The real way to bootstrap fasm with no prior binary

Index > Linux > The real way to bootstrap fasm with no prior binary

Goto page 1, 2 Next

Author

Thread

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 05 Mar 2023, 04:14

It's ugly and stupid, but it works. here is a preview of a bash script that creates fasm v1.0 without needing any pre-existing binary.

Code:

#!/bin/bash
function p () { for BYTE in $@ ; do printf "\x$BYTE" >>fasm.com ; done ; }

;...

p                             # 0000    org     100h
p                             # 0100    use16
p                             #      
p                             #      start:
p                             #      
p B4 4A                       #         mov     ah,4Ah
p BB 10 10                    # 0102    mov     bx,1010h
p CD 21                       # 0105    int     21h
p BA A2 7D                    # 0107    mov     dx,_logo
p B4 09                       # 010A    mov     ah,9

;...

p 0F 87 51 CF                 # 3735    ja      invalid_operand_size
p 67 AC                       # 3739    lods    byte [esi]
p 3C 28                       # 373B    cmp     al,'('
p 0F 85 36 CF                 # 373D    jne     invalid_operand
p E8 81 DD                    # 3741    call    get_byte_value
p 88 C4                       # 3744    mov     ah,al
p B0 CD                       # 3746    mov     al,CDh
p 67 AB                       # 3748    stos    word [edi]

;...

p                             # 7E23 real_code_size dd ?
p                             #      
p                             # 7E27 start_time dd ?
p                             # 7E2B written_size dd ?
p                             #      
p                             # 7E2F params rb 100h
p                             # 7F2F buffer rb 4000h
p                             #

I used another script to create the file automatically.. The generator script is this:

Code:

#!/bin/bash

mkdir -p worktemp || exit 1
pushd worktemp || exit 1

unzip -LL -n -q ../fasm10.zip || exit 1

ln -s expressi.inc source/expressions.inc
ln -s preproce.inc source/preprocessor.inc
ln -s assemble.inc source/assembler.inc

cd source

cat > make.asm <<PREMACROS || exit 1
irp j,ja,jb,jc,je,jg,jl,jo,jp,js,jz,jae,jbe,jge,jle,jmp,jna,jnb,jnc,jne,jng,jnl,jno,jnp,jns,jnz,jpe,jpo {
        macro j args \\{
                match byte, short \\\\{
                        j args
                \\\\}
        \\}
}
irp v,  Ah,Bh,Ch,Dh,Eh,Fh,\\
        A0h,A1h,A2h,A3h,A4h,A5h,A6h,A7h,A8h,A9h,AAh,ABh,ACh,ADh,AEh,AFh,\\
        B0h,B1h,B2h,B3h,B4h,B5h,B6h,B7h,B8h,B9h,BAh,BBh,BCh,BDh,BEh,BFh,\\
        C0h,C1h,C2h,C3h,C4h,C5h,C6h,C7h,C8h,C9h,CAh,CBh,CCh,CDh,CEh,CFh,\\
        D0h,D1h,D2h,D3h,D4h,D5h,D6h,D7h,D8h,D9h,DAh,DBh,DCh,DDh,DEh,DFh,\\
        E0h,E1h,E2h,E3h,E4h,E5h,E6h,E7h,E8h,E9h,EAh,EBh,ECh,EDh,EEh,EFh,\\
        F0h,F1h,F2h,F3h,F4h,F5h,F6h,F7h,F8h,F9h,FAh,FBh,FCh,FDh,FEh,FFh,\\
        AF0Fh,BA0Fh,D9DEh,E0DFh,FFFFh,FFFFFh,FFFFF0h { v equ 0#v }
include 'fasm.asm'
PREMACROS

fasm -s make.fas make.asm || exit 1
../../listing -a -b 9 make.fas make.dis || exit 1

cat > fasm-make.sh <<PREFIX || exit 1
#!/bin/bash
function p () { for BYTE in \$@ ; do printf "\\x\$BYTE" >>fasm.com ; done ; }
PREFIX
sed 's/.......................\(....\)..\(...........................\)\(.*\)/p \2 # \1 \3/g' < make.dis >> fasm-make.sh || exit 1
bash fasm-make.sh || exit 1
diff fasm.com make.com || exit 1
mv fasm-make.sh ../../ || exit 1
popd || exit 1
rm -r worktemp || exit 1

Of course you do need a working binary to do the initial generation. You could use a secure VM or something if there is a concern that needs to be addressed. But once you have the generated script it can be used stand-alone on another system with no binary.

You need to have a working copy of the listing generator, and fasm to create the .fas file. You might need to adjust the paths in the script.

