git.oblomov.eu Git - linux-2.6/blob - 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 /*
  39  * Architectures vary in how they handle caching for addresses
  40  * outside of main memory.
  41  *
  42  */
  43 static inline int uncached_access(struct file *file, unsigned long addr)
  44 {
  45 #if defined(CONFIG_IA64)
  46         /*
  47          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  48          */
  49         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  50 #elif defined(CONFIG_MIPS)
  51         {
  52                 extern int __uncached_access(struct file *file,
  53                                              unsigned long addr);
  54
  55                 return __uncached_access(file, addr);
  56         }
  57 #else
  58         /*
  59          * Accessing memory above the top the kernel knows about or through a file pointer
  60          * that was marked O_SYNC will be done non-cached.
  61          */
  62         if (file->f_flags & O_SYNC)
  63                 return 1;
  64         return addr >= __pa(high_memory);
  65 #endif
  66 }
  67
  68 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  69 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
  70 {
  71         if (addr + count > __pa(high_memory))
  72                 return 0;
  73
  74         return 1;
  75 }
  76
  77 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  78 {
  79         return 1;
  80 }
  81 #endif
  82
  83 #ifdef CONFIG_STRICT_DEVMEM
  84 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  85 {
  86         u64 from = ((u64)pfn) << PAGE_SHIFT;
  87         u64 to = from + size;
  88         u64 cursor = from;
  89
  90         while (cursor < to) {
  91                 if (!devmem_is_allowed(pfn)) {
  92                         printk(KERN_INFO
  93                 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
  94                                 current->comm, from, to);
  95                         return 0;
  96                 }
  97                 cursor += PAGE_SIZE;
  98                 pfn++;
  99         }
 100         return 1;
 101 }
 102 #else
 103 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 104 {
 105         return 1;
 106 }
 107 #endif
 108
 109 void __attribute__((weak)) unxlate_dev_mem_ptr(unsigned long phys, void *addr)
 110 {
 111 }
 112
 113 /*
 114  * This funcion reads the *physical* memory. The f_pos points directly to the
 115  * memory location.
 116  */
 117 static ssize_t read_mem(struct file * file, char __user * buf,
 118                         size_t count, loff_t *ppos)
 119 {
 120         unsigned long p = *ppos;
 121         ssize_t read, sz;
 122         char *ptr;
 123
 124         if (!valid_phys_addr_range(p, count))
 125                 return -EFAULT;
 126         read = 0;
 127 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 128         /* we don't have page 0 mapped on sparc and m68k.. */
 129         if (p < PAGE_SIZE) {
 130                 sz = PAGE_SIZE - p;
 131                 if (sz > count)
 132                         sz = count;
 133                 if (sz > 0) {
 134                         if (clear_user(buf, sz))
 135                                 return -EFAULT;
 136                         buf += sz;
 137                         p += sz;
 138                         count -= sz;
 139                         read += sz;
 140                 }
 141         }
 142 #endif
 143
 144         while (count > 0) {
 145                 /*
 146                  * Handle first page in case it's not aligned
 147                  */
 148                 if (-p & (PAGE_SIZE - 1))
 149                         sz = -p & (PAGE_SIZE - 1);
 150                 else
 151                         sz = PAGE_SIZE;
 152
 153                 sz = min_t(unsigned long, sz, count);
 154
 155                 if (!range_is_allowed(p >> PAGE_SHIFT, count))
 156                         return -EPERM;
 157
 158                 /*
 159                  * On ia64 if a page has been mapped somewhere as
 160                  * uncached, then it must also be accessed uncached
 161                  * by the kernel or data corruption may occur
 162                  */
 163                 ptr = xlate_dev_mem_ptr(p);
 164                 if (!ptr)
 165                         return -EFAULT;
 166
 167                 if (copy_to_user(buf, ptr, sz)) {
 168                         unxlate_dev_mem_ptr(p, ptr);
