route: support IPv6 and use IPv4-mapped addresses

This adds support for IPv6 in ovs-router and route-table. IPv4 is stored in
ovs-router using IPv4-mapped addresses.

Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@redhat.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>

lib/netdev-vport.c

@@ -321,7 +321,7 @@ tunnel_check_status_change__(struct netdev_vport *netdev)
 
     iface[0] = '\0';
     route = netdev->tnl_cfg.ip_dst;
-    if (ovs_router_lookup(route, iface, &gw)) {
+    if (ovs_router_lookup4(route, iface, &gw)) {
         struct netdev *egress_netdev;
 
         if (!netdev_open(iface, "system", &egress_netdev)) {

lib/ovs-router.c

@@ -49,8 +49,8 @@ static struct classifier cls;
 struct ovs_router_entry {
     struct cls_rule cr;
     char output_bridge[IFNAMSIZ];
-    ovs_be32 gw;
-    ovs_be32 nw_addr;
+    struct in6_addr gw;
+    struct in6_addr nw_addr;
     uint8_t plen;
     uint8_t priority;
 };
@@ -66,10 +66,11 @@ ovs_router_entry_cast(const struct cls_rule *cr)
 }
 
 bool
-ovs_router_lookup(ovs_be32 ip_dst, char output_bridge[], ovs_be32 *gw)
+ovs_router_lookup(const struct in6_addr *ip6_dst, char output_bridge[],
+                  struct in6_addr *gw)
 {
     const struct cls_rule *cr;
-    struct flow flow = {.nw_dst = ip_dst};
+    struct flow flow = {.ipv6_dst = *ip6_dst};
 
     cr = classifier_lookup(&cls, CLS_MAX_VERSION, &flow, NULL);
     if (cr) {
@@ -79,6 +80,20 @@ ovs_router_lookup(ovs_be32 ip_dst, char output_bridge[], ovs_be32 *gw)
         *gw = p->gw;
         return true;
     }
+    return false;
+}
+
+bool
+ovs_router_lookup4(ovs_be32 ip_dst, char output_bridge[], ovs_be32 *gw)
+{
+    struct in6_addr ip6_dst;
+    struct in6_addr gw6;
+
+    in6_addr_set_mapped_ipv4(&ip6_dst, ip_dst);
+    if (ovs_router_lookup(&ip6_dst, output_bridge, &gw6)) {
+        *gw = in6_addr_get_mapped_ipv4(&gw6);
+        return true;
+    }
     return route_table_fallback_lookup(ip_dst, output_bridge, gw);
 }
 
@@ -89,33 +104,37 @@ rt_entry_free(struct ovs_router_entry *p)
     free(p);
 }
 
-static void rt_init_match(struct match *match, ovs_be32 ip_dst, uint8_t plen)
+static void rt_init_match(struct match *match, const struct in6_addr *ip6_dst,
+                          uint8_t plen)
 {
-    ovs_be32 mask;
+    struct in6_addr dst;
+    struct in6_addr mask;
 
-    mask = be32_prefix_mask(plen);
+    mask = ipv6_create_mask(plen);
 
-    ip_dst &= mask; /* Clear out insignificant bits. */
+    dst = ipv6_addr_bitand(ip6_dst, &mask);
     memset(match, 0, sizeof *match);
-    match->flow.nw_dst = ip_dst;
-    match->wc.masks.nw_dst = mask;
+    match->flow.ipv6_dst = dst;
+    match->wc.masks.ipv6_dst = mask;
 }
 
 static void
-ovs_router_insert__(uint8_t priority, ovs_be32 ip_dst, uint8_t plen,
-                    const char output_bridge[],
-                    ovs_be32 gw)
+ovs_router_insert__(uint8_t priority, const struct in6_addr *ip6_dst,
+                    uint8_t plen, const char output_bridge[],
+                    const struct in6_addr *gw)
 {
     const struct cls_rule *cr;
     struct ovs_router_entry *p;
     struct match match;
 
