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
 326         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 327                 return -EINVAL;
 328
 329         if (!private_mapping_ok(vma))
 330                 return -ENOSYS;
 331
 332         if (!range_is_allowed(vma->vm_pgoff, size))
 333                 return -EPERM;
 334
 335         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 336                                                 &vma->vm_page_prot))
 337                 return -EINVAL;
 338
 339         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 340                                                  size,
 341                                                  vma->vm_page_prot);
 342
 343         vma->vm_ops = &mmap_mem_ops;
 344
 345         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 346         if (remap_pfn_range(vma,
 347                             vma->vm_start,
 348                             vma->vm_pgoff,
 349                             size,
 350                             vma->vm_page_prot)) {
 351                 return -EAGAIN;
 352         }
 353         return 0;
 354 }
 355
 356 #ifdef CONFIG_DEVKMEM
 357 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 358 {
 359         unsigned long pfn;
 360
 361         /* Turn a kernel-virtual address into a physical page frame */
 362         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 363
 364         /*
 365          * RED-PEN: on some architectures there is more mapped memory than
 366          * available in mem_map which pfn_valid checks for. Perhaps should add a
 367          * new macro here.
 368          *
 369          * RED-PEN: vmalloc is not supported right now.
 370          */
 371         if (!pfn_valid(pfn))
 372                 return -EIO;
 373
 374         vma->vm_pgoff = pfn;
 375         return mmap_mem(file, vma);
 376 }
 377 #endif
 378
 379 #ifdef CONFIG_CRASH_DUMP
 380 /*
 381  * Read memory corresponding to the old kernel.
 382  */
 383 static ssize_t read_oldmem(struct file *file, char __user *buf,
 384                                 size_t count, loff_t *ppos)
 385 {
 386         unsigned long pfn, offset;
 387         size_t read = 0, csize;
 388         int rc = 0;
 389
 390         while (count) {
 391                 pfn = *ppos / PAGE_SIZE;
 392                 if (pfn > saved_max_pfn)
 393                         return read;
 394
 395                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 396                 if (count > PAGE_SIZE - offset)
 397                         csize = PAGE_SIZE - offset;
 398                 else
 399                         csize = count;
 400
 401                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 402                 if (rc < 0)
 403                         return rc;
 404                 buf += csize;
 405                 *ppos += csize;
 406                 read += csize;
 407                 count -= csize;
 408         }
 409         return read;
 410 }
 411 #endif
 412
 413 #ifdef CONFIG_DEVKMEM
 414 /*
 415  * This function reads the *virtual* memory as seen by the kernel.
 416  */
 417 static ssize_t read_kmem(struct file *file, char __user *buf,
 418                          size_t count, loff_t *ppos)
 419 {
 420         unsigned long p = *ppos;
 421         ssize_t low_count, read, sz;
 422         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 423         int err = 0;
 424
 425         read = 0;
 426         if (p < (unsigned long) high_memory) {
 427                 low_count = count;
 428                 if (count > (unsigned long)high_memory - p)
 429                         low_count = (unsigned long)high_memory - p;
 430
 431 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 432                 /* we don't have page 0 mapped on sparc and m68k.. */
 433                 if (p < PAGE_SIZE && low_count > 0) {
 434                         sz = size_inside_page(p, low_count);
 435                         if (clear_user(buf, sz))
 436                                 return -EFAULT;
 437                         buf += sz;
 438                         p += sz;
 439                         read += sz;
 440                         low_count -= sz;
 441                         count -= sz;
 442                 }
 443 #endif
 444                 while (low_count > 0) {
 445                         sz = size_inside_page(p, low_count);
 446
 447                         /*
 448                          * On ia64 if a page has been mapped somewhere as
 449                          * uncached, then it must also be accessed uncached
 450                          * by the kernel or data corruption may occur
 451                          */
 452                         kbuf = xlate_dev_kmem_ptr((char *)p);
 453
 454                         if (copy_to_user(buf, kbuf, sz))
 455                                 return -EFAULT;
