Hrm, somehow the ist was dropped from the reply chain. Adding back. On 5/28/25 3:53 PM, Ilya Maximets wrote: > On 5/28/25 2:47 PM, Aaron Conole via dev wrote: >> Mike Pattrick via dev <ovs-dev@openvswitch.org> writes: >> >>> Previously ct_orig_tuple information was removed from packet metadata >>> initialization with the theory that it would always be set correctly if >>> ct_state was set. >>> >>> However, there are some cases where ct_state is set without setting this >>> value in ipf. This leads to an uninitialized value being used in >>> miniflow_extract. This was identified through ubsan: >>> >>> lib/flow.c:900:35: runtime error: load of value 190, which is not a >>> valid value for type 'bool' >>> lib/flow.c:900 miniflow_extract >>> lib/dpif-netdev.c:8317 dfc_processing >>> lib/dpif-netdev.c:8598 dp_netdev_input__ >>> lib/dpif-netdev.c:8650 dp_netdev_recirculate >>> lib/dpif-netdev.c:9045 dp_execute_cb >>> lib/odp-execute.c:890 odp_actions_impl_set >>> lib/dpif-netdev.c:9231 dp_netdev_execute_actions >>> lib/dpif-netdev.c:8419 handle_packet_upcall >>> lib/dpif-netdev.c:8515 fast_path_processing >>> lib/dpif-netdev.c:8604 dp_netdev_input__ >>> lib/dpif-netdev.c:8642 dp_netdev_input >>> lib/dpif-netdev.c:5423 dp_netdev_process_rxq_port >>> lib/dpif-netdev.c:6758 dpif_netdev_run >>> >>> SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior lib/flow.c:900:35 >>> >>> Reported-at: https://issues.redhat.com/browse/FDP-1433 >>> Fixes: af697f26b51b ("packets: Do not initialize ct_orig_tuple.") >>> Signed-off-by: Mike Pattrick <m...@redhat.com> >>> --- >> >> Good catch. >> >> Acked-by: Aaron Conole <acon...@redhat.com> > > Doesn't write_ct_md(packet, zone, NULL, NULL, NULL); call later clear these > up? > > Also, why are we not clearing mark and label as well here? They are accessed > unconditionally by the miniflow_extract as well and at least label is beyond > the original memset. > > Also, the fixes tag seems wrong. The mentioned commit is way older than the > fragmentation engine. > > Best regards, Ilya Maximets.
_______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