-    rt_init_match(&match, ip_dst, plen);
+    rt_init_match(&match, ip6_dst, plen);
 
     p = xzalloc(sizeof *p);
     ovs_strlcpy(p->output_bridge, output_bridge, sizeof p->output_bridge);
-    p->gw = gw;
-    p->nw_addr = match.flow.nw_dst;
+    if (ipv6_addr_is_set(gw)) {
+        p->gw = *gw;
+    }
+    p->nw_addr = match.flow.ipv6_dst;
     p->plen = plen;
     p->priority = priority;
     /* Longest prefix matches first. */
@@ -134,8 +153,8 @@ ovs_router_insert__(uint8_t priority, ovs_be32 ip_dst, uint8_t plen,
 }
 
 void
-ovs_router_insert(ovs_be32 ip_dst, uint8_t plen, const char output_bridge[],
-                  ovs_be32 gw)
+ovs_router_insert(const struct in6_addr *ip_dst, uint8_t plen,
+                  const char output_bridge[], const struct in6_addr *gw)
 {
     ovs_router_insert__(plen, ip_dst, plen, output_bridge, gw);
 }
@@ -157,14 +176,14 @@ __rt_entry_delete(const struct cls_rule *cr)
 }
 
 static bool
-rt_entry_delete(uint8_t priority, ovs_be32 ip_dst, uint8_t plen)
+rt_entry_delete(uint8_t priority, const struct in6_addr *ip6_dst, uint8_t plen)
 {
     const struct cls_rule *cr;
     struct cls_rule rule;
     struct match match;
     bool res = false;
 
-    rt_init_match(&match, ip_dst, plen);
+    rt_init_match(&match, ip6_dst, plen);
 
     cls_rule_init(&rule, &match, priority);
 
@@ -178,6 +197,27 @@ rt_entry_delete(uint8_t priority, ovs_be32 ip_dst, uint8_t plen)
     return res;
 }
 
+static bool
+scan_ipv6_route(const char *s, struct in6_addr *addr, unsigned int *plen)
+{
+    int len, n;
+    int slen = strlen(s);
+    char ipv6_s[IPV6_SCAN_LEN + 1];
+
+    if (ovs_scan(s, IPV6_SCAN_FMT"%n", ipv6_s, &len)
+        && inet_pton(AF_INET6, ipv6_s, addr) == 1) {
+        if (len == slen) {
+            *plen = 128;
+            return true;
+        }
+        if (ovs_scan(s + len, "/%u%n", plen, &n)
+            && len + n == slen && *plen <= 128) {
+            return true;
+        }
+    }
+    return false;
+}
+
 static bool
 scan_ipv4_route(const char *s, ovs_be32 *addr, unsigned int *plen)
 {
@@ -216,6 +256,8 @@ ovs_router_add(struct unixctl_conn *conn, int argc,
 {
     ovs_be32 ip, gw;
     unsigned int plen;
+    struct in6_addr ip6;
+    struct in6_addr gw6;
 
     if (scan_ipv4_route(argv[1], &ip, &plen)) {
         if (argc > 3) {
@@ -223,11 +265,20 @@ ovs_router_add(struct unixctl_conn *conn, int argc,
         } else {
             gw = 0;
         }
-        ovs_router_insert__(plen + 32, ip, plen, argv[2], gw);
-        unixctl_command_reply(conn, "OK");
+        in6_addr_set_mapped_ipv4(&ip6, ip);
+        in6_addr_set_mapped_ipv4(&gw6, gw);
+        plen += 96;
+    } else if (scan_ipv6_route(argv[1], &ip6, &plen)) {
+        if (argc > 3) {
+            inet_pton(AF_INET6, argv[3], &gw6);
+        } else {
+            gw6 = in6addr_any;
+        }
     } else {
         unixctl_command_reply(conn, "Invalid parameters");
     }
+    ovs_router_insert__(plen + 32, &ip6, plen, argv[2], &gw6);
+    unixctl_command_reply(conn, "OK");
 }
 
