[ovs-dev] sFlow extension for tunnels / MPLS - question about user-space flow-cache
Jesse Gross
jesse at nicira.com
Fri Apr 24 22:02:13 UTC 2015
On Thu, Apr 23, 2015 at 9:06 PM, Neil McKee <neil.mckee at inmon.com> wrote:
> I looked into what it would take to have the kernel pass up the actions when
> it samples a packet. The changes seemed straightforward (see attached
> diff) and worked well. I prefer this approach because:
>
> (1) These really are the actions that the kernel is applying to the sampled
> packet. No assumptions. No circuitous lookup. No race conditions as flows
> are added and removed. It's straight from the horse's mouth.
> (2) This avoids burdening any of the complex userspace data-structures with
> new requirements. This way the revalidator is not under any pressure to
> have the compiled actions at his fingertips for any packet that might be
> sampled.
>
> Thoughts?
I didn't carefully review the attached diff but it looks conceptually
fine to me. One thing I did notice though - there are two paths for
userspace actions with one of them being a common case optimization
for sampling. This currently returns the actions only in that case but
I would do it in both for consistency.
Also, when you formally submit this, can you do it against the
upstream kernel and sent to the netdev mailing list, as that is the
current process?
More information about the dev
mailing list