Bug#757711: netcfg: promptly kills dhclient, deconfigures interface
> On Tue, Aug 12, 2014 at 02:45:25AM +0200, Cyril Brulebois wrote:
>> That I don't know. Maybe try and compare with netcfg + dhcp client du
>> jour on Linux, and see whether netcfg's behaviour is different in both
>> cases, or the dhcp client's one, or the kernel's one.
netcfg kills the running DHCP client on linux too.
On 12/08/14 20:19, Philipp Kern wrote:
> In the end there's a lot of sadness around what we do, what Ubuntu does,
> and what the end result is. Ubuntu uses isc-dhcp-client, so the same as
> on kfreebsd-* and might as well have some additional patches to netcfg
> because of this.
I could maybe look into porting udhcpc to kfreebsd; that is if netcfg
maintainers still think it is preferred over ISC DHCP? (udhcpc saves
about 2MiB in the install image, if that's reason enough). Or otherwise
go the other way and try to have all architectures using
isc-dhcp-client-udeb.
Except - I'm not sure how heavily systemd is going to feature in d-i for
jessie or jessie+1? If systemd's own DHCP client is going to take over
this task someday we'd end up with different implementations again,
making either of the above changes a waste of time, and personally I'd
prefer to keep the fully-featured ISC DHCP in that case.
(Releasing the DHCP lease and deconfiguring the interface seems to be a
feature udhcpc *lacks*, hence why the problem is not seen yet on linux).
> But at work we do use Ubuntu and found that dhclient is called with -1.
> If the lease expires during installation and your switch does DHCP
> snooping, you lose network during installation because the DHCP client
> does not continue to renew it.
FWIW the netcfg patch in my last message seems to have fixed this nicely
for kfreebsd. And it keeps the dhclient/udhcpc running throughout the
install - I think that is desirable whichever one is used, to avoid the
issue described above.
> This is anecdotal experience but divergence between Linux and kfreebsd
> wrt DHCP might mean that certain problems need to be solved twice or
> certain failure modes only happen in one or another, with userspace
> being the reason, not even the kernel difference.
Thanks for sharing that info.
Regards,
--
Steven Chamberlain
steven@pyro.eu.org
Reply to: