From: db1@ukc.ac.uk (D.Bolla)
Newsgroups: alt.os.linux
Subject: Patches to have reboot and /dev/kmem working
Summary: How to add a syscall. And various considerations
Message-ID: <5375@falcon.ukc.ac.uk>
Date: 22 Feb 92 13:50:56 GMT
Sender: db1@ukc.ac.uk
Organization: Computing Lab, University of Kent at Canterbury, UK.
Lines: 174



Hello 

This time I have some small solutions :-)

The first one is the implentation of the reboot syscall.
The syscall just does a reboot after having checked that you are root.
The old method does not work anymore ( CTRL-ALT-DEL )

The second is an implementation of the /dev/kmem. Most of the code is 
done by linux. I just added FEW lines in the right place ( I hope ).
The point is that with this /unix and nlist(3) it is possible to get 
all information we need from the kernel WITHOUT having to add code to the 
kernel.

It seems to me that people who complain about "normal users cannot read 
/dev/kmem" and "A library to read kernel data is not nice" don't have a 
clear idea about what /dev/kmem is used for.
When you look in /dev/kmem you look for VERY special information about
the system. I am not sayng that normal user should be denied ps :-)
but if you think, normal users use ps just to see what others are doing
( The same thing for w )

Anyway.... under Sun there are three programs that "look" into /dev/kmem
ps, w, pstats. The amount of information you can get out of them is very 
big. The point that I am tryng to make here is:
To obtain the same amount of information from syscalls you need a 
GREAT number of variations and parameters. Is this worth the work ?
( Considering the small amount of programs that need that information )
Again.....

Negative points of kernel implementation.
1) Extra (Unnecessary) code in the kernel -> Kernel bigger.
2) Possible introduction of ERRORS       -> Kernel panic.
3) Increased number of syscall and paramentrs	-> Complexity
4) Problems on returning lists of data in user space
To me point 2 is already enough NOT to use this method. If you can live
with possible kernel panics.......

Positive points of a library implementation.
1) NO extra code in the kernel.		-> Save memory
2) Impossibility to add ERRORS to the kernel code.	-> NO panic
3) Easy to update, modify even by average user.
4) No problems in allcating memory for the result.

What it is important is to DEFINE what information we want to 
extract from /dev/kmem and who can use them. After that is defined it 
is possible to discuss about implementaion. 

And now to the code :-)

-----------------------------------------------------------------
I decided to put the actual routine that does reboot into
kernel/sys.c Probably your sys.c may be different....

522a523,533
> /* This function reboots the system if the current UID is root
>  * No sync is performed. The optional flag is not used yet
>  * The called routine is in kernel/chr_drv/keyboard.S
>  */
> int sys_reboot ( int flag )
> {
> 	if (!suser())
> 		return (-EPERM);
> 	__asm__("call _hard_reset;" );
> 	return (0);
> }

The next step is to add the syscall code in the file 
include/unistd.h

148a149
> #define __NR_reboot	87

And then add the syscall address in the file
include/linux/sys.h

91a92
> extern int sys_reboot();
108c109
< sys_lstat, sys_readlink, sys_uselib };
---
> sys_lstat, sys_readlink, sys_uselib ,sys_reboot };

The last bit is to modify the keyboard.S file
kernel/chr_drv/keyboard.S
This file is different from yours so.... don't just apply the diff
Understand what it does and do it by hand.

25a26
> .globl _hard_reset
203c204
< 	jne reboot
---
> 	jne 1f			/* CTRL-ALT-DEL			*/	
659c660
< reboot:
---
> _hard_reset:	
661,662c662,663
< 	movw $0x1234,0x472	/* don't do memory check */
< 	movb $0xfc,%al		/* pulse reset and A20 low */
---
> 	movw $0x1234,0x472	/* don't do memory check 	*/
> 	movb $0xfc,%al		/* pulse reset and A20 low 	*/

At this point recompile the kernel and all should go together.
Unfortunately the work is NOT finished. You have to add the syscall to
libc.a To do this you need the source for the library. Usually in 
/usr/src/lib. Go into the unistd directory and create the file
reboot.c with the following content.

/* This function will reboot the syestem
 */
#define __LIBRARY__
#include <unistd.h>

_syscall1(int,reboot,int,param)

And then modify the Makefile as follow.
At the end of OBJ=.... list add reboot.o and do a make dep

Before making the new libary you have to change the standard  unistd.h library
file to point to the unistd.h file in the linux directory
cd /usr/include
rm unistd.h
ln -s /usr/src/linux/include/unistd.h unistd.h

At this point you can make the new libc.a 
Then you can experiment with the new syscall.
Eg:

#include <stdio.h>
#include <errno.h>

main ()
   { 
   if ( reboot (1) ==(-EPERM) )
      printf ("Not superuser \n"); 
   exit (-1);
   }

-------------------------------------------------------------
The second part is a patch that gives you access to the kernel memory
in particular kernel DATA memory.
The following diff applies to the file fs/char_dev.c and in particular
to the function rw_kmem .

52,54c52,67
< {
< 	return -EIO;
< }
---
>    {
>    int i = *pos;	/* Current position where to read	*/
>    
>    /* i can go from 0 to LOW_MEM (See include/linux/mm.h	*/
>    /* I am not shure about it but it doesn't mem fault :-)	*/
>    while ( (count-- > 0)  && (i <LOW_MEM) )
>       {
>       if (rw==READ)  put_fs_byte( *(char *)i ,buf++);
>       else return (-EIO);
>       i++;
>       }
>       
>    i -= *pos;		/* Count how many read or write		*/
>    *pos += i;		/* Update position			*/
>    return (i);		/* Return number read			*/
>    }

NOTE: The LOW_MEM value may not be the most appropriate. Linus please
confirm if it is the correct one.

Damiano