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 static inline bool should_stop_iteration(void)
 101 {
 102         if (need_resched())
 103                 cond_resched();
 104         return fatal_signal_pending(current);
 105 }
 106
 107 /*
 108  * This funcion reads the *physical* memory. The f_pos points directly to the
 109  * memory location.
 110  */
 111 static ssize_t read_mem(struct file *file, char __user *buf,
 112                         size_t count, loff_t *ppos)
 113 {
 114         phys_addr_t p = *ppos;
 115         ssize_t read, sz;
 116         void *ptr;
 117         char *bounce;
 118         int err;
 119
 120         if (p != *ppos)
 121                 return 0;
 122
 123         if (!valid_phys_addr_range(p, count))
 124                 return -EFAULT;
 125         read = 0;
 126 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 127         /* we don't have page 0 mapped on sparc and m68k.. */
 128         if (p < PAGE_SIZE) {
 129                 sz = size_inside_page(p, count);
 130                 if (sz > 0) {
 131                         if (clear_user(buf, sz))
 132                                 return -EFAULT;
 133                         buf += sz;
 134                         p += sz;
 135                         count -= sz;
 136                         read += sz;
 137                 }
 138         }
 139 #endif
 140
 141         bounce = kmalloc(PAGE_SIZE, GFP_KERNEL);
 142         if (!bounce)
 143                 return -ENOMEM;
 144
 145         while (count > 0) {
 146                 unsigned long remaining;
 147                 int allowed, probe;
 148
 149                 sz = size_inside_page(p, count);
 150
 151                 err = -EPERM;
 152                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 153                 if (!allowed)
 154                         goto failed;
 155
 156                 err = -EFAULT;
 157                 if (allowed == 2) {
 158                         /* Show zeros for restricted memory. */
 159                         remaining = clear_user(buf, sz);
 160                 } else {
 161                         /*
 162                          * On ia64 if a page has been mapped somewhere as
 163                          * uncached, then it must also be accessed uncached
 164                          * by the kernel or data corruption may occur.
 165                          */
 166                         ptr = xlate_dev_mem_ptr(p);
 167                         if (!ptr)
 168                                 goto failed;
 169
 170                         probe = probe_kernel_read(bounce, ptr, sz);
 171                         unxlate_dev_mem_ptr(p, ptr);
 172                         if (probe)
 173                                 goto failed;
 174
 175                         remaining = copy_to_user(buf, bounce, sz);
 176                 }
 177
 178                 if (remaining)
 179                         goto failed;
 180
 181                 buf += sz;
 182                 p += sz;
 183                 count -= sz;
 184                 read += sz;
 185                 if (should_stop_iteration())
 186                         break;
 187         }
 188         kfree(bounce);
 189
 190         *ppos += read;
 191         return read;
 192
 193 failed:
 194         kfree(bounce);
 195         return err;
 196 }
 197
 198 static ssize_t write_mem(struct file *file, const char __user *buf,
 199                          size_t count, loff_t *ppos)
 200 {
 201         phys_addr_t p = *ppos;
 202         ssize_t written, sz;
 203         unsigned long copied;
 204         void *ptr;
 205
 206         if (p != *ppos)
 207                 return -EFBIG;
 208
 209         if (!valid_phys_addr_range(p, count))
 210                 return -EFAULT;
 211
 212         written = 0;
 213
 214 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 215         /* we don't have page 0 mapped on sparc and m68k.. */
 216         if (p < PAGE_SIZE) {
 217                 sz = size_inside_page(p, count);
 218                 /* Hmm. Do something? */
 219                 buf += sz;
 220                 p += sz;
 221                 count -= sz;
 222                 written += sz;
 223         }
 224 #endif
 225
 226         while (count > 0) {
 227                 int allowed;
 228
 229                 sz = size_inside_page(p, count);
 230
 231                 allowed = page_is_allowed(p >> PAGE_SHIFT);
 232                 if (!allowed)
 233                         return -EPERM;
 234
 235                 /* Skip actual writing when a page is marked as restricted. */
 236                 if (allowed == 1) {
 237                         /*
 238                          * On ia64 if a page has been mapped somewhere as
 239                          * uncached, then it must also be accessed uncached
 240                          * by the kernel or data corruption may occur.
