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 #include <linux/io.h>
  31 #include <linux/aio.h>
  32
  33 #include <asm/uaccess.h>
  34
  35 #ifdef CONFIG_IA64
  36 # include <linux/efi.h>
  37 #endif
  38
  39 #define DEVPORT_MINOR   4
  40
  41 static inline unsigned long size_inside_page(unsigned long start,
  42                                              unsigned long size)
  43 {
  44         unsigned long sz;
  45
  46         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  47
  48         return min(sz, size);
  49 }
  50
  51 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  52 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
  53 {
  54         return addr + count <= __pa(high_memory);
  55 }
  56
  57 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  58 {
  59         return 1;
  60 }
  61 #endif
  62
  63 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM)
  64 #ifdef CONFIG_STRICT_DEVMEM
  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 range_is_allowed(unsigned long pfn, unsigned long size)
  85 {
  86         return 1;
  87 }
  88 #endif
  89 #endif
  90
  91 #ifdef CONFIG_DEVMEM
  92 void __weak unxlate_dev_mem_ptr(unsigned long phys, void *addr)
  93 {
  94 }
  95
  96 /*
  97  * This funcion reads the *physical* memory. The f_pos points directly to the
  98  * memory location.
  99  */
 100 static ssize_t read_mem(struct file *file, char __user *buf,
 101                         size_t count, loff_t *ppos)
 102 {
 103         phys_addr_t p = *ppos;
 104         ssize_t read, sz;
 105         char *ptr;
 106
 107         if (!valid_phys_addr_range(p, count))
 108                 return -EFAULT;
 109         read = 0;
 110 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 111         /* we don't have page 0 mapped on sparc and m68k.. */
 112         if (p < PAGE_SIZE) {
 113                 sz = size_inside_page(p, count);
 114                 if (sz > 0) {
 115                         if (clear_user(buf, sz))
 116                                 return -EFAULT;
 117                         buf += sz;
 118                         p += sz;
 119                         count -= sz;
 120                         read += sz;
 121                 }
 122         }
 123 #endif
 124
 125         while (count > 0) {
 126                 unsigned long remaining;
 127
 128                 sz = size_inside_page(p, count);
 129
 130                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 131                         return -EPERM;
 132
 133                 /*
 134                  * On ia64 if a page has been mapped somewhere as uncached, then
 135                  * it must also be accessed uncached by the kernel or data
 136                  * corruption may occur.
 137                  */
 138                 ptr = xlate_dev_mem_ptr(p);
 139                 if (!ptr)
 140                         return -EFAULT;
 141
 142                 remaining = copy_to_user(buf, ptr, sz);
 143                 unxlate_dev_mem_ptr(p, ptr);
 144                 if (remaining)
 145                         return -EFAULT;
 146
 147                 buf += sz;
 148                 p += sz;
 149                 count -= sz;
 150                 read += sz;
 151         }
 152
 153         *ppos += read;
 154         return read;
 155 }
 156
 157 static ssize_t write_mem(struct file *file, const char __user *buf,
 158                          size_t count, loff_t *ppos)
 159 {
 160         phys_addr_t p = *ppos;
 161         ssize_t written, sz;
 162         unsigned long copied;
 163         void *ptr;
 164
 165         if (!valid_phys_addr_range(p, count))
 166                 return -EFAULT;
 167
 168         written = 0;
 169
 170 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 171         /* we don't have page 0 mapped on sparc and m68k.. */
 172         if (p < PAGE_SIZE) {
 173                 sz = size_inside_page(p, count);
 174                 /* Hmm. Do something? */
 175                 buf += sz;
 176                 p += sz;
 177                 count -= sz;
 178                 written += sz;
 179         }
 180 #endif
 181
 182         while (count > 0) {
 183                 sz = size_inside_page(p, count);
 184
 185                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 186                         return -EPERM;
 187
 188                 /*
 189                  * On ia64 if a page has been mapped somewhere as uncached, then
 190                  * it must also be accessed uncached by the kernel or data
 191                  * corruption may occur.
