inet: fix prolems in ipv6 all-nodes and all-routers address initializ…

…ation - configure the addresses only when ipv6 is enabled - configure the addresses on vlan devices - configure the addresses on ip-tunnel devices (though nothing is done) Signed-off-by: ywc689 <[email protected]>
iqiyi · Sep 13, 2024 · db72d09 · db72d09
1 parent 257d89f
commit db72d09
Show file tree

Hide file tree

Showing 11 changed files with 148 additions and 35 deletions.
diff --git a/include/ctrl.h b/include/ctrl.h
@@ -214,6 +214,7 @@ int msg_dump(const struct dpvs_msg *msg, char *buf, int len);
 #define MSG_TYPE_IPSET_SET                  40
 #define MSG_TYPE_DEST_CHECK_NOTIFY_MASTER   41
 #define MSG_TYPE_DEST_CHECK_NOTIFY_SLAVES   42
+#define MSG_TYPE_IFA_IDEVINIT               43
 #define MSG_TYPE_IPVS_RANGE_START           100
 
 /* for svc per_core, refer to service.h*/

diff --git a/include/inetaddr.h b/include/inetaddr.h
@@ -26,13 +26,19 @@
 #include "dpdk.h"
 #include "list.h"
 
+
+enum {
+    IDEV_F_NO_IPV6  = 0x00000001
+};
+
 struct inet_device {
     struct netif_port   *dev;
     struct list_head    ifa_list[DPVS_MAX_LCORE];   /* inet_ifaddr list */
     struct list_head    ifm_list[DPVS_MAX_LCORE];   /* inet_ifmcaddr list*/
     uint32_t            ifa_cnt[DPVS_MAX_LCORE];
     uint32_t            ifm_cnt[DPVS_MAX_LCORE];
     rte_atomic32_t      refcnt;                     /* not used yet */
+    uint32_t            flags;                      /* IDEV_F_XXX */
 #define this_ifa_list   ifa_list[rte_lcore_id()]
 #define this_ifm_list   ifm_list[rte_lcore_id()]
 #define this_ifa_cnt    ifa_cnt[rte_lcore_id()]
@@ -119,7 +125,7 @@ bool inet_chk_mcast_addr(int af, struct netif_port *dev,
 
 void inet_ifaddr_dad_failure(struct inet_ifaddr *ifa);
 
-int idev_add_mcast_init(struct netif_port *dev);
+int idev_addr_init(struct inet_device *idev);
 
 int inet_addr_init(void);
 int inet_addr_term(void);

diff --git a/include/ip_tunnel.h b/include/ip_tunnel.h
@@ -98,6 +98,8 @@ int ip_tunnel_xmit(struct rte_mbuf *mbuf, struct netif_port *dev,
 
 int ip_tunnel_pull_header(struct rte_mbuf *mbuf, int hlen, __be16 in_proto);
 
+int ip_tunnel_dev_init(struct netif_port *dev);
+int ip_tunnel_set_mc_list(struct netif_port *dev);
 int ip_tunnel_get_link(struct netif_port *dev, struct rte_eth_link *link);
 int ip_tunnel_get_stats(struct netif_port *dev, struct rte_eth_stats *stats);
 int ip_tunnel_get_promisc(struct netif_port *dev, bool *promisc);

diff --git a/include/ipv6.h b/include/ipv6.h
@@ -33,6 +33,13 @@
 #define IPV6
 #define RTE_LOGTYPE_IPV6    RTE_LOGTYPE_USER1
 
+struct ipv6_config {
+    unsigned disable:1;
+    unsigned forwarding:1;
+};
+
+const struct ipv6_config *ip6_config_get(void);
+
 /*
  * helper functions
  */

diff --git a/src/inetaddr.c b/src/inetaddr.c
@@ -24,6 +24,7 @@
 #include "sa_pool.h"
 #include "ndisc.h"
 #include "route.h"
+#include "ipv6.h"
 #include "route6.h"
 #include "inetaddr.h"
 #include "conf/inetaddr.h"
@@ -232,19 +233,13 @@ static int ifa_add_del_mcast(struct inet_ifaddr *ifa, bool add, bool is_master)
 }
 
