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