[Vxlanmgr] vnet netdev cleanup during config reload fix (sonic-net#2191)

* Enhanced the Existing cleanup methods i.e. getAllVxlanNetDevices & clearAllVxlanDevices to also include any stale Brvxlan interfaces and remove them.

Signed-off-by: Vivek Reddy Karri <vkarri@nvidia.com>

cfgmgr/vxlanmgr.cpp

@@ -1031,29 +1031,59 @@ int VxlanMgr::deleteVxlanNetdevice(std::string vxlan_dev_name)
     return swss::exec(cmd, res);
 }
 
+std::vector<std::string> VxlanMgr::parseNetDev(const string& stdout){
+    std::vector<std::string> netdevs;
+    std::regex device_name_pattern("^\\d+:\\s+([^:]+)");
+    std::smatch match_result;
+    auto lines = tokenize(stdout, '\n');
+    for (const std::string & line : lines)
+    {
+        SWSS_LOG_DEBUG("line : %s\n",line.c_str());
+        if (!std::regex_search(line, match_result, device_name_pattern))
+        {
+            continue;
+        }
+        std::string dev_name = match_result[1];
+        netdevs.push_back(dev_name);
+    }
+    return netdevs;
+}
+
 void VxlanMgr::getAllVxlanNetDevices()
 {
     std::string stdout;
-    const std::string cmd = std::string("") + IP_CMD + " link show type vxlan";
+
+    // Get VxLan Netdev Interfaces
+    std::string cmd = std::string("") + IP_CMD + " link show type vxlan";
     int ret = swss::exec(cmd, stdout);
     if (ret != 0)
     {
-        SWSS_LOG_ERROR("Cannot get devices by command : %s", cmd.c_str());
-        return;
+        SWSS_LOG_ERROR("Cannot get vxlan devices by command : %s", cmd.c_str());
+        stdout.clear();
     }
-    std::regex device_name_pattern("^\\d+:\\s+([^:]+)");
-    std::smatch match_result;
-    auto lines = tokenize(stdout, '\n');
-    for (const std::string & line : lines)
+    std::vector<std::string> netdevs = parseNetDev(stdout);
+    for (auto netdev : netdevs)
     {
-        SWSS_LOG_INFO("line : %s\n",line.c_str());
-        if (!std::regex_search(line, match_result, device_name_pattern))
+        m_vxlanNetDevices[netdev] = VXLAN;
+    }
+
+    // Get VxLanIf Netdev Interfaces
+    cmd = std::string("") + IP_CMD + " link show type bridge";
+    ret = swss::exec(cmd, stdout);
+    if (ret != 0)
+    {
+        SWSS_LOG_ERROR("Cannot get vxlanIf devices by command : %s", cmd.c_str());
+        stdout.clear();
+    }
+    netdevs = parseNetDev(stdout);
+    for (auto netdev : netdevs)
+    {
+        if (netdev.find(VXLAN_IF_NAME_PREFIX) == 0)
         {
-            continue;
+            m_vxlanNetDevices[netdev] = VXLAN_IF;
         }
-        std::string vxlan_dev_name = match_result[1];
-        m_vxlanNetDevices[vxlan_dev_name] = vxlan_dev_name;
     }
+
     return;
 }
 
@@ -1150,8 +1180,21 @@ void VxlanMgr::clearAllVxlanDevices()
 {
     for (auto it = m_vxlanNetDevices.begin(); it != m_vxlanNetDevices.end();)
     {
-        SWSS_LOG_INFO("Deleting Stale NetDevice vxlandevname %s\n", (it->first).c_str());
-        deleteVxlanNetdevice(it->first);
+        std::string netdev_name = it->first;
+        std::string netdev_type = it->second;
+        SWSS_LOG_INFO("Deleting Stale NetDevice %s, type: %s\n", netdev_name.c_str(), netdev_type.c_str());
+        VxlanInfo info;
+        std::string res;
+        if (netdev_type.compare(VXLAN))
+        {
+            info.m_vxlan = netdev_name;
+            cmdDeleteVxlan(info, res);
+        }
+        else if(netdev_type.compare(VXLAN_IF))
+        {
+            info.m_vxlanIf = netdev_name;
+            cmdDeleteVxlanIf(info, res);
+        }
         it = m_vxlanNetDevices.erase(it);
     }
 }

cfgmgr/vxlanmgr.h

@@ -6,6 +6,7 @@
 #include "orch.h"
 
