[VRF]: submit vrf feature

cfgmgr/intfmgr.cpp

@@ -15,6 +15,7 @@ using namespace swss;
 #define LAG_PREFIX          "PortChannel"
 #define LOOPBACK_PREFIX     "Loopback"
 #define VNET_PREFIX         "Vnet"
+#define VRF_PREFIX          "Vrf"
 
 IntfMgr::IntfMgr(DBConnector *cfgDb, DBConnector *appDb, DBConnector *stateDb, const vector<string> &tableNames) :
         Orch(cfgDb, tableNames),
@@ -50,7 +51,7 @@ void IntfMgr::setIntfIp(const string &alias, const string &opCmd,
     }
 }
 
-void IntfMgr::setIntfVrf(const string &alias, const string vrfName)
+void IntfMgr::setIntfVrf(const string &alias, const string &vrfName)
 {
     stringstream cmd;
     string res;
@@ -63,7 +64,90 @@ void IntfMgr::setIntfVrf(const string &alias, const string vrfName)
     {
         cmd << IP_CMD << " link set " << alias << " nomaster";
     }
-    EXEC_WITH_ERROR_THROW(cmd.str(), res);
+    int ret = swss::exec(cmd.str(), res);
+    if (ret)
+    {
+        SWSS_LOG_ERROR("Command '%s' failed with rc %d", cmd.str().c_str(), ret);
+    }
+}
+
+void IntfMgr::addLoopbackIntf(const string &alias)
+{
+    stringstream cmd;
+    string res;
+
+    cmd << IP_CMD << " link add " << alias << " mtu 65536 type dummy && ";
+    cmd << IP_CMD << " link set " << alias << " up";
+    int ret = swss::exec(cmd.str(), res);
+    if (ret)
+    {
+        SWSS_LOG_ERROR("Command '%s' failed with rc %d", cmd.str().c_str(), ret);
+    }
+}
+
+void IntfMgr::delLoopbackIntf(const string &alias)
+{
+    stringstream cmd;
+    string res;
+
+    cmd << IP_CMD << " link del " << alias;
+    int ret = swss::exec(cmd.str(), res);
+    if (ret)
+    {
+        SWSS_LOG_ERROR("Command '%s' failed with rc %d", cmd.str().c_str(), ret);
+    }
+}
+
+int IntfMgr::getIntfIpCount(const string &alias)
+{
+    stringstream cmd;
+    string res;
+
+    cmd << IP_CMD << " address show " << alias << " | grep inet | grep -v 'inet6 fe80:' | wc -l";
+    int ret = swss::exec(cmd.str(), res);
+    if (ret)
+    {
+        SWSS_LOG_ERROR("Command '%s' failed with rc %d", cmd.str().c_str(), ret);
+        return 0;
+    }
+
+    return std::stoi(res);
+}
+
+bool IntfMgr::isIntfGeneralDone(const string &alias)
+{
+    vector<FieldValueTuple> temp;
+
+    if (m_stateIntfTable.get(alias, temp))
+    {
+        SWSS_LOG_DEBUG("Intf %s is ready", alias.c_str());
+        return true;
+    }
+
+    return false;
+}
+
+bool IntfMgr::isIntfChangeVrf(const string &alias, const string &vrfName)
+{
+    vector<FieldValueTuple> temp;
+
+    if (m_stateIntfTable.get(alias, temp))
+    {
+        for (auto idx : temp)
+        {
+            const auto &field = fvField(idx);
+            const auto &value = fvValue(idx);
+            if (field == "vrf")
+            {
+                if (value == vrfName)
+                    return false;
+                else
+                    return true;
+            }
+	}
+    }
+
+    return false;
 }
 
 bool IntfMgr::isIntfStateOk(const string &alias)
@@ -94,6 +178,14 @@ bool IntfMgr::isIntfStateOk(const string &alias)
             return true;
         }
     }
+    else if (!alias.compare(0, strlen(VRF_PREFIX), VRF_PREFIX))
+    {
+        if (m_stateVrfTable.get(alias, temp))
+        {
+            SWSS_LOG_DEBUG("Vrf %s is ready", alias.c_str());
+            return true;
+        }
+    }
     else if (m_statePortTable.get(alias, temp))
     {
         SWSS_LOG_DEBUG("Port %s is ready", alias.c_str());
@@ -141,32 +233,38 @@ bool IntfMgr::doIntfGeneralTask(const vector<string>& keys,
             return false;
         }
 
