zebra: Fix wrong vrf change procedure

Currently the vrf change procedure for the deleted interface is after
its deletion, it causes problem for upper daemons.

Here is the problem of `bgp`:

After deletion of one **irrelevant** interface in the same vrf, its
`ifindex` is set to 0. And then, the vrf change procedure will send
"ZEBRA_INTERFACE_DOWN" to `bgpd`.

Normally, `bgp_nht_ifp_table_handle()` should igore this message for
no correlation. However, it wrongly matched `ifindex` of 0, and removed
the related routes for the down `bnc`.

Adjust the location of the vrf change procedure to fix this issue.

Signed-off-by: anlan_cs <vic.lan@pica8.com>
This commit is contained in:
anlan_cs 2023-05-22 18:32:23 +08:00
parent 88236d4e95
commit a99521a26f

View File

@ -787,6 +787,15 @@ void if_delete_update(struct interface **pifp)
/* Delete connected routes from the kernel. */
if_delete_connected(ifp);
/* if the ifp is in a vrf, move it to default so vrf can be deleted if
* desired. This operation is not done for netns implementation to avoid
* collision with interface with the same name in the default vrf (can
* occur with this implementation whereas it is not possible with
* vrf-lite).
*/
if (ifp->vrf->vrf_id && !vrf_is_backend_netns())
if_handle_vrf_change(ifp, VRF_DEFAULT);
/* Send out notification on interface delete. */
zebra_interface_delete_update(ifp);
@ -800,15 +809,6 @@ void if_delete_update(struct interface **pifp)
if_set_index(ifp, IFINDEX_INTERNAL);
ifp->node = NULL;
/* if the ifp is in a vrf, move it to default so vrf can be deleted if
* desired. This operation is not done for netns implementation to avoid
* collision with interface with the same name in the default vrf (can
* occur with this implementation whereas it is not possible with
* vrf-lite).
*/
if (ifp->vrf->vrf_id && !vrf_is_backend_netns())
if_handle_vrf_change(ifp, VRF_DEFAULT);
UNSET_FLAG(ifp->status, ZEBRA_INTERFACE_VRF_LOOPBACK);
/* Reset some zebra interface params to default values. */