If OpenSSL 1.0.2 isn't in old systems, we can leave 2.2.10 alone. For the ftp site, old releases are in the "old" directory. :) Sent from my iPhone

On Oct 10, 2015, at 12:54 PM, Arran Cudbard-Bell <a.cudbardb@freeradius.org> wrote:

...

On 10 Oct 2015, at 08:57, Alan DeKok <aland@deployingradius.com> wrote:

OpenSSL 1.0.2 changes the way it interacts with FreeRADIUS. None of this is documented by OpenSSL. The result is that instead of successful authentication, you get:

(6) eap_ttls: ERROR: Invalid ACK received: 256 (6) eap_ttls: ERROR: [eaptls verify] = invalid (6) eap_ttls: ERROR: [eaptls process] = invalid

The only solution is to apply the patch in commit b7b5493c61. It doesn't fix the underlying OpenSSL problem, but it makes FreeRADIUS ignore the broken API calls.

This problem is serious enough that we may have to issue 3.0.11, and possibly 2.2.10, also.

Have to draw a line on 2.2.x this uncertainty undermines people making the case to move to v3.0.x. 1.0.2 is not included by default in any stable releases of FreeBSD, Ubuntu/Debian, Redhat/Centos, OSX.

We experienced it because homebrew has moved to OpenSSL 1.0.2.

In related news, FTP server seems to be broken.

shinyhead:freeradius-server-fork arr2036$ brew install freeradius-server ==> Downloading ftp://ftp.freeradius.org/pub/freeradius/freeradius-server-3.0.9.tar.bz2

curl: (78) RETR response: 550 Trying a mirror...

-Arran

Arran Cudbard-Bell <a.cudbardb@freeradius.org> FreeRADIUS development team

FD31 3077 42EC 7FCD 32FE 5EE2 56CF 27F9 30A8 CAA2

- List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html

On Oct 10, 2015, at 2:46 PM, Nick Lowe <nick.lowe@gmail.com> wrote:

Would it really be that big a deal to release a 2.2.10 that solves this there?

The problem is we want to encourage people to use version 3. Having version 2 live forever as a zombie release (not quite dead, not quite alive) is a bad idea. Give that this is a major problem, I'm inclined to do another release to fix the OpenSSL idiocy. But it's only under protest, and only because OpenSSL broke a fundamental API used by FreeRADIUS. Alan DeKok.

Am 11.10.2015 um 20:34 schrieb Alain Péan:

Hi Alan,

Would you mind to wait until Ubuntu 16.04 release (LTS) before dropping support for 2.2.x ? I am using at this time Ubuntu 14.04, and am hoping to upgrade to FR 3.x when Ubuntu 16.04 is released (with the hope that FR 3.x will be included in standard ubunytu repository then).

The problem is that Ubuntu is using Debian's FreeRADIUS package (unless they decide to ship their own package for 16.04) and that is pretty much stuck at 2.2.x Though there was some activity at Debian, thanks to the people who have worked on improving the debian packaging files inside the Debian source files. -- Mathieu

On Oct 11, 2015, at 2:34 PM, Alain Péan <alain.pean@lpn.cnrs.fr> wrote:

Would you mind to wait until Ubuntu 16.04 release (LTS) before dropping support for 2.2.x ?

Debian / Ubuntu are free to package software which is a decade old an unmaintained. They do it with multiple other RADIUS servers. Why be different for FreeRADIUS?

I am using at this time Ubuntu 14.04, and am hoping to upgrade to FR 3.x when Ubuntu 16.04 is released (with the hope that FR 3.x will be included in standard ubunytu repository then).

Perhaps hell will freeze over, and Debian will upgrade to a recent version of the server. Alan DeKok.

On Oct 12, 2015, at 7:50 AM, Alain Péan <alain.pean@lpn.cnrs.fr> wrote:

I have also the option to switch to CentOS (RHEL) 7. It seems that freeradius is version 3.0.4 there :

Yeah... RedHat upgrades the server... sometimes.

In general, bug corrections are backported to previous (stable in RHEL) versions. I don't know if it is the case there ?

