drivers/char/mem.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/drivers/char/mem.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  *
   7  *  Added devfs support.
   8  *    Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
   9  *  Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
  10  */
  11
  12 #include <linux/mm.h>
  13 #include <linux/miscdevice.h>
  14 #include <linux/slab.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/mman.h>
  17 #include <linux/random.h>
  18 #include <linux/init.h>
  19 #include <linux/raw.h>
  20 #include <linux/tty.h>
  21 #include <linux/capability.h>
  22 #include <linux/ptrace.h>
  23 #include <linux/device.h>
  24 #include <linux/highmem.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/shmem_fs.h>
  27 #include <linux/splice.h>
  28 #include <linux/pfn.h>
  29 #include <linux/export.h>
  30 #include <linux/io.h>
  31 #include <linux/uio.h>
  32
  33 #include <linux/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 page_is_allowed(unsigned long pfn)
  65 {
  66         return devmem_is_allowed(pfn);
  67 }
  68 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  69 {
  70         u64 from = ((u64)pfn) << PAGE_SHIFT;
  71         u64 to = from + size;
  72         u64 cursor = from;
  73
  74         while (cursor < to) {
  75                 if (!devmem_is_allowed(pfn))
  76                         return 0;
  77                 cursor += PAGE_SIZE;
  78                 pfn++;
  79         }
  80         return 1;
  81 }
  82 #else
  83 static inline int page_is_allowed(unsigned long pfn)
  84 {
  85         return 1;
  86 }
  87 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  88 {
  89         return 1;
  90 }
  91 #endif
  92
  93 #ifndef unxlate_dev_mem_ptr
  94 #define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
  95 void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
  96 {
  97 }
  98 #endif
  99
 100 /*
 101  * This funcion reads the *physical* memory. The f_pos points directly to the
 102  * memory location.
 103  */
 104 static ssize_t read_mem(struct file *file, char __user *buf,
 105                         size_t count, loff_t *ppos)
 106 {
 107         phys_addr_t p = *ppos;
 108         ssize_t read, sz;
 109         void *ptr;
 110
 111         if (p != *ppos)
 112                 return 0;
 113
 114         if (!valid_phys_addr_range(p, count))
 115                 return -EFAULT;
 116         read = 0;
 117 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 118         /* we don't have page 0 mapped on sparc and m68k.. */
 119         if (p < PAGE_SIZE) {
 120                 sz = size_inside_page(p, count);
 121                 if (sz > 0) {
 122                         if (clear_user(buf, sz))
 123                                 return -EFAULT;
 124                         buf += sz;
 125                         p += sz;
 126                         count -= sz;
 127                         read += sz;
 128                 }
 129         }
 130 #endif
 131
 132         while (count > 0) {
 133                 unsigned long remaining;
 134                 int allowed;
 135
 136                 sz = size_inside_page(p, count);
 137
 138                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 139                 if (!allowed)
 140                         return -EPERM;
 141                 if (allowed == 2) {
 142                         /* Show zeros for restricted memory. */
 143                         remaining = clear_user(buf, sz);
 144                 } else {
 145                         /*
 146                          * On ia64 if a page has been mapped somewhere as
 147                          * uncached, then it must also be accessed uncached
 148                          * by the kernel or data corruption may occur.
 149                          */
 150                         ptr = xlate_dev_mem_ptr(p);
 151                         if (!ptr)
 152                                 return -EFAULT;
 153
 154                         remaining = copy_to_user(buf, ptr, sz);
 155
 156                         unxlate_dev_mem_ptr(p, ptr);
 157                 }
 158
 159                 if (remaining)
 160                         return -EFAULT;
 161
 162                 buf += sz;
 163                 p += sz;
 164                 count -= sz;
 165                 read += sz;
 166         }
 167
 168         *ppos += read;
 169         return read;
 170 }
 171
 172 static ssize_t write_mem(struct file *file, const char __user *buf,
 173                          size_t count, loff_t *ppos)
 174 {
 175         phys_addr_t p = *ppos;
 176         ssize_t written, sz;
 177         unsigned long copied;
 178         void *ptr;
 179
 180         if (p != *ppos)
 181                 return -EFBIG;
 182
 183         if (!valid_phys_addr_range(p, count))
 184                 return -EFAULT;
 185
 186         written = 0;
 187
 188 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 189         /* we don't have page 0 mapped on sparc and m68k.. */
 190         if (p < PAGE_SIZE) {
 191                 sz = size_inside_page(p, count);
 192                 /* Hmm. Do something? */
 193                 buf += sz;
 194                 p += sz;
 195                 count -= sz;
 196                 written += sz;
 197         }
 198 #endif
 199
 200         while (count > 0) {
 201                 int allowed;
 202
 203                 sz = size_inside_page(p, count);
 204
 205                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 206                 if (!allowed)
 207                         return -EPERM;
 208
 209                 /* Skip actual writing when a page is marked as restricted. */
 210                 if (allowed == 1) {
 211                         /*
 212                          * On ia64 if a page has been mapped somewhere as
 213                          * uncached, then it must also be accessed uncached
 214                          * by the kernel or data corruption may occur.
