conntrack-tools rpc helper

conntrack-tools rpc helper

[Richard A Nelson]

I'm having a heck of a time getting this all going ;)

Debian Sid/Experimental
Linux 3.7.1
conntrack toolchain downloded yesterday (stable, not git)

I'm following the doc on conntrack-tools to enable RPC tracking (for NFS 
support), and wind up with this state:
# nfct helper list
{
     .name = rpc,
     .queuenum = 0,
     .l3protonum = 2,
     .l4protonum = 6,
     .priv_data_len = 16,
     .status = disabled,
};
{
     .name = rpc,
     .queuenum = 0,
     .l3protonum = 2,
     .l4protonum = 17,
     .priv_data_len = 16,
     .status = disabled,
};

Why disabled?  conntrackd startup went fine, the nfct helper add went fine, what 
am I missing ?

And, of course, it doesn't work ;)

I may be royally screwed anyway - due to the bizarre situation I'm trying to 
support...
LAN:  192.168/16 - mix of Linux, Windows, and AIX clients
Gateway: 192.168.1.205 (Linux Gateway/Firewall)
OpenConnect VPN to work (CISCO ASA client) - IBM
IBM network - GSA NFS Servers

The sticky point for a statefull firewall seems to be twofold:
1) AIX (at least 5.3) still does RPC mount lookups - even if NFSV4 and port=2049 
are specified
     Which, it seems, is totally needless

2) The GSA infrastructure is layered, and littered with High Availability & load 
balancing - but seems to break
     extant rules...
- My LAN client does a RPC GETPORT (MOUNT -V3) to snjgsa.sanjose.ibm.com
- My firewall NATs the request
- The GSA mainline server delegates (forwards, whatever) to a disk sever
- The disk server (snjxgsasd2.sanjose.ibm.com) replies the RCP GETPORT request

At this point, we have two streams, with matching ports, but differing IP sets, 
both in UNREPLIED status.

I am not at all sure, that even were I to get the RPC helper going, and it 
issues and EXPECT, that things will work, but am willing to hack to make it so :)


-- 
Rick Nelson
Life'll kill ya                         -- Warren Zevon
Then you'll be dead                     -- Life'll kill ya

Re: conntrack-tools rpc helper

[Richard A Nelson]

I left out the end of the scenario :(

As a result, afaict, of the IP missmatch, and both RPC GETPORT streams being 
marked UNREPLIED, the Linux gateway box returns an ICMP PORT UNREACHABLE (there 
being no such open port on the gateway box).

On 12/24/2012 09:23 PM, Richard A Nelson wrote:
> I'm having a heck of a time getting this all going ;)
>
> Debian Sid/Experimental
> Linux 3.7.1
> conntrack toolchain downloded yesterday (stable, not git)
>
> I'm following the doc on conntrack-tools to enable RPC tracking (for NFS 
> support), and wind up with this state:
> # nfct helper list
> {
>     .name = rpc,
>     .queuenum = 0,
>     .l3protonum = 2,
>     .l4protonum = 6,
>     .priv_data_len = 16,
>     .status = disabled,
> };
> {
>     .name = rpc,
>     .queuenum = 0,
>     .l3protonum = 2,
>     .l4protonum = 17,
>     .priv_data_len = 16,
>     .status = disabled,
> };
>
> Why disabled?  conntrackd startup went fine, the nfct helper add went fine, 
> what am I missing ?
>
> And, of course, it doesn't work ;)
>
> I may be royally screwed anyway - due to the bizarre situation I'm trying to 
> support...
> LAN:  192.168/16 - mix of Linux, Windows, and AIX clients
> Gateway: 192.168.1.205 (Linux Gateway/Firewall)
> OpenConnect VPN to work (CISCO ASA client) - IBM
> IBM network - GSA NFS Servers
>
> The sticky point for a statefull firewall seems to be twofold:
> 1) AIX (at least 5.3) still does RPC mount lookups - even if NFSV4 and 
> port=2049 are specified
>     Which, it seems, is totally needless
>
> 2) The GSA infrastructure is layered, and littered with High Availability & 
> load balancing - but seems to break
>     extant rules...
> - My LAN client does a RPC GETPORT (MOUNT -V3) to snjgsa.sanjose.ibm.com
> - My firewall NATs the request
> - The GSA mainline server delegates (forwards, whatever) to a disk sever
> - The disk server (snjxgsasd2.sanjose.ibm.com) replies the RCP GETPORT request
>
> At this point, we have two streams, with matching ports, but differing IP 
> sets, both in UNREPLIED status.
>
> I am not at all sure, that even were I to get the RPC helper going, and it 
> issues and EXPECT, that things will work, but am willing to hack to make it so :)
>
>


-- 
Rick Nelson
Life'll kill ya                         -- Warren Zevon
Then you'll be dead                     -- Life'll kill ya

Re: conntrack-tools rpc helper

[Pablo Neira Ayuso]

