While the commit message says "support" 31-bit prefix, this patch is a bug fix by nature. Whether one can actually uses a /31 subnet for *anything* (i.e. not just OpenVPN) pretty much depends entirely on the platform itself. This patch is needed simply because broadcast address does not "apply" in a /31 subnet, and having it set *prevents* it from working. In fact, if openvpn tries to set the broadcast address with `+` instead of an explicit address calculated by itself, `ip` can handle it well. I do it with a prefix length check because I am not sure if there's a reason that `broadcast +` wasn't used instead.
Yeah the removal of the p2p topology was one of the reasons. It's actually the fact that ics-openvpn doesn't really parse point-to-point ifconfig that made me aware of this. I don't know anything about sitnl. Is it available only in 2.5/master? While I have also sent the equivalent fix for that, it's merely a "forwardport". I haven't actually used 2.5/master at all (unless ics-openvpn counts, while I don't see broadcast being set for any case on Android). On Mon, 4 Nov 2019 at 02:33, Gert Doering <g...@greenie.muc.de> wrote: > > Hi, > > On Sun, Nov 03, 2019 at 07:24:43PM +0100, Gert Doering wrote: > > I tend to NAK this, on a number of reasons - we support arbitrary > > point-to-point links "since ever" if you do "topology p2p" (can be > > out of the same /31, or just arbitrary addresses on both ends), so > > I do not see why doing this in "topology subnet" would be beneficial. > > I do see where you're coming from - the man page talks about depreciating > p2p, in which case you need to make sure topology subnet does /31s. > > It might certainly be a useful excercise to investigate our current > ifconfig (etc) calls - in the 2.5 branch - and possibly get rid of all > the "broadcast" settings, across all platforms that do not need them. > > Personally I've never understood why people are so keen on explicitly > configuring broadcast addresses everywhere (like in the network config > files, etc.) - the standard address can be computed and "just works" > (and on a tun interface, there are no link-layer broadcasts anyway, even > if we pretend it were differently). > > The code was that way when David and I inherited the project, so I can't > explain *why* it is - but this might be the opportunity to kick out a bit > of needless garbage. > > > Out of curiosity: does the sitnl code path handle /31s? > > gert > > > -- > "If was one thing all people took for granted, was conviction that if you > feed honest figures into a computer, honest figures come out. Never doubted > it myself till I met a computer with a sense of humor." > Robert A. Heinlein, The Moon is a Harsh Mistress > > Gert Doering - Munich, Germany g...@greenie.muc.de _______________________________________________ Openvpn-devel mailing list Openvpn-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/openvpn-devel
