• Re: Weird failure of dhclient on a bridge

    From Andrew Gideon@110:110/2002 to All on Mon Jul 29 15:09:42 2013
    On Sat, 20 Jul 2013 10:05:19 +0200, Pascal Hambourg wrote:

    Did you capture and compare network traffic on the different interfaces (physical, VLAN, bridge) ? Did you check the kernel and dhclient logs ?


    I captured the traffic on the VLAN. The DHCP requests from the
    workstation did make it to "the wire". The requests from the servers did
    not.

    Did you try different DHCP clients (pump, dhcpcd, udhcpc...) ?

    No. That's a good idea.

    I'm running out of possible differences to explain this.

    Different hardware, different drivers ? Offload optimizations enabled/disabled (see ethtool) ?

    But what optimization could possibly cause this? That's my problem: I
    don't know what to check, as I'm not clear what setting could have this effect.

    Thanks...

    Andrew


    --- MBSE BBS v0.95.15 (GNU/Linux-i386)
    * Origin: UsenetServer - www.usenetserver.com (110:110/2002@linuxnet)
  • From Andrew Gideon@110:110/2002 to All on Mon Jul 15 18:18:12 2013

    I've three CentOS6 machines, all of which are using 802.11q VLAN tagging
    and bridging to support KVM guests. On one of the machines, I am able to
    use dhclient to assign a dynamic IP to a bridge. On the other two, I am
    not.

    I'm running out of possible differences to explain this.

    If I statically assign IPs to the bridges, all is fine. Guests on the
    two "problematic" machines have no trouble getting IPs via DHCP from this particular VLAN, which suggests that the switch fabric isn't an issue.
    I've moved cables around to confirm this.

    The only substantive difference between the one working and the two non- working machines is that the working machine is a workstation with a
    display and keyboard. The other two are servers with serial-port
    consoles.

    While this shouldn't make a difference to the networking, it does mean
    that there's additional software installed on the workstation.
    NetworkManager is an obvious possible culprit, for example, even though
    it is chkconfig-ed off on the workstation.

    I did try installing NetworkManager (and NetworkManager-glib) on one of
    the two servers, and this didn't make a difference. But there's plenty
    of other software installed on the workstation but not on the servers. Perhaps the culprit is one of these components?

    I'm getting close to the fairly desperate step of re-installing one of
    the servers as a workstation just to see if that makes a difference.
    Keep in mind, though, the entire idea of this being an issue of what
    software is/isn't installed is just a guess on my part.

    So...I'm fishing here for ideas. Anyone have any suggestions?

    Thanks...

    Andrew

    --- MBSE BBS v1.0.0 (GNU/Linux-i386)
    * Origin: UsenetServer - www.usenetserver.com (110:110/2002@linuxnet)
  • From Pascal Hambourg@110:110/2002 to All on Sat Jul 20 08:05:18 2013

    Reply-To: pascal.news@plouf.fr.eu.org

    Hello,

    Andrew Gideon a ‚crit :
    I've three CentOS6 machines, all of which are using 802.11q VLAN tagging
    and bridging to support KVM guests. On one of the machines, I am able to use dhclient to assign a dynamic IP to a bridge. On the other two, I am not.

    Did you capture and compare network traffic on the different interfaces (physical, VLAN, bridge) ? Did you check the kernel and dhclient logs ?

    Did you try different DHCP clients (pump, dhcpcd, udhcpc...) ?

    I'm running out of possible differences to explain this.

    Different hardware, different drivers ? Offload optimizations
    enabled/disabled (see ethtool) ?

    --- MBSE BBS v1.0.0 (GNU/Linux-i386)
    * Origin: Plouf ! (110:110/2002@linuxnet)