Hi Richard,

On Mon, Dec 24, 2012 at 09:23:04PM -0500, Richard A Nelson wrote:
> I'm having a heck of a time getting this all going ;)
> 
> Debian Sid/Experimental
> Linux 3.7.1
> conntrack toolchain downloded yesterday (stable, not git)
> 
> I'm following the doc on conntrack-tools to enable RPC tracking (for
> NFS support), and wind up with this state:
> # nfct helper list
> {
>     .name = rpc,
>     .queuenum = 0,
>     .l3protonum = 2,
>     .l4protonum = 6,
>     .priv_data_len = 16,
>     .status = disabled,
> };
> {
>     .name = rpc,
>     .queuenum = 0,
>     .l3protonum = 2,
>     .l4protonum = 17,
>     .priv_data_len = 16,
>     .status = disabled,
> };
> 
> Why disabled?  conntrackd startup went fine, the nfct helper add
> went fine, what am I missing ?

That means conntrackd did not configured the helper so far, so it
remains dormant in the kernel.

Did you add the corresponding Helper clause to conntrackd.conf?

It should be something similar to what you can find under
conntrack-tools/doc/helper/conntrackd.conf.

Please, check /var/log/conntrackd.log (or custom path in case you set
it). It should report some information regarding the helper
configuration.

> And, of course, it doesn't work ;)
> 
> I may be royally screwed anyway - due to the bizarre situation I'm
> trying to support...
> LAN:  192.168/16 - mix of Linux, Windows, and AIX clients
> Gateway: 192.168.1.205 (Linux Gateway/Firewall)
> OpenConnect VPN to work (CISCO ASA client) - IBM
> IBM network - GSA NFS Servers
> 
> The sticky point for a statefull firewall seems to be twofold:
> 1) AIX (at least 5.3) still does RPC mount lookups - even if NFSV4
> and port=2049 are specified
>     Which, it seems, is totally needless
> 
> 2) The GSA infrastructure is layered, and littered with High
> Availability & load balancing - but seems to break
>     extant rules...
> - My LAN client does a RPC GETPORT (MOUNT -V3) to snjgsa.sanjose.ibm.com
> - My firewall NATs the request
> - The GSA mainline server delegates (forwards, whatever) to a disk sever
> - The disk server (snjxgsasd2.sanjose.ibm.com) replies the RCP GETPORT request
> 
> At this point, we have two streams, with matching ports, but
> differing IP sets, both in UNREPLIED status.
> 
> I am not at all sure, that even were I to get the RPC helper going,
> and it issues and EXPECT, that things will work, but am willing to
> hack to make it so :)

This was tested with NFSv3 running both Linux in the client and the
server sides.

I'd suggest to start by getting that "disabled" issue resolved first,
then, move on to see what other problem you find.

Re: conntrack-tools rpc helper

[Richard A Nelson]

On 12/26/2012 05:43 PM, Pablo Neira Ayuso wrote:
> Hi Richard,

Thanks for the quick, and on target note !

>
> {
>      .name = rpc,
>      .queuenum = 0,
>      .l3protonum = 2,
>      .l4protonum = 17,
>      .priv_data_len = 16,
>      .status = disabled,
> };
>
> Why disabled?  conntrackd startup went fine, the nfct helper add
> went fine, what am I missing ?
> That means conntrackd did not configured the helper so far, so it
> remains dormant in the kernel.
>
> Did you add the corresponding Helper clause to conntrackd.conf?
>
> It should be something similar to what you can find under
> conntrack-tools/doc/helper/conntrackd.conf.
>
> Please, check /var/log/conntrackd.log (or custom path in case you set
> it). It should report some information regarding the helper
> configuration.

My confusion seemed to have been the requisite three part dance - even though 
they're pretty well described in the docs.
1) Register helpers via nfct
2) Add iptables rules in -t raw (output & prerouting)
3) (Re)Start conntrackd

I now see:
# nfct helper list
{
     .name = rpc,
     .queuenum = 2,
     .l3protonum = 2,
     .l4protonum = 6,
     .priv_data_len = 16,
     .status = enabled,
};
{
     .name = rpc,
     .queuenum = 1,
     .l3protonum = 2,
     .l4protonum = 17,
     .priv_data_len = 16,
     .status = enabled,
};

I rebooted, and made sure things came back up in the same state... looking good.
>
> This was tested with NFSv3 running both Linux in the client and the
> server sides.
Perfect, then - there is a good chance this'll go...  I'm still somewhat worried 
about the fact that the RPC call is answered by a different machine than the 
request was made to... but trial and error is called for now ;)
>
> I'd suggest to start by getting that "disabled" issue resolved first,
> then, move on to see what other problem you find.
Exactly, one down - and I have removed the old rules allowing arbitrary replies 
from port 111, and things are working well on the server side (NFSv3 mounts work 
fine).

Now, to see if I can coax the AIX box into booting again - otherwise I'll try 
another Linux client behind the server.

Thanks again!

-- 
Rick Nelson
Life'll kill ya                         -- Warren Zevon
Then you'll be dead                     -- Life'll kill ya