drivers/char/mem.c

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