drivers/char/mem.c

   1 /*
   2  *  linux/drivers/char/mem.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *
   6  *  Added devfs support.
   7  *    Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
   8  *  Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
   9  */
  10
  11 #include <linux/mm.h>
  12 #include <linux/miscdevice.h>
  13 #include <linux/slab.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/mman.h>
  16 #include <linux/random.h>
  17 #include <linux/init.h>
  18 #include <linux/raw.h>
  19 #include <linux/tty.h>
  20 #include <linux/capability.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/device.h>
  23 #include <linux/highmem.h>
  24 #include <linux/crash_dump.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/splice.h>
  28 #include <linux/pfn.h>
  29
  30 #include <asm/uaccess.h>
  31 #include <asm/io.h>
  32
  33 #ifdef CONFIG_IA64
  34 # include <linux/efi.h>
  35 #endif
  36
  37 static inline unsigned long size_inside_page(unsigned long start,
  38                                              unsigned long size)
  39 {
  40         unsigned long sz;
  41
  42         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  43
  44         return min(sz, size);
  45 }
  46
  47 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  48 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  49 {
  50         if (addr + count > __pa(high_memory))
  51                 return 0;
  52
  53         return 1;
  54 }
  55
  56 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  57 {
  58         return 1;
  59 }
  60 #endif
  61
  62 #ifdef CONFIG_STRICT_DEVMEM
  63 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  64 {
  65         u64 from = ((u64)pfn) << PAGE_SHIFT;
  66         u64 to = from + size;
  67         u64 cursor = from;
  68
  69         while (cursor < to) {
  70                 if (!devmem_is_allowed(pfn)) {
  71                         printk(KERN_INFO
  72                 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
  73                                 current->comm, from, to);
  74                         return 0;
  75                 }
  76                 cursor += PAGE_SIZE;
  77                 pfn++;
  78         }
  79         return 1;
  80 }
  81 #else
  82 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  83 {
  84         return 1;
  85 }
  86 #endif
  87
  88 void __weak unxlate_dev_mem_ptr(unsigned long phys, void *addr)
  89 {
  90 }
  91
  92 /*
  93  * This funcion reads the *physical* memory. The f_pos points directly to the
  94  * memory location.
  95  */
  96 static ssize_t read_mem(struct file *file, char __user *buf,
  97                         size_t count, loff_t *ppos)
  98 {
  99         unsigned long p = *ppos;
 100         ssize_t read, sz;
 101         char *ptr;
 102
 103         if (!valid_phys_addr_range(p, count))
 104                 return -EFAULT;
 105         read = 0;
 106 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 107         /* we don't have page 0 mapped on sparc and m68k.. */
 108         if (p < PAGE_SIZE) {
 109                 sz = size_inside_page(p, count);
 110                 if (sz > 0) {
 111                         if (clear_user(buf, sz))
 112                                 return -EFAULT;
 113                         buf += sz;
 114                         p += sz;
 115                         count -= sz;
 116                         read += sz;
 117                 }
 118         }
 119 #endif
 120
 121         while (count > 0) {
 122                 unsigned long remaining;
 123
 124                 sz = size_inside_page(p, count);
 125
 126                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 127                         return -EPERM;
 128
 129                 /*
 130                  * On ia64 if a page has been mapped somewhere as uncached, then
 131                  * it must also be accessed uncached by the kernel or data
 132                  * corruption may occur.
