On 27 June 2016 at 11:28, Joe Stringer <j...@ovn.org> wrote: > On 24 June 2016 at 23:47, Jesse Gross <je...@kernel.org> wrote: >> On Wed, Jun 22, 2016 at 6:00 PM, Joe Stringer <j...@ovn.org> wrote: >>> When the backport of ip_do_fragment() was done, it seemed as though the >>> upstream version of this function could always be trusted, provided we >>> ensure >>> that we cannot trigger the ICMP response checks inside the upstream >>> implementation. This seemed to be correct, based on the kernel module tests >>> passing on various platforms. Unfortunately, in each of the kernel module >>> testsuite tests that handle IP fragments, non-fragmented traffic would be >>> sent >>> through the datapath before the fragmented traffic is sent. As a result, the >>> refragmentation path for packets coming from userspace was not well-tested. >>> >>> This series addresses a bug found in the datapath execute path where >>> refragmentation is expected, but not currently carried out. Patches 1-2 are >>> trivial updates for upstream changes to OVS, Patch 3 is the main chunk which >>> backports the upstream ip_do_fragment() function and dependencies, and >>> patch 4 >>> ensures that the kernel module tests will now send ip fragments through the >>> upcall/execute path as well as the regular path. >> >> Other than the minor changes that I already commented on (which are >> pretty trivial), this all looks reasonable to me: >> >> Acked-by: Jesse Gross <je...@kernel.org> > > Thanks, I applied the minor changes and pushed this series to master.
Also branch-2.5, since the issue was present there as well. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