 static void
@@ -236,18 +287,20 @@ ovs_router_del(struct unixctl_conn *conn, int argc OVS_UNUSED,
 {
     ovs_be32 ip;
     unsigned int plen;
+    struct in6_addr ip6;
 
     if (scan_ipv4_route(argv[1], &ip, &plen)) {
-
-        if (rt_entry_delete(plen + 32, ip, plen)) {
-            unixctl_command_reply(conn, "OK");
-            seq_change(tnl_conf_seq);
-        } else {
-            unixctl_command_reply(conn, "Not found");
-        }
-    } else {
+        in6_addr_set_mapped_ipv4(&ip6, ip);
+        plen += 96;
+    } else if (!scan_ipv6_route(argv[1], &ip6, &plen)) {
         unixctl_command_reply(conn, "Invalid parameters");
     }
+    if (rt_entry_delete(plen + 32, &ip6, plen)) {
+        unixctl_command_reply(conn, "OK");
+        seq_change(tnl_conf_seq);
+    } else {
+        unixctl_command_reply(conn, "Not found");
+    }
 }
 
 static void
@@ -259,16 +312,21 @@ ovs_router_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
 
     ds_put_format(&ds, "Route Table:\n");
     CLS_FOR_EACH(rt, cr, &cls) {
+        uint8_t plen;
         if (rt->priority == rt->plen) {
             ds_put_format(&ds, "Cached: ");
         } else {
             ds_put_format(&ds, "User: ");
         }
-        ds_put_format(&ds, IP_FMT"/%"PRIu16" dev %s",
-                      IP_ARGS(rt->nw_addr), rt->plen,
-                      rt->output_bridge);
-        if (rt->gw) {
-            ds_put_format(&ds, " GW "IP_FMT, IP_ARGS(rt->gw));
+        print_ipv6_mapped(&ds, &rt->nw_addr);
+        plen = rt->plen;
+        if (IN6_IS_ADDR_V4MAPPED(&rt->nw_addr)) {
+            plen -= 96;
+        }
+        ds_put_format(&ds, "/%"PRIu16" dev %s", plen, rt->output_bridge);
+        if (ipv6_addr_is_set(&rt->gw)) {
+            ds_put_format(&ds, " GW ");
+            print_ipv6_mapped(&ds, &rt->gw);
         }
         ds_put_format(&ds, "\n");
     }
@@ -281,25 +339,27 @@ ovs_router_lookup_cmd(struct unixctl_conn *conn, int argc OVS_UNUSED,
                       const char *argv[], void *aux OVS_UNUSED)
 {
     ovs_be32 ip;
+    struct in6_addr ip6;
     unsigned int plen;
+    char iface[IFNAMSIZ];
+    struct in6_addr gw;
 
     if (scan_ipv4_route(argv[1], &ip, &plen) && plen == 32) {
-        char iface[IFNAMSIZ];
-        ovs_be32 gw;
-
-        if (ovs_router_lookup(ip, iface, &gw)) {
-            struct ds ds = DS_EMPTY_INITIALIZER;
-
-            ds_put_format(&ds, "gateway " IP_FMT "\n", IP_ARGS(gw));
-            ds_put_format(&ds, "dev %s\n", iface);
-            unixctl_command_reply(conn, ds_cstr(&ds));
-            ds_destroy(&ds);
-        } else {
-            unixctl_command_reply(conn, "Not found");
-        }
-    } else {
+        in6_addr_set_mapped_ipv4(&ip6, ip);
+    } else if (!(scan_ipv6_route(argv[1], &ip6, &plen) && plen == 128)) {
         unixctl_command_reply(conn, "Invalid parameters");
     }
+
+    if (ovs_router_lookup(&ip6, iface, &gw)) {
+        struct ds ds = DS_EMPTY_INITIALIZER;
+        ds_put_format(&ds, "gateway ");
+        print_ipv6_mapped(&ds, &ip6);
+        ds_put_format(&ds, "\ndev %s\n", iface);
+        unixctl_command_reply(conn, ds_cstr(&ds));
+        ds_destroy(&ds);
+    } else {
+        unixctl_command_reply(conn, "Not found");
+    }
 }
 