 133                  */
 134                 ptr = xlate_dev_mem_ptr(p);
 135                 if (!ptr)
 136                         return -EFAULT;
 137
 138                 remaining = copy_to_user(buf, ptr, sz);
 139                 unxlate_dev_mem_ptr(p, ptr);
 140                 if (remaining)
 141                         return -EFAULT;
 142
 143                 buf += sz;
 144                 p += sz;
 145                 count -= sz;
 146                 read += sz;
 147         }
 148
 149         *ppos += read;
 150         return read;
 151 }
 152
 153 static ssize_t write_mem(struct file *file, const char __user *buf,
 154                          size_t count, loff_t *ppos)
 155 {
 156         unsigned long p = *ppos;
 157         ssize_t written, sz;
 158         unsigned long copied;
 159         void *ptr;
 160
 161         if (!valid_phys_addr_range(p, count))
 162                 return -EFAULT;
 163
 164         written = 0;
 165
 166 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 167         /* we don't have page 0 mapped on sparc and m68k.. */
 168         if (p < PAGE_SIZE) {
 169                 sz = size_inside_page(p, count);
 170                 /* Hmm. Do something? */
 171                 buf += sz;
 172                 p += sz;
 173                 count -= sz;
 174                 written += sz;
 175         }
 176 #endif
 177
 178         while (count > 0) {
 179                 sz = size_inside_page(p, count);
 180
 181                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 182                         return -EPERM;
 183
 184                 /*
 185                  * On ia64 if a page has been mapped somewhere as uncached, then
 186                  * it must also be accessed uncached by the kernel or data
 187                  * corruption may occur.
 188                  */
 189                 ptr = xlate_dev_mem_ptr(p);
 190                 if (!ptr) {
 191                         if (written)
 192                                 break;
 193                         return -EFAULT;
 194                 }
 195
 196                 copied = copy_from_user(ptr, buf, sz);
 197                 unxlate_dev_mem_ptr(p, ptr);
 198                 if (copied) {
 199                         written += sz - copied;
 200                         if (written)
 201                                 break;
 202                         return -EFAULT;
 203                 }
 204
 205                 buf += sz;
 206                 p += sz;
 207                 count -= sz;
 208                 written += sz;
 209         }
 210
 211         *ppos += written;
 212         return written;
 213 }
 214
 215 int __weak phys_mem_access_prot_allowed(struct file *file,
 216         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 217 {
 218         return 1;
 219 }
 220
 221 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 222
 223 /*
 224  * Architectures vary in how they handle caching for addresses
 225  * outside of main memory.
 226  *
 227  */
 228 static int uncached_access(struct file *file, unsigned long addr)
 229 {
 230 #if defined(CONFIG_IA64)
 231         /*
 232          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 233          * attribute aliases.
 234          */
 235         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 236 #elif defined(CONFIG_MIPS)
 237         {
 238                 extern int __uncached_access(struct file *file,
 239                                              unsigned long addr);
 240
 241                 return __uncached_access(file, addr);
 242         }
 243 #else
 244         /*
 245          * Accessing memory above the top the kernel knows about or through a
 246          * file pointer
 247          * that was marked O_DSYNC will be done non-cached.
 248          */
 249         if (file->f_flags & O_DSYNC)
 250                 return 1;
 251         return addr >= __pa(high_memory);
 252 #endif
 253 }
 254
 255 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 256                                      unsigned long size, pgprot_t vma_prot)
 257 {
 258 #ifdef pgprot_noncached
 259         unsigned long offset = pfn << PAGE_SHIFT;
 260
 261         if (uncached_access(file, offset))
 262                 return pgprot_noncached(vma_prot);
 263 #endif
