On Wed, Oct 28, 2026 at 9:50 PM MJ Ponsonby via dev <[email protected]> wrote:
> When the VIF plug provider's prepare callback returns false, meaning the > port is not ready to be plugged, consider_plug_lport__() destroys its > port context and returns true without setting *vif_plug_port_ctx_ptr. > Both call sites in consider_plug_lport() then dereference the NULL > context (vif_plug_port_ctx->vif_plug_port_ctx_out.name), crashing > ovn-controller with a SIGSEGV. > > This is easily reproduced with the representor provider when a requested > VF cannot be resolved while the VF representors are being recreated, for > example while the compute host is rebooting. > > Fix it by making consider_plug_lport() defer and return early when > consider_plug_lport__() returns true without producing an output context, > mirroring how consider_unplug_iface() already handles a false return from > the provider's prepare callback. > > Add a vif-plug-dummy-return-false option to the dummy provider so tests > can exercise this path, and a regression test verifying that > ovn-controller defers plugging instead of crashing. > > Assisted-by: OpenCode (deepseek/deepseek-pro-latest) > nit: This is not the proper Assisted-by format, it should be <MODEL>, <TOOL> Signed-off-by: MJ Ponsonby <[email protected]> > --- > Hi MJ, thank you for the patch, I have a couple of small comments below. We are also missing the Fixes tag. > controller/vif-plug.c | 10 +++++ > lib/vif-plug-providers/dummy/vif-plug-dummy.c | 7 +++ > tests/ovn.at | 44 +++++++++++++++++++ > 3 files changed, 61 insertions(+) > > diff --git a/controller/vif-plug.c b/controller/vif-plug.c > index 8a463f6e1..5aaf1af51 100644 > --- a/controller/vif-plug.c > +++ b/controller/vif-plug.c > @@ -438,6 +438,11 @@ consider_plug_lport(const struct sbrec_port_binding > *pb, > vif_plug_ctx_in)) { > return false; > } > + if (!vif_plug_port_ctx) { > + /* The VIF plug provider asked to defer plugging/updating > + * this lport, nothing to do this iteration. */ > I think this comment is unnecessary. > + return true; > + } > > ret = consider_plug_lport_update__(vif_plug, > &iface_external_ids, > pb, lbinding->iface, > @@ -450,6 +455,11 @@ consider_plug_lport(const struct sbrec_port_binding > *pb, > vif_plug_ctx_in)) { > return false; > } > + if (!vif_plug_port_ctx) { > + /* The VIF plug provider asked to defer plugging/updating > + * this lport, nothing to do this iteration. */ > Same here. > + return true; > + } > > const struct ovsrec_interface *iface_rec = > > iface_lookup_by_name(vif_plug_ctx_in->ovsrec_interface_by_name, > diff --git a/lib/vif-plug-providers/dummy/vif-plug-dummy.c > b/lib/vif-plug-providers/dummy/vif-plug-dummy.c > index bc8953e98..23092cd6d 100644 > --- a/lib/vif-plug-providers/dummy/vif-plug-dummy.c > +++ b/lib/vif-plug-providers/dummy/vif-plug-dummy.c > @@ -70,6 +70,13 @@ vif_plug_dummy_port_prepare(const struct > vif_plug_port_ctx_in *ctx_in, > VLOG_DBG("vif_plug_dummy_port_prepare: %s", ctx_in->lport_name); > > if (ctx_in->op_type == PLUG_OP_CREATE) { > + const char *return_false = smap_get( > + &ctx_in->lport_options, "vif-plug-dummy-return-false"); > + if (return_false && !strcmp(return_false, "true")) { > Can be replaced with "if (smap_get_bool(...)) {". > + VLOG_DBG("vif_plug_dummy_port_prepare: %s - returning false > as " > + "requested", ctx_in->lport_name); > + return false; > + } > size_t lport_name_len = strlen(ctx_in->lport_name); > ctx_out->name = xzalloc(IFNAMSIZ); > memcpy(ctx_out->name, ctx_in->lport_name, > diff --git a/tests/ovn.at b/tests/ovn.at > index 780149467..2965c7dc0 100644 > --- a/tests/ovn.at > +++ b/tests/ovn.at > @@ -44042,6 +44042,50 @@ OVN_CLEANUP([hv1]) > AT_CLEANUP > ]) > > +OVN_FOR_EACH_NORTHD([ > +AT_SETUP([ovn-controller - vif-plug provider returns false]) > +AT_KEYWORDS([vif-plug]) > + > +ovn_start > + > +net_add n1 > +sim_add hv1 > +ovs-vsctl add-br br-phys > +ovn_attach n1 br-phys 192.168.0.1 > + > +check ovn-nbctl ls-add lsw0 > + > +check ovn-nbctl lsp-add lsw0 lsp1 > +check ovn-nbctl lsp-set-addresses lsp1 "f0:00:00:00:00:01 172.16.0.101" > + > +# Ask the dummy provider to signal "not ready" from its prepare callback. > +# ovn-controller must defer plugging and keep running instead of crashing. > +check ovn-nbctl lsp-set-options lsp1 \ > + requested-chassis=hv1 \ > + vif-plug-type=dummy \ > + vif-plug-dummy-return-false=true > + > +OVS_WAIT_UNTIL([ > + grep -q "Not plugging/updating lport lsp1 on direction from VIF plug > provider." hv1/ovn-controller.log > +]) > + > +# The controller must still be alive and the port must not be plugged. > +AT_CHECK([as hv1 ovs-appctl -t ovn-controller version >/dev/null], [0], > [], [ignore]) > +AT_CHECK([as hv1 ovs-vsctl get Interface lsp1 _uuid], [1], [], [ignore]) > + > +# Once the provider is ready again, the port gets plugged. > +check ovn-nbctl lsp-set-options lsp1 \ > + requested-chassis=hv1 \ > + vif-plug-type=dummy > + > +OVS_WAIT_UNTIL([ > + as hv1 ovs-vsctl get Interface lsp1 _uuid >/dev/null 2>&1 > +]) > + > +OVN_CLEANUP([hv1]) > +AT_CLEANUP > +]) > + > AT_SETUP([Patch ports not owned by OVN]) > > AT_KEYWORDS([ovn]) > -- > 2.43.0 > > _______________________________________________ > dev mailing list > [email protected] > https://mail.openvswitch.org/mailman/listinfo/ovs-dev > > Regards, Ales _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
