Some query on ingress policing
Subsystems (2)
A message that is not in the archive (sent off-list or not yet received): [email protected]
Hi List,
It is really appreciable, if one of you can clarify me on this issue.
I was going thro the packet journey thro the network stack in Linux. Now if I want enable ingress policing, it looks like packet classification and policing should happen in netif_rx (i.e in the interrupt context) before packet is queued on to the input queue am I missing something.
A message that is not in the archive (sent off-list or not yet received): [email protected]
Yes, That is the point. Now since the ingress policing is done in net_rx_action (i.e) after queueing the packet on the input queue, the packet which was misbehaving would have used the queue resources. One would drop the packet only after you clear the backlog queue, and a greedy flow could potentially over-run the buffer of well behaved flows.
Also another serious issues is that if I ever want to provide priority for the VOIP traffic, that is not feasible since the backlog queue is drained in a FIFO manner.
Quoted 21 lines · Asim Shankar
Quoted 5 lines
I was going thro the packet journey thro the network stack in Linux. Now if I want enable ingress policing, it looks like packet classification and policing should happen in
netif_rx(i.e in the interrupt context) before packet is queued on to the input queue am I missing something.My understanding is that ingress policing (the
qdisc_ingress->enqueue()call) is done in the softint. Seenetif_receive_skb()for a call toing_filter().Alternatively, if
CONFIG_NET_CLS_ACT=nandCONFIG_NETFILTER=ythen the policing happens as anNF_IP_PRE_ROUTINGnetfilter hook instead.Also,
netif_rx()is not the single point of entry for skb's to the stack. NAPI drivers are polled for packets, so packets aren't always picked up from the NIC in the interrupt context.Hope that helps,
-- Asim