The output is a 646710 byte file called fasm-make.sh. The preview of it is above. gzipped it is 100439 bytes.

To not abuse the board with large attachments I'll leave off attaching it for now. If requested I can attach it, but you can create your own instead.

05 Mar 2023, 04:14

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 05 Mar 2023, 04:57

For those that need something other than a DOS bootstrap.

The first Win32 console version is 1.04
The first Linux version is 1.37
The first Win32 GUI is version 1.41
The first libc is version 1.64
The first Linux x64 version is 1.71.58

And special mention for those wanting an even older HDOS version then you can start with version 0.90 and work your way up from there.

05 Mar 2023, 04:57

sylware

Joined: 23 Oct 2020
Posts: 484
Location: Marseille/France

sylware 05 Mar 2023, 16:13

bootstraping is a big subject.

Current "elf/linux" stack is a disaster making "bootstraping" literaly insane.

Now you need a linux-like kernel with a modern gcc toolchain which you cannot build without a c++11 compiler... yeah... c++11 compiler, those guys are seriously toxic for the open source software world, like they want to destroy it.

05 Mar 2023, 16:13

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 05 Mar 2023, 18:02

I'm only bootstrapping fasm here.

It is someone else's problem if they also need to bootstrap their OS from zero.

05 Mar 2023, 18:02

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 06 Mar 2023, 00:20

For compatibility with sh the "PREFIX" here string is this:

Code:

#!/bin/sh
p () { for BYTE in \$@ ; do printf \\\\"\$(printf %o 0x\$BYTE)" >>fasm.com ; done ; }

It is a lot slower with the extra 31995 sub-processes that need to be started. Still only a few seconds though, and gives extra availability to systems that have sh.

06 Mar 2023, 00:20

Furs

Joined: 04 Mar 2016
Posts: 2626

Furs 06 Mar 2023, 14:35

sylware wrote:

bootstraping is a big subject.

Current "elf/linux" stack is a disaster making "bootstraping" literaly insane.

Now you need a linux-like kernel with a modern gcc toolchain which you cannot build without a c++11 compiler... yeah... c++11 compiler, those guys are seriously toxic for the open source software world, like they want to destroy it.

GCC can bootstrap itself, but it will take forever.

06 Mar 2023, 14:35

sylware

Joined: 23 Oct 2020
Posts: 484
Location: Marseille/France

sylware 06 Mar 2023, 18:25

The reality is more complex.

before gcc 4.7.4, plain and simple C compilers could bootstrap gcc c++98 compilers. But some geniuses at GNU steering commitee did change that and made gcc to require a c++98 compiler to build.

And gcc 12.2.0 requires a c++11 compiler...

I guess you get the picture now: inductive vendor-lock-in by complexity and size.

Linux kernel: if you don't have the latest gcc extensions, it won't build...

06 Mar 2023, 18:25

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 06 Mar 2023, 19:31

Furs wrote:

GCC can bootstrap itself, but it will take forever.

Being able to self compile is not bootstrapping IMO.

You need something outside of that thing itself to initially create it.

So for GCC (and C compilers in general also I guess) you need something that is not GCC (and is not a C compiler at all) to make the first stage.

06 Mar 2023, 19:31

sylware

Joined: 23 Oct 2020
Posts: 484
Location: Marseille/France

sylware 06 Mar 2023, 20:11

It seems the convention about "The Bootstrap" is the set of binaries (which can build themselves) required to have the mimimum runtime and SDK to build a modern system.

The problem on elf/linux on PC, this set is just insane and disgusting.

Some are trying to clean that up: https://bootstrappable.org but I am pessimistic about it.

If I am not mistaken, they use an assembly-like byte code, to write a lisp interpreter (which seems to be order of magnitude simpler than a C compiler) and then they write a minimal C compiler with this lisp interpreter in order to build tinycc then gcc 4.7.4(c++98 ) then the last gcc you can compile with c++98 which provides c++11 then gcc 12.2.0.

The interpreter of the byte code is binary written (no assembler) specific to a machine ISA, it requires literaly a near 0 runtime.

Personally, I don't write assembly for that bootstrapable thingy: I write assembly because mainstream "high level" languages failed at pertinent syntax stability on the long run and their obscene syntax complexity kills most real-life alternative development effort right from the start: they are a planned obsolescence scam tainted with vendor lock-in. Saying otherwise would be bluntly hypocritical and would be negating reality.

