vrrp: fix vmac creation issue due to race condition

The VMAC interface sometimes does not reappear when its associated link
interface is quickly re-added after being deleted, a situation caused by
a race condition.

This problem manifests during the operations of cleanup_lost_interface()
This function checks for VMAC interfaces on top of the removed link
interface. It deletes these VMAC interfaces if they are present.
Subsequently, Netlink is invoked to refresh the information of all
interfaces, and this is followed by the cleaning of the link interface
data in memory.

The problem occurs when Netlink, queried indirectly by sub-functions of
cleanup_lost_interface(), detects that the link interface has
reappeared. Although the interface data in memory is updated
accordingly, cleanup_lost_interface() unconditionally clears this
refreshed information. As a result, the data regarding the link
interface is lost, preventing the re-creation of its associated
VMAC interfaces.

Fix the VMAC creation issue by adding a 'cleaning flag' that is set at
the start of the cleanup process. This flag says whether to proceed with
the interface data cleanup. If the interface is refreshed during the
Netlink polling, the flag is unset, thereby preventing the subsequent
clearing of the updated interface information.

Signed-off-by: Louis Scalbert <[email protected]>

keepalived/core/keepalived_netlink.c

@@ -1978,6 +1978,8 @@ netlink_if_link_populate(interface_t *ifp, struct rtattr *tb[], struct ifinfomsg
 #endif /* _HAVE_VRF_ */
 
 	ifp->rp_filter = UINT_MAX;	/* We have not read it yet */
+
+	ifp->cleaning = false;
 #endif /* _HAVE_VRRP_VMAC_ */
 
 	ifp->ifi_flags = ifi->ifi_flags;

keepalived/include/vrrp_if.h

@@ -161,6 +161,7 @@ typedef struct _interface {
 							   otherwise the physical interface */
 	bool			is_ours;		/* keepalived created the interface */
 	bool			deleting;		/* Set when we are deleting the interface */
+	bool			cleaning;		/* Set when we are cleaning the interface */
 	bool			seen_interface;		/* The interface has existed at some point since we started */
 	bool			changeable_type;	/* The interface type or underlying interface can be changed */
 #ifdef _HAVE_VRF_

keepalived/vrrp/vrrp_if.c

@@ -1394,6 +1394,11 @@ cleanup_lost_interface(interface_t *ifp)
 	tracking_obj_t *top;
 	vrrp_t *vrrp;
 
+#ifdef _HAVE_VRRP_VMAC_
+	ifp->cleaning = true;
+#endif /* _HAVE_VRRP_VMAC_ */
+
+
 	list_for_each_entry(top, &ifp->tracking_vrrp, e_list) {
 		vrrp = top->obj.vrrp;
 
@@ -1476,12 +1481,19 @@ cleanup_lost_interface(interface_t *ifp)
 
 	interface_down(ifp);
 
+#ifdef _HAVE_VRRP_VMAC_
+	if (!ifp->cleaning)
+		/* interface has been refreshed. Do not clean */
+		return;
+#endif /* _HAVE_VRRP_VMAC_ */
+
 	ifp->ifindex = 0;
 	ifp->ifi_flags = 0;
 	ifp->seen_up = false;
 #ifdef _HAVE_VRRP_VMAC_
 	if (!ifp->is_ours)
 		ifp->base_ifp = ifp;
+	ifp->cleaning = false;
 #endif
 #ifdef _HAVE_VRF_
 	ifp->vrf_master_ifp = NULL;