 192                  */
 193                 ptr = xlate_dev_mem_ptr(p);
 194                 if (!ptr) {
 195                         if (written)
 196                                 break;
 197                         return -EFAULT;
 198                 }
 199
 200                 copied = copy_from_user(ptr, buf, sz);
 201                 unxlate_dev_mem_ptr(p, ptr);
 202                 if (copied) {
 203                         written += sz - copied;
 204                         if (written)
 205                                 break;
 206                         return -EFAULT;
 207                 }
 208
 209                 buf += sz;
 210                 p += sz;
 211                 count -= sz;
 212                 written += sz;
 213         }
 214
 215         *ppos += written;
 216         return written;
 217 }
 218 #endif  /* CONFIG_DEVMEM */
 219
 220 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM)
 221
 222 int __weak phys_mem_access_prot_allowed(struct file *file,
 223         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 224 {
 225         return 1;
 226 }
 227
 228 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 229
 230 /*
 231  * Architectures vary in how they handle caching for addresses
 232  * outside of main memory.
 233  *
 234  */
 235 #ifdef pgprot_noncached
 236 static int uncached_access(struct file *file, phys_addr_t addr)
 237 {
 238 #if defined(CONFIG_IA64)
 239         /*
 240          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 241          * attribute aliases.
 242          */
 243         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 244 #elif defined(CONFIG_MIPS)
 245         {
 246                 extern int __uncached_access(struct file *file,
 247                                              unsigned long addr);
 248
 249                 return __uncached_access(file, addr);
 250         }
 251 #else
 252         /*
 253          * Accessing memory above the top the kernel knows about or through a
 254          * file pointer
 255          * that was marked O_DSYNC will be done non-cached.
 256          */
 257         if (file->f_flags & O_DSYNC)
 258                 return 1;
 259         return addr >= __pa(high_memory);
 260 #endif
 261 }
 262 #endif
 263
 264 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 265                                      unsigned long size, pgprot_t vma_prot)
 266 {
 267 #ifdef pgprot_noncached
 268         phys_addr_t offset = pfn << PAGE_SHIFT;
 269
 270         if (uncached_access(file, offset))
 271                 return pgprot_noncached(vma_prot);
 272 #endif
 273         return vma_prot;
 274 }
 275 #endif
 276
 277 #ifndef CONFIG_MMU
 278 static unsigned long get_unmapped_area_mem(struct file *file,
 279                                            unsigned long addr,
 280                                            unsigned long len,
 281                                            unsigned long pgoff,
 282                                            unsigned long flags)
 283 {
 284         if (!valid_mmap_phys_addr_range(pgoff, len))
 285                 return (unsigned long) -EINVAL;
 286         return pgoff << PAGE_SHIFT;
 287 }
 288
 289 /* can't do an in-place private mapping if there's no MMU */
 290 static inline int private_mapping_ok(struct vm_area_struct *vma)
 291 {
 292         return vma->vm_flags & VM_MAYSHARE;
 293 }
 294 #else
 295 #define get_unmapped_area_mem   NULL
 296
 297 static inline int private_mapping_ok(struct vm_area_struct *vma)
 298 {
 299         return 1;
 300 }
 301 #endif
 302
 303 static const struct vm_operations_struct mmap_mem_ops = {
 304 #ifdef CONFIG_HAVE_IOREMAP_PROT
 305         .access = generic_access_phys
 306 #endif
 307 };
 308
 309 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 310 {
 311         size_t size = vma->vm_end - vma->vm_start;
 312
 313         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 314                 return -EINVAL;
 315
 316         if (!private_mapping_ok(vma))
 317                 return -ENOSYS;
 318
 319         if (!range_is_allowed(vma->vm_pgoff, size))
 320                 return -EPERM;
 321
 322         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 323                                                 &vma->vm_page_prot))
 324                 return -EINVAL;
 325
 326         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 327                                                  size,
 328                                                  vma->vm_page_prot);
 329
 330         vma->vm_ops = &mmap_mem_ops;
 331
 332         /* Remap-pfn-range will mark the range VM_IO */
 333         if (remap_pfn_range(vma,
 334                             vma->vm_start,
 335                             vma->vm_pgoff,
 336                             size,
 337                             vma->vm_page_prot)) {
 338                 return -EAGAIN;
 339         }
 340         return 0;
 341 }
 342 #endif  /* CONFIG_DEVMEM */
 343
 344 #ifdef CONFIG_DEVKMEM
 345 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 346 {
 347         unsigned long pfn;
 348
 349         /* Turn a kernel-virtual address into a physical page frame */
 350         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 351
 352         /*
 353          * RED-PEN: on some architectures there is more mapped memory than
 354          * available in mem_map which pfn_valid checks for. Perhaps should add a
 355          * new macro here.
