         }
 190 #endif
 191
 192         while (count > 0) {
 193                 int allowed;
 194
 195                 sz = size_inside_page(p, count);
 196
 197                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 198                 if (!allowed)
 199                         return -EPERM;
 200
 201                 /* Skip actual writing when a page is marked as restricted. */
 202                 if (allowed == 1) {
 203                         /*
 204                          * On ia64 if a page has been mapped somewhere as
 205                          * uncached, then it must also be accessed uncached
 206                          * by the kernel or data corruption may occur.
 207                          */
 208                         ptr = xlate_dev_mem_ptr(p);
 209                         if (!ptr) {
 210                                 if (written)
 211                                         break;
 212                                 return -EFAULT;
 213                         }
 214
 215                         copied = copy_from_user(ptr, buf, sz);
 216                         unxlate_dev_mem_ptr(p, ptr);
 217                         if (copied) {
 218                                 written += sz - copied;
 219                                 if (written)
 220                                         break;
 221                                 return -EFAULT;
 222                         }
 223                 }
 224
 225                 buf += sz;
 226                 p += sz;
 227                 count -= sz;
 228                 written += sz;
 229         }
 230
 231         *ppos += written;
 232         return written;
 233 }
 234
 235 int __weak phys_mem_access_prot_allowed(struct file *file,
 236         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 237 {
 238         return 1;
 239 }
 240
 241 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 242
 243 /*
 244  * Architectures vary in how they handle caching for addresses
 245  * outside of main memory.
 246  *
 247  */
 248 #ifdef pgprot_noncached
 249 static int uncached_access(struct file *file, unsigned long addr)
 250 {
 251 #if defined(CONFIG_IA64)
 252         /*
 253          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 254          * attribute aliases.
 255          */
 256         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 257 #elif defined(CONFIG_MIPS)
 258         {
 259                 extern int __uncached_access(struct file *file,
 260                                              unsigned long addr);
 261
 262                 return __uncached_access(file, addr);
 263         }
 264 #else
 265         /*
 266          * Accessing memory above the top the kernel knows about or through a
 267          * file pointer
 268          * that was marked O_DSYNC will be done non-cached.
 269          */
 270         if (file->f_flags & O_DSYNC)
 271                 return 1;
 272         return addr >= __pa(high_memory);
 273 #endif
 274 }
 275 #endif
 276
 277 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 278                                      unsigned long size, pgprot_t vma_prot)
 279 {
 280 #ifdef pgprot_noncached
 281         unsigned long offset = pfn << PAGE_SHIFT;
 282
 283         if (uncached_access(file, offset))
 284                 return pgprot_noncached(vma_prot);
 285 #endif
 286         return vma_prot;
 287 }
 288 #endif
 289
 290 #ifndef CONFIG_MMU
 291 static unsigned long get_unmapped_area_mem(struct file *file,
 292                                            unsigned long addr,
 293                                            unsigned long len,
 294                                            unsigned long pgoff,
 295                                            unsigned long flags)
 296 {
 297         if (!valid_mmap_phys_addr_range(pgoff, len))
 298                 return (unsigned long) -EINVAL;
 299         return pgoff << PAGE_SHIFT;
 300 }
 301
