drivers/char/mem.c

   1 /*
   2  *  linux/drivers/char/mem.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *
   6  *  Added devfs support.
   7  *    Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
   8  *  Shared /dev/zero mmaping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
   9  */
  10
  11 #include <linux/config.h>
  12 #include <linux/mm.h>
  13 #include <linux/miscdevice.h>
  14 #include <linux/tpqic02.h>
  15 #include <linux/ftape.h>
  16 #include <linux/slab.h>
  17 #include <linux/vmalloc.h>
  18 #include <linux/mman.h>
  19 #include <linux/random.h>
  20 #include <linux/init.h>
  21 #include <linux/raw.h>
  22 #include <linux/tty.h>
  23 #include <linux/capability.h>
  24 #include <linux/smp_lock.h>
  25 #include <linux/devfs_fs_kernel.h>
  26 #include <linux/ptrace.h>
  27
  28 #include <asm/uaccess.h>
  29 #include <asm/io.h>
  30 #include <asm/pgalloc.h>
  31
  32 #ifdef CONFIG_IA64
  33 # include <linux/efi.h>
  34 #endif
  35
  36 #ifdef CONFIG_FB
  37 extern void fbmem_init(void);
  38 #endif
  39 #if defined(CONFIG_S390_TAPE) && defined(CONFIG_S390_TAPE_CHAR)
  40 extern void tapechar_init(void);
  41 #endif
  42
  43 /*
  44  * Architectures vary in how they handle caching for addresses
  45  * outside of main memory.
  46  *
  47  */
  48 static inline int uncached_access(struct file *file, unsigned long addr)
  49 {
  50 #if defined(__i386__)
  51         /*
  52          * On the PPro and successors, the MTRRs are used to set
  53          * memory types for physical addresses outside main memory,
  54          * so blindly setting PCD or PWT on those pages is wrong.
  55          * For Pentiums and earlier, the surround logic should disable
  56          * caching for the high addresses through the KEN pin, but
  57          * we maintain the tradition of paranoia in this code.
  58          */
  59         if (file->f_flags & O_SYNC)
  60                 return 1;
  61         return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
  62                   test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
  63                   test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
  64                   test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
  65           && addr >= __pa(high_memory);
  66 #elif defined(CONFIG_IA64)
  67         /*
  68          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
  69          */
  70         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
  71 #else
  72         /*
  73          * Accessing memory above the top the kernel knows about or through a file pointer
  74          * that was marked O_SYNC will be done non-cached.
  75          */
  76         if (file->f_flags & O_SYNC)
  77                 return 1;
  78         return addr >= __pa(high_memory);
  79 #endif
  80 }
  81
  82 static ssize_t do_write_mem(struct file * file, void *p, unsigned long realp,
  83                             const char * buf, size_t count, loff_t *ppos)
  84 {
  85         ssize_t written;
  86
  87         written = 0;
  88 #if defined(__sparc__) || (defined(__mc68000__) && defined(CONFIG_MMU))
  89         /* we don't have page 0 mapped on sparc and m68k.. */
  90         if (realp < PAGE_SIZE) {
  91                 unsigned long sz = PAGE_SIZE-realp;
  92                 if (sz > count) sz = count;
  93                 /* Hmm. Do something? */
  94                 buf+=sz;
  95                 p+=sz;
  96                 count-=sz;
  97                 written+=sz;
  98         }
  99 #endif
 100         if (copy_from_user(p, buf, count))
 101                 return -EFAULT;
 102         written += count;
 103         *ppos += written;
 104         return written;
 105 }
 106
 107
 108 /*
 109  * This funcion reads the *physical* memory. The f_pos points directly to the
 110  * memory location.