 356          *
 357          * RED-PEN: vmalloc is not supported right now.
 358          */
 359         if (!pfn_valid(pfn))
 360                 return -EIO;
 361
 362         vma->vm_pgoff = pfn;
 363         return mmap_mem(file, vma);
 364 }
 365 #endif
 366
 367 #ifdef CONFIG_CRASH_DUMP
 368 /*
 369  * Read memory corresponding to the old kernel.
 370  */
 371 static ssize_t read_oldmem(struct file *file, char __user *buf,
 372                                 size_t count, loff_t *ppos)
 373 {
 374         unsigned long pfn, offset;
 375         size_t read = 0, csize;
 376         int rc = 0;
 377
 378         while (count) {
 379                 pfn = *ppos / PAGE_SIZE;
 380                 if (pfn > saved_max_pfn)
 381                         return read;
 382
 383                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 384                 if (count > PAGE_SIZE - offset)
 385                         csize = PAGE_SIZE - offset;
 386                 else
 387                         csize = count;
 388
 389                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 390                 if (rc < 0)
 391                         return rc;
 392                 buf += csize;
 393                 *ppos += csize;
 394                 read += csize;
 395                 count -= csize;
 396         }
 397         return read;
 398 }
 399 #endif
 400
 401 #ifdef CONFIG_DEVKMEM
 402 /*
 403  * This function reads the *virtual* memory as seen by the kernel.
 404  */
 405 static ssize_t read_kmem(struct file *file, char __user *buf,
 406                          size_t count, loff_t *ppos)
 407 {
 408         unsigned long p = *ppos;
 409         ssize_t low_count, read, sz;
 410         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 411         int err = 0;
 412
 413         read = 0;
 414         if (p < (unsigned long) high_memory) {
 415                 low_count = count;
 416                 if (count > (unsigned long)high_memory - p)
 417                         low_count = (unsigned long)high_memory - p;
 418
 419 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 420                 /* we don't have page 0 mapped on sparc and m68k.. */
 421                 if (p < PAGE_SIZE && low_count > 0) {
 422                         sz = size_inside_page(p, low_count);
 423                         if (clear_user(buf, sz))
 424                                 return -EFAULT;
 425                         buf += sz;
 426                         p += sz;
 427                         read += sz;
 428                         low_count -= sz;
 429                         count -= sz;
 430                 }
 431 #endif
 432                 while (low_count > 0) {
 433                         sz = size_inside_page(p, low_count);
 434
 435                         /*
 436                          * On ia64 if a page has been mapped somewhere as
 437                          * uncached, then it must also be accessed uncached
 438                          * by the kernel or data corruption may occur
 439                          */
 440                         kbuf = xlate_dev_kmem_ptr((char *)p);
 441
 442                         if (copy_to_user(buf, kbuf, sz))
 443                                 return -EFAULT;
 444                         buf += sz;
 445                         p += sz;
 446                         read += sz;
 447                         low_count -= sz;
 448                         count -= sz;
 449                 }
 450         }
 451
 452         if (count > 0) {
 453                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 454                 if (!kbuf)
 455                         return -ENOMEM;
 456                 while (count > 0) {
 457                         sz = size_inside_page(p, count);
 458                         if (!is_vmalloc_or_module_addr((void *)p)) {
 459                                 err = -ENXIO;
 460                                 break;
 461                         }
 462                         sz = vread(kbuf, (char *)p, sz);
 463                         if (!sz)
 464                                 break;
 465                         if (copy_to_user(buf, kbuf, sz)) {
 466                                 err = -EFAULT;
 467                                 break;
 468                         }
 469                         count -= sz;
