Add support for recreate host interfaces tap devices on warm start (sonic-net#392)

Author: kcudnik

vslib/inc/sai_vs_state.h

@@ -135,6 +135,15 @@ typedef struct _fdb_info_t
 
 } fdb_info_t;
 
+extern std::string sai_vs_serialize_fdb_info(
+        _In_ const fdb_info_t& fdb_info);
+
+extern void sai_vs_deserialize_fdb_info(
+        _In_ const std::string& data,
+        _Out_ fdb_info_t& fdb_info);
+
+#define SAI_VS_FDB_INFO "SAI_VS_FDB_INFO"
+
 extern std::set<fdb_info_t> g_fdb_info_set;
 
 class SwitchState
@@ -268,6 +277,9 @@ void vs_clear_switch_ids();
 void vs_free_real_object_id(
         _In_ sai_object_id_t switch_id);
 
+sai_status_t vs_recreate_hostif_tap_interfaces(
+        _In_ sai_object_id_t switch_id);
+
 sai_object_id_t vs_create_real_object_id(
         _In_ sai_object_type_t object_type,
         _In_ sai_object_id_t switch_id);

vslib/src/sai_vs_generic_create.cpp

@@ -304,7 +304,28 @@ std::shared_ptr<SwitchState> vs_read_switch_database_for_warm_restart(
         std::string str_attr_id;
         std::string str_attr_value;
 
-        iss >> str_object_type >> str_object_id >> str_attr_id >> str_attr_value;
+        iss >> str_object_type >> str_object_id;
+
+        if (str_object_type == SAI_VS_FDB_INFO)
+        {
+            /*
+             * If we read line from fdb info set and use tap device is enabled
+             * just parse line and repopulate fdb info set.
+             */
+
+            if (g_vs_hostif_use_tap_device)
+            {
+                fdb_info_t fi;
+
+                sai_vs_deserialize_fdb_info(str_object_id, fi);
+
+                g_fdb_info_set.insert(fi);
+            }
+
+            continue;
+        }
+
+        iss >> str_attr_id >> str_attr_value;
 
         sai_object_meta_key_t meta_key;
 
@@ -362,6 +383,11 @@ std::shared_ptr<SwitchState> vs_read_switch_database_for_warm_restart(
 
     dumpFile.close();
 
+    if (g_vs_hostif_use_tap_device)
+    {
+        SWSS_LOG_NOTICE("loaded %zu fdb infos", g_fdb_info_set.size());
+    }
+
     SWSS_LOG_NOTICE("loaded %zu objects from: %s", count, g_warm_boot_read_file);
 
     return ss;
@@ -580,6 +606,11 @@ sai_status_t internal_vs_generic_create(
             vs_update_real_object_ids(warmBootState);
 
             vs_update_local_metadata(switch_id);
+
+            if (g_vs_hostif_use_tap_device)
+            {
+                vs_recreate_hostif_tap_interfaces(switch_id);
+            }
         }
     }
 

vslib/src/sai_vs_generic_get.cpp

@@ -41,6 +41,13 @@ sai_status_t internal_vs_generic_get(
 {
     SWSS_LOG_ENTER();
 
+    if (g_switch_state_map.find(switch_id) == g_switch_state_map.end())
+    {
+        SWSS_LOG_ERROR("failed to find switch %s in switch state map", sai_serialize_object_id(switch_id).c_str());
+
+        return SAI_STATUS_FAILURE;
+    }
+
     auto &objectHash = g_switch_state_map.at(switch_id)->objectHash.at(object_type);
 
     auto it = objectHash.find(serialized_object_id);

vslib/src/sai_vs_generic_remove.cpp

@@ -95,6 +95,21 @@ void vs_dump_switch_database_for_warm_restart(
         }
     }
 
+    if (g_vs_hostif_use_tap_device)
+    {
+        /*
+         * If user is using tap devices we also need to dump local fdb info
+         * data and restore it on warm start.
+         */
+
+        for (auto fi: g_fdb_info_set)
+        {
+            dumpFile << SAI_VS_FDB_INFO << " " << sai_vs_serialize_fdb_info(fi) << std::endl;
+        }
+
+        SWSS_LOG_NOTICE("dumped %zu fdb infos", g_fdb_info_set.size());
+    }
+
     dumpFile.close();
 
     SWSS_LOG_NOTICE("dumped %zu objects to %s", count, g_warm_boot_write_file);

vslib/src/sai_vs_hostintf.cpp

@@ -21,6 +21,10 @@
 #include <net/if_arp.h>
 #include <linux/if_ether.h>
 
+#include "swss/json.hpp"
+
+using json = nlohmann::json;
+
 // TODO on hostif remove we should stop threads
 
 typedef struct _hostif_info_t
@@ -45,6 +49,43 @@ std::map<std::string, std::shared_ptr<hostif_info_t>> hostif_info_map;
 
 std::set<fdb_info_t> g_fdb_info_set;
 
