Hello Michael,
Michael Tremer:
Hello,
I agree that we should indeed do something here.
But frankly the usual case would be that people will stick with the algorithms they picked at the start and that is it. This is nothing that I would like to defend, but we have many people with hundreds or even thousands of OpenVPN connections and replacing them is not feasible for them. I should be done though, but they won't do it.
Hence I think that a big warning message on the front page is not the right thing to do. A warning on the OpenVPN page would be something that I would find better.
I agree with that.
On Wed, 2015-12-02 at 10:07 +0100, IT Superhack wrote:
Hello Michael,
Michael Tremer:
Hi,
I am probably with Erik on this.
I agree that 1024 bits are not enough any more. It is better to use longer keys and DH params if possible. However we have the same argument here that we had on the Apache thread. It is pretty much not feasible to generate these keys on many systems.
Speaking about the DH params, I agree with that, since the generation of those takes ages on older systems and they can be easily generated on a fast workstation and updated later. Maybe it would be an idea to display a notification on the WebIF page (similar to the "fireinfo is disabled" one), which spells like "The DH parameter is not strong enough. Please update a stronger one to the OpenVPN page, if possible", what do you think?
I am also not sure if labeling key sizes of 1024 bit as "insecure" is the best idea. I would prefer something like "recommended" for all higher key sizes. This is however in conflict with the argument above.
In my opinion, this is kind of a general discussion. Having in mind the patch Larsen submitted a while ago, which marked required fields with a star and not the not-required ones, the question here is the same: Should we mark the secure or the insecure options?
This was a bit different since the entire web did it the way that we didn't do it before. This is just consistent with the rest and just a change in user-experience.
Of course, both is possible, but I think it would be better to settle that for the WebIF in general to avoid inconsistent "insecure" or "recommended" tags.
Of course this should be consistent.
Another point is the use of SHA1 and (3)DES, as well as other algorithms. In my opinion, the following ciphers/algorithms are insecure:
OpenVPN page: DES-EDE3-CBC DESX-CBC DES-EDE-CBC
I have never seen any evidence that 3DES is broken. It may not be as secure as AES, but it is not generally considered weak - as far as I know.
Personally, I don't know a method to crack 3DES with normal time effort, either. But since the effective strength of 3DES is 80 bits only, it should be considered at least as "weak" in my point of view. (Source: http://www.differencebetween.net/technology/difference-between-aes-and-3des/)
For example, Ivan Ristic recommends in his book "Bulletproof SSL and TLS" not to use ciphers with a strength less than 128 bits.
I guess in the next few years we'll see the breakdown of 3DES...
DH Params with sizes 1024 and 2048 bits.
OpenVPN page - advanced options: SHA1
IPSec page - advanced connection options: SHA1
SHA1 is getting weaker and weaker, that is true. However it is the best thing that many systems can do.
In IPsec we do not allow MODP-768 or even lower to be selected. Many other appliances (especially Lancom routers are really really really really bad when it comes to encryption) require MODP-512, MD5 and DES.
And this is not just old hardware that people use. Cisco seems to default to 3DES and MD5 for their IPsec VPNs as well (at least that is what you get when you set up an IPsec connection with someone who is using their stuff).
So my point is: These appliances are out there. And allowing our users to set up a VPN connection with that is still better than nothing. It is very common that weak cryptography is used. Therefore I would like to point out that we should at best have a recommendation and show subtle warnings instead of big red error messages or preventing the users from choosing bad cryptography at all.
I don't really like what I am writing here, but if IPFire should play its role in the industry, this must be possible.
Regrettably, this is true. Even if you turn off SHA1 & Co. in IPFire, users just will set up another firewall software that still supports those ciphers. In the end, nothing is won by that.
However, I'm still happy that IPFire does not support RC4 and MD5 in VPN connections and the internal Apache webserver.
By the way: What about SHA1-suites in Apache? I disabled them recently on my local machine, and nobody has ever complaint about not getting a TLS connection to the firewall in my company. But disabling SHA1 effectively turns off TLS 1.0 in HTTPS, because SHA2/SHA3 is only supported in TLS 1.2 suites. In IPFire, this might cause problems (especially with the Internet Exploder), but it would certainly improve transport security.
Concerning OpenVPN, some of those algorithms are necessary because of older devices (Cisco ASA appliances, for example), so it would not make sense to drop them at all, as well as the 1024-DH-param.
What would you think of a patch which marks those alltogether as "insecure" (or recommends only stronger ones such as AES, CAMELLIA and SHA2)? Maybe this would be an improvement.
If you want to mark the "bad" cryptography, how about the term "weak" instead of "insecure". These are two very different things for me.
Being an cryptography expert, yes.
In my opinion (= a normal, a bit paranoid user), "weak" and "insecure" match. If a cipher is considered weak, it usually takes a short time until it is "insecure" (see RC4 for example). So, if I want to protect a secret, I'd use neither "weak" nor "insecure" ciphers - in ten years, it has the same effect.
Another thought to this topic: How long should we assume the transmitted contents of a VPN connection or a HTTPS connection to Apache to be secure at least? In most scenarios, you pick your algorithms (if you are serious about them) from the answer of that question. For example, if you want a protection of five years, AES128 is plenty (3DES fits in here, too). If you are looking for 30 years or more (including quantum computers), 3DES and AES128 are wrong choices.
