On Mon, Aug 18, 2025 at 1:32 PM Eric Biggers <ebiggers@xxxxxxxxxx> wrote:
>
> On Mon, Aug 18, 2025 at 08:43:45AM -0700, Jakub Kicinski wrote:
> > On Sat, 16 Aug 2025 13:15:12 -0400 Xin Long wrote:
> > > > > Ideally we'd just fail the write and remove the last mentions of md5 and
> > > > > sha1 from the code. But I'm concerned there could be a case where
> > > > > userspace is enabling cookie authentication by setting
> > > > > cookie_hmac_alg=md5 or cookie_hmac_alg=sha1, and by just failing the
> > > > > write the system would end up with cookie authentication not enabled.
> > > > >
> > > > > It would have been nice if this sysctl had just been a boolean toggle.
> > > > >
> > > > > A deprecation warning might be a good idea. How about the following on
> > > > > top of this patch:
> > > >
> > > > No strong opinion but I find the deprecation warnings futile.
> > > > Chances are we'll be printing this until the end of time.
> > > > Either someone hard-cares and we'll need to revert, or nobody
> > > > does and we can deprecate today.
> > > Reviewing past network sysctl changes, several commits have simply
> > > removed or renamed parameters:
> > >
> > > 4a7f60094411 ("tcp: remove thin_dupack feature")
> > > 4396e46187ca ("tcp: remove tcp_tw_recycle")
> > > d8b81175e412 ("tcp: remove sk_{tr}x_skb_cache")
> > > 3e0b8f529c10 ("net/ipv6: Expand and rename accept_unsolicited_na to
> > > accept_untracked_na")
> > > 5027d54a9c30 ("net: change accept_ra_min_rtr_lft to affect all RA lifetimes")
> > >
> > > It seems to me that if we deprecate something, it's okay to change the
> > > sysctls, so I would prefer rejecting writes with md5 or sha1, or even
> > > better following Eric’s suggestion and turn this into a simple boolean
> > > toggle.
> >
> > Slight preference towards reject. bool is worse in case we need to
> > revert (if it takes a few releases for the regression report to appear
> > we may have to maintain backward compat with both string and bool
> > formats going forward).
>
> To be clear, by "It would have been nice if this sysctl had just been a
> boolean toggle", I meant it would have been nice if it had been that way
> *originally*. I wasn't suggesting making that change now.
>
> It would be safest to continue to honor existing attempts to enable
> cookie authentication (by writing md5 or sha1), as this patch does.
>
> If you'd prefer that those attempts be rejected instead, I can do that,
> but how about I do it as a separate patch on top of this one? That way
> if there's a problem with it, we can just revert that patch, instead of
> the entire upgrade to the cookie auth.
>
Sounds good to me.
Thanks.