 264         return vma_prot;
 265 }
 266 #endif
 267
 268 #ifndef CONFIG_MMU
 269 static unsigned long get_unmapped_area_mem(struct file *file,
 270                                            unsigned long addr,
 271                                            unsigned long len,
 272                                            unsigned long pgoff,
 273                                            unsigned long flags)
 274 {
 275         if (!valid_mmap_phys_addr_range(pgoff, len))
 276                 return (unsigned long) -EINVAL;
 277         return pgoff << PAGE_SHIFT;
 278 }
 279
 280 /* can't do an in-place private mapping if there's no MMU */
 281 static inline int private_mapping_ok(struct vm_area_struct *vma)
 282 {
 283         return vma->vm_flags & VM_MAYSHARE;
 284 }
 285 #else
 286 #define get_unmapped_area_mem   NULL
 287
 288 static inline int private_mapping_ok(struct vm_area_struct *vma)
 289 {
 290         return 1;
 291 }
 292 #endif
 293
 294 static const struct vm_operations_struct mmap_mem_ops = {
 295 #ifdef CONFIG_HAVE_IOREMAP_PROT
 296         .access = generic_access_phys
 297 #endif
 298 };
 299
 300 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 301 {
 302         size_t size = vma->vm_end - vma->vm_start;
 303
 304         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 305                 return -EINVAL;
 306
 307         if (!private_mapping_ok(vma))
 308                 return -ENOSYS;
 309
 310         if (!range_is_allowed(vma->vm_pgoff, size))
 311                 return -EPERM;
 312
 313         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 314                                                 &vma->vm_page_prot))
 315                 return -EINVAL;
 316
 317         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 318                                                  size,
 319                                                  vma->vm_page_prot);
 320
 321         vma->vm_ops = &mmap_mem_ops;
 322
 323         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 324         if (remap_pfn_range(vma,
 325                             vma->vm_start,
 326                             vma->vm_pgoff,
 327                             size,
 328                             vma->vm_page_prot)) {
 329                 return -EAGAIN;
 330         }
 331         return 0;
 332 }
 333
 334 #ifdef CONFIG_DEVKMEM
 335 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 336 {
 337         unsigned long pfn;
 338
 339         /* Turn a kernel-virtual address into a physical page frame */
 340         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 341
 342         /*
 343          * RED-PEN: on some architectures there is more mapped memory than
 344          * available in mem_map which pfn_valid checks for. Perhaps should add a
 345          * new macro here.
 346          *
 347          * RED-PEN: vmalloc is not supported right now.
 348          */
 349         if (!pfn_valid(pfn))
 350                 return -EIO;
 351
 352         vma->vm_pgoff = pfn;
 353         return mmap_mem(file, vma);
 354 }
 355 #endif
 356
 357 #ifdef CONFIG_CRASH_DUMP
 358 /*
 359  * Read memory corresponding to the old kernel.
 360  */
 361 static ssize_t read_oldmem(struct file *file, char __user *buf,
 362                                 size_t count, loff_t *ppos)
 363 {
 364         unsigned long pfn, offset;
 365         size_t read = 0, csize;
 366         int rc = 0;
 367
 368         while (count) {
 369                 pfn = *ppos / PAGE_SIZE;
 370                 if (pfn > saved_max_pfn)
 371                         return read;
 372
 373                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 374                 if (count > PAGE_SIZE - offset)
 375                         csize = PAGE_SIZE - offset;
 376                 else
 377                         csize = count;
 378
 379                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 380                 if (rc < 0)
 381                         return rc;
 382                 buf += csize;
 383                 *ppos += csize;
 384                 read += csize;
 385                 count -= csize;
 386         }
 387         return read;
 388 }
 389 #endif
 390
 391 #ifdef CONFIG_DEVKMEM
 392 /*
 393  * This function reads the *virtual* memory as seen by the kernel.
