On 2/9/26 21:49, Guido Falsi wrote:
On 1/28/26 11:00, Brooks Davis wrote:
On Tue, Jan 27, 2026 at 03:35:16AM +0330, Pouria Mousavizadeh Tehrani
wrote:
Hi everyone,
With `net.inet6.ip6.use_stableaddr` now available, I believe we
should enable
it by default in CURRENT at least.
As you may already know, we currently use the EUI64 method for
generating
stable IPv6 addresses, which has serious privacy issues.
IMHO, trying to maintain backward compatibility defeats the purpose of a
privacy RFC.
To be clear, we don't want to change the ip addresses of existing
servers.
However, it's reasonable for users to expect changes during a major
upgrade
(15 -> 16), a fresh install of a new major release, or living on
CURRENT.
So, for obvious reasons, changing the default value would not be MFCed.
What do you think?
I wonder if we should ship an update to 15 (landing in 15.1) explicitly
adding net.inet6.ip6.use_stableaddr=1 and a suitable comment to
/etc/sysctl.conf so people who later upgrade to 16 aren't painfully
surprised when their server disappears. New installs of 16 would get
the new default, but upgrades would keep the old default. The downside
would be that people who have edited sysctl.conf would have a merge
conflict to resolve, but that's a fairly normal thing.
-- Brooks
Hi all, I just committed the change in the default (thanks to zlei for
approving it, and all the reviewers). [1]
I'll also send an heads up to current@ and net@ just in case.
I am replying t this specific message in the thread because I do like
brooks' idea on how to introduce this on stable.
Once I get the MFC approved and committed [2], I could send a further PR
implementing such a change on stable/15 sysctl.conf.
While writing my heads up message I just noticed this plan cannot work,
unluckily.
Due to the nature of the sysctl, enabling it via /etc/sysctl.conf would
cause the change to only affect interfaces created after sourcing the
file. This means that for most machines the default interface would be
unaffected and keep the default to the in kernel one.
To achieve the effect Brooks suggests would require the "soft switch" to
happen via loader.conf. Not sure if this is a good idea though.
--
Guido Falsi <[email protected]>