 241                          */
 242                         ptr = xlate_dev_mem_ptr(p);
 243                         if (!ptr) {
 244                                 if (written)
 245                                         break;
 246                                 return -EFAULT;
 247                         }
 248
 249                         copied = copy_from_user(ptr, buf, sz);
 250                         unxlate_dev_mem_ptr(p, ptr);
 251                         if (copied) {
 252                                 written += sz - copied;
 253                                 if (written)
 254                                         break;
 255                                 return -EFAULT;
 256                         }
 257                 }
 258
 259                 buf += sz;
 260                 p += sz;
 261                 count -= sz;
 262                 written += sz;
 263                 if (should_stop_iteration())
 264                         break;
 265         }
 266
 267         *ppos += written;
 268         return written;
 269 }
 270
 271 int __weak phys_mem_access_prot_allowed(struct file *file,
 272         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 273 {
 274         return 1;
 275 }
 276
 277 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 278
 279 /*
 280  * Architectures vary in how they handle caching for addresses
 281  * outside of main memory.
 282  *
 283  */
 284 #ifdef pgprot_noncached
 285 static int uncached_access(struct file *file, phys_addr_t addr)
 286 {
 287 #if defined(CONFIG_IA64)
 288         /*
 289          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 290          * attribute aliases.
 291          */
 292         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 293 #elif defined(CONFIG_MIPS)
 294         {
 295                 extern int __uncached_access(struct file *file,
 296                                              unsigned long addr);
 297
 298                 return __uncached_access(file, addr);
 299         }
 300 #else
 301         /*
 302          * Accessing memory above the top the kernel knows about or through a
 303          * file pointer
 304          * that was marked O_DSYNC will be done non-cached.
 305          */
 306         if (file->f_flags & O_DSYNC)
 307                 return 1;
 308         return addr >= __pa(high_memory);
 309 #endif
 310 }
 311 #endif
 312
 313 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 314                                      unsigned long size, pgprot_t vma_prot)
 315 {
 316 #ifdef pgprot_noncached
 317         phys_addr_t offset = pfn << PAGE_SHIFT;
 318
 319         if (uncached_access(file, offset))
 320                 return pgprot_noncached(vma_prot);
 321 #endif
 322         return vma_prot;
 323 }
 324 #endif
 325
 326 #ifndef CONFIG_MMU
 327 static unsigned long get_unmapped_area_mem(struct file *file,
 328                                            unsigned long addr,
 329                                            unsigned long len,
 330                                            unsigned long pgoff,
 331                                            unsigned long flags)
 332 {
 333         if (!valid_mmap_phys_addr_range(pgoff, len))
 334                 return (unsigned long) -EINVAL;
 335         return pgoff << PAGE_SHIFT;
 336 }
 337
 338 /* permit direct mmap, for read, write or exec */
 339 static unsigned memory_mmap_capabilities(struct file *file)
 340 {
 341         return NOMMU_MAP_DIRECT |
 342                 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
 343 }
 344
 345 static unsigned zero_mmap_capabilities(struct file *file)
 346 {
 347         return NOMMU_MAP_COPY;
 348 }
 349
 350 /* can't do an in-place private mapping if there's no MMU */
 351 static inline int private_mapping_ok(struct vm_area_struct *vma)
 352 {
 353         return vma->vm_flags & VM_MAYSHARE;
 354 }
 355 #else
 356
 357 static inline int private_mapping_ok(struct vm_area_struct *vma)
 358 {
 359         return 1;
 360 }
 361 #endif
 362
 363 static const struct vm_operations_struct mmap_mem_ops = {
 364 #ifdef CONFIG_HAVE_IOREMAP_PROT
 365         .access = generic_access_phys
 366 #endif
 367 };
 368
 369 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 370 {
 371         size_t size = vma->vm_end - vma->vm_start;
 372         phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
 373
 374         /* Does it even fit in phys_addr_t? */
 375         if (offset >> PAGE_SHIFT != vma->vm_pgoff)
 376                 return -EINVAL;
 377
 378         /* It's illegal to wrap around the end of the physical address space. */
 379         if (offset + (phys_addr_t)size - 1 < offset)
 380                 return -EINVAL;
 381
 382         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 383                 return -EINVAL;
 384
 385         if (!private_mapping_ok(vma))
 386                 return -ENOSYS;
 387
 388         if (!range_is_allowed(vma->vm_pgoff, size))
 389                 return -EPERM;
 390
 391         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 392                                                 &vma->vm_page_prot))
 393                 return -EINVAL;
 394
 395         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 396                                                  size,
 397                                                  vma->vm_page_prot);
 398
 399         vma->vm_ops = &mmap_mem_ops;
 400
 401         /* Remap-pfn-range will mark the range VM_IO */
 402         if (remap_pfn_range(vma,
 403                             vma->vm_start,
 404                             vma->vm_pgoff,
 405                             size,
 406                             vma->vm_page_prot)) {
 407                 return -EAGAIN;
 408         }
 409         return 0;
 410 }
 411
 412 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 413 {
 414         unsigned long pfn;
 415
 416         /* Turn a kernel-virtual address into a physical page frame */
 417         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 418
 419         /*
 420          * RED-PEN: on some architectures there is more mapped memory than
 421          * available in mem_map which pfn_valid checks for. Perhaps should add a
 422          * new macro here.
