drivers/char/mem.c

   1 /*
   2  *  linux/drivers/char/mem.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *
   6  *  Added devfs support.
   7  *    Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
   8  *  Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
   9  */
  10
  11 #include <linux/mm.h>
  12 #include <linux/miscdevice.h>
  13 #include <linux/slab.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/mman.h>
  16 #include <linux/random.h>
  17 #include <linux/init.h>
  18 #include <linux/raw.h>
  19 #include <linux/tty.h>
  20 #include <linux/capability.h>
  21 #include <linux/ptrace.h>
  22 #include <linux/device.h>
  23 #include <linux/highmem.h>
  24 #include <linux/crash_dump.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/splice.h>
  28 #include <linux/pfn.h>
  29 #include <linux/export.h>
  30 #include <linux/io.h>
  31 #include <linux/aio.h>
  32
  33 #include <asm/uaccess.h>
  34
  35 #ifdef CONFIG_IA64
  36 # include <linux/efi.h>
  37 #endif
  38
  39 #define DEVPORT_MINOR   4
  40
  41 static inline unsigned long size_inside_page(unsigned long start,
  42                                              unsigned long size)
  43 {
  44         unsigned long sz;
  45
  46         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  47
  48         return min(sz, size);
  49 }
  50
  51 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  52 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
  53 {
  54         return addr + count <= __pa(high_memory);
  55 }
  56
  57 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  58 {
  59         return 1;
  60 }
  61 #endif
  62
  63 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM)
  64 #ifdef CONFIG_STRICT_DEVMEM
  65 static inline int page_is_allowed(unsigned long pfn)
  66 {
  67         return devmem_is_allowed(pfn);
  68 }
  69 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  70 {
  71         u64 from = ((u64)pfn) << PAGE_SHIFT;
  72         u64 to = from + size;
  73         u64 cursor = from;
  74
  75         while (cursor < to) {
  76                 if (!devmem_is_allowed(pfn))
  77                         return 0;
  78                 cursor += PAGE_SIZE;
  79                 pfn++;
  80         }
  81         return 1;
  82 }
  83 #else
  84 static inline int page_is_allowed(unsigned long pfn)
  85 {
  86         return 1;
  87 }
  88 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  89 {
  90         return 1;
  91 }
  92 #endif
  93 #endif
  94
  95 #ifdef CONFIG_DEVMEM
  96 void __weak unxlate_dev_mem_ptr(unsigned long phys, void *addr)
  97 {
  98 }
  99
 100 /*
 101  * This funcion reads the *physical* memory. The f_pos points directly to the
 102  * memory location.
 103  */
 104 static ssize_t read_mem(struct file *file, char __user *buf,
 105                         size_t count, loff_t *ppos)
 106 {
 107         phys_addr_t p = *ppos;
 108         ssize_t read, sz;
 109         char *ptr;
 110
 111         if (!valid_phys_addr_range(p, count))
 112                 return -EFAULT;
 113         read = 0;
 114 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 115         /* we don't have page 0 mapped on sparc and m68k.. */
 116         if (p < PAGE_SIZE) {
 117                 sz = size_inside_page(p, count);
 118                 if (sz > 0) {
 119                         if (clear_user(buf, sz))
 120                                 return -EFAULT;
 121                         buf += sz;
 122                         p += sz;
 123                         count -= sz;
 124                         read += sz;
 125                 }
 126         }
 127 #endif
 128
 129         while (count > 0) {
 130                 unsigned long remaining;
 131                 int allowed;
 132
 133                 sz = size_inside_page(p, count);
 134
 135                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 136                 if (!allowed)
 137                         return -EPERM;
 138                 if (allowed == 2) {
 139                         /* Show zeros for restricted memory. */
 140                         remaining = clear_user(buf, sz);
 141                 } else {
 142                         /*
 143                          * On ia64 if a page has been mapped somewhere as
 144                          * uncached, then it must also be accessed uncached
 145                          * by the kernel or data corruption may occur.
