On 16-09-25 01:33 PM, Shmulik Ladkani wrote:
On Sun, 25 Sep 2016 09:05:08 -0400 Jamal Hadi Salim <j...@mojatatu.com> wrote:
On 16-09-23 11:40 AM, Shmulik Ladkani wrote:
[off topic]
I think this is still on topic!
Sorry, wasn't too clear on that.
What I meant is that _existing_ "egress redirect" already gets us into
crazy loops - the veth misconfig being just one example of, but
many more exist (many device stacking constructs, with lower dev issuing
an egress-redirect back to the topmost dev).
So this is stopped by the xmit_recursion Daniel mentioned, correct?
Point is, IMO loop detection (whether/how addressed), is orthogonal to
this series implementing "ingress redirect", and doesn't seem as a
strict prerequisite to adding the "ingress redirect" functionality to
act_mirred.
We can later address any loop-detection improvements in mirred.
WDYT?
If indeed the xmit_recursion solves the egress->egress problem then I
would suggest we need to address the egress->ingress issue.
BTW: plans to also address ingress->ingress?
cheers,
jamal
