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         /* It's illegal to wrap around the end of the physical address space. */
 364         if (offset + (phys_addr_t)size - 1 < offset)
 365                 return -EINVAL;
 366
 367         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 368                 return -EINVAL;
 369
 370         if (!private_mapping_ok(vma))
 371                 return -ENOSYS;
 372
 373         if (!range_is_allowed(vma->vm_pgoff, size))
 374                 return -EPERM;
 375
 376         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 377                                                 &vma->vm_page_prot))
 378                 return -EINVAL;
 379
 380         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 381                                                  size,
 382                                                  vma->vm_page_prot);
 383
 384         vma->vm_ops = &mmap_mem_ops;
 385
 386         /* Remap-pfn-range will mark the range VM_IO */
 387         if (remap_pfn_range(vma,
 388                             vma->vm_start,
 389                             vma->vm_pgoff,
 390                             size,
 391                             vma->vm_page_prot)) {
 392                 return -EAGAIN;
 393         }
 394         return 0;
 395 }
 396
 397 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 398 {
 399         unsigned long pfn;
 400
 401         /* Turn a kernel-virtual address into a physical page frame */
 402         pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 403
 404         /*
 405          * RED-PEN: on some architectures there is more mapped memory than
 406          * available in mem_map which pfn_valid checks for. Perhaps should add a
 407          * new macro here.
 408          *
 409          * RED-PEN: vmalloc is not supported right now.
 410          */
 411         if (!pfn_valid(pfn))
 412                 return -EIO;
 413
 414         vma->vm_pgoff = pfn;
 415         return mmap_mem(file, vma);
 416 }
 417
 418 /*
 419  * This function reads the *virtual* memory as seen by the kernel.
 420  */
 421 static ssize_t read_kmem(struct file *file, char __user *buf,
 422                          size_t count, loff_t *ppos)
 423 {
 424         unsigned long p = *ppos;
 425         ssize_t low_count, read, sz;
 426         char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 427         int err = 0;
 428
 429         read = 0;
 430         if (p < (unsigned long) high_memory) {
 431                 low_count = count;
 432                 if (count > (unsigned long)high_memory - p)
 433                         low_count = (unsigned long)high_memory - p;
 434
 435 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 436                 /* we don't have page 0 mapped on sparc and m68k.. */
 437                 if (p < PAGE_SIZE && low_count > 0) {
 438                         sz = size_inside_page(p, low_count);
 439                         if (clear_user(buf, sz))
 440                                 return -EFAULT;
 441                         buf += sz;
 442                         p += sz;
 443                         read += sz;
 444                         low_count -= sz;
 445                         count -= sz;
 446                 }
 447 #endif
 448                 while (low_count > 0) {
 449                         sz = size_inside_page(p, low_count);
 450
 451                         /*
 452                          * On ia64 if a page has been mapped somewhere as
 453                          * uncached, then it must also be accessed uncached
 454                          * by the kernel or data corruption may occur
 455                          */
 456                         kbuf = xlate_dev_kmem_ptr((void *)p);
 457                         if (!virt_addr_valid(kbuf))
 458                                 return -ENXIO;
 459
 460                         if (copy_to_user(buf, kbuf, sz))
 461                                 return -EFAULT;
 462                         buf += sz;
 463                         p += sz;
 464                         read += sz;
 465                         low_count -= sz;
 466                         count -= sz;
 467                 }
 468         }
 469
 470         if (count > 0) {
 471                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 472                 if (!kbuf)
 473                         return -ENOMEM;
 474                 while (count > 0) {
 475                         sz = size_inside_page(p, count);
 476                         if (!is_vmalloc_or_module_addr((void *)p)) {
 477                                 err = -ENXIO;
 478                                 break;
 479                         }
 480                         sz = vread(kbuf, (char *)p, sz);
 481                         if (!sz)
 482                                 break;
 483                         if (copy_to_user(buf, kbuf, sz)) {
 484                                 err = -EFAULT;
 485                                 break;
 486                         }
 487                         count -= sz;
 488                         buf += sz;
 489                         read += sz;
 490                         p += sz;
 491                 }
 492                 free_page((unsigned long)kbuf);
 493         }
 494         *ppos = p;
 495         return read ? read : err;
 496 }
 497
 498
 499 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 500                                 size_t count, loff_t *ppos)
 501 {
 502         ssize_t written, sz;
 503         unsigned long copied;
 504
 505         written = 0;
 506 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 507         /* we don't have page 0 mapped on sparc and m68k.. */
 508         if (p < PAGE_SIZE) {
 509                 sz = size_inside_page(p, count);
 510                 /* Hmm. Do something? */
 511                 buf += sz;
 512                 p += sz;
 513                 count -= sz;
 514                 written += sz;
 515         }
 516 #endif
 517
 518         while (count > 0) {
 519                 void *ptr;
 520
 521                 sz = size_inside_page(p, count);
 522
 523                 /*
 524                  * On ia64 if a page has been mapped somewhere as uncached, then
 525                  * it must also be accessed uncached by the kernel or data
 526                  * corruption may occur.