 void
@@ -324,11 +384,11 @@ void
 ovs_router_init(void)
 {
     classifier_init(&cls, NULL);
-    unixctl_command_register("ovs/route/add", "ipv4_addr/prefix_len out_br_name gw", 2, 3,
+    unixctl_command_register("ovs/route/add", "ip_addr/prefix_len out_br_name gw", 2, 3,
                              ovs_router_add, NULL);
     unixctl_command_register("ovs/route/show", "", 0, 0, ovs_router_show, NULL);
-    unixctl_command_register("ovs/route/del", "ipv4_addr/prefix_len", 1, 1, ovs_router_del,
+    unixctl_command_register("ovs/route/del", "ip_addr/prefix_len", 1, 1, ovs_router_del,
                              NULL);
-    unixctl_command_register("ovs/route/lookup", "ipv4_addr", 1, 1,
+    unixctl_command_register("ovs/route/lookup", "ip_addr", 1, 1,
                              ovs_router_lookup_cmd, NULL);
 }

lib/ovs-router.h

@@ -23,10 +23,12 @@
 extern "C" {
 #endif
 
-bool ovs_router_lookup(ovs_be32 ip_dst, char out_dev[], ovs_be32 *gw);
+bool ovs_router_lookup(const struct in6_addr *ip_dst, char out_dev[],
+                        struct in6_addr *gw);
+bool ovs_router_lookup4(ovs_be32 ip_dst, char out_dev[], ovs_be32 *gw);
 void ovs_router_init(void);
-void ovs_router_insert(ovs_be32 ip_dst, uint8_t plen,
-                       const char output_bridge[], ovs_be32 gw);
+void ovs_router_insert(const struct in6_addr *ip_dst, uint8_t plen,
+                       const char output_bridge[], const struct in6_addr *gw);
 void ovs_router_flush(void);
 #ifdef  __cplusplus
 }

lib/route-table.c

@@ -30,6 +30,7 @@
 #include "netlink-socket.h"
 #include "ofpbuf.h"
 #include "ovs-router.h"
+#include "packets.h"
 #include "rtnetlink.h"
 #include "openvswitch/vlog.h"
 
@@ -40,8 +41,8 @@ struct route_data {
     unsigned char rtm_dst_len;
 
     /* Extracted from Netlink attributes. */
-    ovs_be32 rta_dst; /* 0 if missing. */
-    ovs_be32 rta_gw;
+    struct in6_addr rta_dst; /* 0 if missing. */
+    struct in6_addr rta_gw;
     char ifname[IFNAMSIZ]; /* Interface name. */
 };
 
@@ -61,8 +62,10 @@ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
 static uint64_t rt_change_seq;
 
 static struct nln *nln = NULL;
+static struct nln *nln6 = NULL;
 static struct route_table_msg rtmsg;
 static struct nln_notifier *route_notifier = NULL;
+static struct nln_notifier *route6_notifier = NULL;
 static struct nln_notifier *name_notifier = NULL;
 
 static bool route_table_valid = false;
@@ -90,15 +93,22 @@ route_table_init(void)
 {
     ovs_mutex_lock(&route_table_mutex);
     ovs_assert(!nln);
+    ovs_assert(!nln6);
     ovs_assert(!route_notifier);
+    ovs_assert(!route6_notifier);
 
     ovs_router_init();
     nln = nln_create(NETLINK_ROUTE, RTNLGRP_IPV4_ROUTE,
                      (nln_parse_func *) route_table_parse, &rtmsg);
+    nln6 = nln_create(NETLINK_ROUTE, RTNLGRP_IPV6_ROUTE,
+                      (nln_parse_func *) route_table_parse, &rtmsg);
 
