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