 111  */
 112 static ssize_t read_mem(struct file * file, char * buf,
 113                         size_t count, loff_t *ppos)
 114 {
 115         unsigned long p = *ppos;
 116         unsigned long end_mem;
 117         ssize_t read;
 118
 119         end_mem = __pa(high_memory);
 120         if (p >= end_mem)
 121                 return 0;
 122         if (count > end_mem - p)
 123                 count = end_mem - p;
 124         read = 0;
 125 #if defined(__sparc__) || (defined(__mc68000__) && defined(CONFIG_MMU))
 126         /* we don't have page 0 mapped on sparc and m68k.. */
 127         if (p < PAGE_SIZE) {
 128                 unsigned long sz = PAGE_SIZE-p;
 129                 if (sz > count)
 130                         sz = count;
 131                 if (sz > 0) {
 132                         if (clear_user(buf, sz))
 133                                 return -EFAULT;
 134                         buf += sz;
 135                         p += sz;
 136                         count -= sz;
 137                         read += sz;
 138                 }
 139         }
 140 #endif
 141         if (copy_to_user(buf, __va(p), count))
 142                 return -EFAULT;
 143         read += count;
 144         *ppos += read;
 145         return read;
 146 }
 147
 148 static ssize_t write_mem(struct file * file, const char * buf,
 149                          size_t count, loff_t *ppos)
 150 {
 151         unsigned long p = *ppos;
 152         unsigned long end_mem;
 153
 154         end_mem = __pa(high_memory);
 155         if (p >= end_mem)
 156                 return 0;
 157         if (count > end_mem - p)
 158                 count = end_mem - p;
 159         return do_write_mem(file, __va(p), p, buf, count, ppos);
 160 }
 161
 162 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
 163 {
 164         unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
 165         int uncached;
 166
 167         uncached = uncached_access(file, offset);
 168 #ifdef pgprot_noncached
 169         if (uncached)
 170                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 171 #endif
 172
 173         /* Don't try to swap out physical pages.. */
 174         vma->vm_flags |= VM_RESERVED;
 175
 176         /*
 177          * Don't dump addresses that are not real memory to a core file.
 178          */
 179         if (uncached)
 180                 vma->vm_flags |= VM_IO;
 181
 182         if (remap_page_range(vma, vma->vm_start, offset, vma->vm_end-vma->vm_start,
 183                              vma->vm_page_prot))
 184                 return -EAGAIN;
 185         return 0;
 186 }
 187
 188 extern long vread(char *buf, char *addr, unsigned long count);
 189 extern long vwrite(char *buf, char *addr, unsigned long count);
 190
 191 /*
 192  * This function reads the *virtual* memory as seen by the kernel.
 193  */
 194 static ssize_t read_kmem(struct file *file, char *buf,
 195                          size_t count, loff_t *ppos)
 196 {
 197         unsigned long p = *ppos;
 198         ssize_t read = 0;
 199         ssize_t virtr = 0;
 200         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 201
 202         if (p < (unsigned long) high_memory) {
 203                 read = count;
 204                 if (count > (unsigned long) high_memory - p)
 205                         read = (unsigned long) high_memory - p;
 206
 207 #if defined(__sparc__) || (defined(__mc68000__) && defined(CONFIG_MMU))
 208                 /* we don't have page 0 mapped on sparc and m68k.. */
 209                 if (p < PAGE_SIZE && read > 0) {
 210                         size_t tmp = PAGE_SIZE - p;
 211                         if (tmp > read) tmp = read;
 212                         if (clear_user(buf, tmp))
 213                                 return -EFAULT;
 214                         buf += tmp;
 215                         p += tmp;
 216                         read -= tmp;
 217                         count -= tmp;
 218                 }
 219 #endif
 220                 if (copy_to_user(buf, (char *)p, read))
 221                         return -EFAULT;
 222                 p += read;
 223                 buf += read;
 224                 count -= read;
 225         }
 226
 227         if (count > 0) {
 228                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 229                 if (!kbuf)
 230                         return -ENOMEM;
 231                 while (count > 0) {
 232                         int len = count;
 233
 234                         if (len > PAGE_SIZE)
 235                                 len = PAGE_SIZE;
 236                         len = vread(kbuf, (char *)p, len);
 237                         if (!len)
 238                                 break;
 239                         if (copy_to_user(buf, kbuf, len)) {
 240                                 free_page((unsigned long)kbuf);
 241                                 return -EFAULT;
 242                         }
 243                         count -= len;
 244                         buf += len;
 245                         virtr += len;
 246                         p += len;
 247                 }
 248                 free_page((unsigned long)kbuf);
 249         }
 250         *ppos = p;
 251         return virtr + read;
 252 }
 253
 254 /*
 255  * This function writes to the *virtual* memory as seen by the kernel.