 169                         return -EFAULT;
 170                 }
 171
 172                 unxlate_dev_mem_ptr(p, ptr);
 173
 174                 buf += sz;
 175                 p += sz;
 176                 count -= sz;
 177                 read += sz;
 178         }
 179
 180         *ppos += read;
 181         return read;
 182 }
 183
 184 static ssize_t write_mem(struct file * file, const char __user * buf,
 185                          size_t count, loff_t *ppos)
 186 {
 187         unsigned long p = *ppos;
 188         ssize_t written, sz;
 189         unsigned long copied;
 190         void *ptr;
 191
 192         if (!valid_phys_addr_range(p, count))
 193                 return -EFAULT;
 194
 195         written = 0;
 196
 197 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 198         /* we don't have page 0 mapped on sparc and m68k.. */
 199         if (p < PAGE_SIZE) {
 200                 unsigned long sz = PAGE_SIZE - p;
 201                 if (sz > count)
 202                         sz = count;
 203                 /* Hmm. Do something? */
 204                 buf += sz;
 205                 p += sz;
 206                 count -= sz;
 207                 written += sz;
 208         }
 209 #endif
 210
 211         while (count > 0) {
 212                 /*
 213                  * Handle first page in case it's not aligned
 214                  */
 215                 if (-p & (PAGE_SIZE - 1))
 216                         sz = -p & (PAGE_SIZE - 1);
 217                 else
 218                         sz = PAGE_SIZE;
 219
 220                 sz = min_t(unsigned long, sz, count);
 221
 222                 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 223                         return -EPERM;
 224
 225                 /*
 226                  * On ia64 if a page has been mapped somewhere as
 227                  * uncached, then it must also be accessed uncached
 228                  * by the kernel or data corruption may occur
 229                  */
 230                 ptr = xlate_dev_mem_ptr(p);
 231                 if (!ptr) {
 232                         if (written)
 233                                 break;
 234                         return -EFAULT;
 235                 }
 236
 237                 copied = copy_from_user(ptr, buf, sz);
 238                 if (copied) {
 239                         written += sz - copied;
 240                         unxlate_dev_mem_ptr(p, ptr);
 241                         if (written)
 242                                 break;
 243                         return -EFAULT;
 244                 }
 245
 246                 unxlate_dev_mem_ptr(p, ptr);
 247
 248                 buf += sz;
 249                 p += sz;
 250                 count -= sz;
 251                 written += sz;
 252         }
 253
 254         *ppos += written;
 255         return written;
 256 }
 257
 258 int __attribute__((weak)) phys_mem_access_prot_allowed(struct file *file,
 259         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 260 {
 261         return 1;
 262 }
 263
 264 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 265 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 266                                      unsigned long size, pgprot_t vma_prot)
 267 {
 268 #ifdef pgprot_noncached
 269         unsigned long offset = pfn << PAGE_SHIFT;
 270
 271         if (uncached_access(file, offset))
 272                 return pgprot_noncached(vma_prot);
 273 #endif
 274         return vma_prot;
 275 }
 276 #endif
 277
 278 #ifndef CONFIG_MMU
 279 static unsigned long get_unmapped_area_mem(struct file *file,
 280                                            unsigned long addr,
 281                                            unsigned long len,
 282                                            unsigned long pgoff,
 283                                            unsigned long flags)
 284 {
 285         if (!valid_mmap_phys_addr_range(pgoff, len))
 286                 return (unsigned long) -EINVAL;
 287         return pgoff << PAGE_SHIFT;
 288 }
 289
 290 /* can't do an in-place private mapping if there's no MMU */
 291 static inline int private_mapping_ok(struct vm_area_struct *vma)
 292 {
 293         return vma->vm_flags & VM_MAYSHARE;
 294 }
 295 #else
 296 #define get_unmapped_area_mem   NULL
 297
 298 static inline int private_mapping_ok(struct vm_area_struct *vma)
 299 {
 300         return 1;
