Tutorials/Assembler Tutorial

From ThorstensHome
Revision as of 12:03, 2 January 2015 by ThorstenStaerk (Talk | contribs)

Jump to: navigation, search

Machine language is everywhere. Whether you are playing Call of Duty, surf in the internet or write a document - it is machine language that is being executed inside your computer. No matter if you wrote your software in C, BASIC or Ruby, at execution time it has been translated to machine language. Machine language is the godfather of programming languages. Its commands are binary, for example "wait" is 90h for x86. This is why assembler exists - it maps the machine language commands to mnemonics like "jmp" for "jump" or "nop" for "wait". You see this is very low-level and I like low-level topics because I think the deeper it goes the more interesting it becomes. So here I show you how I deal with machine language and assembler. I am using x86 Linux in the examples.


Your first program

Lesson 1's MindMap (click to enlarge)

A "hello world" program in assembler is already advanced. So as a first lesson we will take a look at a program that does nothing but an endless loop. Here is it:


global _start
jmp _start

This assembler source code contains two commands, "nop" for "no operation" and "jmp" for "jump". The other two lines is a label (_start:) and meta-information (global _start saying that "start" is where the program starts).

compile it

Compile it for 64bit Linux:

nasm -f elf64 endless.asm

link it

ld -s -o endless endless.o

execute it

Call the program like this:


Now you will need to press CTRL_C to stop the program. Note that this is possible because there is an operating system giving time slices to the process and the operating system is watching for keypresses still.

disassemble it

Now we want to take a look at the machine language in this program right? Here it is:

 # objdump -M intel -d endless

endless:     file format elf64-x86-64

Disassembly of section .text:

0000000000400080 <.text>:
  400080:       90                      nop
  400081:       eb fd                   jmp    0x400080

So, the byte "90" (hexadecimal) is machine code for "do nothing" or "wait" or "no operation", its assembler mnemonic is "nop". "eb" is "jump" or "jmp", its parameter is where to jump. It is a relative jump, jumping to ff would mean jumping to the same byte, so to the parameter of jump. "jmp fe" means "jump to the jump command" and "jmp fd" means "jump to the byte before the jump command".

What have we seen

So compiling translated the assembler commands into machine language, for example "nop" got translated to the byte 90h. This machine code got embedded into the default file format for a Linux executable, called elf.

Theory: registers and flags

The x86 processor uses registers to store data in it from RAM. Typically the processor would read a byte from RAM into a register, perform an operation on it (like adding a number) and write it back to RAM. The processor has at least one register to point to the RAM address it is executing the code from. This register cannot be modified. The x86 processor also has "normal" registers called eax, ebx, ecx, edx, ebp, edi and esi. They are 32 bit and you can work with them - working means adding, subtracting, comparing and more. The lower parts (16 bits) of eax, ebx, exc and edx are called ax, bx, cx, dx. The lower half (8 bits, one byte) of them are called al, bl, cl and dl. The higher byte of them are called ah, bh, ch and dh. So if you change the value of ah, you will at the same time change the value of ax and eax.

colspan=4 value of eax
higher 16bit value of eax ax=lower 16bit value of eax
higher 16bit value of eax ah=higher 8bit value of ax al=lower 8bit value of eax

The processor also contains registers that tell it where it put variables onto the stack (stack pointer, esp) and where it is executing instructions in memory (instruction pointer, eip). You cannot or should not access these registers.

The processor also contains flags. They are used e.g. for conditional executions. For example if you compare two number you will typically load one number into one register, say ax, and compare it with the other number using the command cmp. Imagine you have one variable in your RAM segment at position 0x04 and the other at 0x08. To compare them you will first load the first variable into register eax:

mov    eax,DWORD PTR [rbp-0x4]

then you will compare eax with the number at position 0x08:

cmp    eax,DWORD PTR [rbp-0x8]

Now the processor's flag will contain the information if the last comparison yielded equal or not equal. So you can now "jump if not equal" with the command jne:

jne    40051d

In C, the data type to fill the register eax is called int, so this is how you would program it in C:

int main()
  int i=0x23;
  int n=0x25;
  if (i==n) {n=i;};

Hello world

We now create a hello world program in C. Then we compile and disassemble it. So we have the C compiler translate it into machine language and then we use a disassembler to translate it into assembler.

Write a C program

This is the program:


#include <stdio.h> 

int main()
  int i=0x23;
  printf("hello world");

Translate the program to machine language

Now we compile it:

gcc hello.c -o hello

and see that it runs:

hello world

disassemble it

To disassemble it we have to possibilities, disassemble to gcc assembler or the Intel dialect. Here are they:

gcc Assembler Syntax Intel Assembler Syntax
# objdump -d hello
000000000040055d <main>:
 40055d:       55                      push   %rbp
 40055e:       48 89 e5                mov    %rsp,%rbp
 400561:       48 83 ec 10             sub    $0x10,%rsp
 400565:       c7 45 fc 23 00 00 00    movl   $0x23,-0x4(%rbp)
 40056c:       bf 04 06 40 00          mov    $0x400604,%edi
 400571:       b8 00 00 00 00          mov    $0x0,%eax
 400576:       e8 c5 fe ff ff          callq  400440 <printf@plt>
 40057b:       c9                      leaveq 
 40057c:       c3                      retq 
# objdump -Mintel -d hello
000000000040053c <main>:
 40053c:       55                      push   rbp
 40053d:       48 89 e5                mov    rbp,rsp
 400540:       48 83 ec 20             sub    rsp,0x20
 400544:       c7 45 fc 23 00 00 00    mov    DWORD PTR [rbp-0x4],0x23
 40054b:       bf 4c 06 40 00          mov    edi,0x40064c            
 400550:       b8 00 00 00 00          mov    eax,0x0                 
 400555:       e8 d6 fe ff ff          call   400430 <printf@plt>     
 40055a:       c9                      leave                          
 40055b:       c3                      ret

Intel assembler vs GNU assembler

We see two assembler representations of the same machine code. The mnemonics are partly different - movl for "move long value", retq instead of ret for "return", callq instead of call for "execute syscall".

How it works

The actual "hello world" string is stored not in the <main> section but in the data section. Note that the "text" section is the "code" section; it is the section that will be executed:

tweedleburg:~ # strings hello
hello world

Before the program is executed the loader will load the needed library functions into the program's memory segment. Here is how you find the library functions:

# nm hello | grep U
                U __libc_start_main@@GLIBC_2.2.5
                U printf@@GLIBC_2.2.5

Then the loader will start __libc_start_main which will call the main function. The main function will call printf and hand over parameters like the address of the string to output in registers. The return code will be handed over as register content.

translate C to assembler

To learn the syntax of a gcc assembler program, let's write a C program and compile it without assembling it. Here is the C program, hello.c:

#include <stdio.h>

int main()
  int i=0x23;
  printf("hello world");

Now we compile this without assembling it:

# gcc -o hello.asm -S hello.c

Now we have the program transformed to assembler and take a look at it:

# cat hello.asm              
        .file   "hello.c"                        
        .section        .rodata                  
        .string "hello world"                    
.globl main                                      
        .type   main, @function                  
        pushq   %rbp                             
        movq    %rsp, %rbp                       
        subq    $32, %rsp                        
        movl    $35, -4(%rbp)                    
        movl    $.LC0, %edi                      
        movl    $0, %eax                         
        call    printf  

Now we know the syntax of gcc assembler and we can finally write a program that consists of an endless loop:

.globl main
  jmp start

Create a boot sector

Under program your own OS I show how to create a boot sector for your own operating system. The bad thing about this is that all functions of your operating system are not available to you and that you only have 512 bytes at your availability. The good thing is that the processor will execute each of your commands while on a running kernel, your program may not have the necessary privileges to run and abort with a segment violation.

Here is our boot sector:

  • create a file kernel.s


; this should print H
    mov ax, 0xe48
    mov bx, 7
    int 0x10
; E
    mov ax, 0xe45
    int 0x10
; L
    mov ax, 0xe4C
    int 0x10
; L
    mov ax, 0xe4C
    int 0x10
; O
    mov ax, 0xe4F
    int 0x10
    jmp .ende

You may note that we say here "mov ax,..." while in the previous example we have seen "mov eax,...". The reason is that there are so many assembler dialects.

  • translate this assembler code into machine language:
nasm kernel.s
  • the result is the file kernel. Let's look at it:
tweedleburg:~ # ll kernel
-rw-r--r-- 1 root root 30 Nov 27 21:29 kernel
tweedleburg:~ # hexdump -C kernel
00000000  b8 48 0e bb 07 00 cd 10  b8 45 0e cd 10 b8 4c 0e  |.H.......E....L.|
00000010  cd 10 b8 4c 0e cd 10 b8  4f 0e cd 10 eb fe        |...L....O.....|
tweedleburg:~ # 

You see the mov ax (or mov eax) assembler command is again translated to b8 as a byte in machine language. You see all assembler commands are translated and there is nothing but machine language in that file. If you want to use this, see programming your own OS.

x86 assembler

In all of this article we are talking about Intel x86 assembler. Only here I want to give you an example for ARM assembler that I got from http://linuxintro.org/wiki/objDump:

Nokia-N810-43-7:~# objdump -d a.out | head

a.out:     file format elf32-littlearm

Disassembly of section .init:

000084f8 <_init>:
    84f8:	e52de004 	str	lr, [sp, #-4]!
    84fc:	e24dd004 	sub	sp, sp, #4	; 0x4
    8500:	eb000035 	bl	85dc <call_gmon_start>
    8504:	e28dd004 	add	sp, sp, #4	; 0x4

The main difference is, as you can see, that this machine language is has a fixed-width command set - every command consists of 4 bytes. That makes it easier to jump to the command that is one, two or whatever commands apart. While the machine language is completely different, there are many assembler mnemonics that also exist in x86 assembler.

run vlc as root

Knowing assembler can help you in situations where you need to do variations to already compiled programs. For example when you start vlc as root you will get an error message

VLC is not supposed to be run as root. Sorry.
If you need to use real-time priorities and/or privileged TCP ports
you can use vlc-wrapper (make sure it is Set-UID root and
cannot be run by non-trusted users first).