 423          *
 424          * RED-PEN: vmalloc is not supported right now.
 425          */
 426         if (!pfn_valid(pfn))
 427                 return -EIO;
 428
 429         vma->vm_pgoff = pfn;
 430         return mmap_mem(file, vma);
 431 }
 432
 433 /*
 434  * This function reads the *virtual* memory as seen by the kernel.
 435  */
 436 static ssize_t read_kmem(struct file *file, char __user *buf,
 437                          size_t count, loff_t *ppos)
 438 {
 439         unsigned long p = *ppos;
 440         ssize_t low_count, read, sz;
 441         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 442         int err = 0;
 443
 444         read = 0;
 445         if (p < (unsigned long) high_memory) {
 446                 low_count = count;
 447                 if (count > (unsigned long)high_memory - p)
 448                         low_count = (unsigned long)high_memory - p;
 449
 450 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 451                 /* we don't have page 0 mapped on sparc and m68k.. */
 452                 if (p < PAGE_SIZE && low_count > 0) {
 453                         sz = size_inside_page(p, low_count);
 454                         if (clear_user(buf, sz))
 455                                 return -EFAULT;
 456                         buf += sz;
 457                         p += sz;
 458                         read += sz;
 459                         low_count -= sz;
 460                         count -= sz;
 461                 }
 462 #endif
 463                 while (low_count > 0) {
 464                         sz = size_inside_page(p, low_count);
 465
 466                         /*
 467                          * On ia64 if a page has been mapped somewhere as
 468                          * uncached, then it must also be accessed uncached
 469                          * by the kernel or data corruption may occur
 470                          */
 471                         kbuf = xlate_dev_kmem_ptr((void *)p);
 472                         if (!virt_addr_valid(kbuf))
 473                                 return -ENXIO;
 474
 475                         if (copy_to_user(buf, kbuf, sz))
 476                                 return -EFAULT;
 477                         buf += sz;
 478                         p += sz;
 479                         read += sz;
 480                         low_count -= sz;
 481                         count -= sz;
 482                         if (should_stop_iteration()) {
 483                                 count = 0;
 484                                 break;
 485                         }
 486                 }
 487         }
 488
 489         if (count > 0) {
 490                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 491                 if (!kbuf)
 492                         return -ENOMEM;
 493                 while (count > 0) {
 494                         sz = size_inside_page(p, count);
 495                         if (!is_vmalloc_or_module_addr((void *)p)) {
 496                                 err = -ENXIO;
 497                                 break;
 498                         }
 499                         sz = vread(kbuf, (char *)p, sz);
 500                         if (!sz)
 501                                 break;
 502                         if (copy_to_user(buf, kbuf, sz)) {
 503                                 err = -EFAULT;
 504                                 break;
 505                         }
 506                         count -= sz;
 507                         buf += sz;
 508                         read += sz;
 509                         p += sz;
 510                         if (should_stop_iteration())
 511                                 break;
 512                 }
 513                 free_page((unsigned long)kbuf);
 514         }
 515         *ppos = p;
 516         return read ? read : err;
 517 }
 518
 519
 520 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 521                                 size_t count, loff_t *ppos)
 522 {
 523         ssize_t written, sz;
 524         unsigned long copied;
 525
 526         written = 0;
 527 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 528         /* we don't have page 0 mapped on sparc and m68k.. */
 529         if (p < PAGE_SIZE) {
 530                 sz = size_inside_page(p, count);
 531                 /* Hmm. Do something? */
 532                 buf += sz;
 533                 p += sz;
 534                 count -= sz;
 535                 written += sz;
 536         }
 537 #endif
 538
 539         while (count > 0) {
 540                 void *ptr;
 541
 542                 sz = size_inside_page(p, count);
 543
 544                 /*
 545                  * On ia64 if a page has been mapped somewhere as uncached, then
 546                  * it must also be accessed uncached by the kernel or data
 547                  * corruption may occur.