 302 /* can't do an in-place private mapping if there's no MMU */
 303 static inline int private_mapping_ok(struct vm_area_struct *vma)
 304 {
 305         return vma->vm_flags & VM_MAYSHARE;
 306 }
 307 #else
 308 #define get_unmapped_area_mem   NULL
 309
 310 static inline int private_mapping_ok(struct vm_area_struct *vma)
 311 {
 312         return 1;
 313 }
 314 #endif
 315
 316 static const struct vm_operations_struct mmap_mem_ops = {
 317 #ifdef CONFIG_HAVE_IOREMAP_PROT
 318         .access = generic_access_phys
 319 #endif
 320 };
 321
 322 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 323 {
 324         size_t size = vma->vm_end - vma->vm_start;
 325         phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
 326
 327         /* It's illegal to wrap around the end of the physical address space. */
 328         if (offset + (phys_addr_t)size - 1 < offset)
 329                 return -EINVAL;
 330
 331         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 332                 return -EINVAL;
 333
 334         if (!private_mapping_ok(vma))
 335                 return -ENOSYS;
 336
 337         if (!range_is_allowed(vma->vm_pgoff, size))
 338                 return -EPERM;
 339
 340         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 341                                                 &vma->vm_page_prot))
 342                 return -EINVAL;
 343
 344         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 345                                                  size,
 346                                                  vma->vm_page_prot);
 347
 348         vma->vm_ops = &mmap_mem_ops;
 349
 350         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 351         if (remap_pfn_range(vma,
 352                             vma->vm_start,
 353                             vma->vm_pgoff,
 354                             size,
 355                             vma->vm_page_prot)) {
 356                 return -EAGAIN;
 357         }
 358         return 0;
 359 }
 360
 361 #ifdef CONFIG_DEVKMEM
 362 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 363 {
 364         unsigned long pfn;
 365
 366         /* Turn a kernel-virtual address into a physical page frame */
 367         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 368
 369         /*
 370          * RED-PEN: on some architectures there is more mapped memory than
 371          * available in mem_map which pfn_valid checks for. Perhaps should add a
 372          * new macro here.
 373          *
 374          * RED-PEN: vmalloc is not supported right now.
 375          */
 376         if (!pfn_valid(pfn))
 377                 return -EIO;
 378
 379         vma->vm_pgoff = pfn;
 380         return mmap_mem(file, vma);
 381 }
 382 #endif
 383
 384 #ifdef CONFIG_CRASH_DUMP
 385 /*
 386  * Read memory corresponding to the old kernel.
 387  */
 388 static ssize_t read_oldmem(struct file *file, char __user *buf,
 389                                 size_t count, loff_t *ppos)
 390 {
 391         unsigned long pfn, offset;
 392         size_t read = 0, csize;
 393         int rc = 0;
 394
 395         while (count) {
 396                 pfn = *ppos / PAGE_SIZE;
 397                 if (pfn > saved_max_pfn)
 398                         return read;
 399
 400                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 401                 if (count > PAGE_SIZE - offset)
 402                         csize = PAGE_SIZE - offset;
 403                 else
 404                         csize = count;
 405
 406                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 407                 if (rc < 0)
 408                         return rc;
 409                 buf += csize;
 410                 *ppos += csize;
 411                 read += csize;
 412                 count -= csize;
 413         }
 414         return read;
 415 }
 416 #endif
 417
 418 #ifdef CONFIG_DEVKMEM
 419 /*
 420  * This function reads the *virtual* memory as seen by the kernel.