 215                          */
 216                         ptr = xlate_dev_mem_ptr(p);
 217                         if (!ptr) {
 218                                 if (written)
 219                                         break;
 220                                 return -EFAULT;
 221                         }
 222
 223                         copied = copy_from_user(ptr, buf, sz);
 224                         unxlate_dev_mem_ptr(p, ptr);
 225                         if (copied) {
 226                                 written += sz - copied;
 227                                 if (written)
 228                                         break;
 229                                 return -EFAULT;
 230                         }
 231                 }
 232
 233                 buf += sz;
 234                 p += sz;
 235                 count -= sz;
 236                 written += sz;
 237         }
 238
 239         *ppos += written;
 240         return written;
 241 }
 242
 243 int __weak phys_mem_access_prot_allowed(struct file *file,
 244         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 245 {
 246         return 1;
 247 }
 248
 249 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 250
 251 /*
 252  * Architectures vary in how they handle caching for addresses
 253  * outside of main memory.
 254  *
 255  */
 256 #ifdef pgprot_noncached
 257 static int uncached_access(struct file *file, phys_addr_t addr)
 258 {
 259 #if defined(CONFIG_IA64)
 260         /*
 261          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 262          * attribute aliases.
 263          */
 264         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 265 #elif defined(CONFIG_MIPS)
 266         {
 267                 extern int __uncached_access(struct file *file,
 268                                              unsigned long addr);
 269
 270                 return __uncached_access(file, addr);
 271         }
 272 #else
 273         /*
 274          * Accessing memory above the top the kernel knows about or through a
 275          * file pointer
 276          * that was marked O_DSYNC will be done non-cached.
 277          */
 278         if (file->f_flags & O_DSYNC)
 279                 return 1;
 280         return addr >= __pa(high_memory);
 281 #endif
 282 }
 283 #endif
 284
 285 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 286                                      unsigned long size, pgprot_t vma_prot)
 287 {
 288 #ifdef pgprot_noncached
 289         phys_addr_t offset = pfn << PAGE_SHIFT;
 290
 291         if (uncached_access(file, offset))
 292                 return pgprot_noncached(vma_prot);
 293 #endif
 294         return vma_prot;
 295 }
 296 #endif
 297
 298 #ifndef CONFIG_MMU
 299 static unsigned long get_unmapped_area_mem(struct file *file,
 300                                            unsigned long addr,
 301                                            unsigned long len,
 302                                            unsigned long pgoff,
 303                                            unsigned long flags)
 304 {
 305         if (!valid_mmap_phys_addr_range(pgoff, len))
 306                 return (unsigned long) -EINVAL;
 307         return pgoff << PAGE_SHIFT;
 308 }
 309
 310 /* permit direct mmap, for read, write or exec */
 311 static unsigned memory_mmap_capabilities(struct file *file)
 312 {
 313         return NOMMU_MAP_DIRECT |
 314                 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
 315 }
 316
 317 static unsigned zero_mmap_capabilities(struct file *file)
 318 {
 319         return NOMMU_MAP_COPY;
 320 }
 321
 322 /* can't do an in-place private mapping if there's no MMU */
 323 static inline int private_mapping_ok(struct vm_area_struct *vma)
 324 {
 325         return vma->vm_flags & VM_MAYSHARE;
 326 }
 327 #else
 328
 329 static inline int private_mapping_ok(struct vm_area_struct *vma)
 330 {
 331         return 1;
 332 }
 333 #endif
 334
 335 static const struct vm_operations_struct mmap_mem_ops = {
 336 #ifdef CONFIG_HAVE_IOREMAP_PROT
 337         .access = generic_access_phys
 338 #endif
 339 };
 340
 341 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 342 {
 343         size_t size = vma->vm_end - vma->vm_start;
 344         phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
 345
 346         /* Does it even fit in phys_addr_t? */
 347         if (offset >> PAGE_SHIFT != vma->vm_pgoff)
 348                 return -EINVAL;
 349
 350         /* It's illegal to wrap around the end of the physical address space. */
 351         if (offset + (phys_addr_t)size - 1 < offset)
 352                 return -EINVAL;
 353
 354         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 355                 return -EINVAL;
 356
 357         if (!private_mapping_ok(vma))
 358                 return -ENOSYS;
 359
 360         if (!range_is_allowed(vma->vm_pgoff, size))
 361                 return -EPERM;
 362
 363         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 364                                                 &vma->vm_page_prot))
 365                 return -EINVAL;
 366
 367         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 368                                                  size,
 369                                                  vma->vm_page_prot);
 370
 371         vma->vm_ops = &mmap_mem_ops;
 372
 373         /* Remap-pfn-range will mark the range VM_IO */