 256  */
 257 static ssize_t write_kmem(struct file * file, const char * buf,
 258                           size_t count, loff_t *ppos)
 259 {
 260         unsigned long p = *ppos;
 261         ssize_t wrote = 0;
 262         ssize_t virtr = 0;
 263         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 264
 265         if (p < (unsigned long) high_memory) {
 266                 wrote = count;
 267                 if (count > (unsigned long) high_memory - p)
 268                         wrote = (unsigned long) high_memory - p;
 269
 270                 wrote = do_write_mem(file, (void*)p, p, buf, wrote, ppos);
 271
 272                 p += wrote;
 273                 buf += wrote;
 274                 count -= wrote;
 275         }
 276
 277         if (count > 0) {
 278                 kbuf = (char *)__get_free_page(GFP_KERNEL);
 279                 if (!kbuf)
 280                         return -ENOMEM;
 281                 while (count > 0) {
 282                         int len = count;
 283
 284                         if (len > PAGE_SIZE)
 285                                 len = PAGE_SIZE;
 286                         if (len && copy_from_user(kbuf, buf, len)) {
 287                                 free_page((unsigned long)kbuf);
 288                                 return -EFAULT;
 289                         }
 290                         len = vwrite(kbuf, (char *)p, len);
 291                         count -= len;
 292                         buf += len;
 293                         virtr += len;
 294                         p += len;
 295                 }
 296                 free_page((unsigned long)kbuf);
 297         }
 298
 299         *ppos = p;
 300         return virtr + wrote;
 301 }
 302
 303 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 304 static ssize_t read_port(struct file * file, char * buf,
 305                          size_t count, loff_t *ppos)
 306 {
 307         unsigned long i = *ppos;
 308         char *tmp = buf;
 309
 310         if (verify_area(VERIFY_WRITE,buf,count))
 311                 return -EFAULT;
 312         while (count-- > 0 && i < 65536) {
 313                 if (__put_user(inb(i),tmp) < 0)
 314                         return -EFAULT;
 315                 i++;
 316                 tmp++;
 317         }
 318         *ppos = i;
 319         return tmp-buf;
 320 }
 321
 322 static ssize_t write_port(struct file * file, const char * buf,
 323                           size_t count, loff_t *ppos)
 324 {
 325         unsigned long i = *ppos;
 326         const char * tmp = buf;
 327
 328         if (verify_area(VERIFY_READ,buf,count))
 329                 return -EFAULT;
 330         while (count-- > 0 && i < 65536) {
 331                 char c;
 332                 if (__get_user(c, tmp))
 333                         return -EFAULT;
 334                 outb(c,i);
 335                 i++;
 336                 tmp++;
 337         }
 338         *ppos = i;
 339         return tmp-buf;
 340 }
 341 #endif
 342
 343 static ssize_t read_null(struct file * file, char * buf,
 344                          size_t count, loff_t *ppos)
 345 {
 346         return 0;
 347 }
 348
 349 static ssize_t write_null(struct file * file, const char * buf,
 350                           size_t count, loff_t *ppos)
 351 {
 352         return count;
 353 }
 354
 355 #ifdef CONFIG_MMU
 356 /*
 357  * For fun, we are using the MMU for this.