-        // Set Interface VRF except for lo
-        if (!is_lo)
+        /* if to change vrf then skip */
+        if (isIntfChangeVrf(alias, vrf_name))
         {
-            if (!vrf_name.empty())
-            {
-                setIntfVrf(alias, vrf_name);
-            }
-            m_appIntfTableProducer.set(alias, data);
+            SWSS_LOG_ERROR("%s can not change to %s directly, skipping", alias.c_str(), vrf_name.c_str());
+            return true;
         }
-        else
+        if (is_lo)
+        {
+            addLoopbackIntf(alias);
+        }
+        if (!vrf_name.empty())
         {
-            m_appIntfTableProducer.set("lo", data);
+            setIntfVrf(alias, vrf_name);
         }
+        m_appIntfTableProducer.set(alias, data);
+        m_stateIntfTable.hset(alias, "vrf", vrf_name);
     }
     else if (op == DEL_COMMAND)
     {
-        // Set Interface VRF except for lo
-        if (!is_lo)
+        /* make sure all ip addresses associated with interface are removed, otherwise these ip address would
+           be set with global vrf and it may cause ip address confliction. */
+        if (getIntfIpCount(alias))
         {
-            setIntfVrf(alias, "");
-            m_appIntfTableProducer.del(alias);
+            return false;
         }
-        else
+        setIntfVrf(alias, "");
+        if (is_lo)
         {
-            m_appIntfTableProducer.del("lo");
+            delLoopbackIntf(alias);
         }
+        m_appIntfTableProducer.del(alias);
+        m_stateIntfTable.del(alias);
     }
     else
     {
@@ -184,26 +282,21 @@ bool IntfMgr::doIntfAddrTask(const vector<string>& keys,
 
     string alias(keys[0]);
     IpPrefix ip_prefix(keys[1]);
-    bool is_lo = !alias.compare(0, strlen(LOOPBACK_PREFIX), LOOPBACK_PREFIX);
-    string appKey = (is_lo ? "lo" : keys[0]) + ":" + keys[1];
+    string appKey = keys[0] + ":" + keys[1];
 
     if (op == SET_COMMAND)
     {
         /*
-         * Don't proceed if port/LAG/VLAN is not ready yet.
+         * Don't proceed if port/LAG/VLAN and intfGeneral are not ready yet.
          * The pending task will be checked periodically and retried.
          */
-        if (!isIntfStateOk(alias))
+        if (!isIntfStateOk(alias) || !isIntfGeneralDone(alias))
         {
             SWSS_LOG_DEBUG("Interface is not ready, skipping %s", alias.c_str());
             return false;
         }
 
-        // Set Interface IP except for lo
-        if (!is_lo)
-        {
-            setIntfIp(alias, "add", ip_prefix.to_string(), ip_prefix.isV4());
-        }
+        setIntfIp(alias, "add", ip_prefix.to_string(), ip_prefix.isV4());
 
         std::vector<FieldValueTuple> fvVector;
         FieldValueTuple f("family", ip_prefix.isV4() ? IPV4_NAME : IPV6_NAME);
@@ -216,11 +309,8 @@ bool IntfMgr::doIntfAddrTask(const vector<string>& keys,
     }
     else if (op == DEL_COMMAND)
     {
-        // Set Interface IP except for lo
-        if (!is_lo)
-        {
-            setIntfIp(alias, "del", ip_prefix.to_string(), ip_prefix.isV4());
-        }
+        setIntfIp(alias, "del", ip_prefix.to_string(), ip_prefix.isV4());
+
         m_appIntfTableProducer.del(appKey);
         m_stateIntfTable.del(keys[0] + state_db_key_delimiter + keys[1]);
     }

