On Fri, Feb 09, 2018 at 03:39:43AM +0100, Claudio Jeker wrote:
> On netsplits it can happen that on join multiple ospfd end up as DR.
> In my case with 3 routers the one cut off stays DR even though the rest of
> the network already has a DR and BDR.
>
> Looking into this it seems that in some cases we don't issue an
> IF_EVT_NBR_CHNG and so the re-evaluation of DR/BDR does not happen.
> Looking at hello.c and the rfc seems to suggest that the following case is
> currently not handled:
>
> o Bidirectional communication has been established to a
> neighbor. In other words, the state of the neighbor has
> transitioned to 2-Way or higher.
>
> The other cases in the RFC seem to be covered.
> The following diff fixes this and seems to solve the problem I'm seeing.
>
> Since this is one of those bits that always caused trouble I would like
> more tests and maybe someone is brave enough to OK the diff.
> --
> :wq Claudio
I reproduced the issue you describe with VMs in vmm. To simulate a netsplit
I just removed the tap interface from the bridge. The vio in the VM then
still has link but can obviously not reach the others anymore.
I verified that with your patch applied this problem is solved.
No unwanted side effects happened in my testing setup.
OK remi@
>
> Index: hello.c
> ===================================================================
> RCS file: /cvs/src/usr.sbin/ospfd/hello.c,v
> retrieving revision 1.21
> diff -u -p -r1.21 hello.c
> --- hello.c 18 Nov 2014 20:54:29 -0000 1.21
> +++ hello.c 9 Feb 2018 02:11:55 -0000
> @@ -188,7 +188,6 @@ recv_hello(struct iface *iface, struct i
> nbr->dr.s_addr = hello.d_rtr;
> nbr->bdr.s_addr = hello.bd_rtr;
> nbr->priority = hello.rtr_priority;
> - nbr_change = 1;
> }
>
> /* actually the neighbor address shouldn't be stored on virtual links */
> @@ -201,8 +200,10 @@ recv_hello(struct iface *iface, struct i
> memcpy(&nbr_id, buf, sizeof(nbr_id));
> if (nbr_id == ospfe_router_id()) {
> /* seen myself */
> - if (nbr->state & NBR_STA_PRELIM)
> + if (nbr->state & NBR_STA_PRELIM) {
> nbr_fsm(nbr, NBR_EVT_2_WAY_RCVD);
> + nbr_change = 1;
> + }
> break;
> }
> buf += sizeof(nbr_id);