 421  */
 422 static ssize_t read_kmem(struct file *file, char __user *buf,
 423                          size_t count, loff_t *ppos)
 424 {
 425         unsigned long p = *ppos;
 426         ssize_t low_count, read, sz;
 427         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 428         int err = 0;
 429
 430         read = 0;
 431         if (p < (unsigned long) high_memory) {
 432                 low_count = count;
 433                 if (count > (unsigned long)high_memory - p)
 434                         low_count = (unsigned long)high_memory - p;
 435
 436 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 437                 /* we don't have page 0 mapped on sparc and m68k.. */
 438                 if (p < PAGE_SIZE && low_count > 0) {
 439                         sz = size_inside_page(p, low_count);
 440                         if (clear_user(buf, sz))
 441                                 return -EFAULT;
 442                         buf += sz;
 443                         p += sz;
 444                         read += sz;
 445                         low_count -= sz;
 446                         count -= sz;
 447                 }
 448 #endif
 449                 while (low_count > 0) {
 450                         sz = size_inside_page(p, low_count);
 451
 452                         /*
 453                          * On ia64 if a page has been mapped somewhere as
 454                          * uncached, then it must also be accessed uncached
 455                          * by the kernel or data corruption may occur
 456                          */
 457                         kbuf = xlate_dev_kmem_ptr((char *)p);
 458
 459                         if (copy_to_user(buf, kbuf, sz))
 460                                 return -EFAULT;
 461                         buf += sz;
 462                         p += sz;
 463                         read += sz;
 464                         low_count -= sz;
 465                         count -= sz;
 466                 }
 467         }
 468
 469         if (count > 0) {
 470                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 471                 if (!kbuf)
 472                         return -ENOMEM;
 473                 while (count > 0) {
 474                         sz = size_inside_page(p, count);
 475                         if (!is_vmalloc_or_module_addr((void *)p)) {
 476                                 err = -ENXIO;
 477                                 break;
 478                         }
 479                         sz = vread(kbuf, (char *)p, sz);
 480                         if (!sz)
 481                                 break;
 482                         if (copy_to_user(buf, kbuf, sz)) {
 483                                 err = -EFAULT;
 484                                 break;
 485                         }
 486                         count -= sz;
 487                         buf += sz;
 488                         read += sz;
 489                         p += sz;
 490                 }
 491                 free_page((unsigned long)kbuf);
 492         }
 493         *ppos = p;
 494         return read ? read : err;
 495 }
 496
 497
 498 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 499                                 size_t count, loff_t *ppos)
 500 {
 501         ssize_t written, sz;
 502         unsigned long copied;
 503
 504         written = 0;
 505 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 506         /* we don't have page 0 mapped on sparc and m68k.. */
 507         if (p < PAGE_SIZE) {
 508                 sz = size_inside_page(p, count);
 509                 /* Hmm. Do something? */
 510                 buf += sz;
 511                 p += sz;
 512                 count -= sz;
 513                 written += sz;
 514         }
 515 #endif
 516
 517         while (count > 0) {
 518                 char *ptr;
 519
 520                 sz = size_inside_page(p, count);
 521
 522                 /*
 523                  * On ia64 if a page has been mapped somewhere as uncached, then
 524                  * it must also be accessed uncached by the kernel or data
 525                  * corruption may occur.
 526                  */
 527                 ptr = xlate_dev_kmem_ptr((char *)p);
 528
 529                 copied = copy_from_user(ptr, buf, sz);
 530                 if (copied) {
 531                         written += sz - copied;
 532                         if (written)
 533                                 break;
 534                         return -EFAULT;
 535                 }
 536                 buf += sz;
 537                 p += sz;
 538                 count -= sz;
 539                 written += sz;
 540         }
 541
 542         *ppos += written;
 543         return written;
 544 }
 545
 546 /*
 547  * This function writes to the *virtual* memory as seen by the kernel.