 456                         buf += sz;
 457                         p += sz;
 458                         read += sz;
 459                         low_count -= sz;
 460                         count -= sz;
 461                 }
 462         }
 463
 464         if (count > 0) {
 465                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 466                 if (!kbuf)
 467                         return -ENOMEM;
 468                 while (count > 0) {
 469                         sz = size_inside_page(p, count);
 470                         if (!is_vmalloc_or_module_addr((void *)p)) {
 471                                 err = -ENXIO;
 472                                 break;
 473                         }
 474                         sz = vread(kbuf, (char *)p, sz);
 475                         if (!sz)
 476                                 break;
 477                         if (copy_to_user(buf, kbuf, sz)) {
 478                                 err = -EFAULT;
 479                                 break;
 480                         }
 481                         count -= sz;
 482                         buf += sz;
 483                         read += sz;
 484                         p += sz;
 485                 }
 486                 free_page((unsigned long)kbuf);
 487         }
 488         *ppos = p;
 489         return read ? read : err;
 490 }
 491
 492
 493 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 494                                 size_t count, loff_t *ppos)
 495 {
 496         ssize_t written, sz;
 497         unsigned long copied;
 498
 499         written = 0;
 500 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 501         /* we don't have page 0 mapped on sparc and m68k.. */
 502         if (p < PAGE_SIZE) {
 503                 sz = size_inside_page(p, count);
 504                 /* Hmm. Do something? */
 505                 buf += sz;
 506                 p += sz;
 507                 count -= sz;
 508                 written += sz;
 509         }
 510 #endif
 511
 512         while (count > 0) {
 513                 char *ptr;
 514
 515                 sz = size_inside_page(p, count);
 516
 517                 /*
 518                  * On ia64 if a page has been mapped somewhere as uncached, then
 519                  * it must also be accessed uncached by the kernel or data
 520                  * corruption may occur.
 521                  */
 522                 ptr = xlate_dev_kmem_ptr((char *)p);
 523
 524                 copied = copy_from_user(ptr, buf, sz);
 525                 if (copied) {
 526                         written += sz - copied;
 527                         if (written)
 528                                 break;
 529                         return -EFAULT;
 530                 }
 531                 buf += sz;
 532                 p += sz;
 533                 count -= sz;
 534                 written += sz;
 535         }
 536
 537         *ppos += written;
 538         return written;
 539 }
 540
 541 /*
 542  * This function writes to the *virtual* memory as seen by the kernel.
 543  */
 544 static ssize_t write_kmem(struct file *file, const char __user *buf,
 545                           size_t count, loff_t *ppos)
 546 {
 547         unsigned long p = *ppos;
 548         ssize_t wrote = 0;
 549         ssize_t virtr = 0;
 550         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 551         int err = 0;
 552
 553         if (p < (unsigned long) high_memory) {
 554                 unsigned long to_write = min_t(unsigned long, count,
 555                                                (unsigned long)high_memory - p);
 556                 wrote = do_write_kmem(p, buf, to_write, ppos);
 557                 if (wrote != to_write)
 558                         return wrote;
 559                 p += wrote;
 560                 buf += wrote;
 561                 count -= wrote;
 562         }
 563
 564         if (count > 0) {
 565                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 566                 if (!kbuf)
 567                         return wrote ? wrote : -ENOMEM;
 568                 while (count > 0) {
 569                         unsigned long sz = size_inside_page(p, count);
 570                         unsigned long n;
 571
 572                         if (!is_vmalloc_or_module_addr((void *)p)) {
 573                                 err = -ENXIO;
 574                                 break;
 575                         }
 576                         n = copy_from_user(kbuf, buf, sz);
 577                         if (n) {
 578                                 err = -EFAULT;
 579                                 break;
 580                         }
 581                         vwrite(kbuf, (char *)p, sz);
 582                         count -= sz;
 583                         buf += sz;
 584                         virtr += sz;
