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/splice.h>
  26 #include <linux/pfn.h>
  27 #include <linux/export.h>
  28 #include <linux/io.h>
  29 #include <linux/uio.h>
  30
  31 #include <linux/uaccess.h>
  32
  33 #ifdef CONFIG_IA64
  34 # include <linux/efi.h>
  35 #endif
  36
  37 #define DEVPORT_MINOR   4
  38
  39 static inline unsigned long size_inside_page(unsigned long start,
  40                                              unsigned long size)
  41 {
  42         unsigned long sz;
  43
  44         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  45
  46         return min(sz, size);
  47 }
  48
  49 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  50 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
  51 {
  52         return addr + count <= __pa(high_memory);
  53 }
  54
  55 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  56 {
  57         return 1;
  58 }
  59 #endif
  60
  61 #ifdef CONFIG_STRICT_DEVMEM
  62 static inline int page_is_allowed(unsigned long pfn)
  63 {
  64         return devmem_is_allowed(pfn);
  65 }
  66 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  67 {
  68         u64 from = ((u64)pfn) << PAGE_SHIFT;
  69         u64 to = from + size;
  70         u64 cursor = from;
  71
  72         while (cursor < to) {
  73                 if (!devmem_is_allowed(pfn))
  74                         return 0;
  75                 cursor += PAGE_SIZE;
  76                 pfn++;
  77         }
  78         return 1;
  79 }
  80 #else
  81 static inline int page_is_allowed(unsigned long pfn)
  82 {
  83         return 1;
  84 }
  85 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  86 {
  87         return 1;
  88 }
  89 #endif
  90
  91 #ifndef unxlate_dev_mem_ptr
  92 #define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
  93 void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
  94 {
  95 }
  96 #endif
  97
  98 static inline bool should_stop_iteration(void)
  99 {
 100         if (need_resched())
 101                 cond_resched();
 102         return fatal_signal_pending(current);
 103 }
 104
 105 /*
 106  * This funcion reads the *physical* memory. The f_pos points directly to the
 107  * memory location.
 108  */
 109 static ssize_t read_mem(struct file *file, char __user *buf,
 110                         size_t count, loff_t *ppos)
 111 {
 112         phys_addr_t p = *ppos;
 113         ssize_t read, sz;
 114         void *ptr;
 115
 116         if (p != *ppos)
 117                 return 0;
 118
 119         if (!valid_phys_addr_range(p, count))
 120                 return -EFAULT;
 121         read = 0;
 122 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 123         /* we don't have page 0 mapped on sparc and m68k.. */
 124         if (p < PAGE_SIZE) {
 125                 sz = size_inside_page(p, count);
 126                 if (sz > 0) {
 127                         if (clear_user(buf, sz))
 128                                 return -EFAULT;
 129                         buf += sz;
 130                         p += sz;
 131                         count -= sz;
 132                         read += sz;
 133                 }
 134         }
 135 #endif
 136
 137         while (count > 0) {
 138                 unsigned long remaining;
 139                 int allowed;
 140
 141                 sz = size_inside_page(p, count);
 142
 143                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 144                 if (!allowed)
 145                         return -EPERM;
 146                 if (allowed == 2) {
 147                         /* Show zeros for restricted memory. */
 148                         remaining = clear_user(buf, sz);
 149                 } else {
 150                         /*
 151                          * On ia64 if a page has been mapped somewhere as
 152                          * uncached, then it must also be accessed uncached
 153                          * by the kernel or data corruption may occur.
 154                          */
 155                         ptr = xlate_dev_mem_ptr(p);
 156                         if (!ptr)
 157                                 return -EFAULT;
 158
 159                         remaining = copy_to_user(buf, ptr, sz);
 160
 161                         unxlate_dev_mem_ptr(p, ptr);
 162                 }
 163
 164                 if (remaining)
 165                         return -EFAULT;
 166
 167                 buf += sz;
 168                 p += sz;
 169                 count -= sz;
 170                 read += sz;
 171                 if (should_stop_iteration())
 172                         break;
 173         }
 174
 175         *ppos += read;
 176         return read;
 177 }
 178
 179 static ssize_t write_mem(struct file *file, const char __user *buf,
 180                          size_t count, loff_t *ppos)
 181 {
 182         phys_addr_t p = *ppos;
 183         ssize_t written, sz;
 184         unsigned long copied;
 185         void *ptr;
 186
 187         if (p != *ppos)
 188                 return -EFBIG;
 189
 190         if (!valid_phys_addr_range(p, count))
 191                 return -EFAULT;
 192
 193         written = 0;
 194
 195 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 196         /* we don't have page 0 mapped on sparc and m68k.. */
 197         if (p < PAGE_SIZE) {
 198                 sz = size_inside_page(p, count);
 199                 /* Hmm. Do something? */
 200                 buf += sz;
 201                 p += sz;
 202                 count -= sz;
 203                 written += sz;
 204         }
 205 #endif
 206
 207         while (count > 0) {
 208                 int allowed;
 209
 210                 sz = size_inside_page(p, count);
 211
 212                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 213                 if (!allowed)
 214                         return -EPERM;
 215
 216                 /* Skip actual writing when a page is marked as restricted. */
 217                 if (allowed == 1) {
 218                         /*
 219                          * On ia64 if a page has been mapped somewhere as
 220                          * uncached, then it must also be accessed uncached
 221                          * by the kernel or data corruption may occur.
 222                          */
 223                         ptr = xlate_dev_mem_ptr(p);
 224                         if (!ptr) {
 225                                 if (written)
 226                                         break;
 227                                 return -EFAULT;
 228                         }
 229
 230                         copied = copy_from_user(ptr, buf, sz);
 231                         unxlate_dev_mem_ptr(p, ptr);
 232                         if (copied) {
 233                                 written += sz - copied;
 234                                 if (written)
 235                                         break;
 236                                 return -EFAULT;
 237                         }
 238                 }
 239
 240                 buf += sz;
 241                 p += sz;
 242                 count -= sz;
 243                 written += sz;
 244                 if (should_stop_iteration())
 245                         break;
 246         }
 247
 248         *ppos += written;
 249         return written;
 250 }
 251
 252 int __weak phys_mem_access_prot_allowed(struct file *file,
 253         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 254 {
 255         return 1;
 256 }
 257
 258 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 259
 260 /*
 261  * Architectures vary in how they handle caching for addresses
 262  * outside of main memory.
 263  *
 264  */
 265 #ifdef pgprot_noncached
 266 static int uncached_access(struct file *file, phys_addr_t addr)
 267 {
 268 #if defined(CONFIG_IA64)
 269         /*
 270          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 271          * attribute aliases.
 272          */
 273         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 274 #elif defined(CONFIG_MIPS)
 275         {
 276                 extern int __uncached_access(struct file *file,
 277                                              unsigned long addr);
 278
 279                 return __uncached_access(file, addr);
 280         }
 281 #else
 282         /*
 283          * Accessing memory above the top the kernel knows about or through a
 284          * file pointer
 285          * that was marked O_DSYNC will be done non-cached.
 286          */
 287         if (file->f_flags & O_DSYNC)
 288                 return 1;
 289         return addr >= __pa(high_memory);
 290 #endif
 291 }
 292 #endif
 293
 294 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 295                                      unsigned long size, pgprot_t vma_prot)
 296 {
 297 #ifdef pgprot_noncached
 298         phys_addr_t offset = pfn << PAGE_SHIFT;
 299
 300         if (uncached_access(file, offset))
 301                 return pgprot_noncached(vma_prot);
 302 #endif
 303         return vma_prot;
 304 }
 305 #endif
 306
 307 #ifndef CONFIG_MMU
 308 static unsigned long get_unmapped_area_mem(struct file *file,
 309                                            unsigned long addr,
 310                                            unsigned long len,
 311                                            unsigned long pgoff,
 312                                            unsigned long flags)
 313 {
 314         if (!valid_mmap_phys_addr_range(pgoff, len))
 315                 return (unsigned long) -EINVAL;
 316         return pgoff << PAGE_SHIFT;
 317 }
 318
 319 /* permit direct mmap, for read, write or exec */
 320 static unsigned memory_mmap_capabilities(struct file *file)
 321 {
 322         return NOMMU_MAP_DIRECT |
 323                 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
 324 }
 325
 326 static unsigned zero_mmap_capabilities(struct file *file)
 327 {
 328         return NOMMU_MAP_COPY;
 329 }
 330
 331 /* can't do an in-place private mapping if there's no MMU */
 332 static inline int private_mapping_ok(struct vm_area_struct *vma)
 333 {
 334         return vma->vm_flags & VM_MAYSHARE;
 335 }
 336 #else
 337
 338 static inline int private_mapping_ok(struct vm_area_struct *vma)
 339 {
 340         return 1;
 341 }
 342 #endif
 343
 344 static const struct vm_operations_struct mmap_mem_ops = {
 345 #ifdef CONFIG_HAVE_IOREMAP_PROT
 346         .access = generic_access_phys
 347 #endif
 348 };
 349
 350 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 351 {
 352         size_t size = vma->vm_end - vma->vm_start;
 353         phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
 354
 355         /* It's illegal to wrap around the end of the physical address space. */
 356         if (offset + (phys_addr_t)size - 1 < offset)
 357                 return -EINVAL;
 358
 359         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 360                 return -EINVAL;
 361
 362         if (!private_mapping_ok(vma))
 363                 return -ENOSYS;
 364
 365         if (!range_is_allowed(vma->vm_pgoff, size))
 366                 return -EPERM;
 367
 368         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 369                                                 &vma->vm_page_prot))
 370                 return -EINVAL;
 371
 372         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 373                                                  size,
 374                                                  vma->vm_page_prot);
 375
 376         vma->vm_ops = &mmap_mem_ops;
 377
 378         /* Remap-pfn-range will mark the range VM_IO */
 379         if (remap_pfn_range(vma,
 380                             vma->vm_start,
 381                             vma->vm_pgoff,
 382                             size,
 383                             vma->vm_page_prot)) {
 384                 return -EAGAIN;
 385         }
 386         return 0;
 387 }
 388
 389 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 390 {
 391         unsigned long pfn;
 392
 393         /* Turn a kernel-virtual address into a physical page frame */
 394         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 395
 396         /*
 397          * RED-PEN: on some architectures there is more mapped memory than
 398          * available in mem_map which pfn_valid checks for. Perhaps should add a
 399          * new macro here.
 400          *
 401          * RED-PEN: vmalloc is not supported right now.
 402          */
 403         if (!pfn_valid(pfn))
 404                 return -EIO;
 405
 406         vma->vm_pgoff = pfn;
 407         return mmap_mem(file, vma);
 408 }
 409
 410 /*
 411  * This function reads the *virtual* memory as seen by the kernel.
 412  */
 413 static ssize_t read_kmem(struct file *file, char __user *buf,
 414                          size_t count, loff_t *ppos)
 415 {
 416         unsigned long p = *ppos;
 417         ssize_t low_count, read, sz;
 418         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 419         int err = 0;
 420
 421         read = 0;
 422         if (p < (unsigned long) high_memory) {
 423                 low_count = count;
 424                 if (count > (unsigned long)high_memory - p)
 425                         low_count = (unsigned long)high_memory - p;
 426
 427 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 428                 /* we don't have page 0 mapped on sparc and m68k.. */
 429                 if (p < PAGE_SIZE && low_count > 0) {
 430                         sz = size_inside_page(p, low_count);
 431                         if (clear_user(buf, sz))
 432                                 return -EFAULT;
 433                         buf += sz;
 434                         p += sz;
 435                         read += sz;
 436                         low_count -= sz;
 437                         count -= sz;
 438                 }
 439 #endif
 440                 while (low_count > 0) {
 441                         sz = size_inside_page(p, low_count);
 442
 443                         /*
 444                          * On ia64 if a page has been mapped somewhere as
 445                          * uncached, then it must also be accessed uncached
 446                          * by the kernel or data corruption may occur
 447                          */
 448                         kbuf = xlate_dev_kmem_ptr((void *)p);
 449
 450                         if (copy_to_user(buf, kbuf, sz))
 451                                 return -EFAULT;
 452                         buf += sz;
 453                         p += sz;
 454                         read += sz;
 455                         low_count -= sz;
 456                         count -= sz;
 457                         if (should_stop_iteration()) {
 458                                 count = 0;
 459                                 break;
 460                         }
 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                         if (should_stop_iteration())
 486                                 break;
 487                 }
 488                 free_page((unsigned long)kbuf);
 489         }
 490         *ppos = p;
 491         return read ? read : err;
 492 }
 493
 494
 495 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 496                                 size_t count, loff_t *ppos)
 497 {
 498         ssize_t written, sz;
 499         unsigned long copied;
 500
 501         written = 0;
 502 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 503         /* we don't have page 0 mapped on sparc and m68k.. */
 504         if (p < PAGE_SIZE) {
 505                 sz = size_inside_page(p, count);
 506                 /* Hmm. Do something? */
 507                 buf += sz;
 508                 p += sz;
 509                 count -= sz;
 510                 written += sz;
 511         }
 512 #endif
 513
 514         while (count > 0) {
 515                 void *ptr;
 516
 517                 sz = size_inside_page(p, count);
 518
 519                 /*
 520                  * On ia64 if a page has been mapped somewhere as uncached, then
 521                  * it must also be accessed uncached by the kernel or data
 522                  * corruption may occur.
