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/backing-dev.h>
  25 #include <linux/shmem_fs.h>
  26 #include <linux/splice.h>
  27 #include <linux/pfn.h>
  28 #include <linux/export.h>
  29 #include <linux/io.h>
  30 #include <linux/uio.h>
  31
  32 #include <linux/uaccess.h>
  33
  34 #ifdef CONFIG_IA64
  35 # include <linux/efi.h>
  36 #endif
  37
  38 #define DEVPORT_MINOR   4
  39
  40 static inline unsigned long size_inside_page(unsigned long start,
  41                                              unsigned long size)
  42 {
  43         unsigned long sz;
  44
  45         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  46
  47         return min(sz, size);
  48 }
  49
  50 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  51 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
  52 {
  53         return addr + count <= __pa(high_memory);
  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                         return 0;
  72                 cursor += PAGE_SIZE;
  73                 pfn++;
  74         }
  75         return 1;
  76 }
  77 #else
  78 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  79 {
  80         return 1;
  81 }
  82 #endif
  83
  84 #ifndef unxlate_dev_mem_ptr
  85 #define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
  86 void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
  87 {
  88 }
  89 #endif
  90
  91 /*
  92  * This funcion reads the *physical* memory. The f_pos points directly to the
  93  * memory location.
  94  */
  95 static ssize_t read_mem(struct file *file, char __user *buf,
  96                         size_t count, loff_t *ppos)
  97 {
  98         phys_addr_t p = *ppos;
  99         ssize_t read, sz;
 100         void *ptr;
 101
 102         if (p != *ppos)
 103                 return 0;
 104
 105         if (!valid_phys_addr_range(p, count))
 106                 return -EFAULT;
 107         read = 0;
 108 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 109         /* we don't have page 0 mapped on sparc and m68k.. */
 110         if (p < PAGE_SIZE) {
 111                 sz = size_inside_page(p, count);
 112                 if (sz > 0) {
 113                         if (clear_user(buf, sz))
 114                                 return -EFAULT;
 115                         buf += sz;
 116                         p += sz;
 117                         count -= sz;
 118                         read += sz;
 119                 }
 120         }
 121 #endif
 122
 123         while (count > 0) {
 124                 unsigned long remaining;
 125
 126                 sz = size_inside_page(p, count);
 127
 128                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 129                         return -EPERM;
 130
 131                 /*
 132                  * On ia64 if a page has been mapped somewhere as uncached, then
 133                  * it must also be accessed uncached by the kernel or data
 134                  * corruption may occur.
 135                  */
 136                 ptr = xlate_dev_mem_ptr(p);
 137                 if (!ptr)
 138                         return -EFAULT;
 139
 140                 remaining = copy_to_user(buf, ptr, sz);
 141                 unxlate_dev_mem_ptr(p, ptr);
 142                 if (remaining)
 143                         return -EFAULT;
 144
 145                 buf += sz;
 146                 p += sz;
 147                 count -= sz;
 148                 read += sz;
 149         }
 150
 151         *ppos += read;
 152         return read;
 153 }
 154
 155 static ssize_t write_mem(struct file *file, const char __user *buf,
 156                          size_t count, loff_t *ppos)
 157 {
 158         phys_addr_t p = *ppos;
 159         ssize_t written, sz;
 160         unsigned long copied;
 161         void *ptr;
 162
 163         if (p != *ppos)
 164                 return -EFBIG;
 165
 166         if (!valid_phys_addr_range(p, count))
 167                 return -EFAULT;
 168
 169         written = 0;
 170
 171 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 172         /* we don't have page 0 mapped on sparc and m68k.. */
 173         if (p < PAGE_SIZE) {
 174                 sz = size_inside_page(p, count);
 175                 /* Hmm. Do something? */
 176                 buf += sz;
 177                 p += sz;
 178                 count -= sz;
 179                 written += sz;
 180         }
 181 #endif
 182
 183         while (count > 0) {
 184                 sz = size_inside_page(p, count);
 185
 186                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 187                         return -EPERM;
 188
 189                 /*
 190                  * On ia64 if a page has been mapped somewhere as uncached, then
 191                  * it must also be accessed uncached by the kernel or data
 192                  * corruption may occur.