 146                          */
 147                         ptr = xlate_dev_mem_ptr(p);
 148                         if (!ptr)
 149                                 return -EFAULT;
 150
 151                         remaining = copy_to_user(buf, ptr, sz);
 152
 153                         unxlate_dev_mem_ptr(p, ptr);
 154                 }
 155
 156                 if (remaining)
 157                         return -EFAULT;
 158
 159                 buf += sz;
 160                 p += sz;
 161                 count -= sz;
 162                 read += sz;
 163         }
 164
 165         *ppos += read;
 166         return read;
 167 }
 168
 169 static ssize_t write_mem(struct file *file, const char __user *buf,
 170                          size_t count, loff_t *ppos)
 171 {
 172         phys_addr_t p = *ppos;
 173         ssize_t written, sz;
 174         unsigned long copied;
 175         void *ptr;
 176
 177         if (!valid_phys_addr_range(p, count))
 178                 return -EFAULT;
 179
 180         written = 0;
 181
 182 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 183         /* we don't have page 0 mapped on sparc and m68k.. */
 184         if (p < PAGE_SIZE) {
 185                 sz = size_inside_page(p, count);
 186                 /* Hmm. Do something? */
 187                 buf += sz;
 188                 p += sz;
 189                 count -= sz;
 190                 written += sz;
 191         }
 192 #endif
 193
 194         while (count > 0) {
 195                 int allowed;
 196
 197                 sz = size_inside_page(p, count);
 198
 199                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 200                 if (!allowed)
 201                         return -EPERM;
 202
 203                 /* Skip actual writing when a page is marked as restricted. */
 204                 if (allowed == 1) {
 205                         /*
 206                          * On ia64 if a page has been mapped somewhere as
 207                          * uncached, then it must also be accessed uncached
 208                          * by the kernel or data corruption may occur.
 209                          */
 210                         ptr = xlate_dev_mem_ptr(p);
 211                         if (!ptr) {
 212                                 if (written)
 213                                         break;
 214                                 return -EFAULT;
 215                         }
 216
 217                         copied = copy_from_user(ptr, buf, sz);
 218                         unxlate_dev_mem_ptr(p, ptr);
 219                         if (copied) {
 220                                 written += sz - copied;
 221                                 if (written)
 222                                         break;
 223                                 return -EFAULT;
 224                         }
 225                 }
 226
 227                 buf += sz;
 228                 p += sz;
 229                 count -= sz;
 230                 written += sz;
 231         }
 232
 233         *ppos += written;
 234         return written;
 235 }
 236 #endif  /* CONFIG_DEVMEM */
 237
 238 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM)
 239
 240 int __weak phys_mem_access_prot_allowed(struct file *file,
 241         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 242 {
 243         return 1;
 244 }
 245
 246 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 247
 248 /*
 249  * Architectures vary in how they handle caching for addresses
 250  * outside of main memory.
 251  *
 252  */
 253 #ifdef pgprot_noncached
 254 static int uncached_access(struct file *file, phys_addr_t addr)
 255 {
 256 #if defined(CONFIG_IA64)
 257         /*
 258          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 259          * attribute aliases.
 260          */
 261         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 262 #elif defined(CONFIG_MIPS)
 263         {
 264                 extern int __uncached_access(struct file *file,
 265                                              unsigned long addr);
 266
 267                 return __uncached_access(file, addr);
 268         }
 269 #else
 270         /*
 271          * Accessing memory above the top the kernel knows about or through a
 272          * file pointer
 273          * that was marked O_DSYNC will be done non-cached.
