Tuning GnuTLS

Tuning GnuTLS

[Lars Magne Ingebrigtsen]

We should strive to make TLS connections as painless as possible, and
involving as little user intervention as possible, while preserving a
reasonable level of security.

So far, two failure points have been identified:

1) Some servers sends a prime with fewer bits than the accepted default.
I think the right thing to do here is to just default
`gnutls-min-prime-bits' to a lower number than the default GnuTLS
number.  I don't know what that number should be, but I think people who
want better bits than that can adjust this number upwards.

2) Servers presenting broken, er, certificates with certain algorithms.
If negotiation with DHE-RSA has failed, then negotiation without that
algorithm should be attempted.  But is it possible to fall back to
plain-text?  I don't really know how that works.  But if that's
possible, the fall-back should obviously stop before it gets that far.

After a priority has been established, I then think that the priority
for this specific server/port pair should be saved via Customize, so
that the next connection can be done faster automatically, without the
need for all this negotiation.

-- 
(domestic pets only, the antidote for overdose, milk.)
  bloggy blog http://lars.ingebrigtsen.no/

Re: Tuning GnuTLS

[Ted Zlatanov]

On Mon, 18 Jul 2011 05:23:16 +0200 Lars Magne Ingebrigtsen <larsi@gnus.org> wrote: 

LMI> We should strive to make TLS connections as painless as possible, and
LMI> involving as little user intervention as possible, while preserving a
LMI> reasonable level of security.

LMI> So far, two failure points have been identified:

LMI> 1) Some servers sends a prime with fewer bits than the accepted default.
LMI> I think the right thing to do here is to just default
LMI> `gnutls-min-prime-bits' to a lower number than the default GnuTLS
LMI> number.  I don't know what that number should be, but I think people who
LMI> want better bits than that can adjust this number upwards.

LMI> 2) Servers presenting broken, er, certificates with certain algorithms.
LMI> If negotiation with DHE-RSA has failed, then negotiation without that
LMI> algorithm should be attempted.  But is it possible to fall back to
LMI> plain-text?  I don't really know how that works.  But if that's
LMI> possible, the fall-back should obviously stop before it gets that far.

LMI> After a priority has been established, I then think that the priority
LMI> for this specific server/port pair should be saved via Customize, so
LMI> that the next connection can be done faster automatically, without the
LMI> need for all this negotiation.

Could you ask on the GnuTLS dev list?  Both of these are possible AFAICT
but perhaps they have suggestions for the implementation specifics.

Ted

Re: Tuning GnuTLS

[Lars Magne Ingebrigtsen]

Ted Zlatanov <tzz@lifelogs.com> writes:

> Could you ask on the GnuTLS dev list?  Both of these are possible AFAICT
> but perhaps they have suggestions for the implementation specifics.

Is the gnutls-devel@gnu.org the right list?

-- 
(domestic pets only, the antidote for overdose, milk.)
  bloggy blog http://lars.ingebrigtsen.no/

Re: Tuning GnuTLS

[Ted Zlatanov]

On Thu, 06 Oct 2011 21:07:49 +0200 Lars Magne Ingebrigtsen <larsi@gnus.org> wrote: 

LMI> Ted Zlatanov <tzz@lifelogs.com> writes:
>> Could you ask on the GnuTLS dev list?  Both of these are possible AFAICT
>> but perhaps they have suggestions for the implementation specifics.

LMI> Is the gnutls-devel@gnu.org the right list?

Yes (you'd use gmane.comp.encryption.gpg.gnutls.devel though)

Ted

Re: Tuning GnuTLS

[Ted Zlatanov]

On Wed, 28 Sep 2011 12:32:56 -0500 Ted Zlatanov <tzz@lifelogs.com> wrote: 

TZ> On Mon, 18 Jul 2011 05:23:16 +0200 Lars Magne Ingebrigtsen <larsi@gnus.org> wrote: 
LMI> We should strive to make TLS connections as painless as possible, and
LMI> involving as little user intervention as possible, while preserving a
LMI> reasonable level of security.

LMI> So far, two failure points have been identified:

LMI> 1) Some servers sends a prime with fewer bits than the accepted default.
LMI> I think the right thing to do here is to just default
LMI> `gnutls-min-prime-bits' to a lower number than the default GnuTLS
LMI> number.  I don't know what that number should be, but I think people who
LMI> want better bits than that can adjust this number upwards.

LMI> 2) Servers presenting broken, er, certificates with certain algorithms.
LMI> If negotiation with DHE-RSA has failed, then negotiation without that
LMI> algorithm should be attempted.  But is it possible to fall back to
LMI> plain-text?  I don't really know how that works.  But if that's
LMI> possible, the fall-back should obviously stop before it gets that far.

LMI> After a priority has been established, I then think that the priority
LMI> for this specific server/port pair should be saved via Customize, so
LMI> that the next connection can be done faster automatically, without the
LMI> need for all this negotiation.

TZ> Could you ask on the GnuTLS dev list?  Both of these are possible AFAICT
TZ> but perhaps they have suggestions for the implementation specifics.

Ping? I haven't seen anything on the GnuTLS dev list.  Do you want me to
ask?

Thanks
Ted

Re: Tuning GnuTLS

[Lars Magne Ingebrigtsen]

Ted Zlatanov <tzz@lifelogs.com> writes:

> Ping? I haven't seen anything on the GnuTLS dev list.  Do you want me to
> ask?

Yes, please.

-- 
(domestic pets only, the antidote for overdose, milk.)
  bloggy blog http://lars.ingebrigtsen.no/