 470                         buf += sz;
 471                         read += sz;
 472                         p += sz;
 473                 }
 474                 free_page((unsigned long)kbuf);
 475         }
 476         *ppos = p;
 477         return read ? read : err;
 478 }
 479
 480
 481 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 482                                 size_t count, loff_t *ppos)
 483 {
 484         ssize_t written, sz;
 485         unsigned long copied;
 486
 487         written = 0;
 488 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 489         /* we don't have page 0 mapped on sparc and m68k.. */
 490         if (p < PAGE_SIZE) {
 491                 sz = size_inside_page(p, count);
 492                 /* Hmm. Do something? */
 493                 buf += sz;
 494                 p += sz;
 495                 count -= sz;
 496                 written += sz;
 497         }
 498 #endif
 499
 500         while (count > 0) {
 501                 char *ptr;
 502
 503                 sz = size_inside_page(p, count);
 504
 505                 /*
 506                  * On ia64 if a page has been mapped somewhere as uncached, then
 507                  * it must also be accessed uncached by the kernel or data
 508                  * corruption may occur.
 509                  */
 510                 ptr = xlate_dev_kmem_ptr((char *)p);
 511
 512                 copied = copy_from_user(ptr, buf, sz);
 513                 if (copied) {
 514                         written += sz - copied;
 515                         if (written)
 516                                 break;
 517                         return -EFAULT;
 518                 }
 519                 buf += sz;
 520                 p += sz;
 521                 count -= sz;
 522                 written += sz;
 523         }
 524
 525         *ppos += written;
 526         return written;
 527 }
 528
 529 /*
 530  * This function writes to the *virtual* memory as seen by the kernel.
 531  */
 532 static ssize_t write_kmem(struct file *file, const char __user *buf,
 533                           size_t count, loff_t *ppos)
 534 {
 535         unsigned long p = *ppos;
 536         ssize_t wrote = 0;
 537         ssize_t virtr = 0;
 538         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 539         int err = 0;
 540
 541         if (p < (unsigned long) high_memory) {
 542                 unsigned long to_write = min_t(unsigned long, count,
 543                                                (unsigned long)high_memory - p);
 544                 wrote = do_write_kmem(p, buf, to_write, ppos);
 545                 if (wrote != to_write)
 546                         return wrote;
 547                 p += wrote;
 548                 buf += wrote;
 549                 count -= wrote;
 550         }
 551
 552         if (count > 0) {
 553                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 554                 if (!kbuf)
 555                         return wrote ? wrote : -ENOMEM;
 556                 while (count > 0) {
 557                         unsigned long sz = size_inside_page(p, count);
 558                         unsigned long n;
 559
 560                         if (!is_vmalloc_or_module_addr((void *)p)) {
 561                                 err = -ENXIO;
 562                                 break;
 563                         }
 564                         n = copy_from_user(kbuf, buf, sz);
 565                         if (n) {
 566                                 err = -EFAULT;
 567                                 break;
 568                         }
 569                         vwrite(kbuf, (char *)p, sz);
 570                         count -= sz;
 571                         buf += sz;
 572                         virtr += sz;
 573                         p += sz;
 574                 }
 575                 free_page((unsigned long)kbuf);
 576         }
 577
 578         *ppos = p;
 579         return virtr + wrote ? : err;
 580 }
 581 #endif
 582
 583 #ifdef CONFIG_DEVPORT
 584 static ssize_t read_port(struct file *file, char __user *buf,
 585                          size_t count, loff_t *ppos)
 586 {
 587         unsigned long i = *ppos;
 588         char __user *tmp = buf;
 589
 590         if (!access_ok(VERIFY_WRITE, buf, count))
 591                 return -EFAULT;
 592         while (count-- > 0 && i < 65536) {
