Bug 1129500

Created attachment 926231 [details]
commands to reproduce the bug

(In reply to Ondřej Svoboda from comment #0)
> A simple way to fix this issue is to busy-wait until 'ip -6 -o addr show
> veth_45' stops reporting the 'tentative' flag. A harder one is to wait in
> dhclient.

We might add such a waiting into dhclient-script after we 'up' the interface in PREINIT6, see line 707 in
http://pkgs.fedoraproject.org/cgit/dhcp.git/tree/dhclient-script

Something like this works for me:

    PREINIT6)
        # ensure interface is up
        ip link set dev ${interface} up
        
        # remove any stale addresses from aborted clients
        ip -6 addr flush dev ${interface} scope global permanent

        for i in $(seq 3); do
            # tentative flag == DAD is still not complete
            tentative=$(ip -6 addr show dev ${interface} scope link | grep tentative)
            if [ -z "${tentative}" ] ; then
                # DAD is over
                exit_with_hooks 0
            fi
            sleep 1
        done

        exit_with_hooks 0
        ;;

Could you try to modify PREINIT6 in /usr/sbin/dhclient-script like that and see if it fixes the problem for you Ondrej ?

I confirm this approach works. It takes ~1.5 second for the link-local address to lose the tentative flag and approx. 2 seconds for the global address, "measured" using the following snippet in the same environment as originally.

 while true; do date +%X.%N; ip addr show dev veth_45; sleep 0.1; done;

To avoid waiting too much, could we shorten the period to less than half a second? If it would not be too resource-intensive (which I doubt), even go down to 0.1 s and have 30–50 cycles (in a similar situation, there is a 5s limit in dhclient-script [1]).

[1] http://pkgs.fedoraproject.org/cgit/dhcp.git/tree/dhclient-script#n563

Fixed in rawhide/F21 with:
http://pkgs.fedoraproject.org/cgit/dhcp.git/commit/?h=f21&id=991bd354d956cc2f31ba75689ddcca0021706f0b

I'd rather not push this into F20.

It does not happen 100% of time but the very existence of the link-local address is not guaranteed that early (after 100ms). We also have to check that

 linklocal=$(ip -6 addr show dev ${interface} scope link)
 …
 if [ ! -z "${linklocal}" -a -z "${tentative}" ] ; then

Dan, I assume we need this change backported to EL because that is where VDSM is supposed to run. What systems do we want to support DHCPv6 in?

Ondrej, what about this ?

for i in $(seq 50); do
    linklocal=$(ip -6 addr show dev ${interface} scope link)
    tentative=$(echo "${linklocal}" | grep tentative)
    [[ -n "${linklocal}" &&  -z "${tentative}" ]] && exit_with_hooks 0
    sleep 0.1
done

Jirka, thanks, this is it exactly and work reliably :-)

Jiri, can we have this backported to f20, el7 and el6? We would oVirt's Vdsm to start using dhcp6.

(In reply to Dan Kenigsberg from comment #9)
> Jiri, can we have this backported to f20

Sure, dhcp-4.2.7-2.fc20 should do the trick:
http://pkgs.fedoraproject.org/cgit/dhcp.git/commit/?h=f20&id=88bb3be48ee88795f43ba90c4ef1bf3a33a03c96

> el7 and el6? 

You have to clone this bug for these products if you want it to be fixed there.

dhcp-4.2.7-2.fc20 has been submitted as an update for Fedora 20.
https://admin.fedoraproject.org/updates/FEDORA-2014-8287/dhcp-4.2.7-2.fc20

Package dhcp-4.2.7-2.fc20:
* should fix your issue,
* was pushed to the Fedora 20 testing repository,
* should be available at your local mirror within two days.
Update it with:
# su -c 'yum update --enablerepo=updates-testing dhcp-4.2.7-2.fc20'
as soon as you are able to.
Please go to the following url:
https://admin.fedoraproject.org/updates/FEDORA-2014-8287/dhcp-4.2.7-2.fc20
then log in and leave karma (feedback).

4.2.7-2.fc20 works fine! :-) I cloned the bug to EL6 and EL7.

dhcp-4.2.7-2.fc20 has been pushed to the Fedora 20 stable repository.  If problems still persist, please make note of it in this bug report.