On Wed, Dec 01, 2010 at 08:08:54AM +0000, Jan Beulich wrote:
On 30.11.10 at 23:05, Jeff Mahoney <jeffm@suse.de> wrote: On 11/30/2010 04:26 PM, Jiri Bohac wrote: It is important to keep CONFIG_IPV6=y everywhere or it will break again. I need to see why this is currently not true for x86_64/ec2, sparc64/net and i386/ec2.
The ec2 flavors should be adjusted but sparc64 is under Jan Engelhardt's maintainership.
I'm actually opposed to changing this back without knowing that it's actually used in the EC2 environment.
I guess you can set it to =n and see if someone complains then. If it turns out it is used, please make it =y. See below why.
Even more, I was actually considering making it (and other stuff - crypto comes to mind immediately) =m again in the -xen configs
Please don't...
unless it is known that these bits are used on the vast majority of systems. After all that's what modules are for, and I don't see the point of a bloated kernel (not even with the argument of boot speed, as I doubt anyone has ever tried to compare the run time effect [accumulated over the perhaps long periods of run time] with the boot time effect) when only a subset of users actually uses the features in question.
The argument of sysconfig not being able to cope with IPV6=m is certainly bogus - if so, the tool needs to be fixed.
... the sysctl configuration interface of IPv6 (and IPv4 as well, FWIW) is really really frustrating wrt the effect and availability of the tunables during boot. A simple example of the horror: Say you want to set accept_ra to 0 to disable autoconfiguration on an interface. To do this, you need to : - have IPv6 available (i.e. loaded if it is a module) - the interface must not yet be up, otherwise it is too late to stop it from autoconfiguring If you want to enable the interface with something like ifplugd/NM/udev when it appears, doing everything in the right order is quite complicated. If the scripts needed to check/load the ipv6 module, this would be a nightmare. Even worse, you may need to configure the network interface in initrd, etc... Disabling IPv6 (through YaST/sysconfig) is another problem. The old method of blacklisting the module does no longer work, because e.g. the bonding module depends on the ipv6 module. Disabling IPv6 with the sysctl cannot be done unless the module is already loaded, at which point, interfaces may have started autoconfiguring. The only option left would be a kernel command-line option. Having ipv6 built-in everywhere minimises the frustration of people putting their own ipv6 setting to /etc/sysctl.conf. Please keep it consistent for all archs/flavours. -- Jiri Bohac <jbohac@suse.cz> SUSE Labs, SUSE CZ -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org