 523                  */
 524                 ptr = xlate_dev_kmem_ptr((void *)p);
 525
 526                 copied = copy_from_user(ptr, buf, sz);
 527                 if (copied) {
 528                         written += sz - copied;
 529                         if (written)
 530                                 break;
 531                         return -EFAULT;
 532                 }
 533                 buf += sz;
 534                 p += sz;
 535                 count -= sz;
 536                 written += sz;
 537                 if (should_stop_iteration())
 538                         break;
 539         }
 540
 541         *ppos += written;
 542         return written;
 543 }
 544
 545 /*
 546  * This function writes to the *virtual* memory as seen by the kernel.
 547  */
 548 static ssize_t write_kmem(struct file *file, const char __user *buf,
 549                           size_t count, loff_t *ppos)
 550 {
 551         unsigned long p = *ppos;
 552         ssize_t wrote = 0;
 553         ssize_t virtr = 0;
 554         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 555         int err = 0;
 556
 557         if (p < (unsigned long) high_memory) {
 558                 unsigned long to_write = min_t(unsigned long, count,
 559                                                (unsigned long)high_memory - p);
 560                 wrote = do_write_kmem(p, buf, to_write, ppos);
 561                 if (wrote != to_write)
 562                         return wrote;
 563                 p += wrote;
 564                 buf += wrote;
 565                 count -= wrote;
 566         }
 567
 568         if (count > 0) {
 569                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 570                 if (!kbuf)
 571                         return wrote ? wrote : -ENOMEM;
 572                 while (count > 0) {
 573                         unsigned long sz = size_inside_page(p, count);
 574                         unsigned long n;
 575
 576                         if (!is_vmalloc_or_module_addr((void *)p)) {
 577                                 err = -ENXIO;
 578                                 break;
 579                         }
 580                         n = copy_from_user(kbuf, buf, sz);
 581                         if (n) {
 582                                 err = -EFAULT;
 583                                 break;
 584                         }
 585                         vwrite(kbuf, (char *)p, sz);
 586                         count -= sz;
 587                         buf += sz;
 588                         virtr += sz;
 589                         p += sz;
 590                         if (should_stop_iteration())
 591                                 break;
 592                 }
 593                 free_page((unsigned long)kbuf);
 594         }
 595
 596         *ppos = p;
 597         return virtr + wrote ? : err;
 598 }
 599
 600 static ssize_t read_port(struct file *file, char __user *buf,
 601                          size_t count, loff_t *ppos)
 602 {
 603         unsigned long i = *ppos;
 604         char __user *tmp = buf;
 605
 606         if (!access_ok(VERIFY_WRITE, buf, count))
 607                 return -EFAULT;
 608         while (count-- > 0 && i < 65536) {
 609                 if (__put_user(inb(i), tmp) < 0)
 610                         return -EFAULT;
 611                 i++;
 612                 tmp++;
 613         }
 614         *ppos = i;
 615         return tmp-buf;
 616 }
 617
 618 static ssize_t write_port(struct file *file, const char __user *buf,
 619                           size_t count, loff_t *ppos)
 620 {
 621         unsigned long i = *ppos;
 622         const char __user *tmp = buf;
 623
 624         if (!access_ok(VERIFY_READ, buf, count))
 625                 return -EFAULT;
 626         while (count-- > 0 && i < 65536) {
 627                 char c;
 628
 629                 if (__get_user(c, tmp)) {
 630                         if (tmp > buf)
 631                                 break;
 632                         return -EFAULT;
 633                 }
 634                 outb(c, i);
 635                 i++;
 636                 tmp++;
 637         }
 638         *ppos = i;
 639         return tmp-buf;
 640 }
 641
 642 static ssize_t read_null(struct file *file, char __user *buf,
 643                          size_t count, loff_t *ppos)
 644 {
 645         return 0;
 646 }
 647