     route_notifier =
         nln_notifier_create(nln, (nln_notify_func *) route_table_change,
                             NULL);
+    route6_notifier =
+        nln_notifier_create(nln6, (nln_notify_func *) route_table_change,
+                            NULL);
 
     route_table_reset();
     name_table_init();
@@ -112,9 +122,14 @@ route_table_run(void)
     OVS_EXCLUDED(route_table_mutex)
 {
     ovs_mutex_lock(&route_table_mutex);
-    if (nln) {
+    if (nln || nln6) {
         rtnetlink_run();
-        nln_run(nln);
+        if (nln) {
+            nln_run(nln);
+        }
+        if (nln6) {
+            nln_run(nln6);
+        }
 
         if (!route_table_valid) {
             route_table_reset();
@@ -129,9 +144,14 @@ route_table_wait(void)
     OVS_EXCLUDED(route_table_mutex)
 {
     ovs_mutex_lock(&route_table_mutex);
-    if (nln) {
+    if (nln || nln6) {
         rtnetlink_wait();
-        nln_wait(nln);
+        if (nln) {
+            nln_wait(nln);
+        }
+        if (nln6) {
+            nln_wait(nln6);
+        }
     }
     ovs_mutex_unlock(&route_table_mutex);
 }
@@ -153,7 +173,7 @@ route_table_reset(void)
     nl_msg_put_nlmsghdr(&request, sizeof *rtmsg, RTM_GETROUTE, NLM_F_REQUEST);
 
     rtmsg = ofpbuf_put_zeros(&request, sizeof *rtmsg);
-    rtmsg->rtgen_family = AF_INET;
+    rtmsg->rtgen_family = AF_UNSPEC;
 
     nl_dump_start(&dump, NETLINK_ROUTE, &request);
     ofpbuf_uninit(&request);
@@ -171,11 +191,10 @@ route_table_reset(void)
     return nl_dump_done(&dump);
 }
 
-
 static bool
 route_table_parse(struct ofpbuf *buf, struct route_table_msg *change)
 {
-    bool parsed;
+    bool parsed, ipv4 = false;
 
     static const struct nl_policy policy[] = {
         [RTA_DST] = { .type = NL_A_U32, .optional = true  },
@@ -183,23 +202,34 @@ route_table_parse(struct ofpbuf *buf, struct route_table_msg *change)
         [RTA_GATEWAY] = { .type = NL_A_U32, .optional = true },
     };
 
-    struct nlattr *attrs[ARRAY_SIZE(policy)];
+    static const struct nl_policy policy6[] = {
+        [RTA_DST] = { .type = NL_A_IPV6, .optional = true },
+        [RTA_OIF] = { .type = NL_A_U32, .optional = true },
+        [RTA_GATEWAY] = { .type = NL_A_IPV6, .optional = true },
+    };
 
-    parsed = nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct rtmsg),
-                             policy, attrs, ARRAY_SIZE(policy));
+    struct nlattr *attrs[ARRAY_SIZE(policy)];
+    const struct rtmsg *rtm;
+
+    rtm = ofpbuf_at(buf, NLMSG_HDRLEN, sizeof *rtm);
+
+    if (rtm->rtm_family == AF_INET) {
+        parsed = nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct rtmsg),
+                                 policy, attrs, ARRAY_SIZE(policy));
+        ipv4 = true;
+    } else if (rtm->rtm_family == AF_INET6) {
+        parsed = nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct rtmsg),
+                                 policy6, attrs, ARRAY_SIZE(policy6));
+    } else {
+        VLOG_DBG_RL(&rl, "received non AF_INET rtnetlink route message");
+        return false;
+    }
 
     if (parsed) {
-        const struct rtmsg *rtm;
         const struct nlmsghdr *nlmsg;
         int rta_oif;      /* Output interface index. */
 
         nlmsg = buf->data;
-        rtm = ofpbuf_at(buf, NLMSG_HDRLEN, sizeof *rtm);
-
-        if (rtm->rtm_family != AF_INET) {
-            VLOG_DBG_RL(&rl, "received non AF_INET rtnetlink route message");
-            return false;
-        }
 