 358  */
 359 static inline size_t read_zero_pagealigned(char * buf, size_t size)
 360 {
 361         struct mm_struct *mm;
 362         struct vm_area_struct * vma;
 363         unsigned long addr=(unsigned long)buf;
 364
 365         mm = current->mm;
 366         /* Oops, this was forgotten before. -ben */
 367         down_read(&mm->mmap_sem);
 368
 369         /* For private mappings, just map in zero pages. */
 370         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
 371                 unsigned long count;
 372
 373                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
 374                         goto out_up;
 375                 if (vma->vm_flags & VM_SHARED)
 376                         break;
 377                 count = vma->vm_end - addr;
 378                 if (count > size)
 379                         count = size;
 380
 381                 zap_page_range(vma, addr, count);
 382                 zeromap_page_range(vma, addr, count, PAGE_COPY);
 383
 384                 size -= count;
 385                 buf += count;
 386                 addr += count;
 387                 if (size == 0)
 388                         goto out_up;
 389         }
 390
 391         up_read(&mm->mmap_sem);
 392
 393         /* The shared case is hard. Let's do the conventional zeroing. */
 394         do {
 395                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
 396                 if (unwritten)
 397                         return size + unwritten - PAGE_SIZE;
 398                 cond_resched();
 399                 buf += PAGE_SIZE;
 400                 size -= PAGE_SIZE;
 401         } while (size);
 402
 403         return size;
 404 out_up:
 405         up_read(&mm->mmap_sem);
 406         return size;
 407 }
 408
 409 static ssize_t read_zero(struct file * file, char * buf,
 410                          size_t count, loff_t *ppos)
 411 {
 412         unsigned long left, unwritten, written = 0;
 413
 414         if (!count)
 415                 return 0;
 416
 417         if (!access_ok(VERIFY_WRITE, buf, count))
 418                 return -EFAULT;
 419
 420         left = count;
 421
 422         /* do we want to be clever? Arbitrary cut-off */
 423         if (count >= PAGE_SIZE*4) {
 424                 unsigned long partial;
 425
 426                 /* How much left of the page? */
 427                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
 428                 unwritten = clear_user(buf, partial);
 429                 written = partial - unwritten;
 430                 if (unwritten)
 431                         goto out;
 432                 left -= partial;
 433                 buf += partial;
 434                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
 435                 written += (left & PAGE_MASK) - unwritten;
 436                 if (unwritten)
 437                         goto out;
 438                 buf += left & PAGE_MASK;
 439                 left &= ~PAGE_MASK;
 440         }
 441         unwritten = clear_user(buf, left);
 442         written += left - unwritten;
 443 out:
 444         return written ? written : -EFAULT;
 445 }
 446
 447 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 448 {
 449         if (vma->vm_flags & VM_SHARED)
 450                 return shmem_zero_setup(vma);
 451         if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
 452                 return -EAGAIN;
 453         return 0;
 454 }
 455 #else /* CONFIG_MMU */
 456 static ssize_t read_zero(struct file * file, char * buf,
 457                          size_t count, loff_t *ppos)
 458 {
 459         size_t todo = count;
 460
 461         while (todo) {
 462                 size_t chunk = todo;
 463
 464                 if (chunk > 4096)
 465                         chunk = 4096;   /* Just for latency reasons */
 466                 if (clear_user(buf, chunk))
 467                         return -EFAULT;
 468                 buf += chunk;
 469                 todo -= chunk;
 470                 cond_resched();
 471         }
 472         return count;
 473 }
 474
 475 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
 476 {
 477         return -ENOSYS;
 478 }
 479 #endif /* CONFIG_MMU */
 480
 481 static ssize_t write_full(struct file * file, const char * buf,
 482                           size_t count, loff_t *ppos)
 483 {
 484         return -ENOSPC;
 485 }
 486
 487 /*
 488  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 489  * can fopen() both devices with "a" now.  This was previously impossible.
 490  * -- SRB.
 491  */
 492
 493 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
 494 {
 495         return file->f_pos = 0;
 496 }
 497
 498 /*
 499  * The memory devices use the full 32/64 bits of the offset, and so we cannot
 500  * check against negative addresses: they are ok. The return value is weird,
 501  * though, in that case (0).
 502  *
 503  * also note that seeking relative to the "end of file" isn't supported:
 504  * it has no meaning, so it returns -EINVAL.