 527                  */
 528                 ptr = xlate_dev_kmem_ptr((void *)p);
 529                 if (!virt_addr_valid(ptr))
 530                         return -ENXIO;
 531
 532                 copied = copy_from_user(ptr, buf, sz);
 533                 if (copied) {
 534                         written += sz - copied;
 535                         if (written)
 536                                 break;
 537                         return -EFAULT;
 538                 }
 539                 buf += sz;
 540                 p += sz;
 541                 count -= sz;
 542                 written += sz;
 543         }
 544
 545         *ppos += written;
 546         return written;
 547 }
 548
 549 /*
 550  * This function writes to the *virtual* memory as seen by the kernel.
 551  */
 552 static ssize_t write_kmem(struct file *file, const char __user *buf,
 553                           size_t count, loff_t *ppos)
 554 {
 555         unsigned long p = *ppos;
 556         ssize_t wrote = 0;
 557         ssize_t virtr = 0;
 558         char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 559         int err = 0;
 560
 561         if (p < (unsigned long) high_memory) {
 562                 unsigned long to_write = min_t(unsigned long, count,
 563                                                (unsigned long)high_memory - p);
 564                 wrote = do_write_kmem(p, buf, to_write, ppos);
 565                 if (wrote != to_write)
 566                         return wrote;
 567                 p += wrote;
 568                 buf += wrote;
 569                 count -= wrote;
 570         }
 571
 572         if (count > 0) {
 573                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 574                 if (!kbuf)
 575                         return wrote ? wrote : -ENOMEM;
 576                 while (count > 0) {
 577                         unsigned long sz = size_inside_page(p, count);
 578                         unsigned long n;
 579
 580                         if (!is_vmalloc_or_module_addr((void *)p)) {
 581                                 err = -ENXIO;
 582                                 break;
 583                         }
 584                         n = copy_from_user(kbuf, buf, sz);
 585                         if (n) {
 586                                 err = -EFAULT;
 587                                 break;
 588                         }
 589                         vwrite(kbuf, (char *)p, sz);
 590                         count -= sz;
 591                         buf += sz;
 592                         virtr += sz;
 593                         p += sz;
 594                 }
 595                 free_page((unsigned long)kbuf);
 596         }
 597
 598         *ppos = p;
 599         return virtr + wrote ? : err;
 600 }
 601
 602 static ssize_t read_port(struct file *file, char __user *buf,
 603                          size_t count, loff_t *ppos)
 604 {
 605         unsigned long i = *ppos;
 606         char __user *tmp = buf;
 607
 608         if (!access_ok(VERIFY_WRITE, buf, count))
 609                 return -EFAULT;
 610         while (count-- > 0 && i < 65536) {
 611                 if (__put_user(inb(i), tmp) < 0)
 612                         return -EFAULT;
 613                 i++;
 614                 tmp++;
 615         }
 616         *ppos = i;
 617         return tmp-buf;
 618 }
 619
 620 static ssize_t write_port(struct file *file, const char __user *buf,
 621                           size_t count, loff_t *ppos)
 622 {
 623         unsigned long i = *ppos;
 624         const char __user *tmp = buf;
 625
 626         if (!access_ok(VERIFY_READ, buf, count))
 627                 return -EFAULT;
 628         while (count-- > 0 && i < 65536) {
 629                 char c;
 630
 631                 if (__get_user(c, tmp)) {
 632                         if (tmp > buf)
 633                                 break;
 634                         return -EFAULT;
 635                 }
 636                 outb(c, i);
 637                 i++;
 638                 tmp++;
 639         }
 640         *ppos = i;
 641         return tmp-buf;
 642 }
 643
 644 static ssize_t read_null(struct file *file, char __user *buf,
 645                          size_t count, loff_t *ppos)
 646 {
 647         return 0;
 648 }
 649
 650 static ssize_t write_null(struct file *file, const char __user *buf,
 651                           size_t count, loff_t *ppos)
 652 {
 653         return count;
 654 }
 655
 656 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
 657 {
 658         return 0;
 659 }
 660
 661 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
 662 {
 663         size_t count = iov_iter_count(from);