 593                 if (__put_user(inb(i), tmp) < 0)
 594                         return -EFAULT;
 595                 i++;
 596                 tmp++;
 597         }
 598         *ppos = i;
 599         return tmp-buf;
 600 }
 601
 602 static ssize_t write_port(struct file *file, const char __user *buf,
 603                           size_t count, loff_t *ppos)
 604 {
 605         unsigned long i = *ppos;
 606         const char __user *tmp = buf;
 607
 608         if (!access_ok(VERIFY_READ, buf, count))
 609                 return -EFAULT;
 610         while (count-- > 0 && i < 65536) {
 611                 char c;
 612                 if (__get_user(c, tmp)) {
 613                         if (tmp > buf)
 614                                 break;
 615                         return -EFAULT;
 616                 }
 617                 outb(c, i);
 618                 i++;
 619                 tmp++;
 620         }
 621         *ppos = i;
 622         return tmp-buf;
 623 }
 624 #endif
 625
 626 static ssize_t read_null(struct file *file, char __user *buf,
 627                          size_t count, loff_t *ppos)
 628 {
 629         return 0;
 630 }
 631
 632 static ssize_t write_null(struct file *file, const char __user *buf,
 633                           size_t count, loff_t *ppos)
 634 {
 635         return count;
 636 }
 637
 638 static ssize_t aio_read_null(struct kiocb *iocb, const struct iovec *iov,
 639                              unsigned long nr_segs, loff_t pos)
 640 {
 641         return 0;
 642 }
 643
 644 static ssize_t aio_write_null(struct kiocb *iocb, const struct iovec *iov,
 645                               unsigned long nr_segs, loff_t pos)
 646 {
 647         return iov_length(iov, nr_segs);
 648 }
 649
 650 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 651                         struct splice_desc *sd)
 652 {
 653         return sd->len;
 654 }
 655
 656 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 657                                  loff_t *ppos, size_t len, unsigned int flags)
 658 {
 659         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 660 }
 661
 662 static ssize_t read_zero(struct file *file, char __user *buf,
 663                          size_t count, loff_t *ppos)
 664 {
 665         size_t written;
 666
 667         if (!count)
 668                 return 0;
 669
 670         if (!access_ok(VERIFY_WRITE, buf, count))
 671                 return -EFAULT;
 672
 673         written = 0;
 674         while (count) {
 675                 unsigned long unwritten;
 676                 size_t chunk = count;
 677
 678                 if (chunk > PAGE_SIZE)
 679                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 680                 unwritten = __clear_user(buf, chunk);
 681                 written += chunk - unwritten;
 682                 if (unwritten)
 683                         break;
 684                 if (signal_pending(current))
 685                         return written ? written : -ERESTARTSYS;
 686                 buf += chunk;
 687                 count -= chunk;
 688                 cond_resched();
 689         }
 690         return written ? written : -EFAULT;
 691 }
 692
 693 static ssize_t aio_read_zero(struct kiocb *iocb, const struct iovec *iov,
 694                              unsigned long nr_segs, loff_t pos)
 695 {
 696         size_t written = 0;
 697         unsigned long i;
 698         ssize_t ret;
 699
 700         for (i = 0; i < nr_segs; i++) {
 701                 ret = read_zero(iocb->ki_filp, iov[i].iov_base, iov[i].iov_len,
 702                                 &pos);
 703                 if (ret < 0)
 704                         break;
 705                 written += ret;
 706         }
 707
 708         return written ? written : -EFAULT;
 709 }
 710
 711 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 712 {
 713 #ifndef CONFIG_MMU
 714         return -ENOSYS;
 715 #endif
 716         if (vma->vm_flags & VM_SHARED)
 717                 return shmem_zero_setup(vma);
 718         return 0;
 719 }
 720
 721 static ssize_t write_full(struct file *file, const char __user *buf,
 722                           size_t count, loff_t *ppos)
 723 {
 724         return -ENOSPC;
 725 }
 726
 727 /*
 728  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 729  * can fopen() both devices with "a" now.  This was previously impossible.