 193                  */
 194                 ptr = xlate_dev_mem_ptr(p);
 195                 if (!ptr) {
 196                         if (written)
 197                                 break;
 198                         return -EFAULT;
 199                 }
 200
 201                 copied = copy_from_user(ptr, buf, sz);
 202                 unxlate_dev_mem_ptr(p, ptr);
 203                 if (copied) {
 204                         written += sz - copied;
 205                         if (written)
 206                                 break;
 207                         return -EFAULT;
 208                 }
 209
 210                 buf += sz;
 211                 p += sz;
 212                 count -= sz;
 213                 written += sz;
 214         }
 215
 216         *ppos += written;
 217         return written;
 218 }
 219
 220 int __weak phys_mem_access_prot_allowed(struct file *file,
 221         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 222 {
 223         return 1;
 224 }
 225
 226 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 227
 228 /*
 229  * Architectures vary in how they handle caching for addresses
 230  * outside of main memory.
 231  *
 232  */
 233 #ifdef pgprot_noncached
 234 static int uncached_access(struct file *file, phys_addr_t addr)
 235 {
 236 #if defined(CONFIG_IA64)
 237         /*
 238          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 239          * attribute aliases.
 240          */
 241         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 242 #elif defined(CONFIG_MIPS)
 243         {
 244                 extern int __uncached_access(struct file *file,
 245                                              unsigned long addr);
 246
 247                 return __uncached_access(file, addr);
 248         }
 249 #else
 250         /*
 251          * Accessing memory above the top the kernel knows about or through a
 252          * file pointer
 253          * that was marked O_DSYNC will be done non-cached.
 254          */
 255         if (file->f_flags & O_DSYNC)
 256                 return 1;
 257         return addr >= __pa(high_memory);
 258 #endif
 259 }
 260 #endif
 261
 262 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 263                                      unsigned long size, pgprot_t vma_prot)
 264 {
 265 #ifdef pgprot_noncached
 266         phys_addr_t offset = pfn << PAGE_SHIFT;
 267
 268         if (uncached_access(file, offset))
 269                 return pgprot_noncached(vma_prot);
 270 #endif
 271         return vma_prot;
 272 }
 273 #endif
 274
 275 #ifndef CONFIG_MMU
 276 static unsigned long get_unmapped_area_mem(struct file *file,
 277                                            unsigned long addr,
 278                                            unsigned long len,
 279                                            unsigned long pgoff,
 280                                            unsigned long flags)
 281 {
 282         if (!valid_mmap_phys_addr_range(pgoff, len))
 283                 return (unsigned long) -EINVAL;
 284         return pgoff << PAGE_SHIFT;
 285 }
 286
 287 /* permit direct mmap, for read, write or exec */
 288 static unsigned memory_mmap_capabilities(struct file *file)
 289 {
 290         return NOMMU_MAP_DIRECT |
 291                 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
 292 }
 293
 294 static unsigned zero_mmap_capabilities(struct file *file)
 295 {
 296         return NOMMU_MAP_COPY;
 297 }
 298
 299 /* can't do an in-place private mapping if there's no MMU */
 300 static inline int private_mapping_ok(struct vm_area_struct *vma)
 301 {
 302         return vma->vm_flags & VM_MAYSHARE;
 303 }
 304 #else
 305
 306 static inline int private_mapping_ok(struct vm_area_struct *vma)
 307 {
 308         return 1;
 309 }
 310 #endif
 311
 312 static const struct vm_operations_struct mmap_mem_ops = {
 313 #ifdef CONFIG_HAVE_IOREMAP_PROT
 314         .access = generic_access_phys
 315 #endif
 316 };
 317
 318 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 319 {
 320         size_t size = vma->vm_end - vma->vm_start;
 321
 322         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 323                 return -EINVAL;
 324
 325         if (!private_mapping_ok(vma))
 326                 return -ENOSYS;
 327
 328         if (!range_is_allowed(vma->vm_pgoff, size))
 329                 return -EPERM;
 330
 331         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 332                                                 &vma->vm_page_prot))
 333                 return -EINVAL;
 334
 335         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 336                                                  size,
 337                                                  vma->vm_page_prot);
 338
 339         vma->vm_ops = &mmap_mem_ops;
 340
 341         /* Remap-pfn-range will mark the range VM_IO */
 342         if (remap_pfn_range(vma,
 343                             vma->vm_start,
 344                             vma->vm_pgoff,
 345                             size,
 346                             vma->vm_page_prot)) {
 347                 return -EAGAIN;
 348         }
 349         return 0;
 350 }
 351
 352 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 353 {
 354         unsigned long pfn;
 355
 356         /* Turn a kernel-virtual address into a physical page frame */
 357         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 358
 359         /*
 360          * RED-PEN: on some architectures there is more mapped memory than
 361          * available in mem_map which pfn_valid checks for. Perhaps should add a
 362          * new macro here.
