Dear Nitin,
> On 22 Sep 2018, at 20:30, Nitin Saxena <[email protected]> wrote:
>
> Hi Damjan/Dave.
>
>> You will end up with layering violation and bloatig driver code. Please dont
>> go there.
> Ok. My motivation is the error packet should be dropped at the earliest. i.e
> - In device-input node, if interface is not in L2 brdge or in L2XC config
> then drop all error packets in “device-input” node itself. Except TTL expiry,
> all ipv4-input checks drops the packet.
Device input node should not drop packets. However it can provide hint (like we
do with checksum today) so next node can deal with it faster.
Few reasons why doing that is bad idea:
- if packet have errors, it should still be subject of port mirroring (SPAN)
- in some cases such packet should be sent to punt path
> - If interface is in L2 bridge or L2XC config, packet can take L2 path and
> ipv4-input stays in that path
Actually right now I'm working on completely decoupling device driver code from
l2/l3 path selection/ethertype lookup.
So that logic is likely be removed from device driver.
>
>> No, it is not, ip4 packet may also go to l2 path.
> Do you mean to say that in some cases L2 nodes are called after ip4-input?
> Could you please elaborate? “show vlib graph” output do not give any hint.
no, see above
>
>> Ip4-input is also punt node, responsible for sending specific packets to the
>> control plane. Iz should stay in the path…
> I fail to find the punt code that call ip4-input. Could you please point me
> to the code?
see src/vnet/ip/ip4_input.[ch]
--
Damjan
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#10613): https://lists.fd.io/g/vpp-dev/message/10613
Mute This Topic: https://lists.fd.io/mt/25907990/21656
Group Owner: [email protected]
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-