 301 }
 302 #endif
 303
 304 void __attribute__((weak))
 305 map_devmem(unsigned long pfn, unsigned long len, pgprot_t prot)
 306 {
 307         /* nothing. architectures can override. */
 308 }
 309
 310 void __attribute__((weak))
 311 unmap_devmem(unsigned long pfn, unsigned long len, pgprot_t prot)
 312 {
 313         /* nothing. architectures can override. */
 314 }
 315
 316 static void mmap_mem_open(struct vm_area_struct *vma)
 317 {
 318         map_devmem(vma->vm_pgoff,  vma->vm_end - vma->vm_start,
 319                         vma->vm_page_prot);
 320 }
 321
 322 static void mmap_mem_close(struct vm_area_struct *vma)
 323 {
 324         unmap_devmem(vma->vm_pgoff,  vma->vm_end - vma->vm_start,
 325                         vma->vm_page_prot);
 326 }
 327
 328 static struct vm_operations_struct mmap_mem_ops = {
 329         .open  = mmap_mem_open,
 330         .close = mmap_mem_close,
 331 #ifdef CONFIG_HAVE_IOREMAP_PROT
 332         .access = generic_access_phys
 333 #endif
 334 };
 335
 336 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 337 {
 338         size_t size = vma->vm_end - vma->vm_start;
 339
 340         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 341                 return -EINVAL;
 342
 343         if (!private_mapping_ok(vma))
 344                 return -ENOSYS;
 345
 346         if (!range_is_allowed(vma->vm_pgoff, size))
 347                 return -EPERM;
 348
 349         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 350                                                 &vma->vm_page_prot))
 351                 return -EINVAL;
 352
 353         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 354                                                  size,
 355                                                  vma->vm_page_prot);
 356
 357         vma->vm_ops = &mmap_mem_ops;
 358
 359         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
 360         if (remap_pfn_range(vma,
 361                             vma->vm_start,
 362                             vma->vm_pgoff,
 363                             size,
 364                             vma->vm_page_prot)) {
 365                 unmap_devmem(vma->vm_pgoff, size, vma->vm_page_prot);
 366                 return -EAGAIN;
 367         }
 368         return 0;
 369 }
 370
 371 #ifdef CONFIG_DEVKMEM
 372 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
 373 {
 374         unsigned long pfn;
 375
 376         /* Turn a kernel-virtual address into a physical page frame */
 377         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 378
 379         /*
 380          * RED-PEN: on some architectures there is more mapped memory
 381          * than available in mem_map which pfn_valid checks
 382          * for. Perhaps should add a new macro here.
 383          *
 384          * RED-PEN: vmalloc is not supported right now.
 385          */
 386         if (!pfn_valid(pfn))
 387                 return -EIO;
 388
 389         vma->vm_pgoff = pfn;
 390         return mmap_mem(file, vma);
 391 }
 392 #endif
 393
 394 #ifdef CONFIG_CRASH_DUMP
 395 /*
 396  * Read memory corresponding to the old kernel.
 397  */
 398 static ssize_t read_oldmem(struct file *file, char __user *buf,
 399                                 size_t count, loff_t *ppos)
 400 {
 401         unsigned long pfn, offset;
 402         size_t read = 0, csize;
 403         int rc = 0;
 404
 405         while (count) {
 406                 pfn = *ppos / PAGE_SIZE;
 407                 if (pfn > saved_max_pfn)
 408                         return read;
 409
 410                 offset = (unsigned long)(*ppos % PAGE_SIZE);
 411                 if (count > PAGE_SIZE - offset)
 412                         csize = PAGE_SIZE - offset;
 413                 else
 414                         csize = count;
 415
 416                 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
 417                 if (rc < 0)
 418                         return rc;
 419                 buf += csize;
 420                 *ppos += csize;
 421                 read += csize;
 422                 count -= csize;
 423         }
 424         return read;
 425 }
 426 #endif
 427
 428 extern long vread(char *buf, char *addr, unsigned long count);
 429 extern long vwrite(char *buf, char *addr, unsigned long count);
 430
 431 #ifdef CONFIG_DEVKMEM
 432 /*
 433  * This function reads the *virtual* memory as seen by the kernel.
