[systemd-devel] systemd-resolved cannot survive router reinstall

[Tom Gundersen]

On Thu, Aug 21, 2014 at 8:22 AM, Anatol Pomozov
<anatol.pomozov at gmail.com> wrote:
> Hi
>
> Sorry for the delay, just got back home.
>
> On Tue, Aug 12, 2014 at 12:19 PM, Tom Gundersen <teg at jklm.no> wrote:
>> On Tue, Aug 12, 2014 at 2:48 AM, Anatol Pomozov
>> <anatol.pomozov at gmail.com> wrote:
>>> Thanks for the instructions, here is the log
>>> https://gist.github.com/anatol/cb497118cfad6af3dc60
>>>
>>> The interesting entries are:
>>> Aug 11 17:30:25 theserver systemd-networkd[32648]: ICMPv6 CLIENT:
>>> Received Router Advertisment flags MANAGED/OTHER
>>> Aug 11 17:30:25 theserver systemd-networkd[32648]: ICMPv6 CLIENT:
>>> Received Router Advertisment flags MANAGED/OTHER
>>> Aug 11 17:30:30 theserver systemd-networkd[32648]: ICMPv6 CLIENT:
>>> Received Router Advertisment flags MANAGED/OTHER
>>>
>>> Is it the "heartbeat" message from the router?
>>
>> This is IPv6 router advertisment. Not related to your DHCPv4 problems.
>>
>> When the router is rebooted, we should expect to get -LOWER_UP
>> followed by +LOWER_UP in the logs (same as when unplugging the cable).
>> Could you try running "ip monitor link" when rebooting the rooter and
>> paste whatever events you get? How about when you unplug/replug the
>> cable?
>
> I just realized that all my computers (except the home server that has
> the dhcp problem) are either connected directly to router via cable or
> use wifi (i.e. also connected directly). This means when router power
> cycles all directly connected machines see it and thus renew the dhcp
> lease. But the home server is connected via switch so it is located in
> its own network segment. Even if router reboots, that segment stays
> healthy and the server does not notice any "problem" with the router.
> It thinks that the lease is OK. I need to unplug/plug the cable
> connected to the home server - this brings networkd down/up and
> eventually updates the lease.
>
> So the problem is not in networkd, sorry for the noise. The problem is
> in OpenWRT that uses dnsmasq and stores dhcp lease information on
> tmpfs. If I reboot the router it looses all the leases and machines
> should update the information somehow. This is a problem for machines
> in a separate network segment. I can not believe that OpenWRT
> developers never tested such configuration. I am going to contact
> openwrt/dnsmasq maillist and ask what is the right way to solve the
> issue.

Thanks for getting back to this. That explains the situation.

> PS According to off-list discussion with Tom FORCERENEW dhcp option
> http://tools.ietf.org/html/rfc3203 is what suppose to fix this issue.
> But dnsmasq does not support it.
>
> PPS NetworkManager has a higher-level connectivity checking. It can be
> configured to read an url content periodically and if it returns
> unexpected result then connection considered broken.

We intend to add such connectivity checking to networkd too, but I
doubt we should use it to renew the DHCP lease, but keep the DHCP
logic to only renewing when connectivity to the local link is lost.

Cheers,

Tom