 #include <map>
+#include <vector>
 #include <memory>
 #include <string>
 #include <utility>
@@ -70,6 +71,7 @@ class VxlanMgr : public Orch
     int createVxlanNetdevice(std::string vxlanTunnelName, std::string vni_id,
                              std::string src_ip, std::string dst_ip, std::string vlan_id);
     int deleteVxlanNetdevice(std::string vxlan_dev_name);
+    std::vector<std::string> parseNetDev(const std::string& stdout);
     void getAllVxlanNetDevices();
 
     /*

tests/test_vxlan_tunnel.py

@@ -26,6 +26,18 @@ def create_entry_pst(db, table, separator, key, pairs):
     create_entry(tbl, key, pairs)
 
 
+def delete_entry_pst(db, table, key):
+    tbl = swsscommon.ProducerStateTable(db, table)
+    tbl._del(key)
+    time.sleep(1)
+
+
+def delete_entry_tbl(db, table, key):
+    tbl = swsscommon.Table(db, table)
+    tbl._del(key)
+    time.sleep(1)
+
+
 def how_many_entries_exist(db, table):
     tbl =  swsscommon.Table(db, table)
     return len(tbl.getKeys())
@@ -324,6 +336,66 @@ def test_vxlan_term_orch(self, dvs, testlog):
         create_vxlan_tunnel_entry(dvs, 'tunnel_4', 'entry_2', tunnel_map_map, 'Vlan57', '857',
                                   tunnel_map_ids, tunnel_map_entry_ids, tunnel_ids, tunnel_term_ids)
 
+def apply_test_vnet_cfg(cfg_db):
+
+    # create VXLAN Tunnel
+    create_entry_tbl(
+        cfg_db,
+        "VXLAN_TUNNEL", '|', "tunnel1",
+        [
+            ("src_ip", "1.1.1.1")
+        ],
+    )
+
+    # create VNET
+    create_entry_tbl(
+        cfg_db,
+        "VNET", '|', "tunnel1",
+        [
+            ("vxlan_tunnel", "tunnel1"),
+            ("vni", "1")
+        ],
+    )
+
+    return 
+
+
+@pytest.fixture
+def env_setup(dvs):
+    cfg_db = swsscommon.DBConnector(swsscommon.CONFIG_DB, dvs.redis_sock, 0)
+    app_db = swsscommon.DBConnector(swsscommon.APPL_DB, dvs.redis_sock, 0)
+
+    create_entry_pst(
+        app_db,
+        "SWITCH_TABLE", ':', "switch",
+        [
+            ("vxlan_router_mac", "00:01:02:03:04:05")
+        ],
+    )
+
+    apply_test_vnet_cfg(cfg_db)
+
+    yield 
+
+    delete_entry_pst(app_db, "SWITCH_TABLE", "switch")
+    delete_entry_tbl(cfg_db, "VXLAN_TUNNEL", "tunnel1")
+    delete_entry_tbl(cfg_db, "VNET", "Vnet1")
+
+def test_vnet_cleanup_config_reload(dvs, env_setup):
+
+    # Restart vxlanmgrd Process
+    dvs.runcmd(["systemctl", "restart", "vxlanmgrd"])
+
+    # Reapply cfg to simulate cfg reload
+    cfg_db = swsscommon.DBConnector(swsscommon.CONFIG_DB, dvs.redis_sock, 0)
+    apply_test_vnet_cfg(cfg_db)
+
+    time.sleep(0.5)
+
+    # Check if the netdevices is created as expected
+    ret, stdout = dvs.runcmd(["ip", "link", "show"])
+    assert "Vxlan1" in stdout
+    assert "Brvxlan1" in stdout
 
 # Add Dummy always-pass test at end as workaroud
 # for issue when Flaky fail on final test it invokes module tear-down before retrying