 548  */
 549 static ssize_t write_kmem(struct file *file, const char __user *buf,
 550                           size_t count, loff_t *ppos)
 551 {
 552         unsigned long p = *ppos;
 553         ssize_t wrote = 0;
 554         ssize_t virtr = 0;
 555         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 556         int err = 0;
 557
 558         if (p < (unsigned long) high_memory) {
 559                 unsigned long to_write = min_t(unsigned long, count,
 560                                                (unsigned long)high_memory - p);
 561                 wrote = do_write_kmem(p, buf, to_write, ppos);
 562                 if (wrote != to_write)
 563                         return wrote;
 564                 p += wrote;
 565                 buf += wrote;
 566                 count -= wrote;
 567         }
 568
 569         if (count > 0) {
 570                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 571                 if (!kbuf)
 572                         return wrote ? wrote : -ENOMEM;
 573                 while (count > 0) {
 574                         unsigned long sz = size_inside_page(p, count);
 575                         unsigned long n;
 576
 577                         if (!is_vmalloc_or_module_addr((void *)p)) {
 578                                 err = -ENXIO;
 579                                 break;
 580                         }
 581                         n = copy_from_user(kbuf, buf, sz);
 582                         if (n) {
 583                                 err = -EFAULT;
 584                                 break;
 585                         }
 586                         vwrite(kbuf, (char *)p, sz);
 587                         count -= sz;
 588                         buf += sz;
 589                         virtr += sz;
 590                         p += sz;
 591                 }
 592                 free_page((unsigned long)kbuf);
 593         }
 594
 595         *ppos = p;
 596         return virtr + wrote ? : err;
 597 }
 598 #endif
 599
 600 #ifdef CONFIG_DEVPORT
 601 static ssize_t read_port(struct file *file, char __user *buf,
 602                          size_t count, loff_t *ppos)
 603 {
 604         unsigned long i = *ppos;
 605         char __user *tmp = buf;
 606
 607         if (!access_ok(VERIFY_WRITE, buf, count))
 608                 return -EFAULT;
 609         while (count-- > 0 && i < 65536) {
 610                 if (__put_user(inb(i), tmp) < 0)
 611                         return -EFAULT;
 612                 i++;
 613                 tmp++;
 614         }
 615         *ppos = i;
 616         return tmp-buf;
 617 }
 618
 619 static ssize_t write_port(struct file *file, const char __user *buf,
 620                           size_t count, loff_t *ppos)
 621 {
 622         unsigned long i = *ppos;
 623         const char __user * tmp = buf;
 624
 625         if (!access_ok(VERIFY_READ, buf, count))
 626                 return -EFAULT;
 627         while (count-- > 0 && i < 65536) {
 628                 char c;
 629                 if (__get_user(c, tmp)) {
 630                         if (tmp > buf)
 631                                 break;
 632                         return -EFAULT;
 633                 }
 634                 outb(c, i);
 635                 i++;
 636                 tmp++;
 637         }
 638         *ppos = i;
 639         return tmp-buf;
 640 }
 641 #endif
 642
 643 static ssize_t read_null(struct file *file, char __user *buf,
 644                          size_t count, loff_t *ppos)
 645 {
 646         return 0;
 647 }
 648
 649 static ssize_t write_null(struct file *file, const char __user *buf,
 650                           size_t count, loff_t *ppos)
 651 {
 652         return count;
 653 }
 654
 655 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 656                         struct splice_desc *sd)
 657 {
 658         return sd->len;
 659 }
 660
 661 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 662                                  loff_t *ppos, size_t len, unsigned int flags)
 663 {
 664         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 665 }
 666
 667 static ssize_t read_zero(struct file *file, char __user *buf,
 668                          size_t count, loff_t *ppos)
 669 {
 670         size_t written;
 671
 672         if (!count)
 673                 return 0;
 674
 675         if (!access_ok(VERIFY_WRITE, buf, count))
 676                 return -EFAULT;
 677
 678         written = 0;
 679         while (count) {
 680                 unsigned long unwritten;
 681                 size_t chunk = count;
 682
 683                 if (chunk > PAGE_SIZE)
 684                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 685                 unwritten = __clear_user(buf, chunk);
 686                 written += chunk - unwritten;
 687                 if (unwritten)
 688                         break;
 689                 if (signal_pending(current))
 690                         return written ? written : -ERESTARTSYS;
 691                 buf += chunk;
 692                 count -= chunk;
 693                 cond_resched();
 694         }
 695         return written ? written : -EFAULT;
 696 }
 697
 698 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 699 {
 700 #ifndef CONFIG_MMU
 701         return -ENOSYS;
 702 #endif
 703         if (vma->vm_flags & VM_SHARED)
 704                 return shmem_zero_setup(vma);
 705         return 0;
 706 }
 707
 708 static ssize_t write_full(struct file *file, const char __user *buf,
 709                           size_t count, loff_t *ppos)
 710 {
 711         return -ENOSPC;
 712 }
 713
 714 /*
 715  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 716  * can fopen() both devices with "a" now.  This was previously impossible.
