[question] ipt_CLUSTERIP and address length

[question] ipt_CLUSTERIP and address length

[Jiri Pirko]

Hi all.

I want to ask if there is any particular reason for ipt_CLUSTERIP to support
only address length of 6 (ETH_ALEN)? It seems to me reasonable for this to work
even with another types of network hw with different addr_len.

Thanks.

Jirka

Re: [question] ipt_CLUSTERIP and address length

[Patrick McHardy]

Jiri Pirko wrote:
> Hi all.
> 
> I want to ask if there is any particular reason for ipt_CLUSTERIP to support
> only address length of 6 (ETH_ALEN)? It seems to me reasonable for this to work
> even with another types of network hw with different addr_len.

None that I'm aware of, but the length is also used in the ABI,
so you presently can't supply larger addresses.

Re: [question] ipt_CLUSTERIP and address length

[Jan Engelhardt]

On Thursday 2010-02-25 11:23, Patrick McHardy wrote:
>Jiri Pirko wrote:
>> Hi all.
>> 
>> I want to ask if there is any particular reason for ipt_CLUSTERIP
>> to support only address length of 6 (ETH_ALEN)? It seems to me
>> reasonable for this to work even with another types of network hw
>> with different addr_len.
>
>None that I'm aware of, but the length is also used in the ABI,
>so you presently can't supply larger addresses.

The larger picture is that there seems to be quite a bit of ABI
even outside of the kernel that uses ETH_ALEN and/or ETH_HLEN
in many places; for example, the tcpdump cooked interface IIRC.

Is there an _actual_ use case however? Is it suddenly possible
to use EUI-64s? Or a run on IPX? :-)

Re: [question] ipt_CLUSTERIP and address length

[Jiri Pirko]

Thu, Feb 25, 2010 at 12:04:23PM CET, jengelh@medozas.de wrote:
>
>On Thursday 2010-02-25 11:23, Patrick McHardy wrote:
>>Jiri Pirko wrote:
>>> Hi all.
>>> 
>>> I want to ask if there is any particular reason for ipt_CLUSTERIP
>>> to support only address length of 6 (ETH_ALEN)? It seems to me
>>> reasonable for this to work even with another types of network hw
>>> with different addr_len.
>>
>>None that I'm aware of, but the length is also used in the ABI,
>>so you presently can't supply larger addresses.
>
>The larger picture is that there seems to be quite a bit of ABI
>even outside of the kernel that uses ETH_ALEN and/or ETH_HLEN
>in many places; for example, the tcpdump cooked interface IIRC.

Hmm, thats sad :( IMHO this should be ETH independent...

>
>Is there an _actual_ use case however? Is it suddenly possible
>to use EUI-64s? Or a run on IPX? :-)

Well for example Infiniband has 20-bytes address. But I do not know much about
this kinds of hw & iptables use. I'm just looking at code and wondering :)

Jirka

Re: [question] ipt_CLUSTERIP and address length

[Jan Engelhardt]

On Thursday 2010-02-25 12:37, Jiri Pirko wrote:
>>
>>Is there an _actual_ use case however? Is it suddenly possible
>>to use EUI-64s? Or a run on IPX? :-)
>
>Well for example Infiniband has 20-bytes address. But I do not know much about
>this kinds of hw & iptables use. I'm just looking at code and wondering :)

First have IPv4/v6 running over InfiniBand, then we can talk ;-)

I did something like that once, but what it came down to was that
an Ethernet layer was inserted between Myrinet and IPv4.
(`ip link` will show.)

Re: [question] ipt_CLUSTERIP and address length

[Jiri Pirko]

Thu, Feb 25, 2010 at 01:05:06PM CET, jengelh@medozas.de wrote:
>
>On Thursday 2010-02-25 12:37, Jiri Pirko wrote:
>>>
>>>Is there an _actual_ use case however? Is it suddenly possible
>>>to use EUI-64s? Or a run on IPX? :-)
>>
>>Well for example Infiniband has 20-bytes address. But I do not know much about
>>this kinds of hw & iptables use. I'm just looking at code and wondering :)
>
>First have IPv4/v6 running over InfiniBand, then we can talk ;-)
>
>I did something like that once, but what it came down to was that
>an Ethernet layer was inserted between Myrinet and IPv4.
>(`ip link` will show.)

