On Mon, Jan 07, 2019 at 06:25:45PM +0100, Håkon Bugge wrote: > > > > On 5 Jan 2019, at 04:37, Jason Gunthorpe <jgg@xxxxxxxx> wrote: > > > > On Fri, Jan 04, 2019 at 09:19:26PM +0100, Håkon Bugge wrote: > >> > >> > >>> On 4 Jan 2019, at 17:27, Jason Gunthorpe <jgg@xxxxxxxxxxxx> wrote: > >>> > >>> On Fri, Jan 04, 2019 at 02:17:20PM +0100, Håkon Bugge wrote: > >>>> When an IB port has been brought back to Active state, after being > >>>> down, ibacm gets an event about it. It will then (re) enumerate the > >>>> devices, and does so by executing an ioctl with SIOCGIFCONF. This > >>>> particular ioctl will only return interfaces that are "running". > >>> > >>> This seems like an ugly hack, should this be using netlink instead of > >>> SIOCGIFCONF to get the interface list? > >> > >> You mean the SIOCGIFCONF ioctl or the commit that is ugly? > > > > I think trying to recover from the error that is caused by not > > enumerating the interfaces correctly is pretty ugly > > OK. I take that as a Nay to my proposal: > > > I am in favour of pursuing this patch. Down the road, the ioctl method can be hardened by using netlink instead, and this patch can be reverted. > > I am looking into reading /proc/net/dev in addition to the existing > SIOCGIFCONF method. Give a hint if that can be accepted, or if a > pure netlink implementation is required. I think you should just use netlink, it is very simple and we already have libnl3 available in the build system for this purpose Jason