 505  */
 506 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
 507 {
 508         loff_t ret;
 509
 510         down(&file->f_dentry->d_inode->i_sem);
 511         switch (orig) {
 512                 case 0:
 513                         file->f_pos = offset;
 514                         ret = file->f_pos;
 515                         force_successful_syscall_return();
 516                         break;
 517                 case 1:
 518                         file->f_pos += offset;
 519                         ret = file->f_pos;
 520                         force_successful_syscall_return();
 521                         break;
 522                 default:
 523                         ret = -EINVAL;
 524         }
 525         up(&file->f_dentry->d_inode->i_sem);
 526         return ret;
 527 }
 528
 529 static int open_port(struct inode * inode, struct file * filp)
 530 {
 531         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
 532 }
 533
 534 #define mmap_kmem       mmap_mem
 535 #define zero_lseek      null_lseek
 536 #define full_lseek      null_lseek
 537 #define write_zero      write_null
 538 #define read_full       read_zero
 539 #define open_mem        open_port
 540 #define open_kmem       open_mem
 541
 542 static struct file_operations mem_fops = {
 543         .llseek         = memory_lseek,
 544         .read           = read_mem,
 545         .write          = write_mem,
 546         .mmap           = mmap_mem,
 547         .open           = open_mem,
 548 };
 549
 550 static struct file_operations kmem_fops = {
 551         .llseek         = memory_lseek,
 552         .read           = read_kmem,
 553         .write          = write_kmem,
 554         .mmap           = mmap_kmem,
 555         .open           = open_kmem,
 556 };
 557
 558 static struct file_operations null_fops = {
 559         .llseek         = null_lseek,
 560         .read           = read_null,
 561         .write          = write_null,
 562 };
 563
 564 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 565 static struct file_operations port_fops = {
 566         .llseek         = memory_lseek,
 567         .read           = read_port,
 568         .write          = write_port,
 569         .open           = open_port,
 570 };
 571 #endif
 572
 573 static struct file_operations zero_fops = {
 574         .llseek         = zero_lseek,
 575         .read           = read_zero,
 576         .write          = write_zero,
 577         .mmap           = mmap_zero,
 578 };
 579
 580 static struct file_operations full_fops = {
 581         .llseek         = full_lseek,
 582         .read           = read_full,
 583         .write          = write_full,
 584 };
 585
 586 static ssize_t kmsg_write(struct file * file, const char * buf,
 587                           size_t count, loff_t *ppos)
 588 {
 589         char *tmp;
 590         int ret;
 591
 592         tmp = kmalloc(count + 1, GFP_KERNEL);
 593         if (tmp == NULL)
 594                 return -ENOMEM;
 595         ret = -EFAULT;
 596         if (!copy_from_user(tmp, buf, count)) {
 597                 tmp[count] = 0;
 598                 ret = printk("%s", tmp);
 599         }
 600         kfree(tmp);
 601         return ret;
 602 }
 603
 604 static struct file_operations kmsg_fops = {
 605         .write =        kmsg_write,
 606 };
 607
 608 static int memory_open(struct inode * inode, struct file * filp)
 609 {
 610         switch (minor(inode->i_rdev)) {
 611                 case 1:
 612                         filp->f_op = &mem_fops;
 613                         break;
 614                 case 2:
 615                         filp->f_op = &kmem_fops;
 616                         break;
 617                 case 3:
 618                         filp->f_op = &null_fops;
 619                         break;
 620 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 621                 case 4:
 622                         filp->f_op = &port_fops;
 623                         break;
 624 #endif
 625                 case 5:
 626                         filp->f_op = &zero_fops;
 627                         break;
 628                 case 7:
 629                         filp->f_op = &full_fops;
 630                         break;
 631                 case 8:
 632                         filp->f_op = &random_fops;
 633                         break;
 634                 case 9:
 635                         filp->f_op = &urandom_fops;
 636                         break;
 637                 case 11:
 638                         filp->f_op = &kmsg_fops;
 639                         break;
 640                 default:
 641                         return -ENXIO;
 642         }
 643         if (filp->f_op && filp->f_op->open)
 644                 return filp->f_op->open(inode,filp);
 645         return 0;
 646 }
 647
 648 static struct file_operations memory_fops = {
 649         .open           = memory_open,  /* just a selector for the real open */
 650 };
 651
 652 static const struct {
 653         unsigned int            minor;
 654         char                    *name;
 655         umode_t                 mode;
 656         struct file_operations  *fops;
 657 } devlist[] = { /* list of minor devices */
 658         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
 659         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
 660         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
 661 #if defined(CONFIG_ISA) || !defined(__mc68000__)
 662         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
 663 #endif
 664         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
 665         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
 666         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
 667         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
 668         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
 669 };
 670
 671 static int __init chr_dev_init(void)
 672 {
 673         int i;
 674
 675         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
 676                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
 677
 678         for (i = 0; i < ARRAY_SIZE(devlist); i++) {
 679                 devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor),
 680                                 S_IFCHR | devlist[i].mode, devlist[i].name);
 681         }
 682
 683         rand_initialize();
 684 #if defined (CONFIG_FB)
 685         fbmem_init();
 686 #endif
 687         tty_init();
 688 #ifdef CONFIG_M68K_PRINTER
 689         lp_m68k_init();
 690 #endif
 691         misc_init();
 692 #ifdef CONFIG_FTAPE
 693         ftape_init();
 694 #endif
 695         return 0;
 696 }
 697
 698 fs_initcall(chr_dev_init);