cfgmgr/intfmgr.h

@@ -22,11 +22,16 @@ class IntfMgr : public Orch
     Table m_statePortTable, m_stateLagTable, m_stateVlanTable, m_stateVrfTable, m_stateIntfTable;
 
     void setIntfIp(const string &alias, const string &opCmd, const string &ipPrefixStr, const bool ipv4 = true);
-    void setIntfVrf(const string &alias, const string vrfName);
+    void setIntfVrf(const string &alias, const string &vrfName);
     bool doIntfGeneralTask(const vector<string>& keys, const vector<FieldValueTuple>& data, const string& op);
     bool doIntfAddrTask(const vector<string>& keys, const vector<FieldValueTuple>& data, const string& op);
     void doTask(Consumer &consumer);
     bool isIntfStateOk(const string &alias);
+    bool isIntfGeneralDone(const string &alias);
+    bool isIntfChangeVrf(const string &alias, const string &vrfName);
+    int getIntfIpCount(const string &alias);
+    void addLoopbackIntf(const string &alias);
+    void delLoopbackIntf(const string &alias);
 };
 
 }

cfgmgr/nbrmgr.cpp

@@ -13,9 +13,6 @@
 
 using namespace swss;
 
-#define VLAN_PREFIX "Vlan"
-#define LAG_PREFIX  "PortChannel"
-
 static bool send_message(struct nl_sock *sk, struct nl_msg *msg)
 {
     bool rc = false;
@@ -66,25 +63,9 @@ bool NbrMgr::isIntfStateOk(const string &alias)
 {
     vector<FieldValueTuple> temp;
 
-    if (!alias.compare(0, strlen(VLAN_PREFIX), VLAN_PREFIX))
-    {
-        if (m_stateVlanTable.get(alias, temp))
-        {
-            SWSS_LOG_DEBUG("Vlan %s is ready", alias.c_str());
-            return true;
-        }
-    }
-    else if (!alias.compare(0, strlen(LAG_PREFIX), LAG_PREFIX))
-    {
-        if (m_stateLagTable.get(alias, temp))
-        {
-            SWSS_LOG_DEBUG("Lag %s is ready", alias.c_str());
-            return true;
-        }
-    }
-    else if (m_statePortTable.get(alias, temp))
+    if (m_stateIntfTable.get(alias, temp))
     {
-        SWSS_LOG_DEBUG("Port %s is ready", alias.c_str());
+        SWSS_LOG_DEBUG("Intf %s is ready", alias.c_str());
         return true;
     }
 

cfgmgr/vrfmgr.cpp

@@ -139,6 +139,17 @@ bool VrfMgr::setLink(const string& vrfName)
     return true;
 }
 
+int VrfMgr::getVrfIntfCount(const string& vrfName)
+{
+    stringstream cmd;
+    string res;
+
+    cmd << IP_CMD << " link show " << " | grep 'master " << vrfName << "' | wc -l";
+    EXEC_WITH_ERROR_THROW(cmd.str(), res);
+
+    return std::stoi(res);
+}
+
 void VrfMgr::doTask(Consumer &consumer)
 {
     SWSS_LOG_ENTER();
@@ -173,6 +184,12 @@ void VrfMgr::doTask(Consumer &consumer)
         }
         else if (op == DEL_COMMAND)
         {
+            if (getVrfIntfCount(vrfName))
+            {
+                it++;
+                continue;
+            }
+
             if (!delLink(vrfName))
             {
                 SWSS_LOG_ERROR("Failed to remove vrf netdev %s", vrfName.c_str());

cfgmgr/vrfmgr.h

@@ -21,6 +21,7 @@ class VrfMgr : public Orch
 private:
     bool delLink(const string& vrfName);
     bool setLink(const string& vrfName);
+    int getVrfIntfCount(const string& vrfName);
     void recycleTable(uint32_t table);
     uint32_t getFreeTable(void);
     void handleVnetConfigSet(KeyOpFieldsValuesTuple &t);

fpmsyncd/routesync.cpp

@@ -16,6 +16,7 @@ using namespace swss;
 