 394  */
 395 static ssize_t read_kmem(struct file *file, char __user *buf,
 396                          size_t count, loff_t *ppos)
 397 {
 398         unsigned long p = *ppos;
 399         ssize_t low_count, read, sz;
 400         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 401         int err = 0;
 402
 403         read = 0;
 404         if (p < (unsigned long) high_memory) {
 405                 low_count = count;
 406                 if (count > (unsigned long)high_memory - p)
 407                         low_count = (unsigned long)high_memory - p;
 408
 409 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 410                 /* we don't have page 0 mapped on sparc and m68k.. */
 411                 if (p < PAGE_SIZE && low_count > 0) {
 412                         sz = size_inside_page(p, low_count);
 413                         if (clear_user(buf, sz))
 414                                 return -EFAULT;
 415                         buf += sz;
 416                         p += sz;
 417                         read += sz;
 418                         low_count -= sz;
 419                         count -= sz;
 420                 }
 421 #endif
 422                 while (low_count > 0) {
 423                         sz = size_inside_page(p, low_count);
 424
 425                         /*
 426                          * On ia64 if a page has been mapped somewhere as
 427                          * uncached, then it must also be accessed uncached
 428                          * by the kernel or data corruption may occur
 429                          */
 430                         kbuf = xlate_dev_kmem_ptr((char *)p);
 431
 432                         if (copy_to_user(buf, kbuf, sz))
 433                                 return -EFAULT;
 434                         buf += sz;
 435                         p += sz;
 436                         read += sz;
 437                         low_count -= sz;
 438                         count -= sz;
 439                 }
 440         }
 441
 442         if (count > 0) {
 443                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 444                 if (!kbuf)
 445                         return -ENOMEM;
 446                 while (count > 0) {
 447                         sz = size_inside_page(p, count);
 448                         if (!is_vmalloc_or_module_addr((void *)p)) {
 449                                 err = -ENXIO;
 450                                 break;
 451                         }
 452                         sz = vread(kbuf, (char *)p, sz);
 453                         if (!sz)
 454                                 break;
 455                         if (copy_to_user(buf, kbuf, sz)) {
 456                                 err = -EFAULT;
 457                                 break;
 458                         }
 459                         count -= sz;
 460                         buf += sz;
 461                         read += sz;
 462                         p += sz;
 463                 }
 464                 free_page((unsigned long)kbuf);
 465         }
 466         *ppos = p;
 467         return read ? read : err;
 468 }
 469
 470
 471 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 472                                 size_t count, loff_t *ppos)
 473 {
 474         ssize_t written, sz;
 475         unsigned long copied;
 476
 477         written = 0;
 478 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 479         /* we don't have page 0 mapped on sparc and m68k.. */
 480         if (p < PAGE_SIZE) {
 481                 sz = size_inside_page(p, count);
 482                 /* Hmm. Do something? */
 483                 buf += sz;
 484                 p += sz;
 485                 count -= sz;
 486                 written += sz;
 487         }
 488 #endif
 489
 490         while (count > 0) {
 491                 char *ptr;
 492
 493                 sz = size_inside_page(p, count);
 494
 495                 /*
 496                  * On ia64 if a page has been mapped somewhere as uncached, then
 497                  * it must also be accessed uncached by the kernel or data
 498                  * corruption may occur.
 499                  */
 500                 ptr = xlate_dev_kmem_ptr((char *)p);
 501
 502                 copied = copy_from_user(ptr, buf, sz);
 503                 if (copied) {
 504                         written += sz - copied;
 505                         if (written)
 506                                 break;
 507                         return -EFAULT;
 508                 }
 509                 buf += sz;
 510                 p += sz;
 511                 count -= sz;
 512                 written += sz;
 513         }
 514
 515         *ppos += written;
 516         return written;
 517 }
 518
 519 /*
 520  * This function writes to the *virtual* memory as seen by the kernel.