 274          */
 275         if (file->f_flags & O_DSYNC)
 276                 return 1;
 277         return addr >= __pa(high_memory);
 278 #endif
 279 }
 280 #endif
 281
 282 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 283                                      unsigned long size, pgprot_t vma_prot)
 284 {
 285 #ifdef pgprot_noncached
 286         phys_addr_t offset = pfn << PAGE_SHIFT;
 287
 288         if (uncached_access(file, offset))
 289                 return pgprot_noncached(vma_prot);
 290 #endif
 291         return vma_prot;
 292 }
 293 #endif
 294
 295 #ifndef CONFIG_MMU
 296 static unsigned long get_unmapped_area_mem(struct file *file,
 297                                            unsigned long addr,
 298                                            unsigned long len,
 299                                            unsigned long pgoff,
 300                                            unsigned long flags)
 301 {
 302         if (!valid_mmap_phys_addr_range(pgoff, len))
 303                 return (unsigned long) -EINVAL;
 304         return pgoff << PAGE_SHIFT;
 305 }
 306
 307 /* can't do an in-place private mapping if there's no MMU */
 308 static inline int private_mapping_ok(struct vm_area_struct *vma)
 309 {
 310         return vma->vm_flags & VM_MAYSHARE;
 311 }
 312 #else
 313 #define get_unmapped_area_mem   NULL
 314
 315 static inline int private_mapping_ok(struct vm_area_struct *vma)
 316 {
 317         return 1;
 318 }
 319 #endif
 320
 321 static const struct vm_operations_struct mmap_mem_ops = {
 322 #ifdef CONFIG_HAVE_IOREMAP_PROT
 323         .access = generic_access_phys
 324 #endif
 325 };
 326
 327 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 328 {
 329         size_t size = vma->vm_end - vma->vm_start;
 330
 331         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 332                 return -EINVAL;
 333
 334         if (!private_mapping_ok(vma))
 335                 return -ENOSYS;
 336
 337         if (!range_is_allowed(vma->vm_pgoff, size))
 338                 return -EPERM;
 339
 340         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 341                                                 &vma->vm_page_prot))
 342                 return -EINVAL;
 343
 344         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 345                                                  size,
 346                                                  vma->vm_page_prot);
 347
 348         vma->vm_ops = &mmap_mem_ops;
 349
 350         /* Remap-pfn-range will mark the range VM_IO */
 351         if (remap_pfn_range(vma,
 352                             vma->vm_start,
 353                             vma->vm_pgoff,
 354                             size,
 355                             vma->vm_page_prot)) {
 356                 return -EAGAIN;
 357         }
 358         return 0;
 359 }
 360 #endif  /* CONFIG_DEVMEM */
 361
 362 #ifdef CONFIG_DEVKMEM
 363 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 364 {
 365         unsigned long pfn;
 366
 367         /* Turn a kernel-virtual address into a physical page frame */
 368         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 369
 370         /*
 371          * RED-PEN: on some architectures there is more mapped memory than
 372          * available in mem_map which pfn_valid checks for. Perhaps should add a
 373          * new macro here.
 374          *
 375          * RED-PEN: vmalloc is not supported right now.
 376          */
 377         if (!pfn_valid(pfn))
 378                 return -EIO;
 379
 380         vma->vm_pgoff = pfn;
 381         return mmap_mem(file, vma);
 382 }
 383 #endif
 384
 385 #ifdef CONFIG_CRASH_DUMP
 386 /*
 387  * Read memory corresponding to the old kernel.
 388  */
 389 static ssize_t read_oldmem(struct file *file, char __user *buf,
 390                                 size_t count, loff_t *ppos)
 391 {
 392         unsigned long pfn, offset;
 393         size_t read = 0, csize;
 394         int rc = 0;
 395
 396         while (count) {
 397                 pfn = *ppos / PAGE_SIZE;
 398                 if (pfn > saved_max_pfn)
 399                         return read;
 400
 401                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 402                 if (count > PAGE_SIZE - offset)
 403                         csize = PAGE_SIZE - offset;
 404                 else
 405                         csize = count;
 406
 407                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 408                 if (rc < 0)
 409                         return rc;
 410                 buf += csize;
 411                 *ppos += csize;
 412                 read += csize;
 413                 count -= csize;
 414         }
 415         return read;
 416 }
 417 #endif
 418
 419 #ifdef CONFIG_DEVKMEM
 420 /*
 421  * This function reads the *virtual* memory as seen by the kernel.
