Client frequent disconnection + strange logs

[François Désarménien]

David Collier-Brown wrote:
> 
> François Désarménien wrote:
> | in the log.machine, I find at the same timestamp close requests,
> | followed by two line of the form:
> | [timestamp]lib/doscalls.c:dos_ChDir(336) dos_ChDir to
> /var/sauvegarde
> | [timestamp]connexion closed for user compta
> | and then a reconnection to the same user, at the same timestamp.
> 
>         Hmmn, people seem to be reporting that a lot these days...
>         Normally it indicates one of
>                 - an error on the client
>                 - the client timing out
>                 - an error seen by the client
>         One commentator had a bad hub, another only saw
>         it on a particular machine, others didn'd see
>         anything unique that might cause it...
> 

I thought of that, of course. Well, if "people seem to be reporting
that a lot these days", I'll first try to downgrade to 2.0.5a, which
I have running without trouble on many sites.

>         Can you look at your net with a snooper (eg, etherial,
>         tcpdump) and look for suspicious packets just before the
>         disconnection?  The chap with the bad hub was seeing (if
>         memory serves) truncated packets.

Yes, this will be the second try, as it is *much* more time (and disk:)
consuming. I've been hurt with a bad hub one day, this can effectively
leads to strange phenomenon...
> 
> | [timestamp]PAM_pwdb: (login) authentication failure;
>         (uid 0) -> compta for samba service
> 
>         This look like a normal failure reported by the pam module
> 

Except that nobody's trying to conect on 'compta' account, and it is
the samba account the wind'oz client's trying to reconnect...

And the dos_ChDir to /var/sauvegarde really puzzles me :-o

Many thanks for your time and answer,

François