 521  */
 522 static ssize_t write_kmem(struct file *file, const char __user *buf,
 523                           size_t count, loff_t *ppos)
 524 {
 525         unsigned long p = *ppos;
 526         ssize_t wrote = 0;
 527         ssize_t virtr = 0;
 528         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 529         int err = 0;
 530
 531         if (p < (unsigned long) high_memory) {
 532                 unsigned long to_write = min_t(unsigned long, count,
 533                                                (unsigned long)high_memory - p);
 534                 wrote = do_write_kmem(p, buf, to_write, ppos);
 535                 if (wrote != to_write)
 536                         return wrote;
 537                 p += wrote;
 538                 buf += wrote;
 539                 count -= wrote;
 540         }
 541
 542         if (count > 0) {
 543                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 544                 if (!kbuf)
 545                         return wrote ? wrote : -ENOMEM;
 546                 while (count > 0) {
 547                         unsigned long sz = size_inside_page(p, count);
 548                         unsigned long n;
 549
 550                         if (!is_vmalloc_or_module_addr((void *)p)) {
 551                                 err = -ENXIO;
 552                                 break;
 553                         }
 554                         n = copy_from_user(kbuf, buf, sz);
 555                         if (n) {
 556                                 err = -EFAULT;
 557                                 break;
 558                         }
 559                         vwrite(kbuf, (char *)p, sz);
 560                         count -= sz;
 561                         buf += sz;
 562                         virtr += sz;
 563                         p += sz;
 564                 }
 565                 free_page((unsigned long)kbuf);
 566         }
 567
 568         *ppos = p;
 569         return virtr + wrote ? : err;
 570 }
 571 #endif
 572
 573 #ifdef CONFIG_DEVPORT
 574 static ssize_t read_port(struct file *file, char __user *buf,
 575                          size_t count, loff_t *ppos)
 576 {
 577         unsigned long i = *ppos;
 578         char __user *tmp = buf;
 579
 580         if (!access_ok(VERIFY_WRITE, buf, count))
 581                 return -EFAULT;
 582         while (count-- > 0 && i < 65536) {
 583                 if (__put_user(inb(i), tmp) < 0)
 584                         return -EFAULT;
 585                 i++;
 586                 tmp++;
 587         }
 588         *ppos = i;
 589         return tmp-buf;
 590 }
 591
 592 static ssize_t write_port(struct file *file, const char __user *buf,
 593                           size_t count, loff_t *ppos)
 594 {
 595         unsigned long i = *ppos;
 596         const char __user * tmp = buf;
 597
 598         if (!access_ok(VERIFY_READ, buf, count))
 599                 return -EFAULT;
 600         while (count-- > 0 && i < 65536) {
 601                 char c;
 602                 if (__get_user(c, tmp)) {
 603                         if (tmp > buf)
 604                                 break;
 605                         return -EFAULT;
 606                 }
 607                 outb(c, i);
 608                 i++;
 609                 tmp++;
 610         }
 611         *ppos = i;
 612         return tmp-buf;
 613 }
 614 #endif
 615
 616 static ssize_t read_null(struct file *file, char __user *buf,
 617                          size_t count, loff_t *ppos)
 618 {
 619         return 0;
 620 }
 621
 622 static ssize_t write_null(struct file *file, const char __user *buf,
 623                           size_t count, loff_t *ppos)
 624 {
 625         return count;
 626 }
 627
 628 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 629                         struct splice_desc *sd)
 630 {
 631         return sd->len;
 632 }
 633
 634 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 635                                  loff_t *ppos, size_t len, unsigned int flags)
 636 {
 637         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 638 }
 639
 640 static ssize_t read_zero(struct file *file, char __user *buf,
 641                          size_t count, loff_t *ppos)
 642 {
 643         size_t written;
 644
 645         if (!count)
 646                 return 0;
 647
 648         if (!access_ok(VERIFY_WRITE, buf, count))
 649                 return -EFAULT;
 650
 651         written = 0;
 652         while (count) {
 653                 unsigned long unwritten;
 654                 size_t chunk = count;
 655
 656                 if (chunk > PAGE_SIZE)
 657                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 658                 unwritten = __clear_user(buf, chunk);
 659                 written += chunk - unwritten;
 660                 if (unwritten)
 661                         break;
 662                 if (signal_pending(current))
 663                         return written ? written : -ERESTARTSYS;
 664                 buf += chunk;
 665                 count -= chunk;
 666                 cond_resched();
 667         }
 668         return written ? written : -EFAULT;
 669 }
 670
 671 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 672 {
 673 #ifndef CONFIG_MMU
 674         return -ENOSYS;
 675 #endif
 676         if (vma->vm_flags & VM_SHARED)
 677                 return shmem_zero_setup(vma);
 678         return 0;
 679 }
 680
 681 static ssize_t write_full(struct file *file, const char __user *buf,
 682                           size_t count, loff_t *ppos)
 683 {
 684         return -ENOSPC;
 685 }
 686
 687 /*
 688  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 689  * can fopen() both devices with "a" now.  This was previously impossible.