 434  */
 435 static ssize_t read_kmem(struct file *file, char __user *buf,
 436                          size_t count, loff_t *ppos)
 437 {
 438         unsigned long p = *ppos;
 439         ssize_t low_count, read, sz;
 440         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 441
 442         read = 0;
 443         if (p < (unsigned long) high_memory) {
 444                 low_count = count;
 445                 if (count > (unsigned long) high_memory - p)
 446                         low_count = (unsigned long) high_memory - p;
 447
 448 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 449                 /* we don't have page 0 mapped on sparc and m68k.. */
 450                 if (p < PAGE_SIZE && low_count > 0) {
 451                         size_t tmp = PAGE_SIZE - p;
 452                         if (tmp > low_count) tmp = low_count;
 453                         if (clear_user(buf, tmp))
 454                                 return -EFAULT;
 455                         buf += tmp;
 456                         p += tmp;
 457                         read += tmp;
 458                         low_count -= tmp;
 459                         count -= tmp;
 460                 }
 461 #endif
 462                 while (low_count > 0) {
 463                         /*
 464                          * Handle first page in case it's not aligned
 465                          */
 466                         if (-p & (PAGE_SIZE - 1))
 467                                 sz = -p & (PAGE_SIZE - 1);
 468                         else
 469                                 sz = PAGE_SIZE;
 470
 471                         sz = min_t(unsigned long, sz, low_count);
 472
 473                         /*
 474                          * On ia64 if a page has been mapped somewhere as
 475                          * uncached, then it must also be accessed uncached
 476                          * by the kernel or data corruption may occur
 477                          */
 478                         kbuf = xlate_dev_kmem_ptr((char *)p);
 479
 480                         if (copy_to_user(buf, kbuf, sz))
 481                                 return -EFAULT;
 482                         buf += sz;
 483                         p += sz;
 484                         read += sz;
 485                         low_count -= sz;
 486                         count -= sz;
 487                 }
 488         }
 489
 490         if (count > 0) {
 491                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 492                 if (!kbuf)
 493                         return -ENOMEM;
 494                 while (count > 0) {
 495                         int len = count;
 496
 497                         if (len > PAGE_SIZE)
 498                                 len = PAGE_SIZE;
 499                         len = vread(kbuf, (char *)p, len);
 500                         if (!len)
 501                                 break;
 502                         if (copy_to_user(buf, kbuf, len)) {
 503                                 free_page((unsigned long)kbuf);
 504                                 return -EFAULT;
 505                         }
 506                         count -= len;
 507                         buf += len;
 508                         read += len;
 509                         p += len;
 510                 }
 511                 free_page((unsigned long)kbuf);
 512         }
 513         *ppos = p;
 514         return read;
 515 }
 516
 517
 518 static inline ssize_t
 519 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
 520               size_t count, loff_t *ppos)
 521 {
 522         ssize_t written, sz;
 523         unsigned long copied;
 524
 525         written = 0;
 526 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 527         /* we don't have page 0 mapped on sparc and m68k.. */
 528         if (realp < PAGE_SIZE) {
 529                 unsigned long sz = PAGE_SIZE - realp;
 530                 if (sz > count)
 531                         sz = count;
 532                 /* Hmm. Do something? */
 533                 buf += sz;
 534                 p += sz;
 535                 realp += sz;
 536                 count -= sz;
 537                 written += sz;
 538         }
 539 #endif
 540
 541         while (count > 0) {
 542                 char *ptr;
 543                 /*
 544                  * Handle first page in case it's not aligned
 545                  */
 546                 if (-realp & (PAGE_SIZE - 1))
 547                         sz = -realp & (PAGE_SIZE - 1);
 548                 else
 549                         sz = PAGE_SIZE;
 550
 551                 sz = min_t(unsigned long, sz, count);
 552
 553                 /*
 554                  * On ia64 if a page has been mapped somewhere as
 555                  * uncached, then it must also be accessed uncached
 556                  * by the kernel or data corruption may occur
 557                  */
 558                 ptr = xlate_dev_kmem_ptr(p);
 559
 560                 copied = copy_from_user(ptr, buf, sz);
 561                 if (copied) {
 562                         written += sz - copied;
 563                         if (written)
 564                                 break;
 565                         return -EFAULT;
 566                 }
 567                 buf += sz;
 568                 p += sz;
 569                 realp += sz;
 570                 count -= sz;
 571                 written += sz;
 572         }
 573
 574         *ppos += written;
 575         return written;
 576 }
 577
 578
 579 /*
 580  * This function writes to the *virtual* memory as seen by the kernel.
