[ovs-dev] [PATCH] dpif-netdev: Free packets on TUNNEL_PUSH if may_steal.

Stokes, Ian ian.stokes at intel.com
Thu May 24 10:02:17 UTC 2018


> On 24.05.2018 11:52, Ilya Maximets wrote:
> > On 15.05.2018 20:18, Ben Pfaff wrote:
> >> On Tue, May 15, 2018 at 02:23:38PM +0300, Ilya Maximets wrote:
> >>> Unconditional return may cause packet leak in case of 'may_steal ==
> >>> true'.
> >>>
> >>> Additionally, removed redundant checking for depth level and
> >>> clarified ignoring of the 'false' value of 'may_steal'.
> >>>
> >>> CC: Sugesh Chandran <sugesh.chandran at intel.com>
> >>> Fixes: 7c12dfc527a5 ("tunneling: Avoid datapath-recirc by
> >>>                       combining recirc actions at xlate.")
> >>> Signed-off-by: Ilya Maximets <i.maximets at samsung.com>
> >>
> >> Thanks.  This seems reasonable to me.
> >>
> >> Did you take a look at the other cases in the function to see whether
> >> they have the same problem?
> >>
> >> Since this is in dpif-netdev I'll leave the final review to Ian for
> >> his branch.
> >
> > Hello, Ian.
> > What do you think about this patch?
> > Sorry for pinging, but this is a really bad issue which drains the
> > mempool in case of badly configured OpenFlow rules. We're faced this
> > on one of SDN setups in our testing lab. Is it possible to include the
> > fix in one of the next pull requests?
> 
> I've sent v2 rebased on top of current master with updates related to
> recent patches from Darrell.
> v1 should be used for stable branches: 2.8 and 2.9.
> 

Thanks Ilya, I'll look at this now, we'll be able to include it in the new releases for branch 2.8 and 2.9 also this week.

Ian
> >
> > Best regards, Ilya Maximets.
> >


More information about the dev mailing list