Have your favorite orchestrator prepopulate arp / ND tables. APIs are available.

From: Andreas Schultz <[email protected]>
Sent: Friday, December 14, 2018 8:32 AM
To: Dave Barach (dbarach) <[email protected]>
Cc: [email protected]
Subject: Re: [vpp-dev] first packet to yet unknown IP is lost

Dave Barach (dbarach) <[email protected]<mailto:[email protected]>> schrieb am 
Fr., 14. Dez. 2018 um 14:19 Uhr:
This is a conscious design decision / best practice used in routers for 
decades. We’re not going to change this behavior.

Imagine what happens when a DDoS attack sends one packet to each of a large 
number of nonexistent hosts behind a certain router: all of its packet buffers 
end up on a reinject queue waiting for ARP replies. Even with ARP throttling 
[which vpp also uses] and with aggressive drop timers, it’s a bad idea.

Mhh, I do understand your argument for packets routed through VPP. But I'm not 
talking about packets arriving from somewhere and being destined to a host on a 
local net, what I'm interested in are packets originating in VPP (either 
through a memif socket or some other API within VPP. These packets are either 
generated by the memif socket application wanting to talk to someone, an VPP 
component wanting to talk to someone or something responding to a host. In the 
first two cases, no DDoS can trigger that and the later case other mechanisms 
should protect the API from overload.

Would it be possible to mark local originating packets and treat them 
accordingly?

Andreas


D.

From: [email protected]<mailto:[email protected]> 
<[email protected]<mailto:[email protected]>> On Behalf Of Andreas Schultz
Sent: Friday, December 14, 2018 5:04 AM
To: [email protected]<mailto:[email protected]>
Subject: [vpp-dev] first packet to yet unknown IP is lost

Hi,

There seems to some problem with ARP resolution and IP forwarding.

The first packet destined to an yet unknown IP is dropped. It does triggers a 
ARP request, but the packet should be held in some kind of queue till then ARP 
response arrives and the be forwarded. Clearly this does not happen (at least 
with AF-Packet interfaces).

The simplest way to observe this to start a new VPP instance and try to ping 
its interfaces. The first ICMP request will not be answered.

Regards
Andreas
--
--
Dipl.-Inform. Andreas Schultz

----------------------- enabling your networks ----------------------
Travelping GmbH                     Phone:  +49-391-81 90 99 0
Roentgenstr. 13                     Fax:    +49-391-81 90 99 
299<tel:+49%20391%20819099299>
39108 Magdeburg                     Email:  
[email protected]<mailto:[email protected]>
GERMANY                             Web:    http://www.travelping.com
Company Registration: Amtsgericht Stendal        Reg No.:   HRB 10578
Geschaeftsfuehrer: Holger Winkelmann          VAT ID No.: DE236673780
---------------------------------------------------------------------
--
--
Dipl.-Inform. Andreas Schultz

----------------------- enabling your networks ----------------------
Travelping GmbH                     Phone:  +49-391-81 90 99 0
Roentgenstr. 13                     Fax:    +49-391-81 90 99 299
39108 Magdeburg                     Email:  
[email protected]<mailto:[email protected]>
GERMANY                             Web:    http://www.travelping.com
Company Registration: Amtsgericht Stendal        Reg No.:   HRB 10578
Geschaeftsfuehrer: Holger Winkelmann          VAT ID No.: DE236673780
---------------------------------------------------------------------
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.

View/Reply Online (#11605): https://lists.fd.io/g/vpp-dev/message/11605
Mute This Topic: https://lists.fd.io/mt/28751128/21656
Group Owner: [email protected]
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub  [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

Reply via email to