|
From: Keller, J. E <jac...@in...> - 2019-03-15 18:24:38
|
> -----Original Message----- > From: Keller, Jacob E > Sent: Friday, March 15, 2019 11:23 AM > To: 'Robert Carter' <rob...@oc...>; linuxptp- > de...@li... > Subject: RE: [Linuxptp-devel] mcast_bind patch > > > -----Original Message----- > > From: Robert Carter [mailto:rob...@oc...] > > Sent: Friday, March 15, 2019 11:18 AM > > To: Keller, Jacob E <jac...@in...>; linuxptp- > de...@li... > > Subject: Re: [Linuxptp-devel] mcast_bind patch > > > > > It sounds like the problem isn't "it picked the subnet I didn't want" > > > but rather "different instances of ptp4l didn't consistently pick the > > > same subnet, but I don't really care which one they pick as long as > > > it's stable"? > > > > > > If it's the former, I'd rather see some sort of option that lets you > > > specify the subnet. However, being deterministic is reasonably > > > useful. > > > > > > Is my description of the second option accurate? Can you describe the > > > symptoms that happen when the wrong subnet is chosen? > > > > This issue and patch precede me, and the historical information is > > incomplete, so I recompiled without our patch and ran some tests. > > > > I'd put the problem in the "it picked the subnet I didn't want" bucket. > > Worse, it (ptp4l) also picked the "wrong" interface. > > > > Without the patch, ptpl4 is sending Sync and Announce messages out the > > incorrect interface and using a source IP address that isn't on the "-i" > > argument interface. > > > > ptp4l is bound to eno1 which has two address: > > 169.254.52.4/16 > > 169.254.53.4/16 > > > > But Sync and Announce traffic is being sourced from 10.10.100.188, which > > is on en02. Oy vey! > > > > Note, I'm building off the 2.0.0 tag, but that makes no difference. The > > multicast binding code hasn't changed. > > > > That seems weird. We should be obtaining the interface index using > sk_interface_index. > > I suspect that code is what is wrong. > What version of the Linux kernel are you using? Thanks, Jake |