 374         if (remap_pfn_range(vma,
 375                             vma->vm_start,
 376                             vma->vm_pgoff,
 377                             size,
 378                             vma->vm_page_prot)) {
 379                 return -EAGAIN;
 380         }
 381         return 0;
 382 }
 383
 384 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 385 {
 386         unsigned long pfn;
 387
 388         /* Turn a kernel-virtual address into a physical page frame */
 389         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 390
 391         /*
 392          * RED-PEN: on some architectures there is more mapped memory than
 393          * available in mem_map which pfn_valid checks for. Perhaps should add a
 394          * new macro here.
 395          *
 396          * RED-PEN: vmalloc is not supported right now.
 397          */
 398         if (!pfn_valid(pfn))
 399                 return -EIO;
 400
 401         vma->vm_pgoff = pfn;
 402         return mmap_mem(file, vma);
 403 }
 404
 405 /*
 406  * This function reads the *virtual* memory as seen by the kernel.
 407  */
 408 static ssize_t read_kmem(struct file *file, char __user *buf,
 409                          size_t count, loff_t *ppos)
 410 {
 411         unsigned long p = *ppos;
 412         ssize_t low_count, read, sz;
 413         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 414         int err = 0;
 415
 416         read = 0;
 417         if (p < (unsigned long) high_memory) {
 418                 low_count = count;
 419                 if (count > (unsigned long)high_memory - p)
 420                         low_count = (unsigned long)high_memory - p;
 421
 422 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 423                 /* we don't have page 0 mapped on sparc and m68k.. */
 424                 if (p < PAGE_SIZE && low_count > 0) {
 425                         sz = size_inside_page(p, low_count);
 426                         if (clear_user(buf, sz))
 427                                 return -EFAULT;
 428                         buf += sz;
 429                         p += sz;
 430                         read += sz;
 431                         low_count -= sz;
 432                         count -= sz;
 433                 }
 434 #endif
 435                 while (low_count > 0) {
 436                         sz = size_inside_page(p, low_count);
 437
 438                         /*
 439                          * On ia64 if a page has been mapped somewhere as
 440                          * uncached, then it must also be accessed uncached
 441                          * by the kernel or data corruption may occur
 442                          */
 443                         kbuf = xlate_dev_kmem_ptr((void *)p);
 444                         if (!virt_addr_valid(kbuf))
 445                                 return -ENXIO;
 446
 447                         if (copy_to_user(buf, kbuf, sz))
 448                                 return -EFAULT;
 449                         buf += sz;
 450                         p += sz;
 451                         read += sz;
 452                         low_count -= sz;
 453                         count -= sz;
 454                 }
 455         }
 456
 457         if (count > 0) {
 458                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 459                 if (!kbuf)
 460                         return -ENOMEM;
 461                 while (count > 0) {
 462                         sz = size_inside_page(p, count);
 463                         if (!is_vmalloc_or_module_addr((void *)p)) {
 464                                 err = -ENXIO;
 465                                 break;
 466                         }
 467                         sz = vread(kbuf, (char *)p, sz);
 468                         if (!sz)
 469                                 break;
 470                         if (copy_to_user(buf, kbuf, sz)) {
 471                                 err = -EFAULT;
 472                                 break;
 473                         }
 474                         count -= sz;
 475                         buf += sz;
 476                         read += sz;
 477                         p += sz;
 478                 }
 479                 free_page((unsigned long)kbuf);
 480         }
 481         *ppos = p;
 482         return read ? read : err;
 483 }
 484
 485
 486 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 487                                 size_t count, loff_t *ppos)
 488 {
 489         ssize_t written, sz;
 490         unsigned long copied;
 491
 492         written = 0;
 493 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 494         /* we don't have page 0 mapped on sparc and m68k.. */
 495         if (p < PAGE_SIZE) {
 496                 sz = size_inside_page(p, count);
 497                 /* Hmm. Do something? */
 498                 buf += sz;
 499                 p += sz;
 500                 count -= sz;
 501                 written += sz;
 502         }
 503 #endif
 504
 505         while (count > 0) {
 506                 void *ptr;
 507
 508                 sz = size_inside_page(p, count);
 509
 510                 /*
 511                  * On ia64 if a page has been mapped somewhere as uncached, then
 512                  * it must also be accessed uncached by the kernel or data
 513                  * corruption may occur.