 717  * -- SRB.
 718  */
 719 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 720 {
 721         return file->f_pos = 0;
 722 }
 723
 724 /*
 725  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 726  * check against negative addresses: they are ok. The return value is weird,
 727  * though, in that case (0).
 728  *
 729  * also note that seeking relative to the "end of file" isn't supported:
 730  * it has no meaning, so it returns -EINVAL.
 731  */
 732 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 733 {
 734         loff_t ret;
 735
 736         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 737         switch (orig) {
 738         case SEEK_CUR:
 739                 offset += file->f_pos;
 740         case SEEK_SET:
 741                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 742                 if ((unsigned long long)offset >= ~0xFFFULL) {
 743                         ret = -EOVERFLOW;
 744                         break;
 745                 }
 746                 file->f_pos = offset;
 747                 ret = file->f_pos;
 748                 force_successful_syscall_return();
 749                 break;
 750         default:
 751                 ret = -EINVAL;
 752         }
 753         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 754         return ret;
 755 }
 756
 757 static int open_port(struct inode * inode, struct file * filp)
 758 {
 759         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 760 }
 761
 762 #define zero_lseek      null_lseek
 763 #define full_lseek      null_lseek
 764 #define write_zero      write_null
 765 #define read_full       read_zero
 766 #define open_mem        open_port
 767 #define open_kmem       open_mem
 768 #define open_oldmem     open_mem
 769
 770 static const struct file_operations mem_fops = {
 771         .llseek         = memory_lseek,
 772         .read           = read_mem,
 773         .write          = write_mem,
 774         .mmap           = mmap_mem,
 775         .open           = open_mem,
 776         .get_unmapped_area = get_unmapped_area_mem,
 777 };
 778
 779 #ifdef CONFIG_DEVKMEM
 780 static const struct file_operations kmem_fops = {
 781         .llseek         = memory_lseek,
 782         .read           = read_kmem,
 783         .write          = write_kmem,
 784         .mmap           = mmap_kmem,
 785         .open           = open_kmem,
 786         .get_unmapped_area = get_unmapped_area_mem,
 787 };
 788 #endif
 789
 790 static const struct file_operations null_fops = {
 791         .llseek         = null_lseek,
 792         .read           = read_null,
 793         .write          = write_null,
 794         .splice_write   = splice_write_null,
 795 };
 796
 797 #ifdef CONFIG_DEVPORT
 798 static const struct file_operations port_fops = {
 799         .llseek         = memory_lseek,
 800         .read           = read_port,
 801         .write          = write_port,
 802         .open           = open_port,
 803 };
 804 #endif
 805
 806 static const struct file_operations zero_fops = {
 807         .llseek         = zero_lseek,
 808         .read           = read_zero,
 809         .write          = write_zero,
 810         .mmap           = mmap_zero,
 811 };
 812
 813 /*
 814  * capabilities for /dev/zero
 815  * - permits private mappings, "copies" are taken of the source of zeros
 816  * - no writeback happens
 817  */
 818 static struct backing_dev_info zero_bdi = {
 819         .name           = "char/mem",
 820         .capabilities   = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
 821 };
 822
 823 static const struct file_operations full_fops = {
 824         .llseek         = full_lseek,
 825         .read           = read_full,
 826         .write          = write_full,
 827 };
 828
 829 #ifdef CONFIG_CRASH_DUMP
 830 static const struct file_operations oldmem_fops = {
 831         .read   = read_oldmem,
 832         .open   = open_oldmem,
 833         .llseek = default_llseek,
 834 };
 835 #endif
 836
 837 static ssize_t kmsg_writev(struct kiocb *iocb, const struct iovec *iv,
 838                            unsigned long count, loff_t pos)
 839 {