 730  * -- SRB.
 731  */
 732 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 733 {
 734         return file->f_pos = 0;
 735 }
 736
 737 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM) || defined(CONFIG_DEVPORT)
 738
 739 /*
 740  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 741  * check against negative addresses: they are ok. The return value is weird,
 742  * though, in that case (0).
 743  *
 744  * also note that seeking relative to the "end of file" isn't supported:
 745  * it has no meaning, so it returns -EINVAL.
 746  */
 747 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 748 {
 749         loff_t ret;
 750
 751         mutex_lock(&file_inode(file)->i_mutex);
 752         switch (orig) {
 753         case SEEK_CUR:
 754                 offset += file->f_pos;
 755         case SEEK_SET:
 756                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 757                 if ((unsigned long long)offset >= ~0xFFFULL) {
 758                         ret = -EOVERFLOW;
 759                         break;
 760                 }
 761                 file->f_pos = offset;
 762                 ret = file->f_pos;
 763                 force_successful_syscall_return();
 764                 break;
 765         default:
 766                 ret = -EINVAL;
 767         }
 768         mutex_unlock(&file_inode(file)->i_mutex);
 769         return ret;
 770 }
 771
 772 #endif
 773
 774 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM) || defined(CONFIG_DEVPORT)
 775 static int open_port(struct inode *inode, struct file *filp)
 776 {
 777         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 778 }
 779 #endif
 780
 781 #define zero_lseek      null_lseek
 782 #define full_lseek      null_lseek
 783 #define write_zero      write_null
 784 #define read_full       read_zero
 785 #define aio_write_zero  aio_write_null
 786 #define open_mem        open_port
 787 #define open_kmem       open_mem
 788 #define open_oldmem     open_mem
 789
 790 #ifdef CONFIG_DEVMEM
 791 static const struct file_operations mem_fops = {
 792         .llseek         = memory_lseek,
 793         .read           = read_mem,
 794         .write          = write_mem,
 795         .mmap           = mmap_mem,
 796         .open           = open_mem,
 797         .get_unmapped_area = get_unmapped_area_mem,
 798 };
 799 #endif
 800
 801 #ifdef CONFIG_DEVKMEM
 802 static const struct file_operations kmem_fops = {
 803         .llseek         = memory_lseek,
 804         .read           = read_kmem,
 805         .write          = write_kmem,
 806         .mmap           = mmap_kmem,
 807         .open           = open_kmem,
 808         .get_unmapped_area = get_unmapped_area_mem,
 809 };
 810 #endif
 811
 812 static const struct file_operations null_fops = {
 813         .llseek         = null_lseek,
 814         .read           = read_null,
 815         .write          = write_null,
 816         .aio_read       = aio_read_null,
 817         .aio_write      = aio_write_null,
 818         .splice_write   = splice_write_null,
 819 };
 820
 821 #ifdef CONFIG_DEVPORT
 822 static const struct file_operations port_fops = {
 823         .llseek         = memory_lseek,
 824         .read           = read_port,
 825         .write          = write_port,
 826         .open           = open_port,
 827 };
 828 #endif
 829
 830 static const struct file_operations zero_fops = {
 831         .llseek         = zero_lseek,
 832         .read           = read_zero,
 833         .write          = write_zero,
 834         .aio_read       = aio_read_zero,
 835         .aio_write      = aio_write_zero,
 836         .mmap           = mmap_zero,
 837 };
 838
 839 /*
 840  * capabilities for /dev/zero
 841  * - permits private mappings, "copies" are taken of the source of zeros
 842  * - no writeback happens
 843  */
 844 static struct backing_dev_info zero_bdi = {
 845         .name           = "char/mem",