 422  */
 423 static ssize_t read_kmem(struct file *file, char __user *buf,
 424                          size_t count, loff_t *ppos)
 425 {
 426         unsigned long p = *ppos;
 427         ssize_t low_count, read, sz;
 428         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 429         int err = 0;
 430
 431         read = 0;
 432         if (p < (unsigned long) high_memory) {
 433                 low_count = count;
 434                 if (count > (unsigned long)high_memory - p)
 435                         low_count = (unsigned long)high_memory - p;
 436
 437 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 438                 /* we don't have page 0 mapped on sparc and m68k.. */
 439                 if (p < PAGE_SIZE && low_count > 0) {
 440                         sz = size_inside_page(p, low_count);
 441                         if (clear_user(buf, sz))
 442                                 return -EFAULT;
 443                         buf += sz;
 444                         p += sz;
 445                         read += sz;
 446                         low_count -= sz;
 447                         count -= sz;
 448                 }
 449 #endif
 450                 while (low_count > 0) {
 451                         sz = size_inside_page(p, low_count);
 452
 453                         /*
 454                          * On ia64 if a page has been mapped somewhere as
 455                          * uncached, then it must also be accessed uncached
 456                          * by the kernel or data corruption may occur
 457                          */
 458                         kbuf = xlate_dev_kmem_ptr((char *)p);
 459
 460                         if (copy_to_user(buf, kbuf, sz))
 461                                 return -EFAULT;
 462                         buf += sz;
 463                         p += sz;
 464                         read += sz;
 465                         low_count -= sz;
 466                         count -= sz;
 467                 }
 468         }
 469
 470         if (count > 0) {
 471                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 472                 if (!kbuf)
 473                         return -ENOMEM;
 474                 while (count > 0) {
 475                         sz = size_inside_page(p, count);
 476                         if (!is_vmalloc_or_module_addr((void *)p)) {
 477                                 err = -ENXIO;
 478                                 break;
 479                         }
 480                         sz = vread(kbuf, (char *)p, sz);
 481                         if (!sz)
 482                                 break;
 483                         if (copy_to_user(buf, kbuf, sz)) {
 484                                 err = -EFAULT;
 485                                 break;
 486                         }
 487                         count -= sz;
 488                         buf += sz;
 489                         read += sz;
 490                         p += sz;
 491                 }
 492                 free_page((unsigned long)kbuf);
 493         }
 494         *ppos = p;
 495         return read ? read : err;
 496 }
 497
 498
 499 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 500                                 size_t count, loff_t *ppos)
 501 {
 502         ssize_t written, sz;
 503         unsigned long copied;
 504
 505         written = 0;
 506 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 507         /* we don't have page 0 mapped on sparc and m68k.. */
 508         if (p < PAGE_SIZE) {
 509                 sz = size_inside_page(p, count);
 510                 /* Hmm. Do something? */
 511                 buf += sz;
 512                 p += sz;
 513                 count -= sz;
 514                 written += sz;
 515         }
 516 #endif
 517
 518         while (count > 0) {
 519                 char *ptr;
 520
 521                 sz = size_inside_page(p, count);
 522
 523                 /*
 524                  * On ia64 if a page has been mapped somewhere as uncached, then
 525                  * it must also be accessed uncached by the kernel or data
 526                  * corruption may occur.
 527                  */
 528                 ptr = xlate_dev_kmem_ptr((char *)p);
 529
 530                 copied = copy_from_user(ptr, buf, sz);
 531                 if (copied) {
 532                         written += sz - copied;
 533                         if (written)
 534                                 break;
 535                         return -EFAULT;
 536                 }
 537                 buf += sz;
 538                 p += sz;
 539                 count -= sz;
 540                 written += sz;
 541         }
 542
 543         *ppos += written;
 544         return written;
 545 }
 546
 547 /*
 548  * This function writes to the *virtual* memory as seen by the kernel.