 585                         p += sz;
 586                 }
 587                 free_page((unsigned long)kbuf);
 588         }
 589
 590         *ppos = p;
 591         return virtr + wrote ? : err;
 592 }
 593 #endif
 594
 595 #ifdef CONFIG_DEVPORT
 596 static ssize_t read_port(struct file *file, char __user *buf,
 597                          size_t count, loff_t *ppos)
 598 {
 599         unsigned long i = *ppos;
 600         char __user *tmp = buf;
 601
 602         if (!access_ok(VERIFY_WRITE, buf, count))
 603                 return -EFAULT;
 604         while (count-- > 0 && i < 65536) {
 605                 if (__put_user(inb(i), tmp) < 0)
 606                         return -EFAULT;
 607                 i++;
 608                 tmp++;
 609         }
 610         *ppos = i;
 611         return tmp-buf;
 612 }
 613
 614 static ssize_t write_port(struct file *file, const char __user *buf,
 615                           size_t count, loff_t *ppos)
 616 {
 617         unsigned long i = *ppos;
 618         const char __user * tmp = buf;
 619
 620         if (!access_ok(VERIFY_READ, buf, count))
 621                 return -EFAULT;
 622         while (count-- > 0 && i < 65536) {
 623                 char c;
 624                 if (__get_user(c, tmp)) {
 625                         if (tmp > buf)
 626                                 break;
 627                         return -EFAULT;
 628                 }
 629                 outb(c, i);
 630                 i++;
 631                 tmp++;
 632         }
 633         *ppos = i;
 634         return tmp-buf;
 635 }
 636 #endif
 637
 638 static ssize_t read_null(struct file *file, char __user *buf,
 639                          size_t count, loff_t *ppos)
 640 {
 641         return 0;
 642 }
 643
 644 static ssize_t write_null(struct file *file, const char __user *buf,
 645                           size_t count, loff_t *ppos)
 646 {
 647         return count;
 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 int mmap_zero(struct file *file, struct vm_area_struct *vma)
 694 {
 695 #ifndef CONFIG_MMU
 696         return -ENOSYS;
 697 #endif
 698         if (vma->vm_flags & VM_SHARED)
 699                 return shmem_zero_setup(vma);
 700         return 0;
 701 }
 702
 703 static ssize_t write_full(struct file *file, const char __user *buf,
 704                           size_t count, loff_t *ppos)
 705 {
 706         return -ENOSPC;
 707 }
 708
 709 /*
 710  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 711  * can fopen() both devices with "a" now.  This was previously impossible.
 712  * -- SRB.
 713  */
 714 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 715 {
 716         return file->f_pos = 0;
 717 }
 718
 719 /*
 720  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 721  * check against negative addresses: they are ok. The return value is weird,
 722  * though, in that case (0).
 723  *
 724  * also note that seeking relative to the "end of file" isn't supported:
 725  * it has no meaning, so it returns -EINVAL.
 726  */
 727 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 728 {
 729         loff_t ret;
 730
 731         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 732         switch (orig) {
 733         case SEEK_CUR:
 734                 offset += file->f_pos;
 735         case SEEK_SET:
 736                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 737                 if ((unsigned long long)offset >= ~0xFFFULL) {
 738                         ret = -EOVERFLOW;
 739                         break;
 740                 }
 741                 file->f_pos = offset;
 742                 ret = file->f_pos;
 743                 force_successful_syscall_return();
 744                 break;
 745         default:
 746                 ret = -EINVAL;
 747         }
 748         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 749         return ret;
 750 }
 751
 752 static int open_port(struct inode * inode, struct file * filp)
 753 {
 754         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 755 }
 756
 757 #define zero_lseek      null_lseek
 758 #define full_lseek      null_lseek
 759 #define write_zero      write_null
 760 #define read_full       read_zero
 761 #define open_mem        open_port
 762 #define open_kmem       open_mem
 763 #define open_oldmem     open_mem
 764
 765 static const struct file_operations mem_fops = {
 766         .llseek         = memory_lseek,
 767         .read           = read_mem,
 768         .write          = write_mem,
 769         .mmap           = mmap_mem,
 770         .open           = open_mem,
 771         .get_unmapped_area = get_unmapped_area_mem,
 772 };
 773
 774 #ifdef CONFIG_DEVKMEM
 775 static const struct file_operations kmem_fops = {
 776         .llseek         = memory_lseek,
 777         .read           = read_kmem,
 778         .write          = write_kmem,
 779         .mmap           = mmap_kmem,