06 Mar 2023, 20:11

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 06 Mar 2023, 20:46

In simple terms, if A needs B and B needs A then it isn't a bootstrap IMO. The chain could be of arbitrary length A-B-C-D-A etc. Any time you return back to a tool you started with then you are stuck in the loop forever. If your C compiler needs a C compiler then you need a C compiler. Erm, yeah, good luck with that. Smile

bootstrappable.org appear to have the right idea.

06 Mar 2023, 20:46

Furs

Joined: 04 Mar 2016
Posts: 2626

Furs 07 Mar 2023, 14:57

revolution wrote:

Furs wrote:
GCC can bootstrap itself, but it will take forever.
Being able to self compile is not bootstrapping IMO.

You need something outside of that thing itself to initially create it.

So for GCC (and C compilers in general also I guess) you need something that is not GCC (and is not a C compiler at all) to make the first stage.

Yeah. You can bootstrap GCC, starting with the very first version, and progressively build more versions of it. That's why I said it will take ages.

07 Mar 2023, 14:57

al_Fazline

Joined: 24 Oct 2018
Posts: 54

al_Fazline 07 Mar 2023, 16:29

Regarding the script, I'm not sure it really qualifies as bootstrapping, the bash script in question is more like fancier version of a hexdump.

Suppose you use base64 of a fasm binary, would it qualify as no prior binary.

07 Mar 2023, 16:29

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 07 Mar 2023, 17:39

A plain hex dump, or base64, aren't auditable. If it is only raw hex then you still need a program on the target system to convert it to binary. I'm not sure there is any way to avoid that, unless you want to manually toggle switches by hand or something. So it might as well be [ba]sh.

With the script that also includes all the assembly and labels and things all in plain text, then any person suitably knowledgable in x86 encoding can certify the correctness, or at a minimum satisfy themselves that they are getting what they expect.

Last edited by revolution on 07 Mar 2023, 22:10; edited 2 times in total

07 Mar 2023, 17:39

sylware

Joined: 23 Oct 2020
Posts: 484
Location: Marseille/France

sylware 07 Mar 2023, 21:52

Yep, and the smaller the better, that's why the binary bootstrap (whic is supposed to be miniscule) is a "hexdump" with comments.

07 Mar 2023, 21:52

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 15 Mar 2023, 21:54

For DOS and Windows it is less clear to me how to make a nicely auditable bootstrap.

debug can be used to create a binary file from an input script, but it doesn't support comments. Sad

Code:

; in bootfasm.txt:
e100 b4 4a bb 10 10 cd 21 ba a2 7d b4 09 cd 21 fc e8
e110 04 01 e8 8a 01 e8 c7 00 80 3e 2f 7e 00 0f 84 b2
;...
rcx
7cfb
n bootfasm.com
w
q

Then:

Code:

debug < bootfasm.txt

That works. But it is just plain hex. How to show it isn't malware by including the accompanying source?

15 Mar 2023, 21:54

Tomasz Grysztar

Joined: 16 Jun 2003
Posts: 8397
Location: Kraków, Poland

Tomasz Grysztar 15 Mar 2023, 21:59

revolution wrote:

That works. But it is just plain hex. How to show it isn't malware by including the accompanying source?

Maybe a batch file with interleaved comments and ECHO commands that would generate the bootfasm.txt and finally launch DEBUG?

15 Mar 2023, 21:59

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 15 Mar 2023, 22:22

Like this?

Code:

rem                                     0000    org     100h
rem                                     0100    use16
rem
rem                                          start:
rem
rem                                             mov     ah,4Ah
echo e0100 B4 4A          > bootfasm.txt
rem                                     0102    mov     bx,1010h
echo e0102 BB 10 10      >> bootfasm.txt
rem                                     0105    int     21h
echo e0105 CD 21         >> bootfasm.txt
rem                                     0107    mov     dx,_logo
echo e0107 BA A2 7D      >> bootfasm.txt
rem                                     010A    mov     ah,9
echo e010A B4 09         >> bootfasm.txt
;...
echo rcx                 >> bootfasm.txt
echo 7CFB                >> bootfasm.txt
echo n bootfasm.com      >> bootfasm.txt
echo w                   >> bootfasm.txt
echo q                   >> bootfasm.txt
debug < bootfasm.txt

The formatting needs some work. But, yes, all the elements are there. Smile

15 Mar 2023, 22:22

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 17 Mar 2023, 21:47

So we went with this script using awk.