Ask RedHat. If you're running RHEL, you have a support contract with them. If they can't answer your questions (and if they don't bother upgrading to fix bugs), I suggest getting support from someone else. Alan DeKok.

Hi Nick, It seems to me that both rhel 6.7 and centos 6.7 have same last patched freeradius version: https://rhn.redhat.com/errata/RHBA-2015-1829.html https://lists.centos.org/pipermail/centos-announce/2015-September/021404.htm... freeradius.x86_64 2.2.6-6.el6_7 as well as same openssl openssl.x86_64 1.0.1e-42.el6 Regards, Paolo.

On 13 ott 2015, at 10:20, Nick Lowe <nick.lowe@gmail.com> wrote:

Red Hat have an update available for their affected 6.7 release:

https://rhn.redhat.com/errata/RHBA-2015-1829.html

If you're running CentOS 6.7, you're likely out of luck until 6.8 becomes available:

https://bugs.centos.org/view.php?id=9295

(That bug report has seemingly been ignored.)

As this is making a lot of noise with Android 6.0 (Marshmallow), please consider releasing a 3.x and discretionary, under protest 2.x release sooner rather than later to close the loop of known possible issues:

https://code.google.com/p/android/issues/detail?id=188867

https://code.google.com/p/android-developer-preview/issues/detail?id=2136

I agree that there aren't likely to be many sites using OpenSSL 1.0.2 at this point in time, worth nipping in the bud though. - List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html

------------------------------------------------------------------------------------------------ Paolo Barbato Consorzio RFX corso Stati Uniti,4 35127 Padova - Italy Network Administrator phone: +39 049 8295097 fax: +39 049 8700718 ------------------------------------------------------------------------------------------------

On Tue, Oct 13, 2015 at 3:20 PM, Nick Lowe <nick.lowe@gmail.com> wrote:

Red Hat have an update available for their affected 6.7 release:

https://rhn.redhat.com/errata/RHBA-2015-1829.html

If you're running CentOS 6.7, you're likely out of luck until 6.8 becomes available:

https://bugs.centos.org/view.php?id=9295

(That bug report has seemingly been ignored.)

As this is making a lot of noise with Android 6.0 (Marshmallow), please consider releasing a 3.x and discretionary, under protest 2.x release sooner rather than later to close the loop of known possible issues:

https://code.google.com/p/android/issues/detail?id=188867

https://code.google.com/p/android-developer-preview/issues/detail?id=2136

I agree that there aren't likely to be many sites using OpenSSL 1.0.2 at this point in time, worth nipping in the bud though.

@Nick: aren't the links you posted a different issues from Alan's original post? Looking at the dates, as well as Arran's comment on https://code.google.com/p/android/issues/detail?id=188867#c63, the bug affecting clients using TLS-1.2 (including android Marshmallow) is fixed in 2.2.9 and 3.0.10. This should also the one addressed with RH's errata. The warning Allan posted was about servers that use openssl-1.0.2, like debian testing and ubuntu wily (which should be released this month). The fix is present in git, and will be present in the next 3.0.11 (whenever that is), but there probably won't be any 2.2.10 due to EOL policy. In which case the "fix" that admins can use is to ensure: - NOT use openssl-1.0.2. Not an issue if they already stick to LTS release anyway. OR - Build their own FR version from git, and later upgrade to 3.0.11 when that is released. I agree with Arran here. Admins wishing to use openssl-1.0.2 on their server should use the git version or 3.0.11+. -- Fajar

On Tue, Oct 13, 2015 at 3:49 PM, Nick Lowe <nick.lowe@gmail.com> wrote:

It's all mixed in together as they're all service affecting issues.

Read down that Google thread, Arran commented about this very issue.

I did. The last of his comment (which I linked above) was "FreeRADIUS 2.2.9 and FreeRADIUS 3.0.10 contain fixes for the above issues" What I DON'T see is how it is related to Alan's warning. Of course it might be because I missed something. -- Fajar

Sorry, I realise I've framed my email really poorly... I should be more careful. To clarify: The update that Red Hat and CentOS has put out solve connectivity issues -while- they're using OpenSSL 1.0.1. A 3.0.11 and potential 2.0.10 update would be needed for people who end up using OpenSSL 1.0.2. Nick

Arran Cudbard-Bell wrote:

...