 548                  */
 549                 ptr = xlate_dev_kmem_ptr((void *)p);
 550                 if (!virt_addr_valid(ptr))
 551                         return -ENXIO;
 552
 553                 copied = copy_from_user(ptr, buf, sz);
 554                 if (copied) {
 555                         written += sz - copied;
 556                         if (written)
 557                                 break;
 558                         return -EFAULT;
 559                 }
 560                 buf += sz;
 561                 p += sz;
 562                 count -= sz;
 563                 written += sz;
 564                 if (should_stop_iteration())
 565                         break;
 566         }
 567
 568         *ppos += written;
 569         return written;
 570 }
 571
 572 /*
 573  * This function writes to the *virtual* memory as seen by the kernel.
 574  */
 575 static ssize_t write_kmem(struct file *file, const char __user *buf,
 576                           size_t count, loff_t *ppos)
 577 {
 578         unsigned long p = *ppos;
 579         ssize_t wrote = 0;
 580         ssize_t virtr = 0;
 581         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 582         int err = 0;
 583
 584         if (p < (unsigned long) high_memory) {
 585                 unsigned long to_write = min_t(unsigned long, count,
 586                                                (unsigned long)high_memory - p);
 587                 wrote = do_write_kmem(p, buf, to_write, ppos);
 588                 if (wrote != to_write)
 589                         return wrote;
 590                 p += wrote;
 591                 buf += wrote;
 592                 count -= wrote;
 593         }
 594
 595         if (count > 0) {
 596                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 597                 if (!kbuf)
 598                         return wrote ? wrote : -ENOMEM;
 599                 while (count > 0) {
 600                         unsigned long sz = size_inside_page(p, count);
 601                         unsigned long n;
 602
 603                         if (!is_vmalloc_or_module_addr((void *)p)) {
 604                                 err = -ENXIO;
 605                                 break;
 606                         }
 607                         n = copy_from_user(kbuf, buf, sz);
 608                         if (n) {
 609                                 err = -EFAULT;
 610                                 break;
 611                         }
 612                         vwrite(kbuf, (char *)p, sz);
 613                         count -= sz;
 614                         buf += sz;
 615                         virtr += sz;
 616                         p += sz;
 617                         if (should_stop_iteration())
 618                                 break;
 619                 }
 620                 free_page((unsigned long)kbuf);
 621         }
 622
 623         *ppos = p;
 624         return virtr + wrote ? : err;
 625 }
 626
 627 static ssize_t read_port(struct file *file, char __user *buf,
 628                          size_t count, loff_t *ppos)
 629 {
 630         unsigned long i = *ppos;
 631         char __user *tmp = buf;
 632
 633         if (!access_ok(buf, count))
 634                 return -EFAULT;
 635         while (count-- > 0 && i < 65536) {
 636                 if (__put_user(inb(i), tmp) < 0)
 637                         return -EFAULT;
 638                 i++;
 639                 tmp++;
 640         }
 641         *ppos = i;
 642         return tmp-buf;
 643 }
 644
 645 static ssize_t write_port(struct file *file, const char __user *buf,
 646                           size_t count, loff_t *ppos)
 647 {
 648         unsigned long i = *ppos;
 649         const char __user *tmp = buf;
 650
 651         if (!access_ok(buf, count))
 652                 return -EFAULT;
 653         while (count-- > 0 && i < 65536) {
 654                 char c;
 655
 656                 if (__get_user(c, tmp)) {
 657                         if (tmp > buf)
 658                                 break;
 659                         return -EFAULT;
 660                 }
 661                 outb(c, i);
 662                 i++;
 663                 tmp++;
 664         }
 665         *ppos = i;
 666         return tmp-buf;
 667 }
 668
 669 static ssize_t read_null(struct file *file, char __user *buf,
 670                          size_t count, loff_t *ppos)
 671 {
 672         return 0;
 673 }
 674
 675 static ssize_t write_null(struct file *file, const char __user *buf,
 676                           size_t count, loff_t *ppos)
 677 {
 678         return count;
 679 }
 680
 681 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
 682 {
 683         return 0;
 684 }
 685
 686 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