Code:

#!/bin/bash

mkdir -p worktemp || exit 1
pushd worktemp || exit 1

unzip -LL -n -q ../fasm10.zip || exit 1

ln -s expressi.inc source/expressions.inc
ln -s preproce.inc source/preprocessor.inc
ln -s assemble.inc source/assembler.inc

cd source || exit 1

cat > make.asm <<PREMACROS || exit 1
irp j,ja,jb,jc,je,jg,jl,jo,jp,js,jz,jae,jbe,jge,jle,jmp,jna,jnb,\\
    jnc,jne,jng,jnl,jno,jnp,jns,jnz,jpe,jpo,jnae,jnbe,jnge,jnle {
        macro j args \\{
                match byte, short \\\\{
                        j args
                \\\\}
        \\}
}
irp v,  Ah,Bh,Ch,Dh,Eh,Fh,\\
        A0h,A1h,A2h,A3h,A4h,A5h,A6h,A7h,A8h,A9h,AAh,ABh,ACh,ADh,AEh,AFh,\\
        B0h,B1h,B2h,B3h,B4h,B5h,B6h,B7h,B8h,B9h,BAh,BBh,BCh,BDh,BEh,BFh,\\
        C0h,C1h,C2h,C3h,C4h,C5h,C6h,C7h,C8h,C9h,CAh,CBh,CCh,CDh,CEh,CFh,\\
        D0h,D1h,D2h,D3h,D4h,D5h,D6h,D7h,D8h,D9h,DAh,DBh,DCh,DDh,DEh,DFh,\\
        E0h,E1h,E2h,E3h,E4h,E5h,E6h,E7h,E8h,E9h,EAh,EBh,ECh,EDh,EEh,EFh,\\
        F0h,F1h,F2h,F3h,F4h,F5h,F6h,F7h,F8h,F9h,FAh,FBh,FCh,FDh,FEh,FFh,\\
        AF0Fh,BA0Fh,D9DEh,E0DFh,FFFFh,FFFFFh,FFFFF0h { v equ 0#v }
include 'fasm.asm'
PREMACROS

fasm -s make.fas make.asm || exit 1
../../listing -a -b 9 make.fas make.lst || exit 1

awk 'BEGIN{
    print "@echo off                              >fasmmake.txt\r";
  }{
    addr=substr($0,24,4); hex=substr($0,30,26); source=gensub("\r","","g",substr($0,57)); bytes=int(length(gensub(" ","","g",hex))/2);
    eaddr=addr; if (addr == "    ") eaddr=prev_addr;
    prev_addr=sprintf("%04X",strtonum("0x" eaddr)+bytes);
    if (bytes == 0)
      printf "%58s %s %s\r\n", "rem", addr, source;
    else{
      print "echo e" eaddr, hex, ">>fasmmake.txt & rem", addr, source "\r";
      end_addr=prev_addr;}
  }END{
    print  "echo r cx                             >>fasmmake.txt\r";
    printf "echo %04X                             >>fasmmake.txt\r\n", strtonum("0x" end_addr)-0x100;
    print  "echo n fasm10.com                     >>fasmmake.txt\r";
    print  "echo w                                >>fasmmake.txt\r";
    print  "echo q                                >>fasmmake.txt\r";
    print  "debug < fasmmake.txt\r";
  }' < make.lst >> fasmmake.bat || exit 1

mv fasmmake.bat ../../ || exit 1
popd || exit 1
rm -r worktemp || exit 1

It generates a reasonably readable output IMO.

Code:

;...
echo e010A B4 09                      >>fasmmake.txt & rem 010A         mov     ah,9
echo e010C CD 21                      >>fasmmake.txt & rem 010C         int     21h
                                                       rem      
echo e010E FC                         >>fasmmake.txt & rem 010E         cld
                                                       rem      
echo e010F E8 04 01                   >>fasmmake.txt & rem 010F         call    init_flatrm
echo e0112 E8 8A 01                   >>fasmmake.txt & rem 0112         call    init_memory
                                                       rem      
echo e0115 E8 C7 00                   >>fasmmake.txt & rem 0115         call    get_params
echo e0118 80 3E 2F 7E 00             >>fasmmake.txt & rem 0118         cmp     [params],0
echo e011D 0F 84 B2 00                >>fasmmake.txt & rem 011D         je      information
echo e0121 66 8D 06 30 7E             >>fasmmake.txt & rem 0121         lea     eax,[params+1]
;...
                                                       rem 7E2F params rb 100h
                                                       rem 7F2F buffer rb 4000h
                                                       rem      
                                                       rem      
echo r cx                             >>fasmmake.txt
echo 7CFB                             >>fasmmake.txt
echo n fasm10.com                     >>fasmmake.txt
echo w                                >>fasmmake.txt
echo q                                >>fasmmake.txt
debug < fasmmake.txt

Now to submit and see if we can get this across the hatchway. Confused

17 Mar 2023, 21:47

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 27 Mar 2023, 07:16

Assemble fasm v1.0 with an 8kB bootstrap.