 581  */
 582 static ssize_t write_kmem(struct file * file, const char __user * buf,
 583                           size_t count, loff_t *ppos)
 584 {
 585         unsigned long p = *ppos;
 586         ssize_t wrote = 0;
 587         ssize_t virtr = 0;
 588         ssize_t written;
 589         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 590
 591         if (p < (unsigned long) high_memory) {
 592
 593                 wrote = count;
 594                 if (count > (unsigned long) high_memory - p)
 595                         wrote = (unsigned long) high_memory - p;
 596
 597                 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
 598                 if (written != wrote)
 599                         return written;
 600                 wrote = written;
 601                 p += wrote;
 602                 buf += wrote;
 603                 count -= wrote;
 604         }
 605
 606         if (count > 0) {
 607                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 608                 if (!kbuf)
 609                         return wrote ? wrote : -ENOMEM;
 610                 while (count > 0) {
 611                         int len = count;
 612
 613                         if (len > PAGE_SIZE)
 614                                 len = PAGE_SIZE;
 615                         if (len) {
 616                                 written = copy_from_user(kbuf, buf, len);
 617                                 if (written) {
 618                                         if (wrote + virtr)
 619                                                 break;
 620                                         free_page((unsigned long)kbuf);
 621                                         return -EFAULT;
 622                                 }
 623                         }
 624                         len = vwrite(kbuf, (char *)p, len);
 625                         count -= len;
 626                         buf += len;
 627                         virtr += len;
 628                         p += len;
 629                 }
 630                 free_page((unsigned long)kbuf);
 631         }
 632
 633         *ppos = p;
 634         return virtr + wrote;
 635 }
 636 #endif
 637
 638 #ifdef CONFIG_DEVPORT
 639 static ssize_t read_port(struct file * file, char __user * buf,
 640                          size_t count, loff_t *ppos)
 641 {
 642         unsigned long i = *ppos;
 643         char __user *tmp = buf;
 644
 645         if (!access_ok(VERIFY_WRITE, buf, count))
 646                 return -EFAULT;
 647         while (count-- > 0 && i < 65536) {
 648                 if (__put_user(inb(i),tmp) < 0)
 649                         return -EFAULT;
 650                 i++;
 651                 tmp++;
 652         }
 653         *ppos = i;
 654         return tmp-buf;
 655 }
 656
 657 static ssize_t write_port(struct file * file, const char __user * buf,
 658                           size_t count, loff_t *ppos)
 659 {
 660         unsigned long i = *ppos;
 661         const char __user * tmp = buf;
 662
 663         if (!access_ok(VERIFY_READ,buf,count))
 664                 return -EFAULT;
 665         while (count-- > 0 && i < 65536) {
 666                 char c;
 667                 if (__get_user(c, tmp)) {
 668                         if (tmp > buf)
 669                                 break;
 670                         return -EFAULT;
 671                 }
 672                 outb(c,i);
 673                 i++;
 674                 tmp++;
 675         }
 676         *ppos = i;
 677         return tmp-buf;
 678 }
 679 #endif
 680
 681 static ssize_t read_null(struct file * file, char __user * buf,
 682                          size_t count, loff_t *ppos)
 683 {
 684         return 0;
 685 }
 686
 687 static ssize_t write_null(struct file * file, const char __user * buf,
 688                           size_t count, loff_t *ppos)
 689 {
 690         return count;
 691 }
 692
 693 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 694                         struct splice_desc *sd)
 695 {
 696         return sd->len;
 697 }
 698
 699 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
 700                                  loff_t *ppos, size_t len, unsigned int flags)
 701 {
 702         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 703 }
 704
 705 static ssize_t read_zero(struct file * file, char __user * buf,
 706                          size_t count, loff_t *ppos)
 707 {
 708         size_t written;
 709
 710         if (!count)
 711                 return 0;
 712
 713         if (!access_ok(VERIFY_WRITE, buf, count))
 714                 return -EFAULT;
 715
 716         written = 0;
 717         while (count) {
 718                 unsigned long unwritten;
 719                 size_t chunk = count;
 720
 721                 if (chunk > PAGE_SIZE)
 722                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 723                 unwritten = clear_user(buf, chunk);
 724                 written += chunk - unwritten;
 725                 if (unwritten)
 726                         break;
 727                 buf += chunk;
 728                 count -= chunk;
 729                 cond_resched();
 730         }
 731         return written ? written : -EFAULT;
 732 }
 733
 734 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 735 {
 736 #ifndef CONFIG_MMU
 737         return -ENOSYS;
 738 #endif
 739         if (vma->vm_flags & VM_SHARED)
 740                 return shmem_zero_setup(vma);
 741         return 0;
 742 }
 743
 744 static ssize_t write_full(struct file * file, const char __user * buf,
 745                           size_t count, loff_t *ppos)
 746 {
 747         return -ENOSPC;
 748 }
 749
 750 /*
 751  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 752  * can fopen() both devices with "a" now.  This was previously impossible.