 690  * -- SRB.
 691  */
 692 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 693 {
 694         return file->f_pos = 0;
 695 }
 696
 697 /*
 698  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 699  * check against negative addresses: they are ok. The return value is weird,
 700  * though, in that case (0).
 701  *
 702  * also note that seeking relative to the "end of file" isn't supported:
 703  * it has no meaning, so it returns -EINVAL.
 704  */
 705 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 706 {
 707         loff_t ret;
 708
 709         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 710         switch (orig) {
 711         case SEEK_CUR:
 712                 offset += file->f_pos;
 713                 if ((unsigned long long)offset <
 714                     (unsigned long long)file->f_pos) {
 715                         ret = -EOVERFLOW;
 716                         break;
 717                 }
 718         case SEEK_SET:
 719                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 720                 if ((unsigned long long)offset >= ~0xFFFULL) {
 721                         ret = -EOVERFLOW;
 722                         break;
 723                 }
 724                 file->f_pos = offset;
 725                 ret = file->f_pos;
 726                 force_successful_syscall_return();
 727                 break;
 728         default:
 729                 ret = -EINVAL;
 730         }
 731         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 732         return ret;
 733 }
 734
 735 static int open_port(struct inode * inode, struct file * filp)
 736 {
 737         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 738 }
 739
 740 #define zero_lseek      null_lseek
 741 #define full_lseek      null_lseek
 742 #define write_zero      write_null
 743 #define read_full       read_zero
 744 #define open_mem        open_port
 745 #define open_kmem       open_mem
 746 #define open_oldmem     open_mem
 747
 748 static const struct file_operations mem_fops = {
 749         .llseek         = memory_lseek,
 750         .read           = read_mem,
 751         .write          = write_mem,
 752         .mmap           = mmap_mem,
 753         .open           = open_mem,
 754         .get_unmapped_area = get_unmapped_area_mem,
 755 };
 756
 757 #ifdef CONFIG_DEVKMEM
 758 static const struct file_operations kmem_fops = {
 759         .llseek         = memory_lseek,
 760         .read           = read_kmem,
 761         .write          = write_kmem,
 762         .mmap           = mmap_kmem,
 763         .open           = open_kmem,
 764         .get_unmapped_area = get_unmapped_area_mem,
 765 };
 766 #endif
 767
 768 static const struct file_operations null_fops = {
 769         .llseek         = null_lseek,
 770         .read           = read_null,
 771         .write          = write_null,
 772         .splice_write   = splice_write_null,
 773 };
 774
 775 #ifdef CONFIG_DEVPORT
 776 static const struct file_operations port_fops = {
 777         .llseek         = memory_lseek,
 778         .read           = read_port,
 779         .write          = write_port,
 780         .open           = open_port,
 781 };
 782 #endif
 783
 784 static const struct file_operations zero_fops = {
 785         .llseek         = zero_lseek,
 786         .read           = read_zero,
 787         .write          = write_zero,
 788         .mmap           = mmap_zero,
 789 };
 790
 791 /*
 792  * capabilities for /dev/zero
 793  * - permits private mappings, "copies" are taken of the source of zeros
 794  */
 795 static struct backing_dev_info zero_bdi = {
 796         .name           = "char/mem",
 797         .capabilities   = BDI_CAP_MAP_COPY,
 798 };
 799
 800 static const struct file_operations full_fops = {
 801         .llseek         = full_lseek,