 363          *
 364          * RED-PEN: vmalloc is not supported right now.
 365          */
 366         if (!pfn_valid(pfn))
 367                 return -EIO;
 368
 369         vma->vm_pgoff = pfn;
 370         return mmap_mem(file, vma);
 371 }
 372
 373 /*
 374  * This function reads the *virtual* memory as seen by the kernel.
 375  */
 376 static ssize_t read_kmem(struct file *file, char __user *buf,
 377                          size_t count, loff_t *ppos)
 378 {
 379         unsigned long p = *ppos;
 380         ssize_t low_count, read, sz;
 381         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 382         int err = 0;
 383
 384         if (!pfn_valid(PFN_DOWN(p)))
 385                 return -EIO;
 386
 387         read = 0;
 388         if (p < (unsigned long) high_memory) {
 389                 low_count = count;
 390                 if (count > (unsigned long)high_memory - p)
 391                         low_count = (unsigned long)high_memory - p;
 392
 393 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 394                 /* we don't have page 0 mapped on sparc and m68k.. */
 395                 if (p < PAGE_SIZE && low_count > 0) {
 396                         sz = size_inside_page(p, low_count);
 397                         if (clear_user(buf, sz))
 398                                 return -EFAULT;
 399                         buf += sz;
 400                         p += sz;
 401                         read += sz;
 402                         low_count -= sz;
 403                         count -= sz;
 404                 }
 405 #endif
 406                 while (low_count > 0) {
 407                         sz = size_inside_page(p, low_count);
 408
 409                         /*
 410                          * On ia64 if a page has been mapped somewhere as
 411                          * uncached, then it must also be accessed uncached
 412                          * by the kernel or data corruption may occur
 413                          */
 414                         kbuf = xlate_dev_kmem_ptr((void *)p);
 415
 416                         if (copy_to_user(buf, kbuf, sz))
 417                                 return -EFAULT;
 418                         buf += sz;
 419                         p += sz;
 420                         read += sz;
 421                         low_count -= sz;
 422                         count -= sz;
 423                 }
 424         }
 425
 426         if (count > 0) {
 427                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 428                 if (!kbuf)
 429                         return -ENOMEM;
 430                 while (count > 0) {
 431                         sz = size_inside_page(p, count);
 432                         if (!is_vmalloc_or_module_addr((void *)p)) {
 433                                 err = -ENXIO;
 434                                 break;
 435                         }
 436                         sz = vread(kbuf, (char *)p, sz);
 437                         if (!sz)
 438                                 break;
 439                         if (copy_to_user(buf, kbuf, sz)) {
 440                                 err = -EFAULT;
 441                                 break;
 442                         }
 443                         count -= sz;
 444                         buf += sz;
 445                         read += sz;
 446                         p += sz;
 447                 }
 448                 free_page((unsigned long)kbuf);
 449         }
 450         *ppos = p;
 451         return read ? read : err;
 452 }
 453
 454
 455 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 456                                 size_t count, loff_t *ppos)
 457 {
 458         ssize_t written, sz;
 459         unsigned long copied;
 460
 461         written = 0;
 462 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 463         /* we don't have page 0 mapped on sparc and m68k.. */
 464         if (p < PAGE_SIZE) {
 465                 sz = size_inside_page(p, count);
 466                 /* Hmm. Do something? */
 467                 buf += sz;
 468                 p += sz;
 469                 count -= sz;
 470                 written += sz;
 471         }
 472 #endif
 473
 474         while (count > 0) {
 475                 void *ptr;
 476
 477                 sz = size_inside_page(p, count);
 478
 479                 /*
 480                  * On ia64 if a page has been mapped somewhere as uncached, then
 481                  * it must also be accessed uncached by the kernel or data
 482                  * corruption may occur.