 780         .open           = open_kmem,
 781         .get_unmapped_area = get_unmapped_area_mem,
 782 };
 783 #endif
 784
 785 static const struct file_operations null_fops = {
 786         .llseek         = null_lseek,
 787         .read           = read_null,
 788         .write          = write_null,
 789         .splice_write   = splice_write_null,
 790 };
 791
 792 #ifdef CONFIG_DEVPORT
 793 static const struct file_operations port_fops = {
 794         .llseek         = memory_lseek,
 795         .read           = read_port,
 796         .write          = write_port,
 797         .open           = open_port,
 798 };
 799 #endif
 800
 801 static const struct file_operations zero_fops = {
 802         .llseek         = zero_lseek,
 803         .read           = read_zero,
 804         .write          = write_zero,
 805         .mmap           = mmap_zero,
 806 };
 807
 808 /*
 809  * capabilities for /dev/zero
 810  * - permits private mappings, "copies" are taken of the source of zeros
 811  * - no writeback happens
 812  */
 813 static struct backing_dev_info zero_bdi = {
 814         .name           = "char/mem",
 815         .capabilities   = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
 816 };
 817
 818 static const struct file_operations full_fops = {
 819         .llseek         = full_lseek,
 820         .read           = read_full,
 821         .write          = write_full,
 822 };
 823
 824 #ifdef CONFIG_CRASH_DUMP
 825 static const struct file_operations oldmem_fops = {
 826         .read   = read_oldmem,
 827         .open   = open_oldmem,
 828         .llseek = default_llseek,
 829 };
 830 #endif
 831
 832 static ssize_t kmsg_writev(struct kiocb *iocb, const struct iovec *iv,
 833                            unsigned long count, loff_t pos)
 834 {
 835         char *line, *p;
 836         int i;
 837         ssize_t ret = -EFAULT;
 838         size_t len = iov_length(iv, count);
 839
 840         line = kmalloc(len + 1, GFP_KERNEL);
 841         if (line == NULL)
 842                 return -ENOMEM;
 843
 844         /*
 845          * copy all vectors into a single string, to ensure we do
 846          * not interleave our log line with other printk calls
 847          */
 848         p = line;
 849         for (i = 0; i < count; i++) {
 850                 if (copy_from_user(p, iv[i].iov_base, iv[i].iov_len))
 851                         goto out;
 852                 p += iv[i].iov_len;
 853         }
 854         p[0] = '\0';
 855
 856         ret = printk("%s", line);
 857         /* printk can add a prefix */
 858         if (ret > len)
 859                 ret = len;
 860 out:
 861         kfree(line);
 862         return ret;
 863 }
 864
 865 static const struct file_operations kmsg_fops = {
 866         .aio_write = kmsg_writev,
 867         .llseek = noop_llseek,
 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
 899         minor = iminor(inode);
 900         if (minor >= ARRAY_SIZE(devlist))
 901                 return -ENXIO;
 902
 903         dev = &devlist[minor];
 904         if (!dev->fops)
 905                 return -ENXIO;
 906
 907         filp->f_op = dev->fops;
 908         if (dev->dev_info)
 909                 filp->f_mapping->backing_dev_info = dev->dev_info;
 910
 911         /* Is /dev/mem or /dev/kmem ? */
 912         if (dev->dev_info == &directly_mappable_cdev_bdi)
 913                 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
 914
 915         if (dev->fops->open)
 916                 return dev->fops->open(inode, filp);
 917
 918         return 0;
 919 }
 920
 921 static const struct file_operations memory_fops = {
 922         .open = memory_open,
 923         .llseek = noop_llseek,
 924 };
 925
 926 static char *mem_devnode(struct device *dev, mode_t *mode)
 927 {
 928         if (mode && devlist[MINOR(dev->devt)].mode)
 929                 *mode = devlist[MINOR(dev->devt)].mode;
 930         return NULL;
 931 }
 932
 933 static struct class *mem_class;
 934
 935 static int __init chr_dev_init(void)
 936 {
 937         int minor;
 938         int err;
 939
 940         err = bdi_init(&zero_bdi);
 941         if (err)
 942                 return err;
 943
 944         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 945                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 946
 947         mem_class = class_create(THIS_MODULE, "mem");
 948         if (IS_ERR(mem_class))
 949                 return PTR_ERR(mem_class);
 950
 951         mem_class->devnode = mem_devnode;
 952         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 953                 if (!devlist[minor].name)
 954                         continue;
 955                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 956                               NULL, devlist[minor].name);
 957         }
 958
 959         return tty_init();
 960 }
 961
 962 fs_initcall(chr_dev_init);