         memset(change, 0, sizeof *change);
         change->relevant = true;
@@ -213,22 +243,38 @@ route_table_parse(struct ofpbuf *buf, struct route_table_msg *change)
             change->relevant = false;
         }
         change->nlmsg_type     = nlmsg->nlmsg_type;
-        change->rd.rtm_dst_len = rtm->rtm_dst_len;
-        rta_oif = nl_attr_get_u32(attrs[RTA_OIF]);
+        change->rd.rtm_dst_len = rtm->rtm_dst_len + (ipv4 ? 96 : 0);
+        if (attrs[RTA_OIF]) {
+            rta_oif = nl_attr_get_u32(attrs[RTA_OIF]);
 
-        if (!if_indextoname(rta_oif, change->rd.ifname)) {
-            int error = errno;
+            if (!if_indextoname(rta_oif, change->rd.ifname)) {
+                int error = errno;
 
-            VLOG_DBG_RL(&rl, "Could not find interface name[%u]: %s",
-                        rta_oif, ovs_strerror(error));
-            return false;
+                VLOG_DBG_RL(&rl, "Could not find interface name[%u]: %s",
+                            rta_oif, ovs_strerror(error));
+                return false;
+            }
         }
 
         if (attrs[RTA_DST]) {
-            change->rd.rta_dst = nl_attr_get_be32(attrs[RTA_DST]);
+            if (ipv4) {
+                ovs_be32 dst;
+                dst = nl_attr_get_be32(attrs[RTA_DST]);
+                in6_addr_set_mapped_ipv4(&change->rd.rta_dst, dst);
+            } else {
+                change->rd.rta_dst = nl_attr_get_in6_addr(attrs[RTA_DST]);
+            }
+        } else if (ipv4) {
+            in6_addr_set_mapped_ipv4(&change->rd.rta_dst, 0);
         }
         if (attrs[RTA_GATEWAY]) {
-            change->rd.rta_gw = nl_attr_get_be32(attrs[RTA_GATEWAY]);
+            if (ipv4) {
+                ovs_be32 gw;
+                gw = nl_attr_get_be32(attrs[RTA_GATEWAY]);
+                in6_addr_set_mapped_ipv4(&change->rd.rta_gw, gw);
+            } else {
+                change->rd.rta_gw = nl_attr_get_in6_addr(attrs[RTA_GATEWAY]);
+            }
         }
 
 
@@ -252,8 +298,8 @@ route_table_handle_msg(const struct route_table_msg *change)
     if (change->relevant && change->nlmsg_type == RTM_NEWROUTE) {
         const struct route_data *rd = &change->rd;
 
-        ovs_router_insert(rd->rta_dst, rd->rtm_dst_len,
-                          rd->ifname, rd->rta_gw);
+        ovs_router_insert(&rd->rta_dst, rd->rtm_dst_len,
+                          rd->ifname, &rd->rta_gw);
     }
 }
 

ofproto/ofproto-dpif-sflow.c

@@ -455,7 +455,7 @@ sflow_choose_agent_address(const char *agent_device,
             && sa.ss.ss_family == AF_INET) {
             ovs_be32 gw;
 
-            if (ovs_router_lookup(sa.sin.sin_addr.s_addr, name, &gw)
+            if (ovs_router_lookup4(sa.sin.sin_addr.s_addr, name, &gw)
                 && !netdev_get_in4_by_name(name, &in4)) {
                 goto success;
             }

ofproto/ofproto-dpif-xlate.c

@@ -2661,7 +2661,7 @@ tnl_route_lookup_flow(const struct flow *oflow,
     struct xlate_cfg *xcfg;
     ovs_be32 gw;
 
-    if (!ovs_router_lookup(oflow->tunnel.ip_dst, out_dev, &gw)) {
+    if (!ovs_router_lookup4(oflow->tunnel.ip_dst, out_dev, &gw)) {
         return -ENOENT;
     }