+std::string sai_vs_serialize_fdb_info(
+        _In_ const fdb_info_t& fi)
+{
+    SWSS_LOG_ENTER();
+
+    json j;
+
+    j["port_id"] = sai_serialize_object_id(fi.port_id);
+    j["vlan_id"] = sai_serialize_vlan_id(fi.vlan_id);
+    j["bridge_port_id"] = sai_serialize_object_id(fi.bridge_port_id);
+    j["fdb_entry"] = sai_serialize_fdb_entry(fi.fdb_entry);
+    j["timestamp"] = sai_serialize_number(fi.timestamp);
+
+    SWSS_LOG_INFO("item: %s", j.dump().c_str());
+
+    return j.dump();
+}
+
+void sai_vs_deserialize_fdb_info(
+        _In_ const std::string& data,
+        _Out_ fdb_info_t& fi)
+{
+    SWSS_LOG_ENTER();
+
+    SWSS_LOG_INFO("item: %s", data.c_str());
+
+    const json& j = json::parse(data);
+
+    memset(&fi, 0, sizeof(fi));
+
+    sai_deserialize_object_id(j["port_id"], fi.port_id);
+    sai_deserialize_vlan_id(j["vlan_id"], fi.vlan_id);
+    sai_deserialize_object_id(j["bridge_port_id"], fi.bridge_port_id);
+    sai_deserialize_fdb_entry(j["fdb_entry"], fi.fdb_entry);
+    sai_deserialize_number(j["timestamp"], fi.timestamp);
+}
+
 void updateLocalDB(
         _In_ const sai_fdb_event_notification_data_t &data,
         _In_ sai_fdb_event_t fdb_event)
@@ -961,24 +1002,13 @@ bool hostif_create_tap_veth_forwarding(
     return true;
 }
 
-sai_status_t vs_create_hostif_int(
-        _In_ sai_object_type_t object_type,
-        _Out_ sai_object_id_t *hostif_id,
+sai_status_t vs_create_hostif_tap_interface(
         _In_ sai_object_id_t switch_id,
         _In_ uint32_t attr_count,
         _In_ const sai_attribute_t *attr_list)
 {
     SWSS_LOG_ENTER();
 
-    if (g_vs_hostif_use_tap_device == false)
-    {
-        return vs_generic_create(object_type,
-                hostif_id,
-                switch_id,
-                attr_count,
-                attr_list);
-    }
-
     // validate SAI_HOSTIF_ATTR_TYPE
 
     auto attr_type = sai_metadata_get_attr_by_id(SAI_HOSTIF_ATTR_TYPE, attr_count, attr_list);
@@ -1061,6 +1091,8 @@ sai_status_t vs_create_hostif_int(
 
     sai_attribute_t attr;
 
+    memset(&attr, 0, sizeof(attr));
+
     attr.id = SAI_SWITCH_ATTR_SRC_MAC_ADDRESS;
 
     sai_status_t status = vs_generic_get(SAI_OBJECT_TYPE_SWITCH, switch_id, 1, &attr);
@@ -1105,7 +1137,69 @@ sai_status_t vs_create_hostif_int(
 
     SWSS_LOG_INFO("created tap interface %s", name.c_str());
 
-    return vs_generic_create(object_type,
+    return SAI_STATUS_SUCCESS;
+}
+
+sai_status_t vs_recreate_hostif_tap_interfaces(
+        _In_ sai_object_id_t switch_id)
+{
+    SWSS_LOG_ENTER();
+
+    if (g_vs_hostif_use_tap_device == false)
+    {
+        return SAI_STATUS_SUCCESS;
+    }
+
+    if (g_switch_state_map.find(switch_id) == g_switch_state_map.end())
+    {
+        SWSS_LOG_ERROR("failed to find switch %s in switch state map", sai_serialize_object_id(switch_id).c_str());
+
+        return SAI_STATUS_FAILURE;
+    }
+
+    auto &objectHash = g_switch_state_map.at(switch_id)->objectHash.at(SAI_OBJECT_TYPE_HOSTIF);
+
+    SWSS_LOG_NOTICE("attempt to recreate %zu tap devices for host interfaces", objectHash.size());
+
+    for (auto okvp: objectHash)
+    {
+        std::vector<sai_attribute_t> attrs;
+
+        for (auto akvp: okvp.second)
+        {
+            attrs.push_back(*akvp.second->getAttr());
+        }
+
+        vs_create_hostif_tap_interface(switch_id, (uint32_t)attrs.size(), attrs.data());
+    }
+
+    return SAI_STATUS_SUCCESS;
+}
+
+sai_status_t vs_create_hostif_int(
+        _In_ sai_object_type_t object_type,
+        _Out_ sai_object_id_t *hostif_id,
+        _In_ sai_object_id_t switch_id,
+        _In_ uint32_t attr_count,
+        _In_ const sai_attribute_t *attr_list)
+{
+    SWSS_LOG_ENTER();
+
+    if (g_vs_hostif_use_tap_device == true)
+    {
+        sai_status_t status = vs_create_hostif_tap_interface(
+                switch_id,
+                attr_count,
+                attr_list);
+
+        if (status != SAI_STATUS_SUCCESS)
+        {
+            return status;
+        }
+    }
+
+    return vs_generic_create(
+            object_type,
             hostif_id,
             switch_id,
             attr_count,

vslib/src/sai_vs_interfacequery.cpp

@@ -387,6 +387,8 @@ void processFdbEntriesForAging()
         return;
     }
 
+    SWSS_LOG_INFO("fdb infos to process: %zu", g_fdb_info_set.size());
+
     uint32_t current = (uint32_t)time(NULL);
 
     // find aged fdb entries
@@ -716,6 +718,7 @@ sai_status_t sai_api_initialize(
 
     g_fdbAgingThreadRun = true;
 
+    // TODO should this be moved to create switch and SwitchState?
     g_fdbAgingThread = std::make_shared<std::thread>(std::thread(fdbAgingThreadProc));
 
     g_api_initialized = true;

vslib/src/sai_vs_switch.cpp

@@ -253,8 +253,6 @@ void vs_create_netlink_message_listener(
     std::shared_ptr<std::thread> linkThread =
         std::make_shared<std::thread>(linkmsg_thread, switch_id);
 
-    linkThread->detach();
-
     sw->setLinkThread(linkThread);
 }