 #define VXLAN_IF_NAME_PREFIX    "Brvxlan"
 #define VNET_PREFIX             "Vnet"
+#define VRF_PREFIX              "Vrf"
 
 RouteSync::RouteSync(RedisPipeline *pipeline) :
     m_routeTable(pipeline, APP_ROUTE_TABLE_NAME, true),
@@ -70,10 +71,40 @@ void RouteSync::onRouteMsg(int nlmsg_type, struct nl_object *obj)
 {
     struct rtnl_route *route_obj = (struct rtnl_route *)obj;
     struct nl_addr *dip;
-    char destipprefix[MAX_ADDR_SIZE + 1] = {0};
+    char destipprefix[IFNAMSIZ + MAX_ADDR_SIZE + 2] = {0};
+
+    /*
+     * Here vrf_index is not real table id but vrf_id in zebra.
+     * It is vrf interface index in kernel for vrf route and VRF_DEFAULT(0) for global route.
+     * Now zebra fpm only fill in a uchar, ifindex limited to 255.
+     */
+    unsigned int vrf_index = rtnl_route_get_table(route_obj);
+    if (vrf_index)
+    {
+        if (!getIfName(vrf_index, destipprefix, IFNAMSIZ))
+        {
+            SWSS_LOG_ERROR("Fail to get the VRF name (ifindex %u)", vrf_index);
+            return;
+        }
+        /*
+         * Now vrf device name is required to start with VRF_PREFIX,
+         * it is difficult to split vrf_name:ipv6_addr.
+         */
+        if (memcmp(destipprefix, VRF_PREFIX, strlen(VRF_PREFIX)))
+        {
+            SWSS_LOG_ERROR("Invalid VRF name %s (ifindex %u)", destipprefix, vrf_index);
+            return;
+        }
+        destipprefix[strlen(destipprefix)] = ':';
+    }
 
     dip = rtnl_route_get_dst(route_obj);
-    nl_addr2str(dip, destipprefix, MAX_ADDR_SIZE);
+    nl_addr2str(dip, destipprefix + strlen(destipprefix), MAX_ADDR_SIZE);
+    /* Full mask route append prefix length, or else resync cannot match. */
+    if (nl_addr_get_prefixlen(dip) == (8 * nl_addr_get_len(dip)))
+    {
+        snprintf(destipprefix + strlen(destipprefix), sizeof(destipprefix) - strlen(destipprefix), "/%u", nl_addr_get_prefixlen(dip));
+    }
     SWSS_LOG_DEBUG("Receive new route message dest ip prefix: %s", destipprefix);
 
     /*
@@ -186,6 +217,11 @@ void RouteSync::onVnetRouteMsg(int nlmsg_type, struct nl_object *obj, string vne
     struct nl_addr *dip = rtnl_route_get_dst(route_obj);
     char destipprefix[MAX_ADDR_SIZE + 1] = {0};
     nl_addr2str(dip, destipprefix, MAX_ADDR_SIZE);
+    /* Full mask route append prefix length, or else resync cannot match. */
+    if (nl_addr_get_prefixlen(dip) == (8 * nl_addr_get_len(dip)))
+    {
+        snprintf(destipprefix + strlen(destipprefix), sizeof(destipprefix) - strlen(destipprefix), "/%u", nl_addr_get_prefixlen(dip));
+    }
 
     string vnet_dip =  vnet + string(":") + destipprefix;
     SWSS_LOG_DEBUG("Receive new vnet route message %s", vnet_dip.c_str());
@@ -325,6 +361,17 @@ string RouteSync::getNextHopGw(struct rtnl_route *route_obj)
             nl_addr2str(addr, gw_ip, MAX_ADDR_SIZE);
             result += gw_ip;
         }
+        else
+        {
+            if (rtnl_route_get_family(route_obj) == AF_INET)
+            {
+                result += "0.0.0.0";
+            }
+            else
+            {
+                result += "::";
+            }
+        }
 
         if (i + 1 < rtnl_route_get_nnexthops(route_obj))
         {