On 10 Oct 2015, at 08:57, Alan DeKok <aland@deployingradius.com> wrote:

OpenSSL 1.0.2 changes the way it interacts with FreeRADIUS. None of this is documented by OpenSSL. The result is that instead of successful authentication, you get:

(6) eap_ttls: ERROR: Invalid ACK received: 256 (6) eap_ttls: ERROR: [eaptls verify] = invalid (6) eap_ttls: ERROR: [eaptls process] = invalid

The only solution is to apply the patch in commit b7b5493c61. It doesn't fix the underlying OpenSSL problem, but it makes FreeRADIUS ignore the broken API calls.

This problem is serious enough that we may have to issue 3.0.11, and possibly 2.2.10, also.

Have to draw a line on 2.2.x this uncertainty undermines people making the case to move to v3.0.x. 1.0.2 is not included by default in any stable releases of FreeBSD, Ubuntu/Debian, Redhat/Centos, OSX.

We experienced it because homebrew has moved to OpenSSL 1.0.2.

Which exact version of OpenSSL 1.0.2? I'm asking because I'm running FreeRADIUS 3.0.10 (formerly 3.0.9) on openSUSE Tumbleweed (x86_64 and armv6l) with package openssl-1.0.2d-1.1 using EAP-TTLS/PAP without issue. Maybe you're hitting the HMAC ABI incompatibility? It was fixed in 1.0.2c: https://www.openssl.org/news/changelog.html#x2 Ciao, Michael.

On Oct 10, 2015, at 2:40 PM, Michael Ströder <michael@stroeder.com> wrote:

Which exact version of OpenSSL 1.0.2?

I'm not sure... I haven't seen people posting which version of OpenSSL they're using.

I'm asking because I'm running FreeRADIUS 3.0.10 (formerly 3.0.9) on openSUSE Tumbleweed (x86_64 and armv6l) with package openssl-1.0.2d-1.1 using EAP-TTLS/PAP without issue.

Maybe you're hitting the HMAC ABI incompatibility? It was fixed in 1.0.2c: https://www.openssl.org/news/changelog.html#x2

I don't think so. But who knows... If OpenSSL 1.0.2d fixes the problem, then we can leave 2.2.9 alone, and not do 2.2.10. It's almost like the OpenSSL people just write code, and release it without doing any tests. Alan DeKok.

On Oct 11, 2015, at 11:03 AM, Axel Luttgens <axel.luttgens@skynet.be> wrote:

I got the problem with FR 3.0.10/OpenSSL 1.0.2d, before Arran told me to pull the v3.0.x HEAD.

Well, that's that, then. I've tried wandering through the OpenSSL code to see when / why / where the change was made. The code is not just complicated, it's needlessly complicated. Endless layers of obfuscation. The FreeRADIUS code looks beautiful in comparison. :( Alan DeKok.

On Oct 11, 2015, at 3:38 PM, Jouni Malinen <jkmalinen@gmail.com> wrote:

Don't ask me how to find this, but this commit is the most likely reason: https://github.com/openssl/openssl/commit/bc200e691cd68870c2062d3c1e74280a59... ('SSL/TLS record tracing code (backport from HEAD). ')

Yeah... that's probably it.

The msg_callback() calls with hardcoded version 0 for SSL3_RT_HEADER in ssl/s3_pkt.c is the extension that you are likely seeing here. The other lovely ones that you have unlikely seen yet are the msg_callback() calls with write_p == 2 (instead of the documented 0/1).

<sigh> They added tracing mechanisms... by changing their public API. That's *terrible*. It's like they have no comprehension that anyone *uses* their software.

Though, I'd assume and hope that these write_p == 2 cases do not show up without a special OpenSSL build (OPENSSL_SSL_TRACE_CRYPTO enabled with enable-ssl-trace). Anyway, you may want to be ready for them just in case and return from the callback function is write_p == 2 is seen (or maybe more robustly: if write_p is not 0 or 1).

Yes, I'll push a fix for that, too. Ugh. Alan DeKok.