 514                  */
 515                 ptr = xlate_dev_kmem_ptr((void *)p);
 516                 if (!virt_addr_valid(ptr))
 517                         return -ENXIO;
 518
 519                 copied = copy_from_user(ptr, buf, sz);
 520                 if (copied) {
 521                         written += sz - copied;
 522                         if (written)
 523                                 break;
 524                         return -EFAULT;
 525                 }
 526                 buf += sz;
 527                 p += sz;
 528                 count -= sz;
 529                 written += sz;
 530         }
 531
 532         *ppos += written;
 533         return written;
 534 }
 535
 536 /*
 537  * This function writes to the *virtual* memory as seen by the kernel.
 538  */
 539 static ssize_t write_kmem(struct file *file, const char __user *buf,
 540                           size_t count, loff_t *ppos)
 541 {
 542         unsigned long p = *ppos;
 543         ssize_t wrote = 0;
 544         ssize_t virtr = 0;
 545         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 546         int err = 0;
 547
 548         if (p < (unsigned long) high_memory) {
 549                 unsigned long to_write = min_t(unsigned long, count,
 550                                                (unsigned long)high_memory - p);
 551                 wrote = do_write_kmem(p, buf, to_write, ppos);
 552                 if (wrote != to_write)
 553                         return wrote;
 554                 p += wrote;
 555                 buf += wrote;
 556                 count -= wrote;
 557         }
 558
 559         if (count > 0) {
 560                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 561                 if (!kbuf)
 562                         return wrote ? wrote : -ENOMEM;
 563                 while (count > 0) {
 564                         unsigned long sz = size_inside_page(p, count);
 565                         unsigned long n;
 566
 567                         if (!is_vmalloc_or_module_addr((void *)p)) {
 568                                 err = -ENXIO;
 569                                 break;
 570                         }
 571                         n = copy_from_user(kbuf, buf, sz);
 572                         if (n) {
 573                                 err = -EFAULT;
 574                                 break;
 575                         }
 576                         vwrite(kbuf, (char *)p, sz);
 577                         count -= sz;
 578                         buf += sz;
 579                         virtr += sz;
 580                         p += sz;
 581                 }
 582                 free_page((unsigned long)kbuf);
 583         }
 584
 585         *ppos = p;
 586         return virtr + wrote ? : err;
 587 }
 588
 589 static ssize_t read_port(struct file *file, char __user *buf,
 590                          size_t count, loff_t *ppos)
 591 {
 592         unsigned long i = *ppos;
 593         char __user *tmp = buf;
 594
 595         if (!access_ok(VERIFY_WRITE, buf, count))
 596                 return -EFAULT;
 597         while (count-- > 0 && i < 65536) {
 598                 if (__put_user(inb(i), tmp) < 0)
 599                         return -EFAULT;
 600                 i++;
 601                 tmp++;
 602         }
 603         *ppos = i;
 604         return tmp-buf;
 605 }
 606
 607 static ssize_t write_port(struct file *file, const char __user *buf,
 608                           size_t count, loff_t *ppos)
 609 {
 610         unsigned long i = *ppos;
 611         const char __user *tmp = buf;
 612
 613         if (!access_ok(VERIFY_READ, buf, count))
 614                 return -EFAULT;
 615         while (count-- > 0 && i < 65536) {
 616                 char c;
 617
 618                 if (__get_user(c, tmp)) {
 619                         if (tmp > buf)
 620                                 break;
 621                         return -EFAULT;
 622                 }
 623                 outb(c, i);
 624                 i++;
 625                 tmp++;
 626         }
 627         *ppos = i;
 628         return tmp-buf;
 629 }
 630
 631 static ssize_t read_null(struct file *file, char __user *buf,
 632                          size_t count, loff_t *ppos)
 633 {
 634         return 0;
 635 }
 636
 637 static ssize_t write_null(struct file *file, const char __user *buf,
 638                           size_t count, loff_t *ppos)
 639 {
 640         return count;
 641 }
 642
 643 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
 644 {
 645         return 0;
 646 }
 647
 648 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
 649 {
 650         size_t count = iov_iter_count(from);