 483                  */
 484                 ptr = xlate_dev_kmem_ptr((void *)p);
 485
 486                 copied = copy_from_user(ptr, buf, sz);
 487                 if (copied) {
 488                         written += sz - copied;
 489                         if (written)
 490                                 break;
 491                         return -EFAULT;
 492                 }
 493                 buf += sz;
 494                 p += sz;
 495                 count -= sz;
 496                 written += sz;
 497         }
 498
 499         *ppos += written;
 500         return written;
 501 }
 502
 503 /*
 504  * This function writes to the *virtual* memory as seen by the kernel.
 505  */
 506 static ssize_t write_kmem(struct file *file, const char __user *buf,
 507                           size_t count, loff_t *ppos)
 508 {
 509         unsigned long p = *ppos;
 510         ssize_t wrote = 0;
 511         ssize_t virtr = 0;
 512         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 513         int err = 0;
 514
 515         if (!pfn_valid(PFN_DOWN(p)))
 516                 return -EIO;
 517
 518         if (p < (unsigned long) high_memory) {
 519                 unsigned long to_write = min_t(unsigned long, count,
 520                                                (unsigned long)high_memory - p);
 521                 wrote = do_write_kmem(p, buf, to_write, ppos);
 522                 if (wrote != to_write)
 523                         return wrote;
 524                 p += wrote;
 525                 buf += wrote;
 526                 count -= wrote;
 527         }
 528
 529         if (count > 0) {
 530                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 531                 if (!kbuf)
 532                         return wrote ? wrote : -ENOMEM;
 533                 while (count > 0) {
 534                         unsigned long sz = size_inside_page(p, count);
 535                         unsigned long n;
 536
 537                         if (!is_vmalloc_or_module_addr((void *)p)) {
 538                                 err = -ENXIO;
 539                                 break;
 540                         }
 541                         n = copy_from_user(kbuf, buf, sz);
 542                         if (n) {
 543                                 err = -EFAULT;
 544                                 break;
 545                         }
 546                         vwrite(kbuf, (char *)p, sz);
 547                         count -= sz;
 548                         buf += sz;
 549                         virtr += sz;
 550                         p += sz;
 551                 }
 552                 free_page((unsigned long)kbuf);
 553         }
 554
 555         *ppos = p;
 556         return virtr + wrote ? : err;
 557 }
 558
 559 static ssize_t read_port(struct file *file, char __user *buf,
 560                          size_t count, loff_t *ppos)
 561 {
 562         unsigned long i = *ppos;
 563         char __user *tmp = buf;
 564
 565         if (!access_ok(VERIFY_WRITE, buf, count))
 566                 return -EFAULT;
 567         while (count-- > 0 && i < 65536) {
 568                 if (__put_user(inb(i), tmp) < 0)
 569                         return -EFAULT;
 570                 i++;
 571                 tmp++;
 572         }
 573         *ppos = i;
 574         return tmp-buf;
 575 }
 576
 577 static ssize_t write_port(struct file *file, const char __user *buf,
 578                           size_t count, loff_t *ppos)
 579 {
 580         unsigned long i = *ppos;
 581         const char __user *tmp = buf;
 582
 583         if (!access_ok(VERIFY_READ, buf, count))
 584                 return -EFAULT;
 585         while (count-- > 0 && i < 65536) {
 586                 char c;
 587
 588                 if (__get_user(c, tmp)) {
 589                         if (tmp > buf)
 590                                 break;
 591                         return -EFAULT;
 592                 }
 593                 outb(c, i);
 594                 i++;
 595                 tmp++;
 596         }
 597         *ppos = i;
 598         return tmp-buf;
 599 }
 600
 601 static ssize_t read_null(struct file *file, char __user *buf,
 602                          size_t count, loff_t *ppos)
 603 {
 604         return 0;
 605 }
 606
 607 static ssize_t write_null(struct file *file, const char __user *buf,
 608                           size_t count, loff_t *ppos)
 609 {
 610         return count;
 611 }
 612
 613 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
 614 {
 615         return 0;
 616 }
 617
 618 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
 619 {
 620         size_t count = iov_iter_count(from);