Ok, fair enough :)

Re: [question] ipt_CLUSTERIP and address length

[Pablo Neira Ayuso]

Patrick McHardy wrote:
> Jiri Pirko wrote:
>> Hi all.
>>
>> I want to ask if there is any particular reason for ipt_CLUSTERIP to support
>> only address length of 6 (ETH_ALEN)? It seems to me reasonable for this to work
>> even with another types of network hw with different addr_len.
> 
> None that I'm aware of, but the length is also used in the ABI,
> so you presently can't supply larger addresses.

Not directly related to this but I wanted to discuss this time ago. Now
that we have xt_CLUSTER I think that we can deprecate ipt_CLUSTERIP.

With regards to this issue, it seems arptables only support EUI-48 (6
bytes) for ethernet addresses, so xt_CLUSTER would inherit the same
problem but the point would be to fix arptables (not sure if possible
now without breaking ABI or adding some versioning like iptables).

Re: [question] ipt_CLUSTERIP and address length

[Patrick McHardy]

Pablo Neira Ayuso wrote:
> Patrick McHardy wrote:
>> Jiri Pirko wrote:
>>> Hi all.
>>>
>>> I want to ask if there is any particular reason for ipt_CLUSTERIP to support
>>> only address length of 6 (ETH_ALEN)? It seems to me reasonable for this to work
>>> even with another types of network hw with different addr_len.
>> None that I'm aware of, but the length is also used in the ABI,
>> so you presently can't supply larger addresses.
> 
> Not directly related to this but I wanted to discuss this time ago. Now
> that we have xt_CLUSTER I think that we can deprecate ipt_CLUSTERIP.

If xt_cluster supports everything ipt_CLUSTERIP does, thats fine
with me.

> With regards to this issue, it seems arptables only support EUI-48 (6
> bytes) for ethernet addresses, so xt_CLUSTER would inherit the same
> problem but the point would be to fix arptables (not sure if possible
> now without breaking ABI or adding some versioning like iptables).

arptables currently supports up to 16 byte long addresses. Increasing
this is difficult since the addresses are embedded in struct arpt_arp.

Re: [question] ipt_CLUSTERIP and address length

[Pablo Neira Ayuso]

Patrick McHardy wrote:
> Pablo Neira Ayuso wrote:
>> Patrick McHardy wrote:
>>> Jiri Pirko wrote:
>>>> Hi all.
>>>>
>>>> I want to ask if there is any particular reason for ipt_CLUSTERIP to support
>>>> only address length of 6 (ETH_ALEN)? It seems to me reasonable for this to work
>>>> even with another types of network hw with different addr_len.
>>> None that I'm aware of, but the length is also used in the ABI,
>>> so you presently can't supply larger addresses.
>> Not directly related to this but I wanted to discuss this time ago. Now
>> that we have xt_CLUSTER I think that we can deprecate ipt_CLUSTERIP.
> 
> If xt_cluster supports everything ipt_CLUSTERIP does, thats fine
> with me.

Yes, xt_cluster supports gateway and back-end clustering while
ipt_CLUSTERIP only works for back-end setup.

I wanted to have some time to document xt_cluster, I have some scripts
lying here and some unfinished documents. I think that we can deprecated
as soon as I have that doc ready.

>> With regards to this issue, it seems arptables only support EUI-48 (6
>> bytes) for ethernet addresses, so xt_CLUSTER would inherit the same
>> problem but the point would be to fix arptables (not sure if possible
>> now without breaking ABI or adding some versioning like iptables).
> 
> arptables currently supports up to 16 byte long addresses. Increasing
> this is difficult since the addresses are embedded in struct arpt_arp.

