bgpd: Fix EVPN-MH route cleanup race condition during interfaces flap

[krishna-samy]

Issue:
In EVPN-MH setup, when all the interfaces are flapped(using 'ifdown/ifup'), the VTEP ends up having stale route entries(via vxlan) to the multi-homed hosts even though they are connected MH hosts.

The events sequencing is as below:
ifdown br_default:

• Brings down all the host facing interfaces/vni/es
• So, new route entries(via vxlan) created for the hosts, because the multi-homed hosts(before ifdown) are not connected hosts now.

ifup br_default:

• The ES becomes active now.
• We should clear/uninstall the above created routes as the hosts are multi-homed now. This does not happen.
• This is because of the events sequencing where a local ES add event arrives bgpd before before the VNI-VRF association becomes available.
• When this happens, the current code that handles the es_evi add is skipping the unistall_local_routes (bgp_evpn_local_es_evi_add->bgp_evpn_local_es_evi_unistall_local_routes_in_vrfs).
• This is because the explicit check 'if(es_evi->es_vrf)' fails in this path as the VRF mapping is not yet received.
• Also, we never go for an attempt to clear the global routes in the same ES when the 'es_evi->es_vrf' mapping is received later.
• So, This causes stale routes in the system and traffic drop as well.

Fix:
Add 'sweep_local_routes' flag in es-evi to track when cleanup is needed but VRF is not yet available.
Once we receive vni-vrf update, clean up the local routes that were installed earlier for remote ES(now local ES hosts).

[greptile-apps]

Greptile Overview

Greptile Summary

This PR fixes a race condition in EVPN multi-homing (MH) that causes stale route entries when interfaces are flapped. The issue occurs when a local ES add event arrives before the VNI-VRF association is available, causing route cleanup to be skipped.

The fix introduces a deferred cleanup mechanism using a sweep_local_routes flag:

• When bgp_evpn_local_es_evi_uninstall_local_routes_in_vrfs() is called but VRF is not yet available, the flag is set to true
• When VNI-VRF association is later received via bgp_evpn_es_evi_vrf_ref(), the deferred cleanup is executed

Key changes:

• Added sweep_local_routes boolean flag to bgp_evpn_es_evi structure
• Modified bgp_evpn_local_es_evi_uninstall_local_routes_in_vrfs() to set flag when VRF unavailable
• Modified bgp_evpn_es_evi_vrf_ref() to check flag and execute deferred cleanup
• Initialized flag to false in bgp_evpn_es_evi_new()

Confidence Score: 4/5

• This PR is safe to merge with low risk - it addresses a specific race condition with a targeted fix
• The implementation is straightforward and correctly addresses the race condition. The flag-based deferred cleanup approach is sound. Minor concern: flag is cleared before cleanup executes, but this is acceptable given the function design. The fix is well-scoped to the specific issue.
• No files require special attention - changes are minimal and well-contained

Important Files Changed

Filename	Overview
bgpd/bgp_evpn_mh.h	Added sweep_local_routes boolean flag to bgp_evpn_es_evi struct for deferred route cleanup tracking
bgpd/bgp_evpn_mh.c	Implemented deferred route cleanup logic using flag: sets flag when VRF unavailable, clears routes when VRF becomes available

Sequence Diagram

sequenceDiagram
    participant User
    participant System
    participant ES as Ethernet Segment (ES)
    participant ESEVI as ES-EVI
    participant VRF

    Note over User,VRF: Interface Flap Scenario (ifdown/ifup)
    
    User->>System: ifdown br_default
    System->>ES: Deactivate ES
    System->>ESEVI: Remove local host interfaces
    Note over ESEVI: Multi-homed hosts become remote<br/>Create vxlan routes for hosts
    
    User->>System: ifup br_default
    System->>ES: ES becomes active (local)
    ES->>ESEVI: bgp_evpn_local_es_evi_add()
    
    alt VRF not yet available
        ESEVI->>ESEVI: bgp_evpn_local_es_evi_uninstall_local_routes_in_vrfs()
        Note over ESEVI: es_vrf is NULL
        ESEVI->>ESEVI: Set sweep_local_routes = true
        Note over ESEVI: Defer route cleanup
    end
    
    System->>VRF: VNI-VRF association received
    VRF->>ESEVI: bgp_evpn_es_evi_vrf_ref()
    ESEVI->>ESEVI: bgp_evpn_es_vrf_ref()
    Note over ESEVI: es_vrf is now set
    
    alt sweep_local_routes is true
        ESEVI->>ESEVI: Check sweep_local_routes && es_vrf
        ESEVI->>ESEVI: Set sweep_local_routes = false
        ESEVI->>ESEVI: bgp_evpn_local_es_evi_uninstall_local_routes_in_vrfs()
        Note over ESEVI: Clean up stale vxlan routes<br/>Routes now point to local hosts
    end

Loading

[krishna-samy]

ci:rerun

[krishna-samy]

ci:rerun

[ton31337]

Can't we just reuse es_evi->flags?

[krishna-samy]

Yes. Agree
updated the changes, please check.

[ton31337]

LGTM

[krishna-samy]

@ton31337
Is it good to merge ?(I hope it is not blocked by release).