 651         iov_iter_advance(from, count);
 652         return count;
 653 }
 654
 655 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 656                         struct splice_desc *sd)
 657 {
 658         return sd->len;
 659 }
 660
 661 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 662                                  loff_t *ppos, size_t len, unsigned int flags)
 663 {
 664         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 665 }
 666
 667 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
 668 {
 669         size_t written = 0;
 670
 671         while (iov_iter_count(iter)) {
 672                 size_t chunk = iov_iter_count(iter), n;
 673
 674                 if (chunk > PAGE_SIZE)
 675                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 676                 n = iov_iter_zero(chunk, iter);
 677                 if (!n && iov_iter_count(iter))
 678                         return written ? written : -EFAULT;
 679                 written += n;
 680                 if (signal_pending(current))
 681                         return written ? written : -ERESTARTSYS;
 682                 cond_resched();
 683         }
 684         return written;
 685 }
 686
 687 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 688 {
 689 #ifndef CONFIG_MMU
 690         return -ENOSYS;
 691 #endif
 692         if (vma->vm_flags & VM_SHARED)
 693                 return shmem_zero_setup(vma);
 694         return 0;
 695 }
 696
 697 static unsigned long get_unmapped_area_zero(struct file *file,
 698                                 unsigned long addr, unsigned long len,
 699                                 unsigned long pgoff, unsigned long flags)
 700 {
 701 #ifdef CONFIG_MMU
 702         if (flags & MAP_SHARED) {
 703                 /*
 704                  * mmap_zero() will call shmem_zero_setup() to create a file,
 705                  * so use shmem's get_unmapped_area in case it can be huge;
 706                  * and pass NULL for file as in mmap.c's get_unmapped_area(),
 707                  * so as not to confuse shmem with our handle on "/dev/zero".
 708                  */
 709                 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
 710         }
 711
 712         /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
 713         return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
 714 #else
 715         return -ENOSYS;
 716 #endif
 717 }
 718
 719 static ssize_t write_full(struct file *file, const char __user *buf,
 720                           size_t count, loff_t *ppos)
 721 {
 722         return -ENOSPC;
 723 }
 724
 725 /*
 726  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 727  * can fopen() both devices with "a" now.  This was previously impossible.
 728  * -- SRB.
 729  */
 730 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 731 {
 732         return file->f_pos = 0;
 733 }
 734
 735 /*
 736  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 737  * check against negative addresses: they are ok. The return value is weird,
 738  * though, in that case (0).
 739  *
 740  * also note that seeking relative to the "end of file" isn't supported:
 741  * it has no meaning, so it returns -EINVAL.
 742  */
 743 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 744 {
 745         loff_t ret;
 746
 747         inode_lock(file_inode(file));
 748         switch (orig) {
 749         case SEEK_CUR:
 750                 offset += file->f_pos;
 751         case SEEK_SET:
 752                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 753                 if ((unsigned long long)offset >= -MAX_ERRNO) {
 754                         ret = -EOVERFLOW;
 755                         break;
 756                 }
 757                 file->f_pos = offset;
 758                 ret = file->f_pos;
 759                 force_successful_syscall_return();
 760                 break;
 761         default:
 762                 ret = -EINVAL;
 763         }
 764         inode_unlock(file_inode(file));
 765         return ret;
 766 }
 767
 768 static int open_port(struct inode *inode, struct file *filp)
 769 {
 770         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 771 }
 772
 773 #define zero_lseek      null_lseek
 774 #define full_lseek      null_lseek
 775 #define write_zero      write_null
 776 #define write_iter_zero write_iter_null
 777 #define open_mem        open_port
 778 #define open_kmem       open_mem
 779
 780 static const struct file_operations __maybe_unused mem_fops = {
 781         .llseek         = memory_lseek,
 782         .read           = read_mem,
 783         .write          = write_mem,
 784         .mmap           = mmap_mem,
 785         .open           = open_mem,