Hm, so the problem seems to be user-space then:

# arptables -I OUTPUT -o eth1 --h-length 8 \
> -j mangle --mangle-mac-s 01:00:5e:00:01:01:00:00
arptables v0.0.3.3: only --h-length 6 supported
Try `arptables -h' or 'arptables --help' for more information.

As soon as this is fixed. Are 16 bytes long addresses long enough by now?

Re: [question] ipt_CLUSTERIP and address length

[Jan Engelhardt]

On Thursday 2010-02-25 13:18, Pablo Neira Ayuso wrote:
>>>
>>> I want to ask if there is any particular reason for ipt_CLUSTERIP to support
>>> only address length of 6 (ETH_ALEN)? It seems to me reasonable for this to work
>>> even with another types of network hw with different addr_len.
>> 
>> None that I'm aware of, but the length is also used in the ABI,
>> so you presently can't supply larger addresses.
>
>Not directly related to this but I wanted to discuss this time ago. Now
>that we have xt_CLUSTER I think that we can deprecate ipt_CLUSTERIP.

xt_CLUSTER - where in the tree would that be?

Re: [question] ipt_CLUSTERIP and address length

[Jiri Pirko]

Thu, Feb 25, 2010 at 01:54:50PM CET, jengelh@medozas.de wrote:
>On Thursday 2010-02-25 13:18, Pablo Neira Ayuso wrote:
>>>>
>>>> I want to ask if there is any particular reason for ipt_CLUSTERIP to support
>>>> only address length of 6 (ETH_ALEN)? It seems to me reasonable for this to work
>>>> even with another types of network hw with different addr_len.
>>> 
>>> None that I'm aware of, but the length is also used in the ABI,
>>> so you presently can't supply larger addresses.
>>
>>Not directly related to this but I wanted to discuss this time ago. Now
>>that we have xt_CLUSTER I think that we can deprecate ipt_CLUSTERIP.
>
>xt_CLUSTER - where in the tree would that be?

I was trying to find it too. I guess it stands out of it (at least net-next).

Re: [question] ipt_CLUSTERIP and address length

[Pablo Neira Ayuso]

Jiri Pirko wrote:
> Thu, Feb 25, 2010 at 01:54:50PM CET, jengelh@medozas.de wrote:
>> On Thursday 2010-02-25 13:18, Pablo Neira Ayuso wrote:
>>>>> I want to ask if there is any particular reason for ipt_CLUSTERIP to support
>>>>> only address length of 6 (ETH_ALEN)? It seems to me reasonable for this to work
>>>>> even with another types of network hw with different addr_len.
>>>> None that I'm aware of, but the length is also used in the ABI,
>>>> so you presently can't supply larger addresses.
>>> Not directly related to this but I wanted to discuss this time ago. Now
>>> that we have xt_CLUSTER I think that we can deprecate ipt_CLUSTERIP.
>> xt_CLUSTER - where in the tree would that be?
> 
> I was trying to find it too. I guess it stands out of it (at least net-next).

sorry, it's xt_cluster, no capitalization.

pablo@decadence:~/devel/scm/git/nf-2.6$ ls net/netfilter/xt_cluster.c

It's there since quite some time.

Re: [question] ipt_CLUSTERIP and address length

[Jan Engelhardt]

On Thursday 2010-02-25 13:54, Jan Engelhardt wrote:

>On Thursday 2010-02-25 13:18, Pablo Neira Ayuso wrote:
>>>>
>>>> I want to ask if there is any particular reason for ipt_CLUSTERIP to support
>>>> only address length of 6 (ETH_ALEN)? It seems to me reasonable for this to work
>>>> even with another types of network hw with different addr_len.
>>> 
>>> None that I'm aware of, but the length is also used in the ABI,
>>> so you presently can't supply larger addresses.
>>
>>Not directly related to this but I wanted to discuss this time ago. Now
>>that we have xt_CLUSTER I think that we can deprecate ipt_CLUSTERIP.
>
>xt_CLUSTER - where in the tree would that be?

Oh nevermind, it's xt_cluster actually.