 648 static ssize_t write_null(struct file *file, const char __user *buf,
 649                           size_t count, loff_t *ppos)
 650 {
 651         return count;
 652 }
 653
 654 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
 655 {
 656         return 0;
 657 }
 658
 659 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
 660 {
 661         size_t count = iov_iter_count(from);
 662         iov_iter_advance(from, count);
 663         return count;
 664 }
 665
 666 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 667                         struct splice_desc *sd)
 668 {
 669         return sd->len;
 670 }
 671
 672 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 673                                  loff_t *ppos, size_t len, unsigned int flags)
 674 {
 675         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 676 }
 677
 678 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
 679 {
 680         size_t written = 0;
 681
 682         while (iov_iter_count(iter)) {
 683                 size_t chunk = iov_iter_count(iter), n;
 684
 685                 if (chunk > PAGE_SIZE)
 686                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 687                 n = iov_iter_zero(chunk, iter);
 688                 if (!n && iov_iter_count(iter))
 689                         return written ? written : -EFAULT;
 690                 written += n;
 691                 if (signal_pending(current))
 692                         return written ? written : -ERESTARTSYS;
 693                 cond_resched();
 694         }
 695         return written;
 696 }
 697
 698 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 699 {
 700 #ifndef CONFIG_MMU
 701         return -ENOSYS;
 702 #endif
 703         if (vma->vm_flags & VM_SHARED)
 704                 return shmem_zero_setup(vma);
 705         return 0;
 706 }
 707
 708 static ssize_t write_full(struct file *file, const char __user *buf,
 709                           size_t count, loff_t *ppos)
 710 {
 711         return -ENOSPC;
 712 }
 713
 714 /*
 715  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 716  * can fopen() both devices with "a" now.  This was previously impossible.
 717  * -- SRB.
 718  */
 719 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 720 {
 721         return file->f_pos = 0;
 722 }
 723
 724 /*
 725  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 726  * check against negative addresses: they are ok. The return value is weird,
 727  * though, in that case (0).
 728  *
 729  * also note that seeking relative to the "end of file" isn't supported:
 730  * it has no meaning, so it returns -EINVAL.
 731  */
 732 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 733 {
 734         loff_t ret;
 735
 736         mutex_lock(&file_inode(file)->i_mutex);
 737         switch (orig) {
 738         case SEEK_CUR:
 739                 offset += file->f_pos;
 740         case SEEK_SET:
 741                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 742                 if (IS_ERR_VALUE((unsigned long long)offset)) {
 743                         ret = -EOVERFLOW;
 744                         break;
 745                 }
 746                 file->f_pos = offset;
 747                 ret = file->f_pos;
 748                 force_successful_syscall_return();
 749                 break;
 750         default:
 751                 ret = -EINVAL;
 752         }
 753         mutex_unlock(&file_inode(file)->i_mutex);
 754         return ret;
 755 }
 756
 757 static int open_port(struct inode *inode, struct file *filp)
 758 {
 759         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 760 }
 761
 762 #define zero_lseek      null_lseek
 763 #define full_lseek      null_lseek
 764 #define write_zero      write_null
 765 #define write_iter_zero write_iter_null
 766 #define open_mem        open_port
 767 #define open_kmem       open_mem
 768
 769 static const struct file_operations __maybe_unused mem_fops = {
 770         .llseek         = memory_lseek,
 771         .read           = read_mem,
 772         .write          = write_mem,
 773         .mmap           = mmap_mem,
 774         .open           = open_mem,
 775 #ifndef CONFIG_MMU
 776         .get_unmapped_area = get_unmapped_area_mem,
 777         .mmap_capabilities = memory_mmap_capabilities,
 778 #endif
 779 };
 780
 781 static const struct file_operations __maybe_unused kmem_fops = {