 687 {
 688         size_t count = iov_iter_count(from);
 689         iov_iter_advance(from, count);
 690         return count;
 691 }
 692
 693 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 694                         struct splice_desc *sd)
 695 {
 696         return sd->len;
 697 }
 698
 699 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 700                                  loff_t *ppos, size_t len, unsigned int flags)
 701 {
 702         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 703 }
 704
 705 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
 706 {
 707         size_t written = 0;
 708
 709         while (iov_iter_count(iter)) {
 710                 size_t chunk = iov_iter_count(iter), n;
 711
 712                 if (chunk > PAGE_SIZE)
 713                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 714                 n = iov_iter_zero(chunk, iter);
 715                 if (!n && iov_iter_count(iter))
 716                         return written ? written : -EFAULT;
 717                 written += n;
 718                 if (signal_pending(current))
 719                         return written ? written : -ERESTARTSYS;
 720                 cond_resched();
 721         }
 722         return written;
 723 }
 724
 725 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 726 {
 727 #ifndef CONFIG_MMU
 728         return -ENOSYS;
 729 #endif
 730         if (vma->vm_flags & VM_SHARED)
 731                 return shmem_zero_setup(vma);
 732         vma_set_anonymous(vma);
 733         return 0;
 734 }
 735
 736 static unsigned long get_unmapped_area_zero(struct file *file,
 737                                 unsigned long addr, unsigned long len,
 738                                 unsigned long pgoff, unsigned long flags)
 739 {
 740 #ifdef CONFIG_MMU
 741         if (flags & MAP_SHARED) {
 742                 /*
 743                  * mmap_zero() will call shmem_zero_setup() to create a file,
 744                  * so use shmem's get_unmapped_area in case it can be huge;
 745                  * and pass NULL for file as in mmap.c's get_unmapped_area(),
 746                  * so as not to confuse shmem with our handle on "/dev/zero".
 747                  */
 748                 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
 749         }
 750
 751         /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
 752         return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
 753 #else
 754         return -ENOSYS;
 755 #endif
 756 }
 757
 758 static ssize_t write_full(struct file *file, const char __user *buf,
 759                           size_t count, loff_t *ppos)
 760 {
 761         return -ENOSPC;
 762 }
 763
 764 /*
 765  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 766  * can fopen() both devices with "a" now.  This was previously impossible.
 767  * -- SRB.
 768  */
 769 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 770 {
 771         return file->f_pos = 0;
 772 }
 773
 774 /*
 775  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 776  * check against negative addresses: they are ok. The return value is weird,
 777  * though, in that case (0).
 778  *
 779  * also note that seeking relative to the "end of file" isn't supported:
 780  * it has no meaning, so it returns -EINVAL.
 781  */
 782 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 783 {
 784         loff_t ret;
 785
 786         inode_lock(file_inode(file));
 787         switch (orig) {
 788         case SEEK_CUR:
 789                 offset += file->f_pos;
 790                 /* fall through */
 791         case SEEK_SET:
 792                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 793                 if ((unsigned long long)offset >= -MAX_ERRNO) {
 794                         ret = -EOVERFLOW;
 795                         break;
 796                 }
 797                 file->f_pos = offset;
 798                 ret = file->f_pos;
 799                 force_successful_syscall_return();
 800                 break;
 801         default:
 802                 ret = -EINVAL;
 803         }
 804         inode_unlock(file_inode(file));
 805         return ret;
 806 }
 807
 808 static int open_port(struct inode *inode, struct file *filp)
 809 {
 810         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 811 }
 812
 813 #define zero_lseek      null_lseek
 814 #define full_lseek      null_lseek
 815 #define write_zero      write_null
 816 #define write_iter_zero write_iter_null
 817 #define open_mem        open_port
 818 #define open_kmem       open_mem
 819
 820 static const struct file_operations __maybe_unused mem_fops = {
 821         .llseek         = memory_lseek,
 822         .read           = read_mem,
 823         .write          = write_mem,