And what do we do with all the old certificates? If you have migrated an IPFire installation from IPFire 2.1 to now, you will use MD5 for your certificates and you will have key lengths of 1024. If we get warnings, I think we should have these too and make a guide on the wiki how to regenerate the certificates.
Yes, indeed. I was not thinking about certificates, but they are important, too.
Speaking about certificates: At the moment, IPFire uses RSA-2048 and RSA-4096 for them. How about ECDSA? (I am not sure if OpenVPN and StrongSwan support ECDSA keys yet, but Apache certainly does so.) ECDSA is my personal favorite because it is fast and secure. An ECDSA-256 private key is equal to an RSA-2048 private key, talking about security. The generation and use of ECDSA, however, is much faster, which might be comfortable for VPN connections.
Is anybody up for doing this?
I could take care of those certificate issue, yes. Submitting a patch which adds "weak" to ciphers mentioned above (I'll leave out 3DES for this time) should also be easy, that is, if I can get send-email working. :-(
Best regards, Timmothy Wilson
Best, -Michael
Best regards, Timmothy Wilson
Best, -Michael
On Tue, 2015-11-24 at 15:14 +0100, ue wrote:
Hi Timmothy Wilson, we left the 1024 bit choice at this time in cause it provides a shortened time for the whole X509 generation. On slow boards or systems with less entropy the DH generation can take also with 2048 bit DH-parameter a long time (measured at this time up to 10 minutes with 2048 bits) . We´ve made at development time a short list which you can find here --> http://wiki.ipfire.org/en/configuration/servic es/openvpn/extensions/zertkonvert where you can find also the needed time for DH-parameter generation. May 10 Minutes for an e.g. ALIX board is a lot and may too much ? Nevertheless you can upload external generated DH-parameter over the WUI --> http://wiki.ipfire.o rg/en/configuration/services/openvpn/config/upload_gen so a prepackaged DH-parameter can also be uploaded but the generation time can be left short too.
Another thing is, could you may provide more informations about the insecurity of 2048 bit DH-parameters ? On OpenVPN hardening side they called it "Use of 2048-bit is a good minimum." --> https://community.openvpn.net/openvpn/wiki/Hardening . Shurley a longer parameter increases security but needs also lots of more time to generate and with the usage of the upload function may a better way by only hint the 1024 parameter as insecure so both is possible ?
May an "insecure" hint in the flip menu is enough ? A possible "insecure" hint could also be placed for the "Hash algorithm" in "Cryptographic options" for SHA1 --> https://www.schneier.com/blog/archives/2005/02/sha1_broken.html < -- from 2005 :-( .
Some suggestions from here.
Greetings,
Erik
Am 23.11.2015 um 15:18 schrieb IT Superhack:
The OpenVPN CGI offers to create a DH param. The patch below disables the generation of 1024 bit params and marks 2048 bit params as weak/insecure.
It is recommended to use DH params with at least 3072 bits, shorter ones are considered as insecure. The patch does not affect systems where already DH params were created.
Sorry for the crappy line breaks by my mail agent, but it cannot switch this off and git send-email does not work on my system (starttls issues).
Signed-off-by: Timmothy Wilson itsuperhack@web.de
html/cgi-bin/ovpnmain.cgi | 3 +-- langs/de/cgi-bin/de.pl | 1 + 2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/html/cgi-bin/ovpnmain.cgi b/html/cgi -bin/ovpnmain.cgi index 62af54e..4813128 100644 --- a/html/cgi-bin/ovpnmain.cgi +++ b/html/cgi-bin/ovpnmain.cgi @@ -1313,8 +1313,7 @@ END <form method='post'><input type='hidden' name='AREUSURE' value='yes' /> <input type='hidden' name='KEY' value='$cgiparams{'KEY'}' /> <select name='DHLENGHT'>
<option value='1024'$selected{'DHLENGHT'}{'1024'}>1024 $Lang::tr{'bit'}</option>
<option value='2048'$selected{'DHLENGHT'}{'2048'}>2048 $Lang::tr{'bit'}</option>
<option value='2048'$selected{'DHLENGHT'}{'2048'}>2048 $Lang::tr{'bit'} ($Lang::tr{'insecure'})</option> <option value='3072' $selected{'DHLENGHT'}{'3072'}>3072 $Lang::tr{'bit'}</option> <option value='4096' $selected{'DHLENGHT'}{'4096'}>4096 $Lang::tr{'bit'}</option> </select> diff --git a/langs/de/cgi-bin/de.pl b/langs/de/cgi-bin/de.pl index 2bca854..bfed92b 100644 --- a/langs/de/cgi-bin/de.pl +++ b/langs/de/cgi-bin/de.pl @@ -1291,6 +1291,7 @@ 'incorrect password' => 'Fehlerhaftes Passwort', 'info' => 'Info', 'init string' => 'Initialisierung:', +'insecure' => 'unsicher', 'insert floppy' => 'Legen Sie eine formatierte Diskette in das Floppy-Laufwerk in IPFire und klicken auf <i>Datensicherung auf Diskette</i>, um die Systemeinstellungen zu sichern. Überprüfen Sie das Ergebnis sorgfältig, um sicher zu sein, dass die Datensicherung vollständig und erfolgreich abgeschlossen wurde.', 'install' => 'Installieren', 'install new update' => 'Installiere neues Update:',