If you - like me - do not see why root should not be allowed to run vlc, you can disassemble the code using objdump -d

objdump -d -M intel /usr/bin/vlc
 4010f9:       e8 32 0a 00 00          call   401b30 <unsetenv>
 4010fe:       e8 3d fe ff ff          call   400f40 <geteuid@plt>
 401103:       85 c0                   test   eax,eax
 401105:       0f 84 04 06 00 00       je     40170f <fflush@plt+0x66f>
 40110b:       be ca 1f 40 00          mov    esi,0x401fca
 401110:       bf 06 00 00 00          mov    edi,0x6

As you can see, the program calls the syscall geteuid. The return value is stored in register AX. Then AX is compared against 0 (test eax,eax). If it is 0, the "equal" flag in the processor is set. The next instruction is je ("jump if the equal flag is set"), a conditional jump. The solution is to replace the call to geteuid by a command to set AX to another value but 0, for example

b8 00 00 00 01

and then vlc will always run as if your user id was 1. Based on this I could write http://www.linuxintro.org/wiki/Run_vlc_as_root which shows how to do it in 3 lines of bash code.

what's your name in assembler?

We said that everything that is executed in a computer is machine language. Now, can we execute everything as machine language? What does your name mean if you interpret it as machine language? To find out write a program that contains as many NOP operations as your name contains characters. I am showing this for the name Thorsten:


global _start

compile that program:

nasm -f elf64 name.asm

link it:

ld -s -o name name.o

the nop commands will be translated to bytes with the value 90h. Use a hex editor to replace them with your name:

okteta name


Then save and disassemble the file:

 objdump -M intel -d endless

endless:     file format elf64-x86-64

Disassembly of section .text:

0000000000400080 <.text>:
  400080:       54                      push   rsp
  400081:       68 6f 72 73 74          push   0x7473726f
  400086:       65 6e                   outs   dx,BYTE PTR gs:[rsi]

You see, "T" in machine language is a command to push the register rsp, "h" is byte 68, it is a push command that takes "orst" as parameter and "e", byte 65h, is a command to send something to a computer port (like the serial and parallel port). What a beautiful name I have :)

what's this byte in assembler?

We remember the difference between assembler and machine language - assembler contains human-readable commands like NOP, machine language contains bytes that can be executed by the processor like 90h for NOP.

Now we have translated C into machine language (by compiling it), translated it from machine language to assembler (by disassembling it) and translated from C into assembler (by compiling with the -c option). Now it's time to write some arbitrary bytes and translate them to assembler. I assume that the maximum ammount of parameters an assembler command will use is 4. So let's write a program:

# cat >ass
# objdump -D -b binary -mi386 -Maddr16,data16,intel ass

ass:     file format binary

Disassembly of section .data:

00000000 <.data>:
   0:   41                      inc    cx
   1:   41                      inc    cx
   2:   41                      inc    cx
   3:   41                      inc    cx
   4:   41                      inc    cx
   5:   0a                      .byte 0xa

In other words, we have written text into a file ("AAAAA" plus enter) and told objdump "this is an executable program in machine language, tell us what it does". And the answer is: It increases the register CX five times. "A" or byte 65 or 41h has a meaning in machine language - increase register CX. It does not require an argument.

Going further we can try more bytes and make a table what byte means what in machine language:

  J:   4a                      dec    dx
  K:   4b                      dec    bx
  L:   4c                      dec    sp
  M:   4d                      dec    bp
  N:   4e                      dec    si
  O:   4f                      dec    di

C instructions in assembler

noteable C instructions in assembler:

C code:

int main()
  int i=5;

assembler code:

 4004fd:       55                      push   rbp
 4004fe:       48 89 e5                mov    rbp,rsp
 400501:       48 c7 45 f8 05 00 00    mov    QWORD PTR [rbp-0x8],0x5
 400508:       00 
 400509:       5d                      pop    rbp

C code:

int main()
  int *i;
  *i=(int) 5;

assembler code:

 4004fd:       55                      push   rbp
 4004fe:       48 89 e5                mov    rbp,rsp
 400501:       48 8b 45 f8             mov    rax,QWORD PTR [rbp-0x8]
 400505:       c7 00 05 00 00 00       mov    DWORD PTR [rax],0x5
 40050b:       5d                      pop    rbp

See also