 782         .llseek         = memory_lseek,
 783         .read           = read_kmem,
 784         .write          = write_kmem,
 785         .mmap           = mmap_kmem,
 786         .open           = open_kmem,
 787 #ifndef CONFIG_MMU
 788         .get_unmapped_area = get_unmapped_area_mem,
 789         .mmap_capabilities = memory_mmap_capabilities,
 790 #endif
 791 };
 792
 793 static const struct file_operations null_fops = {
 794         .llseek         = null_lseek,
 795         .read           = read_null,
 796         .write          = write_null,
 797         .read_iter      = read_iter_null,
 798         .write_iter     = write_iter_null,
 799         .splice_write   = splice_write_null,
 800 };
 801
 802 static const struct file_operations __maybe_unused port_fops = {
 803         .llseek         = memory_lseek,
 804         .read           = read_port,
 805         .write          = write_port,
 806         .open           = open_port,
 807 };
 808
 809 static const struct file_operations zero_fops = {
 810         .llseek         = zero_lseek,
 811         .write          = write_zero,
 812         .read_iter      = read_iter_zero,
 813         .write_iter     = write_iter_zero,
 814         .mmap           = mmap_zero,
 815 #ifndef CONFIG_MMU
 816         .mmap_capabilities = zero_mmap_capabilities,
 817 #endif
 818 };
 819
 820 static const struct file_operations full_fops = {
 821         .llseek         = full_lseek,
 822         .read_iter      = read_iter_zero,
 823         .write          = write_full,
 824 };
 825
 826 static const struct memdev {
 827         const char *name;
 828         umode_t mode;
 829         const struct file_operations *fops;
 830         fmode_t fmode;
 831 } devlist[] = {
 832 #ifdef CONFIG_DEVMEM
 833          [1] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
 834 #endif
 835 #ifdef CONFIG_DEVKMEM
 836          [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
 837 #endif
 838          [3] = { "null", 0666, &null_fops, 0 },
 839 #ifdef CONFIG_DEVPORT
 840          [4] = { "port", 0, &port_fops, 0 },
 841 #endif
 842          [5] = { "zero", 0666, &zero_fops, 0 },
 843          [7] = { "full", 0666, &full_fops, 0 },
 844          [8] = { "random", 0666, &random_fops, 0 },
 845          [9] = { "urandom", 0666, &urandom_fops, 0 },
 846 #ifdef CONFIG_PRINTK
 847         [11] = { "kmsg", 0644, &kmsg_fops, 0 },
 848 #endif
 849 };
 850
 851 static int memory_open(struct inode *inode, struct file *filp)
 852 {
 853         int minor;
 854         const struct memdev *dev;
 855
 856         minor = iminor(inode);
 857         if (minor >= ARRAY_SIZE(devlist))
 858                 return -ENXIO;
 859
 860         dev = &devlist[minor];
 861         if (!dev->fops)
 862                 return -ENXIO;
 863
 864         filp->f_op = dev->fops;
 865         filp->f_mode |= dev->fmode;
 866
 867         if (dev->fops->open)
 868                 return dev->fops->open(inode, filp);
 869
 870         return 0;
 871 }
 872
 873 static const struct file_operations memory_fops = {
 874         .open = memory_open,
 875         .llseek = noop_llseek,
 876 };
 877
 878 static char *mem_devnode(struct device *dev, umode_t *mode)
 879 {
 880         if (mode && devlist[MINOR(dev->devt)].mode)
 881                 *mode = devlist[MINOR(dev->devt)].mode;
 882         return NULL;
 883 }
 884
 885 static struct class *mem_class;
 886
 887 static int __init chr_dev_init(void)
 888 {
 889         int minor;
 890
 891         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 892                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 893
 894         mem_class = class_create(THIS_MODULE, "mem");
 895         if (IS_ERR(mem_class))
 896                 return PTR_ERR(mem_class);
 897
 898         mem_class->devnode = mem_devnode;
 899         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 900                 if (!devlist[minor].name)
 901                         continue;
 902
 903                 /*
 904                  * Create /dev/port?
 905                  */
 906                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 907                         continue;
 908
 909                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 910                               NULL, devlist[minor].name);
 911         }
 912
 913         return tty_init();
 914 }
 915
 916 fs_initcall(chr_dev_init);