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