 664         iov_iter_advance(from, count);
 665         return count;
 666 }
 667
 668 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 669                         struct splice_desc *sd)
 670 {
 671         return sd->len;
 672 }
 673
 674 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 675                                  loff_t *ppos, size_t len, unsigned int flags)
 676 {
 677         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 678 }
 679
 680 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
 681 {
 682         size_t written = 0;
 683
 684         while (iov_iter_count(iter)) {
 685                 size_t chunk = iov_iter_count(iter), n;
 686
 687                 if (chunk > PAGE_SIZE)
 688                         chunk = PAGE_SIZE;      /* Just for latency reasons */
 689                 n = iov_iter_zero(chunk, iter);
 690                 if (!n && iov_iter_count(iter))
 691                         return written ? written : -EFAULT;
 692                 written += n;
 693                 if (signal_pending(current))
 694                         return written ? written : -ERESTARTSYS;
 695                 cond_resched();
 696         }
 697         return written;
 698 }
 699
 700 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 701 {
 702 #ifndef CONFIG_MMU
 703         return -ENOSYS;
 704 #endif
 705         if (vma->vm_flags & VM_SHARED)
 706                 return shmem_zero_setup(vma);
 707         return 0;
 708 }
 709
 710 static unsigned long get_unmapped_area_zero(struct file *file,
 711                                 unsigned long addr, unsigned long len,
 712                                 unsigned long pgoff, unsigned long flags)
 713 {
 714 #ifdef CONFIG_MMU
 715         if (flags & MAP_SHARED) {
 716                 /*
 717                  * mmap_zero() will call shmem_zero_setup() to create a file,
 718                  * so use shmem's get_unmapped_area in case it can be huge;
 719                  * and pass NULL for file as in mmap.c's get_unmapped_area(),
 720                  * so as not to confuse shmem with our handle on "/dev/zero".
 721                  */
 722                 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
 723         }
 724
 725         /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
 726         return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
 727 #else
 728         return -ENOSYS;
 729 #endif
 730 }
 731
 732 static ssize_t write_full(struct file *file, const char __user *buf,
 733                           size_t count, loff_t *ppos)
 734 {
 735         return -ENOSPC;
 736 }
 737
 738 /*
 739  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 740  * can fopen() both devices with "a" now.  This was previously impossible.
 741  * -- SRB.
 742  */
 743 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 744 {
 745         return file->f_pos = 0;
 746 }
 747
 748 /*
 749  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 750  * check against negative addresses: they are ok. The return value is weird,
 751  * though, in that case (0).
 752  *
 753  * also note that seeking relative to the "end of file" isn't supported:
 754  * it has no meaning, so it returns -EINVAL.
 755  */
 756 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 757 {
 758         loff_t ret;
 759
 760         inode_lock(file_inode(file));
 761         switch (orig) {
 762         case SEEK_CUR:
 763                 offset += file->f_pos;
 764         case SEEK_SET:
 765                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 766                 if ((unsigned long long)offset >= -MAX_ERRNO) {
 767                         ret = -EOVERFLOW;
 768                         break;
 769                 }
 770                 file->f_pos = offset;
 771                 ret = file->f_pos;
 772                 force_successful_syscall_return();
 773                 break;
 774         default:
 775                 ret = -EINVAL;
 776         }
 777         inode_unlock(file_inode(file));
 778         return ret;
 779 }
 780
 781 static int open_port(struct inode *inode, struct file *filp)
 782 {
 783         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 784 }
 785
 786 #define zero_lseek      null_lseek
 787 #define full_lseek      null_lseek
 788 #define write_zero      write_null
 789 #define write_iter_zero write_iter_null
 790 #define open_mem        open_port
 791 #define open_kmem       open_mem
 792
 793 static const struct file_operations __maybe_unused mem_fops = {
 794         .llseek         = memory_lseek,
 795         .read           = read_mem,
 796         .write          = write_mem,
 797         .mmap           = mmap_mem,
 798         .open           = open_mem,