 621         iov_iter_advance(from, count);
 622         return count;
 623 }
 624
 625 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 626                         struct splice_desc *sd)
 627 {
 628         return sd->len;
 629 }
 630
 631 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 632                                  loff_t *ppos, size_t len, unsigned int flags)
 633 {
 634         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 635 }
 636
 637 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
 638 {
 639         size_t written = 0;
 640
 641         while (iov_iter_count(iter)) {
 642                 size_t chunk = iov_iter_count(iter), n;
 643
 644                 if (chunk > PAGE_SIZE)
 645                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 646                 n = iov_iter_zero(chunk, iter);
 647                 if (!n && iov_iter_count(iter))
 648                         return written ? written : -EFAULT;
 649                 written += n;
 650                 if (signal_pending(current))
 651                         return written ? written : -ERESTARTSYS;
 652                 cond_resched();
 653         }
 654         return written;
 655 }
 656
 657 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 658 {
 659 #ifndef CONFIG_MMU
 660         return -ENOSYS;
 661 #endif
 662         if (vma->vm_flags & VM_SHARED)
 663                 return shmem_zero_setup(vma);
 664         return 0;
 665 }
 666
 667 static unsigned long get_unmapped_area_zero(struct file *file,
 668                                 unsigned long addr, unsigned long len,
 669                                 unsigned long pgoff, unsigned long flags)
 670 {
 671 #ifdef CONFIG_MMU
 672         if (flags & MAP_SHARED) {
 673                 /*
 674                  * mmap_zero() will call shmem_zero_setup() to create a file,
 675                  * so use shmem's get_unmapped_area in case it can be huge;
 676                  * and pass NULL for file as in mmap.c's get_unmapped_area(),
 677                  * so as not to confuse shmem with our handle on "/dev/zero".
 678                  */
 679                 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
 680         }
 681
 682         /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
 683         return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
 684 #else
 685         return -ENOSYS;
 686 #endif
 687 }
 688
 689 static ssize_t write_full(struct file *file, const char __user *buf,
 690                           size_t count, loff_t *ppos)
 691 {
 692         return -ENOSPC;
 693 }
 694
 695 /*
 696  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 697  * can fopen() both devices with "a" now.  This was previously impossible.
 698  * -- SRB.
 699  */
 700 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 701 {
 702         return file->f_pos = 0;
 703 }
 704
 705 /*
 706  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 707  * check against negative addresses: they are ok. The return value is weird,
 708  * though, in that case (0).
 709  *
 710  * also note that seeking relative to the "end of file" isn't supported:
 711  * it has no meaning, so it returns -EINVAL.
 712  */
 713 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 714 {
 715         loff_t ret;
 716
 717         inode_lock(file_inode(file));
 718         switch (orig) {
 719         case SEEK_CUR:
 720                 offset += file->f_pos;
 721         case SEEK_SET:
 722                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 723                 if ((unsigned long long)offset >= -MAX_ERRNO) {
 724                         ret = -EOVERFLOW;
 725                         break;
 726                 }
 727                 file->f_pos = offset;
 728                 ret = file->f_pos;
 729                 force_successful_syscall_return();
 730                 break;
 731         default:
 732                 ret = -EINVAL;
 733         }
 734         inode_unlock(file_inode(file));
 735         return ret;
 736 }
 737
 738 static int open_port(struct inode *inode, struct file *filp)
 739 {
 740         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 741 }
 742
 743 #define zero_lseek      null_lseek
 744 #define full_lseek      null_lseek
 745 #define write_zero      write_null
 746 #define write_iter_zero write_iter_null
 747 #define open_mem        open_port
 748 #define open_kmem       open_mem
 749
 750 static const struct file_operations __maybe_unused mem_fops = {
 751         .llseek         = memory_lseek,
 752         .read           = read_mem,
 753         .write          = write_mem,
 754         .mmap           = mmap_mem,
 755         .open           = open_mem,
