RTM_DELROUTE on interface going down - when to send?

RTM_DELROUTE on interface going down - when to send?

[Michal Kubecek]

Hello,

one of our customers noticed that when an interface is switched down, no
RTM_DELROUTE message is sent for IPv4 routes being deleted (but there
are RTM_NEWROUTE messages when the interface is switched up). This is
because rather than deleting the routes with fib_table_delete(), they
are flagged with RTNH_F_DEAD in fib_sync_down_dev() and then deleted in
trie_flush_list().

Before I start with the patch, I would like to ask: would it be more
appropriate to send the message when the route is marked as dead or when
it is actually deleted from the trie?

                                                          Michal Kubecek

Re: RTM_DELROUTE on interface going down - when to send?

[Kristian Evensen]

Hi Michal,

On Tue, Dec 17, 2013 at 12:05 PM, Michal Kubecek <mkubecek@suse.cz> wrote:
>
> Before I start with the patch, I would like to ask: would it be more
> appropriate to send the message when the route is marked as dead or when
> it is actually deleted from the trie?

I noticed the same thing a few months ago and submitted a patch
(http://www.spinics.net/lists/netdev/msg254106.html). However, it was
not applied because it would cause too much traffic on large-scale
routers (see davem's reply here
http://www.spinics.net/lists/netdev/msg254186.html).

-Kristian

Re: RTM_DELROUTE on interface going down - when to send?

[Michal Kubecek]

On Tue, Dec 17, 2013 at 01:11:32PM +0100, Kristian Evensen wrote:
> On Tue, Dec 17, 2013 at 12:05 PM, Michal Kubecek <mkubecek@suse.cz> wrote:
> >
> > Before I start with the patch, I would like to ask: would it be more
> > appropriate to send the message when the route is marked as dead or when
> > it is actually deleted from the trie?
> 
> I noticed the same thing a few months ago and submitted a patch
> (http://www.spinics.net/lists/netdev/msg254106.html). However, it was
> not applied because it would cause too much traffic on large-scale
> routers (see davem's reply here
> http://www.spinics.net/lists/netdev/msg254186.html).

Thank you for the pointer. However, I agree that the inconsistencies
(IPv4 vs. IPv6, RTM_DELROUTE vs. RTM_NEWROUTE) are really unfortunate.

                                                       Michal Kubecek

Re: RTM_DELROUTE on interface going down - when to send?

[Nicolas Dichtel]

Le 17/12/2013 15:01, Michal Kubecek a écrit :
> On Tue, Dec 17, 2013 at 01:11:32PM +0100, Kristian Evensen wrote:
>> On Tue, Dec 17, 2013 at 12:05 PM, Michal Kubecek <mkubecek@suse.cz> wrote:
>>>
>>> Before I start with the patch, I would like to ask: would it be more
>>> appropriate to send the message when the route is marked as dead or when
>>> it is actually deleted from the trie?
>>
>> I noticed the same thing a few months ago and submitted a patch
>> (http://www.spinics.net/lists/netdev/msg254106.html). However, it was
>> not applied because it would cause too much traffic on large-scale
>> routers (see davem's reply here
>> http://www.spinics.net/lists/netdev/msg254186.html).
>
> Thank you for the pointer. However, I agree that the inconsistencies
> (IPv4 vs. IPv6, RTM_DELROUTE vs. RTM_NEWROUTE) are really unfortunate.
Sure. Note that this fix is regularly proposed, another pointer:
http://patchwork.ozlabs.org/patch/195516/