 549  */
 550 static ssize_t write_kmem(struct file *file, const char __user *buf,
 551                           size_t count, loff_t *ppos)
 552 {
 553         unsigned long p = *ppos;
 554         ssize_t wrote = 0;
 555         ssize_t virtr = 0;
 556         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 557         int err = 0;
 558
 559         if (p < (unsigned long) high_memory) {
 560                 unsigned long to_write = min_t(unsigned long, count,
 561                                                (unsigned long)high_memory - p);
 562                 wrote = do_write_kmem(p, buf, to_write, ppos);
 563                 if (wrote != to_write)
 564                         return wrote;
 565                 p += wrote;
 566                 buf += wrote;
 567                 count -= wrote;
 568         }
 569
 570         if (count > 0) {
 571                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 572                 if (!kbuf)
 573                         return wrote ? wrote : -ENOMEM;
 574                 while (count > 0) {
 575                         unsigned long sz = size_inside_page(p, count);
 576                         unsigned long n;
 577
 578                         if (!is_vmalloc_or_module_addr((void *)p)) {
 579                                 err = -ENXIO;
 580                                 break;
 581                         }
 582                         n = copy_from_user(kbuf, buf, sz);
 583                         if (n) {
 584                                 err = -EFAULT;
 585                                 break;
 586                         }
 587                         vwrite(kbuf, (char *)p, sz);
 588                         count -= sz;
 589                         buf += sz;
 590                         virtr += sz;
 591                         p += sz;
 592                 }
 593                 free_page((unsigned long)kbuf);
 594         }
 595
 596         *ppos = p;
 597         return virtr + wrote ? : err;
 598 }
 599 #endif
 600
 601 #ifdef CONFIG_DEVPORT
 602 static ssize_t read_port(struct file *file, char __user *buf,
 603                          size_t count, loff_t *ppos)
 604 {
 605         unsigned long i = *ppos;
 606         char __user *tmp = buf;
 607
 608         if (!access_ok(VERIFY_WRITE, buf, count))
 609                 return -EFAULT;
 610         while (count-- > 0 && i < 65536) {
 611                 if (__put_user(inb(i), tmp) < 0)
 612                         return -EFAULT;
 613                 i++;
 614                 tmp++;
 615         }
 616         *ppos = i;
 617         return tmp-buf;
 618 }
 619
 620 static ssize_t write_port(struct file *file, const char __user *buf,
 621                           size_t count, loff_t *ppos)
 622 {
 623         unsigned long i = *ppos;
 624         const char __user *tmp = buf;
 625
 626         if (!access_ok(VERIFY_READ, buf, count))
 627                 return -EFAULT;
 628         while (count-- > 0 && i < 65536) {
 629                 char c;
 630                 if (__get_user(c, tmp)) {
 631                         if (tmp > buf)
 632                                 break;
 633                         return -EFAULT;
 634                 }
 635                 outb(c, i);
 636                 i++;
 637                 tmp++;
 638         }
 639         *ppos = i;
 640         return tmp-buf;
 641 }
 642 #endif
 643
 644 static ssize_t read_null(struct file *file, char __user *buf,
 645                          size_t count, loff_t *ppos)
 646 {
 647         return 0;
 648 }
 649
 650 static ssize_t write_null(struct file *file, const char __user *buf,
 651                           size_t count, loff_t *ppos)
 652 {
 653         return count;
 654 }
 655
 656 static ssize_t aio_read_null(struct kiocb *iocb, const struct iovec *iov,
 657                              unsigned long nr_segs, loff_t pos)
 658 {
 659         return 0;
 660 }
 661
 662 static ssize_t aio_write_null(struct kiocb *iocb, const struct iovec *iov,
 663                               unsigned long nr_segs, loff_t pos)
 664 {
 665         return iov_length(iov, nr_segs);
 666 }
 667
 668 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 669                         struct splice_desc *sd)
 670 {
 671         return sd->len;
 672 }
 673
 674 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 675                                  loff_t *ppos, size_t len, unsigned int flags)
 676 {
 677         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 678 }
 679
 680 static ssize_t read_zero(struct file *file, char __user *buf,
 681                          size_t count, loff_t *ppos)
 682 {
 683         size_t written;
 684
 685         if (!count)
 686                 return 0;
 687
 688         if (!access_ok(VERIFY_WRITE, buf, count))
 689                 return -EFAULT;
 690
 691         written = 0;
 692         while (count) {
 693                 unsigned long unwritten;
 694                 size_t chunk = count;
 695
 696                 if (chunk > PAGE_SIZE)
 697                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 698                 unwritten = __clear_user(buf, chunk);
 699                 written += chunk - unwritten;
 700                 if (unwritten)
 701                         break;
 702                 if (signal_pending(current))
 703                         return written ? written : -ERESTARTSYS;
 704                 buf += chunk;
 705                 count -= chunk;
 706                 cond_resched();
 707         }
 708         return written ? written : -EFAULT;
 709 }
 710
 711 static ssize_t aio_read_zero(struct kiocb *iocb, const struct iovec *iov,
 712                              unsigned long nr_segs, loff_t pos)
 713 {
 714         size_t written = 0;
 715         unsigned long i;
 716         ssize_t ret;
 717
 718         for (i = 0; i < nr_segs; i++) {
 719                 ret = read_zero(iocb->ki_filp, iov[i].iov_base, iov[i].iov_len,
 720                                 &pos);
 721                 if (ret < 0)
 722                         break;
 723                 written += ret;
 724         }
 725
 726         return written ? written : -EFAULT;
 727 }
 728
 729 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 730 {
 731 #ifndef CONFIG_MMU
 732         return -ENOSYS;
 733 #endif
 734         if (vma->vm_flags & VM_SHARED)
 735                 return shmem_zero_setup(vma);
 736         return 0;
 737 }
 738
 739 static ssize_t write_full(struct file *file, const char __user *buf,
 740                           size_t count, loff_t *ppos)
 741 {
 742         return -ENOSPC;
 743 }
 744
 745 /*
 746  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 747  * can fopen() both devices with "a" now.  This was previously impossible.