 786 #ifndef CONFIG_MMU
 787         .get_unmapped_area = get_unmapped_area_mem,
 788         .mmap_capabilities = memory_mmap_capabilities,
 789 #endif
 790 };
 791
 792 static const struct file_operations __maybe_unused kmem_fops = {
 793         .llseek         = memory_lseek,
 794         .read           = read_kmem,
 795         .write          = write_kmem,
 796         .mmap           = mmap_kmem,
 797         .open           = open_kmem,
 798 #ifndef CONFIG_MMU
 799         .get_unmapped_area = get_unmapped_area_mem,
 800         .mmap_capabilities = memory_mmap_capabilities,
 801 #endif
 802 };
 803
 804 static const struct file_operations null_fops = {
 805         .llseek         = null_lseek,
 806         .read           = read_null,
 807         .write          = write_null,
 808         .read_iter      = read_iter_null,
 809         .write_iter     = write_iter_null,
 810         .splice_write   = splice_write_null,
 811 };
 812
 813 static const struct file_operations __maybe_unused port_fops = {
 814         .llseek         = memory_lseek,
 815         .read           = read_port,
 816         .write          = write_port,
 817         .open           = open_port,
 818 };
 819
 820 static const struct file_operations zero_fops = {
 821         .llseek         = zero_lseek,
 822         .write          = write_zero,
 823         .read_iter      = read_iter_zero,
 824         .write_iter     = write_iter_zero,
 825         .mmap           = mmap_zero,
 826         .get_unmapped_area = get_unmapped_area_zero,
 827 #ifndef CONFIG_MMU
 828         .mmap_capabilities = zero_mmap_capabilities,
 829 #endif
 830 };
 831
 832 static const struct file_operations full_fops = {
 833         .llseek         = full_lseek,
 834         .read_iter      = read_iter_zero,
 835         .write          = write_full,
 836 };
 837
 838 static const struct memdev {
 839         const char *name;
 840         umode_t mode;
 841         const struct file_operations *fops;
 842         fmode_t fmode;
 843 } devlist[] = {
 844 #ifdef CONFIG_DEVMEM
 845          [1] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
 846 #endif
 847 #ifdef CONFIG_DEVKMEM
 848          [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
 849 #endif
 850          [3] = { "null", 0666, &null_fops, 0 },
 851 #ifdef CONFIG_DEVPORT
 852          [4] = { "port", 0, &port_fops, 0 },
 853 #endif
 854          [5] = { "zero", 0666, &zero_fops, 0 },
 855          [7] = { "full", 0666, &full_fops, 0 },
 856          [8] = { "random", 0666, &random_fops, 0 },
 857          [9] = { "urandom", 0666, &urandom_fops, 0 },
 858 #ifdef CONFIG_PRINTK
 859         [11] = { "kmsg", 0644, &kmsg_fops, 0 },
 860 #endif
 861 };
 862
 863 static int memory_open(struct inode *inode, struct file *filp)
 864 {
 865         int minor;
 866         const struct memdev *dev;
 867
 868         minor = iminor(inode);
 869         if (minor >= ARRAY_SIZE(devlist))
 870                 return -ENXIO;
 871
 872         dev = &devlist[minor];
 873         if (!dev->fops)
 874                 return -ENXIO;
 875
 876         filp->f_op = dev->fops;
 877         filp->f_mode |= dev->fmode;
 878
 879         if (dev->fops->open)
 880                 return dev->fops->open(inode, filp);
 881
 882         return 0;
 883 }
 884
 885 static const struct file_operations memory_fops = {
 886         .open = memory_open,
 887         .llseek = noop_llseek,
 888 };
 889
 890 static char *mem_devnode(struct device *dev, umode_t *mode)
 891 {
 892         if (mode && devlist[MINOR(dev->devt)].mode)
 893                 *mode = devlist[MINOR(dev->devt)].mode;
 894         return NULL;
 895 }
 896
 897 static struct class *mem_class;
 898
 899 static int __init chr_dev_init(void)
 900 {
 901         int minor;
 902
 903         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 904                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 905
 906         mem_class = class_create(THIS_MODULE, "mem");
 907         if (IS_ERR(mem_class))
 908                 return PTR_ERR(mem_class);
 909
 910         mem_class->devnode = mem_devnode;
 911         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 912                 if (!devlist[minor].name)
 913                         continue;
 914
 915                 /*
 916                  * Create /dev/port?
 917                  */
 918                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 919                         continue;
 920
 921                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 922                               NULL, devlist[minor].name);
 923         }
 924
 925         return tty_init();
 926 }
 927
 928 fs_initcall(chr_dev_init);