 824         .mmap           = mmap_mem,
 825         .open           = open_mem,
 826 #ifndef CONFIG_MMU
 827         .get_unmapped_area = get_unmapped_area_mem,
 828         .mmap_capabilities = memory_mmap_capabilities,
 829 #endif
 830 };
 831
 832 static const struct file_operations __maybe_unused kmem_fops = {
 833         .llseek         = memory_lseek,
 834         .read           = read_kmem,
 835         .write          = write_kmem,
 836         .mmap           = mmap_kmem,
 837         .open           = open_kmem,
 838 #ifndef CONFIG_MMU
 839         .get_unmapped_area = get_unmapped_area_mem,
 840         .mmap_capabilities = memory_mmap_capabilities,
 841 #endif
 842 };
 843
 844 static const struct file_operations null_fops = {
 845         .llseek         = null_lseek,
 846         .read           = read_null,
 847         .write          = write_null,
 848         .read_iter      = read_iter_null,
 849         .write_iter     = write_iter_null,
 850         .splice_write   = splice_write_null,
 851 };
 852
 853 static const struct file_operations __maybe_unused port_fops = {
 854         .llseek         = memory_lseek,
 855         .read           = read_port,
 856         .write          = write_port,
 857         .open           = open_port,
 858 };
 859
 860 static const struct file_operations zero_fops = {
 861         .llseek         = zero_lseek,
 862         .write          = write_zero,
 863         .read_iter      = read_iter_zero,
 864         .write_iter     = write_iter_zero,
 865         .mmap           = mmap_zero,
 866         .get_unmapped_area = get_unmapped_area_zero,
 867 #ifndef CONFIG_MMU
 868         .mmap_capabilities = zero_mmap_capabilities,
 869 #endif
 870 };
 871
 872 static const struct file_operations full_fops = {
 873         .llseek         = full_lseek,
 874         .read_iter      = read_iter_zero,
 875         .write          = write_full,
 876 };
 877
 878 static const struct memdev {
 879         const char *name;
 880         umode_t mode;
 881         const struct file_operations *fops;
 882         fmode_t fmode;
 883 } devlist[] = {
 884 #ifdef CONFIG_DEVMEM
 885          [1] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
 886 #endif
 887 #ifdef CONFIG_DEVKMEM
 888          [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
 889 #endif
 890          [3] = { "null", 0666, &null_fops, 0 },
 891 #ifdef CONFIG_DEVPORT
 892          [4] = { "port", 0, &port_fops, 0 },
 893 #endif
 894          [5] = { "zero", 0666, &zero_fops, 0 },
 895          [7] = { "full", 0666, &full_fops, 0 },
 896          [8] = { "random", 0666, &random_fops, 0 },
 897          [9] = { "urandom", 0666, &urandom_fops, 0 },
 898 #ifdef CONFIG_PRINTK
 899         [11] = { "kmsg", 0644, &kmsg_fops, 0 },
 900 #endif
 901 };
 902
 903 static int memory_open(struct inode *inode, struct file *filp)
 904 {
 905         int minor;
 906         const struct memdev *dev;
 907
 908         minor = iminor(inode);
 909         if (minor >= ARRAY_SIZE(devlist))
 910                 return -ENXIO;
 911
 912         dev = &devlist[minor];
 913         if (!dev->fops)
 914                 return -ENXIO;
 915
 916         filp->f_op = dev->fops;
 917         filp->f_mode |= dev->fmode;
 918
 919         if (dev->fops->open)
 920                 return dev->fops->open(inode, filp);
 921
 922         return 0;
 923 }
 924
 925 static const struct file_operations memory_fops = {
 926         .open = memory_open,
 927         .llseek = noop_llseek,
 928 };
 929
 930 static char *mem_devnode(struct device *dev, umode_t *mode)
 931 {
 932         if (mode && devlist[MINOR(dev->devt)].mode)
 933                 *mode = devlist[MINOR(dev->devt)].mode;
 934         return NULL;
 935 }
 936
 937 static struct class *mem_class;
 938
 939 static int __init chr_dev_init(void)
 940 {
 941         int minor;
 942
 943         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 944                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 945
 946         mem_class = class_create(THIS_MODULE, "mem");
 947         if (IS_ERR(mem_class))
 948                 return PTR_ERR(mem_class);
 949
 950         mem_class->devnode = mem_devnode;
 951         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 952                 if (!devlist[minor].name)
 953                         continue;
 954
 955                 /*
 956                  * Create /dev/port?
 957                  */
 958                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 959                         continue;
 960
 961                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 962                               NULL, devlist[minor].name);
 963         }
 964
 965         return tty_init();
 966 }
 967
 968 fs_initcall(chr_dev_init);