The original 32002 byte file is too large to easily audit (apparently). It has many unused parts that just take up space with no functional value.

Things like the MMX instructions, and error detection, can be removed to leave only purely functional code where all parts are used and all conditional jumps are exercised in both paths. Only instructions used by fasm v1.0 are included. And further, only the used addressing modes and variants are able to be assembled.

The attached is a ~60kB source file that achieves this. It assembler to 8192 bytes.

Code:

flat assembler  version 1.73.08  (3916144 kilobytes memory)
6 passes, 8192 bytes.

It recreates exactly the original 32002 byte output when used to assemble fasm v1.0. It has a single purpose: to assemble fasm v1.0. If you give it any other input it likely won't work, and will probably crash. It might be fun to watch DOS crash if you want to try. Razz

Code:

Archive:  fasm8k.zip
Zip file size: 12306 bytes, number of entries: 1
-rw-rw-r--  3.0 unx    59388 t- defX 20230327.000000 fasm8k.asm
1 file, 59388 bytes uncompressed, 12188 bytes compressed:  79.5%

Edit: Testing shows that fasm8k can create exactly matching outputs from input sources fasm v1.0, v1.01, v1.02 & v1.03. It fails at v1.04.

Description:	Assemble fasm v1.0 with an 8kB bootstrap	Download
Filename:	fasm8k.zip
Filesize:	12.02 KB
Downloaded:	329 Time(s)

27 Mar 2023, 07:16

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20632
Location: In your JS exploiting you and your system

revolution 28 Mar 2023, 09:02

Here is one path to build fasm from zero all the way to v1.73.30 using just DOS on the target machine.

Code:

rem Build fasm in DOS

rem     folder layout
rem     -------------
rem     fasm8k.com      bootstrapped fasm8k v0.0
rem     makefasm.bat    this batch file
rem     cwsdpmi.exe     in case you need a DPMI server for v1.66+
rem     fasm10\         v1.0 source
rem     fasm111\        v1.11 source
rem     fasm120\        v1.20 source
rem     fasm140\        v1.40 source
rem     fasm166\        v1.66 source
rem     fasm7330\       v1.73.30 source

cd fasm10\source
..\..\fasm8k.com fasm.asm fasm.com

rem bug fix for large jmp/call
echo 8000h equ 10000h    > make10x.asm
echo include 'fasm.asm' >> make10x.asm
fasm.com make10x.asm ..\..\fasm10x.com

cd ..\..\fasm111\source
..\..\fasm10x.com fasm.asm ..\..\fasm111.com

cd ..\..\fasm120\source\dos
rem 'jmp dword ...' to 'jmp ...'
echo saved_dword equ dword      > make120.asm
echo macro jmp dest {dword equ >> make120.asm
echo jmp dest                  >> make120.asm
echo dword equ saved_dword}    >> make120.asm
rem display incuded at v1.13
echo macro display x {}        >> make120.asm
echo include 'fasm.asm'        >> make120.asm
..\..\..\fasm111.com make120.asm ..\..\..\fasm120.com

cd ..\..\..\fasm140\source\dos
..\..\..\fasm120.com fasm.asm ..\..\..\fasm140.com

cd ..\..\..\fasm166\source\dos
rem salc renamed at v1.41
echo salc equ setalc                   > make166.asm
rem align included at v1.50
echo macro align v{rb v-1-($-1)mod v} >> make166.asm
rem 'jcc byte' axed at v1.65.20
echo short equ byte                   >> make166.asm
echo include 'fasm.asm'               >> make166.asm
..\..\..\fasm140.com make166.asm ..\..\..\fasm166.exe

cd ..\..\..\fasm7330\source\dos
..\..\..\cwsdpmi.exe
..\..\..\fasm166.exe fasm.asm ..\..\..\fasm7330.exe

cd ..\..\..\

Each stage of the build fixes a bug, bypasses an enhancement, or overcomes a limitation before we can proceed to the next stage.

The final output is almost an exact match to the v1.73.30 executable. The single difference is an unoptimised forward jump that compiles to 6 bytes, instead of 2, where the target is 0x7e bytes ahead. If you need to get the output precisely matching then the final fasm7330.exe can be used to assemble itself and then it is exactly matching.

28 Mar 2023, 09:02

Goto page 1, 2 Next

< Last Thread | Next Thread >

Forum Rules:

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You cannot attach files in this forum
You can download files in this forum