 748  * -- SRB.
 749  */
 750 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 751 {
 752         return file->f_pos = 0;
 753 }
 754
 755 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM) || defined(CONFIG_DEVPORT)
 756
 757 /*
 758  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 759  * check against negative addresses: they are ok. The return value is weird,
 760  * though, in that case (0).
 761  *
 762  * also note that seeking relative to the "end of file" isn't supported:
 763  * it has no meaning, so it returns -EINVAL.
 764  */
 765 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 766 {
 767         loff_t ret;
 768
 769         mutex_lock(&file_inode(file)->i_mutex);
 770         switch (orig) {
 771         case SEEK_CUR:
 772                 offset += file->f_pos;
 773         case SEEK_SET:
 774                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 775                 if ((unsigned long long)offset >= ~0xFFFULL) {
 776                         ret = -EOVERFLOW;
 777                         break;
 778                 }
 779                 file->f_pos = offset;
 780                 ret = file->f_pos;
 781                 force_successful_syscall_return();
 782                 break;
 783         default:
 784                 ret = -EINVAL;
 785         }
 786         mutex_unlock(&file_inode(file)->i_mutex);
 787         return ret;
 788 }
 789
 790 #endif
 791
 792 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM) || defined(CONFIG_DEVPORT)
 793 static int open_port(struct inode *inode, struct file *filp)
 794 {
 795         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 796 }
 797 #endif
 798
 799 #define zero_lseek      null_lseek
 800 #define full_lseek      null_lseek
 801 #define write_zero      write_null
 802 #define read_full       read_zero
 803 #define aio_write_zero  aio_write_null
 804 #define open_mem        open_port
 805 #define open_kmem       open_mem
 806 #define open_oldmem     open_mem
 807
 808 #ifdef CONFIG_DEVMEM
 809 static const struct file_operations mem_fops = {
 810         .llseek         = memory_lseek,
 811         .read           = read_mem,
 812         .write          = write_mem,
 813         .mmap           = mmap_mem,
 814         .open           = open_mem,
 815         .get_unmapped_area = get_unmapped_area_mem,
 816 };
 817 #endif
 818
 819 #ifdef CONFIG_DEVKMEM
 820 static const struct file_operations kmem_fops = {
 821         .llseek         = memory_lseek,
 822         .read           = read_kmem,
 823         .write          = write_kmem,
 824         .mmap           = mmap_kmem,
 825         .open           = open_kmem,
 826         .get_unmapped_area = get_unmapped_area_mem,
 827 };
 828 #endif
 829
 830 static const struct file_operations null_fops = {
 831         .llseek         = null_lseek,
 832         .read           = read_null,
 833         .write          = write_null,
 834         .aio_read       = aio_read_null,
 835         .aio_write      = aio_write_null,
 836         .splice_write   = splice_write_null,
 837 };
 838
 839 #ifdef CONFIG_DEVPORT
 840 static const struct file_operations port_fops = {
 841         .llseek         = memory_lseek,
 842         .read           = read_port,
 843         .write          = write_port,
 844         .open           = open_port,
 845 };
 846 #endif
 847
 848 static const struct file_operations zero_fops = {
 849         .llseek         = zero_lseek,
 850         .read           = read_zero,
 851         .write          = write_zero,
 852         .aio_read       = aio_read_zero,
 853         .aio_write      = aio_write_zero,
 854         .mmap           = mmap_zero,
 855 };
 856
 857 /*
 858  * capabilities for /dev/zero
 859  * - permits private mappings, "copies" are taken of the source of zeros
 860  * - no writeback happens
 861  */
 862 static struct backing_dev_info zero_bdi = {
 863         .name           = "char/mem",