 799 #ifndef CONFIG_MMU
 800         .get_unmapped_area = get_unmapped_area_mem,
 801         .mmap_capabilities = memory_mmap_capabilities,
 802 #endif
 803 };
 804
 805 static const struct file_operations __maybe_unused kmem_fops = {
 806         .llseek         = memory_lseek,
 807         .read           = read_kmem,
 808         .write          = write_kmem,
 809         .mmap           = mmap_kmem,
 810         .open           = open_kmem,
 811 #ifndef CONFIG_MMU
 812         .get_unmapped_area = get_unmapped_area_mem,
 813         .mmap_capabilities = memory_mmap_capabilities,
 814 #endif
 815 };
 816
 817 static const struct file_operations null_fops = {
 818         .llseek         = null_lseek,
 819         .read           = read_null,
 820         .write          = write_null,
 821         .read_iter      = read_iter_null,
 822         .write_iter     = write_iter_null,
 823         .splice_write   = splice_write_null,
 824 };
 825
 826 static const struct file_operations __maybe_unused port_fops = {
 827         .llseek         = memory_lseek,
 828         .read           = read_port,
 829         .write          = write_port,
 830         .open           = open_port,
 831 };
 832
 833 static const struct file_operations zero_fops = {
 834         .llseek         = zero_lseek,
 835         .write          = write_zero,
 836         .read_iter      = read_iter_zero,
 837         .write_iter     = write_iter_zero,
 838         .mmap           = mmap_zero,
 839         .get_unmapped_area = get_unmapped_area_zero,
 840 #ifndef CONFIG_MMU
 841         .mmap_capabilities = zero_mmap_capabilities,
 842 #endif
 843 };
 844
 845 static const struct file_operations full_fops = {
 846         .llseek         = full_lseek,
 847         .read_iter      = read_iter_zero,
 848         .write          = write_full,
 849 };
 850
 851 static const struct memdev {
 852         const char *name;
 853         umode_t mode;
 854         const struct file_operations *fops;
 855         fmode_t fmode;
 856 } devlist[] = {
 857 #ifdef CONFIG_DEVMEM
 858          [1] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
 859 #endif
 860 #ifdef CONFIG_DEVKMEM
 861          [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
 862 #endif
 863          [3] = { "null", 0666, &null_fops, 0 },
 864 #ifdef CONFIG_DEVPORT
 865          [4] = { "port", 0, &port_fops, 0 },
 866 #endif
 867          [5] = { "zero", 0666, &zero_fops, 0 },
 868          [7] = { "full", 0666, &full_fops, 0 },
 869          [8] = { "random", 0666, &random_fops, 0 },
 870          [9] = { "urandom", 0666, &urandom_fops, 0 },
 871 #ifdef CONFIG_PRINTK
 872         [11] = { "kmsg", 0644, &kmsg_fops, 0 },
 873 #endif
 874 };
 875
 876 static int memory_open(struct inode *inode, struct file *filp)
 877 {
 878         int minor;
 879         const struct memdev *dev;
 880
 881         minor = iminor(inode);
 882         if (minor >= ARRAY_SIZE(devlist))
 883                 return -ENXIO;
 884
 885         dev = &devlist[minor];
 886         if (!dev->fops)
 887                 return -ENXIO;
 888
 889         filp->f_op = dev->fops;
 890         filp->f_mode |= dev->fmode;
 891
 892         if (dev->fops->open)
 893                 return dev->fops->open(inode, filp);
 894
 895         return 0;
 896 }
 897
 898 static const struct file_operations memory_fops = {
 899         .open = memory_open,
 900         .llseek = noop_llseek,
 901 };
 902
 903 static char *mem_devnode(struct device *dev, umode_t *mode)
 904 {
 905         if (mode && devlist[MINOR(dev->devt)].mode)
 906                 *mode = devlist[MINOR(dev->devt)].mode;
 907         return NULL;
 908 }
 909
 910 static struct class *mem_class;
 911
 912 static int __init chr_dev_init(void)
 913 {
 914         int minor;
 915
 916         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
 917                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 918
 919         mem_class = class_create(THIS_MODULE, "mem");
 920         if (IS_ERR(mem_class))
 921                 return PTR_ERR(mem_class);
 922
 923         mem_class->devnode = mem_devnode;
 924         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
 925                 if (!devlist[minor].name)
 926                         continue;
 927
 928                 /*
 929                  * Create /dev/port?
 930                  */
 931                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 932                         continue;
 933
 934                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
 935                               NULL, devlist[minor].name);
 936         }
 937
 938         return tty_init();
 939 }
 940
 941 fs_initcall(chr_dev_init);