 /* add ipv6 multicast address after port start */
-static int __idev_add_mcast_init(void *args)
+static int __idev_inet6_mcast_init(struct netif_port *dev)
 {
     int err;
-    struct inet_device *idev;
     union inet_addr all_nodes, all_routers;
     struct rte_ether_addr eaddr_nodes, eaddr_routers;
-    lcoreid_t cid = rte_lcore_id();
-    bool is_master = (cid == g_master_lcore_id);
-    struct netif_port *dev = (struct netif_port *) args;
-
-    if (cid >= DPVS_MAX_LCORE)
-        return EDPVS_OK;
-    idev = dev_get_idev(dev);
+    struct inet_device *idev = dev_get_idev(dev);
+    bool is_master = (rte_lcore_id() == g_master_lcore_id);
 
     memset(&eaddr_nodes, 0, sizeof(eaddr_nodes));
     memset(&eaddr_routers, 0, sizeof(eaddr_routers));
@@ -290,16 +285,54 @@ static int __idev_add_mcast_init(void *args)
     return err;
 }
 
-int idev_add_mcast_init(struct netif_port *dev)
+static int __idev_addr_init(void *args)
+{
+    struct inet_device *idev = args;
+    assert(idev != NULL && idev->dev != NULL);
+
+    if (rte_lcore_id() >= DPVS_MAX_LCORE)
+        return EDPVS_OK;
+
+    return __idev_inet6_mcast_init(idev->dev);
+}
+
+int idev_addr_init(struct inet_device *idev)
 {
     int err;
     lcoreid_t cid;
+    struct dpvs_msg *msg;
+
+    // only ipv6 needs address initialization now
+    if (ip6_config_get()->disable || (idev->flags & IDEV_F_NO_IPV6))
+        return EDPVS_OK;
+
+    if (rte_lcore_id() != rte_get_main_lcore())
+        return EDPVS_NOTSUPP;
 
-    rte_eal_mp_remote_launch(__idev_add_mcast_init, dev, CALL_MAIN);
-    RTE_LCORE_FOREACH_WORKER(cid) {
-        err = rte_eal_wait_lcore(cid);
-        if (unlikely(err < 0))
+    // do it on master lcore
+    err =  __idev_addr_init(idev);
+    if (err != EDPVS_OK)
+        return err;
+
+    // do it on slave lcores
+    if (dpvs_state_get() == DPVS_STATE_NORMAL) {
+        msg = msg_make(MSG_TYPE_IFA_IDEVINIT, ifa_msg_seq(), DPVS_MSG_MULTICAST,
+                rte_lcore_id(), sizeof(idev), &idev);
+        if (unlikely(!msg))
+            return EDPVS_NOMEM;
+        err = multicast_msg_send(msg, DPVS_MSG_F_ASYNC, NULL);
+        if (err != EDPVS_OK) {
+            msg_destroy(&msg);
             return err;
+        }
+        msg_destroy(&msg);
+    } else {
+        rte_eal_mp_remote_launch(__idev_addr_init, idev, SKIP_MAIN);
+        RTE_LCORE_FOREACH_WORKER(cid) {
+            err = rte_eal_wait_lcore(cid);
+            if (unlikely(err < 0))
+                return err;
+        }
     }
 
     return EDPVS_OK;
@@ -1346,6 +1379,17 @@ static int ifa_msg_sync_cb(struct dpvs_msg *msg)
     return EDPVS_OK;
 }
 
+static int ifa_msg_idevinit_cb(struct dpvs_msg *msg)
+{
+    struct inet_device *idev;
+
+    if (unlikely(!msg || msg->len != sizeof(idev)))
+        return EDPVS_INVAL;
+    idev = *((struct inet_device **)(msg->data));
+
+    return __idev_addr_init(idev);
+}
+
 static int __inet_addr_add(const struct ifaddr_action *param)
 {
     int err;
@@ -2115,6 +2159,13 @@ static struct dpvs_msg_type ifa_msg_types[] = {
         //.cid              = rte_get_main_lcore(),
         .unicast_msg_cb     = ifa_msg_sync_cb,
         .multicast_msg_cb   = NULL
+    },
+    {
+        .type               = MSG_TYPE_IFA_IDEVINIT,
+        .prio               = MSG_PRIO_NORM,
+        .mode               = DPVS_MSG_MULTICAST,
+        .unicast_msg_cb     = ifa_msg_idevinit_cb,
+        .multicast_msg_cb   = NULL,
     }
 };
 

diff --git a/src/ip_gre.c b/src/ip_gre.c
@@ -233,12 +233,13 @@ static int gre_dev_init(struct netif_port *dev)
 
     tnl->hlen = gre_calc_hlen(tnl->params.o_flags);
 