 753  * -- SRB.
 754  */
 755
 756 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 757 {
 758         return file->f_pos = 0;
 759 }
 760
 761 /*
 762  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 763  * check against negative addresses: they are ok. The return value is weird,
 764  * though, in that case (0).
 765  *
 766  * also note that seeking relative to the "end of file" isn't supported:
 767  * it has no meaning, so it returns -EINVAL.
 768  */
 769 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 770 {
 771         loff_t ret;
 772
 773         mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 774         switch (orig) {
 775                 case 0:
 776                         file->f_pos = offset;
 777                         ret = file->f_pos;
 778                         force_successful_syscall_return();
 779                         break;
 780                 case 1:
 781                         file->f_pos += offset;
 782                         ret = file->f_pos;
 783                         force_successful_syscall_return();
 784                         break;
 785                 default:
 786                         ret = -EINVAL;
 787         }
 788         mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 789         return ret;
 790 }
 791
 792 static int open_port(struct inode * inode, struct file * filp)
 793 {
 794         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 795 }
 796
 797 #define zero_lseek      null_lseek
 798 #define full_lseek      null_lseek
 799 #define write_zero      write_null
 800 #define read_full       read_zero
 801 #define open_mem        open_port
 802 #define open_kmem       open_mem
 803 #define open_oldmem     open_mem
 804
 805 static const struct file_operations mem_fops = {
 806         .llseek         = memory_lseek,
 807         .read           = read_mem,
 808         .write          = write_mem,
 809         .mmap           = mmap_mem,
 810         .open           = open_mem,
 811         .get_unmapped_area = get_unmapped_area_mem,
 812 };
 813
 814 #ifdef CONFIG_DEVKMEM
 815 static const struct file_operations kmem_fops = {
 816         .llseek         = memory_lseek,
 817         .read           = read_kmem,
 818         .write          = write_kmem,
 819         .mmap           = mmap_kmem,
 820         .open           = open_kmem,
 821         .get_unmapped_area = get_unmapped_area_mem,
 822 };
 823 #endif
 824
 825 static const struct file_operations null_fops = {
 826         .llseek         = null_lseek,
 827         .read           = read_null,
 828         .write          = write_null,
 829         .splice_write   = splice_write_null,
 830 };
 831
 832 #ifdef CONFIG_DEVPORT
 833 static const struct file_operations port_fops = {
 834         .llseek         = memory_lseek,
 835         .read           = read_port,
 836         .write          = write_port,
 837         .open           = open_port,
 838 };
 839 #endif
 840
 841 static const struct file_operations zero_fops = {
 842         .llseek         = zero_lseek,
 843         .read           = read_zero,
 844         .write          = write_zero,
 845         .mmap           = mmap_zero,
 846 };
 847
 848 /*
 849  * capabilities for /dev/zero
 850  * - permits private mappings, "copies" are taken of the source of zeros
 851  */
 852 static struct backing_dev_info zero_bdi = {
 853         .capabilities   = BDI_CAP_MAP_COPY,
 854 };
 855
 856 static const struct file_operations full_fops = {
 857         .llseek         = full_lseek,
 858         .read           = read_full,
 859         .write          = write_full,
 860 };
 861
 862 #ifdef CONFIG_CRASH_DUMP
 863 static const struct file_operations oldmem_fops = {
 864         .read   = read_oldmem,
 865         .open   = open_oldmem,
 866 };
 867 #endif
 868
 869 static ssize_t kmsg_write(struct file * file, const char __user * buf,
 870                           size_t count, loff_t *ppos)
 871 {
 872         char *tmp;
 873         ssize_t ret;
 874
 875         tmp = kmalloc(count + 1, GFP_KERNEL);
 876         if (tmp == NULL)
 877                 return -ENOMEM;
 878         ret = -EFAULT;
 879         if (!copy_from_user(tmp, buf, count)) {
 880                 tmp[count] = 0;
 881                 ret = printk("%s", tmp);
 882                 if (ret > count)
