wiki:Topics-2011-08-18

Community get-together

A few community get-together options were discussed on the IRC earlier (17th Aug 2011):

  • Long weekend in a major European city (e.g. Vienna) during October-November
  • Next FOSDEM (spring 2012)

Development

Next releases

  • OpenVPN 2.2.1
  • OpenVPN 2.3

SVN merger

Dazo made a heroic merge of James' SVN branch to "master". We need to discuss this in more detail. Here are comments from cron2:

Random ramblings in the order I go through things...

   - I git-clone'd openvpn-testing.git, went to the svn-merger branch, and
     ran "make check" (on Gentoo Linux), with my full-featured t_client.rc
     setup.

     Test ran:
      - p2mp tun udp (ipv4 + ipv6 ok)
      - p2mp tun tcp (ipv4 + ipv6 ok)
      - p2mp tun udp, "topology subnet" (ipv4 + ipv6 ok)
      - p2mp tap udp (ipv4 ok, ipv6 fails, known issue with IPv6 auto-conf
                      on TAP, not related to the svn-merger)

     so the client code (at least) is still working as well as my tests
     cover the code.

   - code alignment needed: for IPv4, the "did_redirect_default_gateway"
     and "spec.remote_endpoint_defined" have been converted to flag bits in
     route_list-iflags, but for IPv6, the old structure elements
     remain - so to make the code more "in-line" for IPv4 and IPv6, this
     needs code adjustments in the IPv6 code.

   - I'm not overly happy about the "default-gateway block-local" changes -
     this is less a code issue (the code might be fine) but a procedural
     issue, with a huge change to route.c coming in without any sort of
     review or discussion.  Gah.  (No response needed).

   - I'm somewhat more annoyed by this one (route.c, line 1280):

 #if defined(TARGET_LINUX)
 #ifdef CONFIG_FEATURE_IPROUTE
    /* FIXME -- add LR_MATCH support for CONFIG_FEATURE_IPROUTE */

     this is implemented only for the "non-iproute2" case, so we have
     differing behaviour for iproute2/non-iproute2 compiles now.
     This MUST be fixed for 2.3

   - implementations of LR_MATCH for most other platforms are missing,
     but this is something that can be documented in the release notes,
     and if someone thinks they need this, they can add it - but having
     support-or-not for Linux, depending on --enable-iproute2, is a no-go

   - the merger has PF_INET6 blocks, and I currently don't run tests over
     IPv6 transport.  So maybe jjo could also take a look at this branch
     and see whether his stuff is still working.

   - the web view of route.c, "-887,13 - +894,10" looks a bit weird,
     with "++add_routes (...", but the code in the branch is fine.

   - there's a functional and potentially-fatal change here:

    void
   -delete_routes (struct route_list *rl, const struct tuntap *tt, unsigned int flags, const struct env_set *es)
   +delete_routes (struct route_list *rl, struct route_ipv6_list *rl6,
   +             const struct tuntap *tt, unsigned int flags, const struct env_set *es)
    {
 -   if (rl&&  rl-routes_added)
 +   if (rl-iflags&  RL_ROUTES_ADDED)


     this new code does not check whether "rl" is non-NULL, but in theory
     it could very well be NULL if we only have IPv6 routes.

     So route.c line 1034 should really be:

       if ( rl&&  rl-iflags&  RL_ROUTES_ADDED)

     and the corresponding code in add_routes (route.c, line 990) should
     read:

       if ( rl&&  !(rl-iflags&  RL_ROUTES_ADDED))

     ... enhancing my tests to add a "--route-nopull --route-ipv6 test"...
     and indeed:

 ./t_client.sh: Zeile 200:  8787 Speicherzugriffsfehler  ./openvpn $openvpn_conf  $LOGDIR/$SUF:openvpn.log

     *bang*

 redirect_default_route_to_vpn (rl=0x0, rl6=0x80f801c, tt=0x8101f90, flags=0,
      es=0x80dc860) at route.c:811
 811       if (rl-flags&  RG_ENABLE)

     a proper patch is attached...

   - ssl.c: it would be useful if andj or d12fk could review that - I'm not
     actually sure I understand what changed, but it seems to be some
     shuffling around of code and #ifdef ENABLE_CLIENT_CR, without actually
     changing much.


 the rest looks ok-ish to me...  (but yes, I can understand that it took
 you a heroic effort to merge that).

Andj was ok with the merge after cron2's changes on #openvpn-devel:

Ok, I can't find anything horrifying in those patches. By just looking at them (that's only ssl.c and ssl.h).
Last modified 7 years ago Last modified on 08/18/11 07:53:36

Attachments (1)

Download all attachments as: .zip