-    return EDPVS_OK;
+    return ip_tunnel_dev_init(dev);
 }
 
 static struct netif_ops gre_dev_ops = {
     .op_init             = gre_dev_init,
     .op_xmit             = gre_xmit,
+    .op_set_mc_list      = ip_tunnel_set_mc_list,
     .op_get_link         = ip_tunnel_get_link,
     .op_get_stats        = ip_tunnel_get_stats,
     .op_get_promisc      = ip_tunnel_get_promisc,

diff --git a/src/ip_tunnel.c b/src/ip_tunnel.c
@@ -207,6 +207,8 @@ static struct netif_port *tunnel_create(struct ip_tunnel_tab *tab,
     dev->flag &= ~NETIF_PORT_FLAG_TX_UDP_CSUM_OFFLOAD;
     dev->flag &= ~NETIF_PORT_FLAG_LLDP;
 
+    dev->in_ptr->flags |= IDEV_F_NO_IPV6;
+
     err = netif_port_register(dev);
     if (err != EDPVS_OK) {
         netif_free(dev);
@@ -900,6 +902,26 @@ int ip_tunnel_pull_header(struct rte_mbuf *mbuf, int hlen, __be16 in_proto)
     return EDPVS_OK;
 }
 
+int ip_tunnel_dev_init(struct netif_port *dev)
+{
+    int err;
+    struct ip_tunnel *tnl = netif_priv(dev);
+
+    err = idev_addr_init(tnl->dev->in_ptr);
+    if (err != EDPVS_OK)
+        return err;
+
+    return EDPVS_OK;
+}
+
+int ip_tunnel_set_mc_list(struct netif_port *dev)
+{
+    // IP tunnel devices need no hw multicast address,
+    // and should always return success
+
+    return EDPVS_OK;
+}
+
 int ip_tunnel_get_link(struct netif_port *dev, struct rte_eth_link *link)
 {
     struct ip_tunnel *tnl = netif_priv(dev);

diff --git a/src/ipip.c b/src/ipip.c
@@ -46,7 +46,9 @@ static int ipip_xmit(struct rte_mbuf *mbuf, struct netif_port *dev)
 }
 
 static struct netif_ops ipip_dev_ops = {
+    .op_init             = ip_tunnel_dev_init,
     .op_xmit             = ipip_xmit,
+    .op_set_mc_list      = ip_tunnel_set_mc_list,
     .op_get_link         = ip_tunnel_get_link,
     .op_get_stats        = ip_tunnel_get_stats,
     .op_get_promisc      = ip_tunnel_get_promisc,

diff --git a/src/ipv6/ipv6.c b/src/ipv6/ipv6.c
@@ -46,8 +46,12 @@ static rte_rwlock_t inet6_prot_lock;
 /*
  * IPv6 configures with default values.
  */
-static bool conf_ipv6_forwarding = false;
-static bool conf_ipv6_disable = false;
+static struct ipv6_config ip6_configs;
+
+const struct ipv6_config *ip6_config_get(void)
+{
+    return &ip6_configs;
+};
 
 /*
  * IPv6 statistics
@@ -115,13 +119,13 @@ static void ip6_conf_forward(vector_t tokens)
     assert(str);
 
     if (strcasecmp(str, "on") == 0)
-        conf_ipv6_forwarding = true;
+        ip6_configs.forwarding = 1;
     else if (strcasecmp(str, "off") == 0)
-        conf_ipv6_forwarding = false;
+        ip6_configs.forwarding = 0;
     else
         RTE_LOG(WARNING, IPV6, "invalid ipv6:forwarding %s\n", str);
 
-    RTE_LOG(INFO, IPV6, "ipv6:forwarding = %s\n", conf_ipv6_forwarding ? "on" : "off");
+    RTE_LOG(INFO, IPV6, "ipv6:forwarding = %s\n", ip6_configs.forwarding ? "on" : "off");
 
     FREE_PTR(str);
 }
@@ -133,13 +137,13 @@ static void ip6_conf_disable(vector_t tokens)
     assert(str);
 
     if (strcasecmp(str, "on") == 0)
-        conf_ipv6_disable = true;
+        ip6_configs.disable = 1;
     else if (strcasecmp(str, "off") == 0)
-        conf_ipv6_disable = false;
+        ip6_configs.disable = 0;
     else
         RTE_LOG(WARNING, IPV6, "invalid ipv6:disable %s\n", str);
 