 756 #ifndef CONFIG_MMU
 757         .get_unmapped_area = get_unmapped_area_mem,
 758         .mmap_capabilities = memory_mmap_capabilities,
 759 #endif
 760 };
 761
 762 static const struct file_operations __maybe_unused kmem_fops = {
 763         .llseek         = memory_lseek,
 764         .read           = read_kmem,
 765         .write          = write_kmem,
 766         .mmap           = mmap_kmem,
 767         .open           = open_kmem,
 768 #ifndef CONFIG_MMU
 769         .get_unmapped_area = get_unmapped_area_mem,
 770         .mmap_capabilities = memory_mmap_capabilities,
 771 #endif
 772 };
 773
 774 static const struct file_operations null_fops = {
 775         .llseek         = null_lseek,
 776         .read           = read_null,
 777         .write          = write_null,
 778         .read_iter      = read_iter_null,
 779         .write_iter     = write_iter_null,
 780         .splice_write   = splice_write_null,
 781 };
 782
 783 static const struct file_operations __maybe_unused port_fops = {
 784         .llseek         = memory_lseek,
 785         .read           = read_port,
 786         .write          = write_port,
 787         .open           = open_port,
 788 };
 789
 790 static const struct file_operations zero_fops = {
 791         .llseek         = zero_lseek,
 792         .write          = write_zero,
 793         .read_iter      = read_iter_zero,
 794         .write_iter     = write_iter_zero,
 795         .mmap           = mmap_zero,
 796         .get_unmapped_area = get_unmapped_area_zero,
 797 #ifndef CONFIG_MMU
 798         .mmap_capabilities = zero_mmap_capabilities,
 799 #endif
 800 };
 801
 802 static const struct file_operations full_fops = {
 803         .llseek         = full_lseek,
 804         .read_iter      = read_iter_zero,
 805         .write          = write_full,
 806 };
 807
 808 static const struct memdev {
 809         const char *name;
 810         umode_t mode;
 811         const struct file_operations *fops;
 812         fmode_t fmode;
 813 } devlist[] = {
 814 #ifdef CONFIG_DEVMEM
 815          [1] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
 816 #endif
 817 #ifdef CONFIG_DEVKMEM
 818          [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
 819 #endif
 820          [3] = { "null", 0666, &null_fops, 0 },
 821 #ifdef CONFIG_DEVPORT
 822          [4] = { "port", 0, &port_fops, 0 },
 823 #endif
 824          [5] = { "zero", 0666, &zero_fops, 0 },
 825          [7] = { "full", 0666, &full_fops, 0 },
 826          [8] = { "random", 0666, &random_fops, 0 },
 827          [9] = { "urandom", 0666, &urandom_fops, 0 },
 828 #ifdef CONFIG_PRINTK
 829         [11] = { "kmsg", 0644, &kmsg_fops, 0 },
 830 #endif
 831 };
 832
 833 static int memory_open(struct inode *inode, struct file *filp)
 834 {
 835         int minor;
 836         const struct memdev *dev;
 837
 838         minor = iminor(inode);
 839         if (minor >= ARRAY_SIZE(devlist))
 840                 return -ENXIO;
 841
 842         dev = &devlist[minor];
 843         if (!dev->fops)
 844                 return -ENXIO;
 845
 846         filp->f_op = dev->fops;
 847         filp->f_mode |= dev->fmode;
 848
 849         if (dev->fops->open)
 850                 return dev->fops->open(inode, filp);
 851
 852         return 0;
 853 }
 854
 855 static const struct file_operations memory_fops = {
 856         .open = memory_open,
 857         .llseek = noop_llseek,
 858 };
 859
 860 static char *mem_devnode(struct device *dev, umode_t *mode)
 861 {
 862         if (mode && devlist[MINOR(dev->devt)].mode)
 863                 *mode = devlist[MINOR(dev->devt)].mode;
 864         return NULL;
 865 }
 866
 867 static struct class *mem_class;
 868
 869 static int __init chr_dev_init(void)
 870 {
 871         int minor;
 872
 873         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 874                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 875
 876         mem_class = class_create(THIS_MODULE, "mem");
 877         if (IS_ERR(mem_class))
 878                 return PTR_ERR(mem_class);
 879
 880         mem_class->devnode = mem_devnode;
 881         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 882                 if (!devlist[minor].name)
 883                         continue;
 884
 885                 /*
 886                  * Create /dev/port?
 887                  */
 888                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 889                         continue;
 890
 891                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 892                               NULL, devlist[minor].name);
 893         }
 894
 895         return tty_init();
 896 }
 897
 898 fs_initcall(chr_dev_init);