samba - Nov 2002 - IP in utmp: smbd/session.c

Dear Samba developers,

Sometime between versions 2.2.2 and 2.2.7 you made the change:

% diff samba-2.2.2/source/smbd/session.c samba-2.2.7/source/smbd/session.c
92,94c92,97
<       hostname = client_name();
<       if (strequal(hostname,"UNKNOWN"))
<               hostname = client_addr();
---
>         /* Don't resolve the hostname in smbd as we can pause for a
long
>            time while waiting for DNS timeouts to occur.  The correct
>            place to do this is in the code that displays the session
>            information. */
> 
>         hostname = client_addr();
This causes utmp entries to be written with the IP address instead of the
hostname. This is ugly, so I reverted to the pre-2.2.7 version.

I disagree that DNS lookups are an issue. You use hostnames at least for
log file names and for "hosts allow" lookups. The fashion lately is to
put
tcp-wrappers around all remote services: does not samba deserve the same?
Sure, avoid repeated lookups (by saving the name upon first lookup); even
implement short timeouts if really needed (but really, just refuse if the
forward/reverse lookups do not "work").


Thanks for this very useful software!

Paul Szabo - psz@maths.usyd.edu.au  http://www.maths.usyd.edu.au:8000/u/psz/
School of Mathematics and Statistics  University of Sydney   2006  Australia

On Mon, 25 Nov 2002 12:48:50 +1100, Paul Szabo wrote:
> I disagree that DNS lookups are an issue. You use hostnames at least for
> log file names and for "hosts allow" lookups. The fashion lately
is to put
DNS lookups are definitely an issue.  If you have a broken DNS server then
mysterious slowdowns and hangs occur when clients try to connect.  Samba
now does not require a working DNS server to clients to successfully
connect, except if you are using the hosts allow/deny functionality.

The log filenames can use the %m parameter which is replaced by the
NetBIOS name.  This doesn't require a DNS lookup.
> tcp-wrappers around all remote services: does not samba deserve the
> same?
The hosts allow/hosts deny code uses the TCP wrapper code so it should be
equivalent to using /etc/hosts.{allow,deny}
> Sure, avoid repeated lookups (by saving the name upon first lookup); even
> implement short timeouts if really needed (but really, just refuse if the
> forward/reverse lookups do not "work").
But you can't tell whether the lookups work without actually doing one,
and if the lookup takes longer than 30 seconds the client times out


Tim.