ipt_MARK/xt_MARK usage problem

ipt_MARK/xt_MARK usage problem

[Jan Engelhardt]

Hello,


kernel 2.6.17 ships with xt_MARK, but iptables 1.3.5 still uses ipt_MARK. 
In essence, I cannot use `iptables -j MARK` giving me 

# iptables -A INPUT -j MARK --set-mark 1
iptables: Unknown error 4294967295

I have seen this before and the problem behind this strange error (-1) is 
that the .targetsize/.matchsize variables in the kernel modules do not 
match their userspace parts.

However, this time it seems to be something different:

# iptables -t mangle -A INPUT -j MARK --set-mark 1

Works without problems. Am I missing something?
How do I get MARK back to work in -t filter -- possibly without hacking in 
xt_MARK.c?




Jan Engelhardt
-- 

Re: ipt_MARK/xt_MARK usage problem

[Patrick McHardy]

Jan Engelhardt wrote:
> kernel 2.6.17 ships with xt_MARK, but iptables 1.3.5 still uses ipt_MARK. 
> In essence, I cannot use `iptables -j MARK` giving me 
> 
> # iptables -A INPUT -j MARK --set-mark 1
> iptables: Unknown error 4294967295
> 
> I have seen this before and the problem behind this strange error (-1) is 
> that the .targetsize/.matchsize variables in the kernel modules do not 
> match their userspace parts.

No, its a bug in the iptables userspace version you're using, which
makes it report any error as "Unknown error 4294967295". The
error itself is that you're using MARK in the filter table.

> However, this time it seems to be something different:
> 
> # iptables -t mangle -A INPUT -j MARK --set-mark 1
> 
> Works without problems. Am I missing something?
> How do I get MARK back to work in -t filter -- possibly without hacking in 
> xt_MARK.c?

You won't, its not supposed to work in the filter table.

Re: ipt_MARK/xt_MARK usage problem

[Jan Engelhardt]

>
>No, its a bug in the iptables userspace version you're using, which
>makes it report any error as "Unknown error 4294967295". The
>error itself is that you're using MARK in the filter table.
>
>> Works without problems. Am I missing something?
>> How do I get MARK back to work in -t filter -- possibly without hacking in 
>> xt_MARK.c?
>
>You won't, its not supposed to work in the filter table.

This worked in 2.6.16, where the ipt_mark_reg_v1 strucutre in xt_MARK.c did 
not have a .table limiter. (It also worked in practice, i.e. packets got 
marked like they should.) I do not see why this was changed.


Jan Engelhardt
-- 

Re: ipt_MARK/xt_MARK usage problem

[Patrick McHardy]

Jan Engelhardt wrote:
>>>How do I get MARK back to work in -t filter -- possibly without hacking in 
>>>xt_MARK.c?
>>
>>You won't, its not supposed to work in the filter table.
> 
> 
> This worked in 2.6.16, where the ipt_mark_reg_v1 strucutre in xt_MARK.c did 
> not have a .table limiter. (It also worked in practice, i.e. packets got 
> marked like they should.) I do not see why this was changed.


That is not true, all versions of the mark target starting in 2.4 had
the same check in the checkentry function.

Re: ipt_MARK/xt_MARK usage problem

[Patrick McHardy]

Jan Engelhardt wrote:
>>>How do I get MARK back to work in -t filter -- possibly without hacking in 
>>>xt_MARK.c?
>>
>>You won't, its not supposed to work in the filter table.
> 
> 
> This worked in 2.6.16, where the ipt_mark_reg_v1 strucutre in xt_MARK.c did 
> not have a .table limiter. (It also worked in practice, i.e. packets got 
> marked like they should.) I do not see why this was changed.


That is not true, all versions of the mark target starting in 2.4 had
the same check in the checkentry function.

Re: ipt_MARK/xt_MARK usage problem

[Jan Engelhardt]

>>>>How do I get MARK back to work in -t filter -- possibly without hacking in 
>>>>xt_MARK.c?
>>>
>>>You won't, its not supposed to work in the filter table.
>> 
>> This worked in 2.6.16, where the ipt_mark_reg_v1 strucutre in xt_MARK.c did 
>> not have a .table limiter. (It also worked in practice, i.e. packets got 
>> marked like they should.) I do not see why this was changed.
>
>That is not true, all versions of the mark target starting in 2.4 had
>the same check in the checkentry function.

Oh sorry I mixed something up. It is CONNMARK which did work (and continues 
to do so) in -t filter... everything fine, sorry for the noise.



Jan Engelhardt
--