Re: autofs reverts to IPv4 for multi-homed IPv6 server ?

[Christof Koehler <christof.koehler@bccms.uni-bremen.de>]

Hello,

I will try to consolidate the answer in a single email without messing
up the quoting too much.

> But we can't really start to work out what needs to be done unless
> libtirpc is being used.

> And I think the best chance is to use 5.1.1 and possibly back port any
> changes.

> As I say, the only other thing we can do is add some targeted debug
> logging and see if we can spot what is going wrong.

I will then stick to the plan to defer this
until ubuntu 16.04 has been released which should contain autofs 5.1.1
and libtirpc 0.25.2. From the libtirpc git repo and the sourceforge page
I see that 0.25.2 is from 2014, but not even debian unstable is
using a newer version (and from the changelog things like gss_api
support are the main changes ?). 

I will use a virtual machine with ubuntu 16.04 (or if preferable debian 
testing or unstable) and then start experimenting. In that environment we 
can do to autofs and also libtirpc (and nfs-common if necessary) whatever 
is needed to get the necessary information. 

I will let you know when I start with that and what the baseline with
the original packages of the distribution and simple rebuild of autofs
--with-libtirpc is.

>> Apr  8 18:05:25 core324 automount[963]: get_nfs_info: called with host
>> core330(fd5f:852:a27c:1261:2000::118) proto 6 version 0x40
>> Apr  8 18:05:25 core324 automount[963]: get_nfs_info: called with host
>> core330(2001:638:708:1261:2000::118) proto 6 version 0x40
>> Apr  8 18:05:25 core324 automount[963]: mount(nfs): no hosts available

> Sadly that doesn't tell us much either, only that the rpc communication
> has failed to get a result in some expected way.

Well, at least it shows that this autofs built at least is somehow aware of
all the different IP adresses and trying the IPv4 one first. Of course
no other conclusions can be drawn from this.

> That's right, as I say the RPC communication isn't failing in an
> unexpected way so we aren't seeing any error messages.
> 
> About all that can be done is to add a patch that adds some extra
> logging to try and get to the bottom of it.

That should be possible with a virtual machine as mentioned above.

> The first thing that stands out is that if libtirpc is not being used
> all IPv6 hosts will be ignored because (my impression is that) glibc RPC
> doesn't support IPv6.

The libtirpc documentation says that libtirpc is needed for IPv6 ready
rpc support, http://nfsv4.bullopensource.org/doc/tirpc_rpcbind.php as
linked from sourceforge.
Might even be that glibc does actually no longer contain any rpc
functionality ?
https://archives.gentoo.org/gentoo-dev/message/186be8dc9753d18aafc9a5a616b3b991
This would be supported by information on the tirpc web page mentioned
above.

As far as I understand your analysis you are saying the IPv6
mount in the "2001" case is just working by accident, right ?

> It's mount.nfs(8) mounting from the IPv6 address (when given a host name
> not an address) in the former case and not autofs that's getting you an
> IPv6 mount. And, AFAICS, the mount.nfs your using does use libtirpc.
Yes.
# ldd /sbin/mount.nfs4|grep tirpc
        libtirpc.so.1 => /lib/x86_64-linux-gnu/libtirpc.so.1 (0x00007ffff7d9d000)
and libtirpc is a hard package dependency of nfs-common.

Best Regards

Christof
-- 
Dr. rer. nat. Christof Köhler       email: c.koehler@bccms.uni-bremen.de
Universitaet Bremen/ BCCMS          phone:  +49-(0)421-218-62334
Am Fallturm 1/ TAB/ Raum 3.12       fax: +49-(0)421-218-62770
28359 Bremen  

PGP: http://www.bccms.uni-bremen.de/cms/people/c_koehler/
--
To unsubscribe from this list: send the line "unsubscribe autofs" in