 840         char *line, *p;
 841         int i;
 842         ssize_t ret = -EFAULT;
 843         size_t len = iov_length(iv, count);
 844
 845         line = kmalloc(len + 1, GFP_KERNEL);
 846         if (line == NULL)
 847                 return -ENOMEM;
 848
 849         /*
 850          * copy all vectors into a single string, to ensure we do
 851          * not interleave our log line with other printk calls
 852          */
 853         p = line;
 854         for (i = 0; i < count; i++) {
 855                 if (copy_from_user(p, iv[i].iov_base, iv[i].iov_len))
 856                         goto out;
 857                 p += iv[i].iov_len;
 858         }
 859         p[0] = '\0';
 860
 861         ret = printk("%s", line);
 862         /* printk can add a prefix */
 863         if (ret > len)
 864                 ret = len;
 865 out:
 866         kfree(line);
 867         return ret;
 868 }
 869
 870 static const struct file_operations kmsg_fops = {
 871         .aio_write = kmsg_writev,
 872         .llseek = noop_llseek,
 873 };
 874
 875 static const struct memdev {
 876         const char *name;
 877         mode_t mode;
 878         const struct file_operations *fops;
 879         struct backing_dev_info *dev_info;
 880 } devlist[] = {
 881          [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
 882 #ifdef CONFIG_DEVKMEM
 883          [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
 884 #endif
 885          [3] = { "null", 0666, &null_fops, NULL },
 886 #ifdef CONFIG_DEVPORT
 887          [4] = { "port", 0, &port_fops, NULL },
 888 #endif
 889          [5] = { "zero", 0666, &zero_fops, &zero_bdi },
 890          [7] = { "full", 0666, &full_fops, NULL },
 891          [8] = { "random", 0666, &random_fops, NULL },
 892          [9] = { "urandom", 0666, &urandom_fops, NULL },
 893         [11] = { "kmsg", 0, &kmsg_fops, NULL },
 894 #ifdef CONFIG_CRASH_DUMP
 895         [12] = { "oldmem", 0, &oldmem_fops, NULL },
 896 #endif
 897 };
 898
 899 static int memory_open(struct inode *inode, struct file *filp)
 900 {
 901         int minor;
 902         const struct memdev *dev;
 903
 904         minor = iminor(inode);
 905         if (minor >= ARRAY_SIZE(devlist))
 906                 return -ENXIO;
 907
 908         dev = &devlist[minor];
 909         if (!dev->fops)
 910                 return -ENXIO;
 911
 912         filp->f_op = dev->fops;
 913         if (dev->dev_info)
 914                 filp->f_mapping->backing_dev_info = dev->dev_info;
 915
 916         /* Is /dev/mem or /dev/kmem ? */
 917         if (dev->dev_info == &directly_mappable_cdev_bdi)
 918                 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
 919
 920         if (dev->fops->open)
 921                 return dev->fops->open(inode, filp);
 922
 923         return 0;
 924 }
 925
 926 static const struct file_operations memory_fops = {
 927         .open = memory_open,
 928         .llseek = noop_llseek,
 929 };
 930
 931 static char *mem_devnode(struct device *dev, mode_t *mode)
 932 {
 933         if (mode && devlist[MINOR(dev->devt)].mode)
 934                 *mode = devlist[MINOR(dev->devt)].mode;
 935         return NULL;
 936 }
 937
 938 static struct class *mem_class;
 939
 940 static int __init chr_dev_init(void)
 941 {
 942         int minor;
 943         int err;
 944
 945         err = bdi_init(&zero_bdi);
 946         if (err)
 947                 return err;
 948
 949         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 950                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 951
 952         mem_class = class_create(THIS_MODULE, "mem");
 953         if (IS_ERR(mem_class))
 954                 return PTR_ERR(mem_class);
 955
 956         mem_class->devnode = mem_devnode;
 957         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 958                 if (!devlist[minor].name)
 959                         continue;
 960                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 961                               NULL, devlist[minor].name);
 962         }
 963
 964         return tty_init();
 965 }
 966
 967 fs_initcall(chr_dev_init);