-    RTE_LOG(INFO, IPV6, "ipv6:disable = %s\n", conf_ipv6_disable ? "on" : "off");
+    RTE_LOG(INFO, IPV6, "ipv6: %s\n", ip6_configs.disable ? "disabled" : "enabled");
 
     FREE_PTR(str);
 }
@@ -371,7 +375,7 @@ int ip6_output(struct rte_mbuf *mbuf)
     mbuf->port = dev->id;
 
     iftraf_pkt_out(AF_INET6, mbuf, dev);
-    if (unlikely(conf_ipv6_disable)) {
+    if (unlikely(ip6_configs.disable)) {
         IP6_INC_STATS(outdiscards);
         if (rt)
             route6_put(rt);
@@ -411,7 +415,7 @@ static int ip6_forward(struct rte_mbuf *mbuf)
     int addrtype;
     uint32_t mtu;
 
-    if (!conf_ipv6_forwarding)
+    if (!ip6_configs.forwarding)
         goto error;
 
     if (mbuf->packet_type != ETH_PKT_HOST)
@@ -539,7 +543,7 @@ static int ip6_rcv(struct rte_mbuf *mbuf, struct netif_port *dev)
     IP6_UPD_PO_STATS(in, mbuf->pkt_len);
     iftraf_pkt_in(AF_INET6, mbuf, dev);
 
-    if (unlikely(conf_ipv6_disable)) {
+    if (unlikely(ip6_configs.disable)) {
         IP6_INC_STATS(indiscards);
         goto drop;
     }
@@ -815,8 +819,8 @@ void ipv6_keyword_value_init(void)
         /* KW_TYPE_INIT keyword */
     }
     /* KW_TYPE NORMAL keyword */
-    conf_ipv6_forwarding = false;
-    conf_ipv6_disable = false;
+    ip6_configs.forwarding = 0;
+    ip6_configs.disable = 0;
 
     route6_keyword_value_init();
 }

diff --git a/src/netif.c b/src/netif.c
@@ -4013,9 +4013,9 @@ int netif_port_start(struct netif_port *port)
     if (port->netif_ops->op_update_addr)
         port->netif_ops->op_update_addr(port);
 
-    /* add in6_addr multicast address */
-    if ((ret = idev_add_mcast_init(port)) != EDPVS_OK) {
-        RTE_LOG(WARNING, NETIF, "%s: idev_add_mcast_init failed -- %d(%s)\n",
+    /* ipv6 default addresses initialization */
+    if ((ret = idev_addr_init(port->in_ptr)) != EDPVS_OK) {
+        RTE_LOG(WARNING, NETIF, "%s: idev_addr_init failed -- %d(%s)\n",
                 __func__, ret, dpvs_strerror(ret));
         return ret;
     }
@@ -4053,7 +4053,7 @@ int netif_port_register(struct netif_port *port)
 {
     struct netif_port *cur;
     int hash, nhash;
-    int err = EDPVS_OK;
+    int err;
 
     if (unlikely(NULL == port))
         return EDPVS_INVAL;
@@ -4076,10 +4076,15 @@ int netif_port_register(struct netif_port *port)
     list_add_tail(&port->nlist, &port_ntab[nhash]);
     g_nports++;
 
-    if (port->netif_ops->op_init)
+    if (port->netif_ops->op_init) {
         err = port->netif_ops->op_init(port);
+        if (err != EDPVS_OK) {
+            netif_port_unregister(port);
+            return err;
+        }
+    }
 
-    return err;
+    return EDPVS_OK;
 }
 
 int netif_port_unregister(struct netif_port *port)

diff --git a/src/vlan.c b/src/vlan.c
@@ -75,6 +75,17 @@ static int alloc_vlan_info(struct netif_port *dev)
     return EDPVS_OK;
 }
 
+static int vlan_dev_init(struct netif_port *dev)
+{
+    int err;
+
+    err = idev_addr_init(dev->in_ptr);
+    if (err != EDPVS_OK)
+        return err;
+
+    return EDPVS_OK;
+}
+
 static int vlan_xmit(struct rte_mbuf *mbuf, struct netif_port *dev)
 {
     struct vlan_dev_priv *vlan = netif_priv(dev);
@@ -171,6 +182,7 @@ static int vlan_get_stats(struct netif_port *dev, struct rte_eth_stats *stats)
 }
 
 static struct netif_ops vlan_netif_ops = {
+    .op_init             = vlan_dev_init,
     .op_xmit             = vlan_xmit,
     .op_set_mc_list      = vlan_set_mc_list,
     .op_get_queue        = vlan_get_queue,