 846         .capabilities   = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
 847 };
 848
 849 static const struct file_operations full_fops = {
 850         .llseek         = full_lseek,
 851         .read           = read_full,
 852         .write          = write_full,
 853 };
 854
 855 #ifdef CONFIG_CRASH_DUMP
 856 static const struct file_operations oldmem_fops = {
 857         .read   = read_oldmem,
 858         .open   = open_oldmem,
 859         .llseek = default_llseek,
 860 };
 861 #endif
 862
 863 static const struct memdev {
 864         const char *name;
 865         umode_t mode;
 866         const struct file_operations *fops;
 867         struct backing_dev_info *dev_info;
 868 } devlist[] = {
 869 #ifdef CONFIG_DEVMEM
 870          [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
 871 #endif
 872 #ifdef CONFIG_DEVKMEM
 873          [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
 874 #endif
 875          [3] = { "null", 0666, &null_fops, NULL },
 876 #ifdef CONFIG_DEVPORT
 877          [4] = { "port", 0, &port_fops, NULL },
 878 #endif
 879          [5] = { "zero", 0666, &zero_fops, &zero_bdi },
 880          [7] = { "full", 0666, &full_fops, NULL },
 881          [8] = { "random", 0666, &random_fops, NULL },
 882          [9] = { "urandom", 0666, &urandom_fops, NULL },
 883 #ifdef CONFIG_PRINTK
 884         [11] = { "kmsg", 0644, &kmsg_fops, NULL },
 885 #endif
 886 #ifdef CONFIG_CRASH_DUMP
 887         [12] = { "oldmem", 0, &oldmem_fops, NULL },
 888 #endif
 889 };
 890
 891 static int memory_open(struct inode *inode, struct file *filp)
 892 {
 893         int minor;
 894         const struct memdev *dev;
 895
 896         minor = iminor(inode);
 897         if (minor >= ARRAY_SIZE(devlist))
 898                 return -ENXIO;
 899
 900         dev = &devlist[minor];
 901         if (!dev->fops)
 902                 return -ENXIO;
 903
 904         filp->f_op = dev->fops;
 905         if (dev->dev_info)
 906                 filp->f_mapping->backing_dev_info = dev->dev_info;
 907
 908         /* Is /dev/mem or /dev/kmem ? */
 909         if (dev->dev_info == &directly_mappable_cdev_bdi)
 910                 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
 911
 912         if (dev->fops->open)
 913                 return dev->fops->open(inode, filp);
 914
 915         return 0;
 916 }
 917
 918 static const struct file_operations memory_fops = {
 919         .open = memory_open,
 920         .llseek = noop_llseek,
 921 };
 922
 923 static char *mem_devnode(struct device *dev, umode_t *mode)
 924 {
 925         if (mode && devlist[MINOR(dev->devt)].mode)
 926                 *mode = devlist[MINOR(dev->devt)].mode;
 927         return NULL;
 928 }
 929
 930 static struct class *mem_class;
 931
 932 static int __init chr_dev_init(void)
 933 {
 934         int minor;
 935         int err;
 936
 937         err = bdi_init(&zero_bdi);
 938         if (err)
 939                 return err;
 940
 941         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 942                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 943
 944         mem_class = class_create(THIS_MODULE, "mem");
 945         if (IS_ERR(mem_class))
 946                 return PTR_ERR(mem_class);
 947
 948         mem_class->devnode = mem_devnode;
 949         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 950                 if (!devlist[minor].name)
 951                         continue;
 952
 953                 /*
 954                  * Create /dev/port?
 955                  */
 956                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 957                         continue;
 958
 959                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 960                               NULL, devlist[minor].name);
 961         }
 962
 963         return tty_init();
 964 }
 965
 966 fs_initcall(chr_dev_init);