Le 30/01/2017 à 16:23, David Ahern a écrit : > On 1/30/17 4:07 AM, Nicolas Dichtel wrote: >> Le 29/01/2017 à 19:02, David Ahern a écrit : >> [snip] >>> Data centers are moving to L3, and multipath is a big part of that. Anyone >>> who looks at ip -6 route enough knows it gets painful mentally pulling the >>> individual routes into a single one. >> I agree, but it's only an iproute2 problem. iproute2 could group routes to >> have >> a better output, there is no need to have a kernel patch for this ;-) >> > > iproute2 is not the only rtnetlink user. The comment above uses ip show as an > example. libnl has a workaround for IPv6 to update route objects versus > replacing them - unnecessary complexity that does not need to replicated to > iproute2, Quagga/FRR or python libraries implementing rtnetlink. Really, > RTA_MULTIPATH support in notifications should have been added when multipath > support was added to the IPv6. Patch 3 is mostly a refactoring of > rt6_fill_node to fill in nexthop information. This could have been done 4+ > years ago when RTA_MULTIPATH route adds was added to the stack. > Like I said, I fully agree that RTA_MULTIPATH is better for the dump. For the notifications, I'm not convinced. I did not this 4 years ago on purpose ;-)
I don't think that ipv4 is the right reference because the implementation is really different. ipv6 is more flexible and this implies differences in the notifications.