 883                         /* printk can add a prefix */
 884                         ret = count;
 885         }
 886         kfree(tmp);
 887         return ret;
 888 }
 889
 890 static const struct file_operations kmsg_fops = {
 891         .write =        kmsg_write,
 892 };
 893
 894 static int memory_open(struct inode * inode, struct file * filp)
 895 {
 896         int ret = 0;
 897
 898         lock_kernel();
 899         switch (iminor(inode)) {
 900                 case 1:
 901                         filp->f_op = &mem_fops;
 902                         filp->f_mapping->backing_dev_info =
 903                                 &directly_mappable_cdev_bdi;
 904                         break;
 905 #ifdef CONFIG_DEVKMEM
 906                 case 2:
 907                         filp->f_op = &kmem_fops;
 908                         filp->f_mapping->backing_dev_info =
 909                                 &directly_mappable_cdev_bdi;
 910                         break;
 911 #endif
 912                 case 3:
 913                         filp->f_op = &null_fops;
 914                         break;
 915 #ifdef CONFIG_DEVPORT
 916                 case 4:
 917                         filp->f_op = &port_fops;
 918                         break;
 919 #endif
 920                 case 5:
 921                         filp->f_mapping->backing_dev_info = &zero_bdi;
 922                         filp->f_op = &zero_fops;
 923                         break;
 924                 case 7:
 925                         filp->f_op = &full_fops;
 926                         break;
 927                 case 8:
 928                         filp->f_op = &random_fops;
 929                         break;
 930                 case 9:
 931                         filp->f_op = &urandom_fops;
 932                         break;
 933                 case 11:
 934                         filp->f_op = &kmsg_fops;
 935                         break;
 936 #ifdef CONFIG_CRASH_DUMP
 937                 case 12:
 938                         filp->f_op = &oldmem_fops;
 939                         break;
 940 #endif
 941                 default:
 942                         unlock_kernel();
 943                         return -ENXIO;
 944         }
 945         if (filp->f_op && filp->f_op->open)
 946                 ret = filp->f_op->open(inode,filp);
 947         unlock_kernel();
 948         return ret;
 949 }
 950
 951 static const struct file_operations memory_fops = {
 952         .open           = memory_open,  /* just a selector for the real open */
 953 };
 954
 955 static const struct {
 956         unsigned int            minor;
 957         char                    *name;
 958         umode_t                 mode;
 959         const struct file_operations    *fops;
 960 } devlist[] = { /* list of minor devices */
 961         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 962 #ifdef CONFIG_DEVKMEM
 963         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 964 #endif
 965         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 966 #ifdef CONFIG_DEVPORT
 967         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 968 #endif
 969         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 970         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 971         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 972         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 973         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 974 #ifdef CONFIG_CRASH_DUMP
 975         {12,"oldmem",    S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
 976 #endif
 977 };
 978
 979 static struct class *mem_class;
 980
 981 static int __init chr_dev_init(void)
 982 {
 983         int i;
 984         int err;
 985
 986         err = bdi_init(&zero_bdi);
 987         if (err)
 988                 return err;
 989
 990         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 991                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 992
 993         mem_class = class_create(THIS_MODULE, "mem");
 994         for (i = 0; i < ARRAY_SIZE(devlist); i++)
 995                 device_create(mem_class, NULL,
 996                               MKDEV(MEM_MAJOR, devlist[i].minor), NULL,
 997                               devlist[i].name);
 998
 999         return 0;
1000 }
1001
1002 fs_initcall(chr_dev_init);