 802         .read           = read_full,
 803         .write          = write_full,
 804 };
 805
 806 #ifdef CONFIG_CRASH_DUMP
 807 static const struct file_operations oldmem_fops = {
 808         .read   = read_oldmem,
 809         .open   = open_oldmem,
 810 };
 811 #endif
 812
 813 static ssize_t kmsg_write(struct file *file, const char __user *buf,
 814                           size_t count, loff_t *ppos)
 815 {
 816         char *tmp;
 817         ssize_t ret;
 818
 819         tmp = kmalloc(count + 1, GFP_KERNEL);
 820         if (tmp == NULL)
 821                 return -ENOMEM;
 822         ret = -EFAULT;
 823         if (!copy_from_user(tmp, buf, count)) {
 824                 tmp[count] = 0;
 825                 ret = printk("%s", tmp);
 826                 if (ret > count)
 827                         /* printk can add a prefix */
 828                         ret = count;
 829         }
 830         kfree(tmp);
 831         return ret;
 832 }
 833
 834 static const struct file_operations kmsg_fops = {
 835         .write = kmsg_write,
 836 };
 837
 838 static const struct memdev {
 839         const char *name;
 840         mode_t mode;
 841         const struct file_operations *fops;
 842         struct backing_dev_info *dev_info;
 843 } devlist[] = {
 844          [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
 845 #ifdef CONFIG_DEVKMEM
 846          [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
 847 #endif
 848          [3] = { "null", 0666, &null_fops, NULL },
 849 #ifdef CONFIG_DEVPORT
 850          [4] = { "port", 0, &port_fops, NULL },
 851 #endif
 852          [5] = { "zero", 0666, &zero_fops, &zero_bdi },
 853          [7] = { "full", 0666, &full_fops, NULL },
 854          [8] = { "random", 0666, &random_fops, NULL },
 855          [9] = { "urandom", 0666, &urandom_fops, NULL },
 856         [11] = { "kmsg", 0, &kmsg_fops, NULL },
 857 #ifdef CONFIG_CRASH_DUMP
 858         [12] = { "oldmem", 0, &oldmem_fops, NULL },
 859 #endif
 860 };
 861
 862 static int memory_open(struct inode *inode, struct file *filp)
 863 {
 864         int minor;
 865         const struct memdev *dev;
 866
 867         minor = iminor(inode);
 868         if (minor >= ARRAY_SIZE(devlist))
 869                 return -ENXIO;
 870
 871         dev = &devlist[minor];
 872         if (!dev->fops)
 873                 return -ENXIO;
 874
 875         filp->f_op = dev->fops;
 876         if (dev->dev_info)
 877                 filp->f_mapping->backing_dev_info = dev->dev_info;
 878
 879         if (dev->fops->open)
 880                 return dev->fops->open(inode, filp);
 881
 882         return 0;
 883 }
 884
 885 static const struct file_operations memory_fops = {
 886         .open = memory_open,
 887 };
 888
 889 static char *mem_devnode(struct device *dev, mode_t *mode)
 890 {
 891         if (mode && devlist[MINOR(dev->devt)].mode)
 892                 *mode = devlist[MINOR(dev->devt)].mode;
 893         return NULL;
 894 }
 895
 896 static struct class *mem_class;
 897
 898 static int __init chr_dev_init(void)
 899 {
 900         int minor;
 901         int err;
 902
 903         err = bdi_init(&zero_bdi);
 904         if (err)
 905                 return err;
 906
 907         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 908                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 909
 910         mem_class = class_create(THIS_MODULE, "mem");
 911         mem_class->devnode = mem_devnode;
 912         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 913                 if (!devlist[minor].name)
 914                         continue;
 915                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 916                               NULL, devlist[minor].name);
 917         }
 918
 919         return 0;
 920 }
 921
 922 fs_initcall(chr_dev_init);