 864         .capabilities   = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
 865 };
 866
 867 static const struct file_operations full_fops = {
 868         .llseek         = full_lseek,
 869         .read           = read_full,
 870         .write          = write_full,
 871 };
 872
 873 #ifdef CONFIG_CRASH_DUMP
 874 static const struct file_operations oldmem_fops = {
 875         .read   = read_oldmem,
 876         .open   = open_oldmem,
 877         .llseek = default_llseek,
 878 };
 879 #endif
 880
 881 static const struct memdev {
 882         const char *name;
 883         umode_t mode;
 884         const struct file_operations *fops;
 885         struct backing_dev_info *dev_info;
 886 } devlist[] = {
 887 #ifdef CONFIG_DEVMEM
 888          [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
 889 #endif
 890 #ifdef CONFIG_DEVKMEM
 891          [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
 892 #endif
 893          [3] = { "null", 0666, &null_fops, NULL },
 894 #ifdef CONFIG_DEVPORT
 895          [4] = { "port", 0, &port_fops, NULL },
 896 #endif
 897          [5] = { "zero", 0666, &zero_fops, &zero_bdi },
 898          [7] = { "full", 0666, &full_fops, NULL },
 899          [8] = { "random", 0666, &random_fops, NULL },
 900          [9] = { "urandom", 0666, &urandom_fops, NULL },
 901 #ifdef CONFIG_PRINTK
 902         [11] = { "kmsg", 0644, &kmsg_fops, NULL },
 903 #endif
 904 #ifdef CONFIG_CRASH_DUMP
 905         [12] = { "oldmem", 0, &oldmem_fops, NULL },
 906 #endif
 907 };
 908
 909 static int memory_open(struct inode *inode, struct file *filp)
 910 {
 911         int minor;
 912         const struct memdev *dev;
 913
 914         minor = iminor(inode);
 915         if (minor >= ARRAY_SIZE(devlist))
 916                 return -ENXIO;
 917
 918         dev = &devlist[minor];
 919         if (!dev->fops)
 920                 return -ENXIO;
 921
 922         filp->f_op = dev->fops;
 923         if (dev->dev_info)
 924                 filp->f_mapping->backing_dev_info = dev->dev_info;
 925
 926         /* Is /dev/mem or /dev/kmem ? */
 927         if (dev->dev_info == &directly_mappable_cdev_bdi)
 928                 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
 929
 930         if (dev->fops->open)
 931                 return dev->fops->open(inode, filp);
 932
 933         return 0;
 934 }
 935
 936 static const struct file_operations memory_fops = {
 937         .open = memory_open,
 938         .llseek = noop_llseek,
 939 };
 940
 941 static char *mem_devnode(struct device *dev, umode_t *mode)
 942 {
 943         if (mode && devlist[MINOR(dev->devt)].mode)
 944                 *mode = devlist[MINOR(dev->devt)].mode;
 945         return NULL;
 946 }
 947
 948 static struct class *mem_class;
 949
 950 static int __init chr_dev_init(void)
 951 {
 952         int minor;
 953         int err;
 954
 955         err = bdi_init(&zero_bdi);
 956         if (err)
 957                 return err;
 958
 959         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 960                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 961
 962         mem_class = class_create(THIS_MODULE, "mem");
 963         if (IS_ERR(mem_class))
 964                 return PTR_ERR(mem_class);
 965
 966         mem_class->devnode = mem_devnode;
 967         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 968                 if (!devlist[minor].name)
 969                         continue;
 970
 971                 /*
 972                  * Create /dev/port?
 973                  */
 974                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 975                         continue;
 976
 977                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 978                               NULL, devlist[minor].name);
 979         }
 980
 981         return tty_init();
 982 }
 983
 984 fs_initcall(chr_dev_init);