vxlan: generalize forwarding tables
authorDavid Stevens <dlstevens@us.ibm.com>
Fri, 15 Mar 2013 04:35:51 +0000 (04:35 +0000)
committerDavid S. Miller <davem@davemloft.net>
Sun, 17 Mar 2013 16:23:46 +0000 (12:23 -0400)
This patch generalizes VXLAN forwarding table entries allowing an administrator
to:
1) specify multiple destinations for a given MAC
2) specify alternate vni's in the VXLAN header
3) specify alternate destination UDP ports
4) use multicast MAC addresses as fdb lookup keys
5) specify multicast destinations
6) specify the outgoing interface for forwarded packets

The combination allows configuration of more complex topologies using VXLAN
encapsulation.

Changes since v1: rebase to 3.9.0-rc2

Signed-Off-By: David L Stevens <dlstevens@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/vxlan.c
include/uapi/linux/neighbour.h
net/core/rtnetlink.c

index db0df07c18dcdbad79f066df3dd50c9c6f5feb23..33427fd62515f0ca51cfe3f5bfeb9ef73919b6e4 100644 (file)
@@ -81,13 +81,22 @@ struct vxlan_net {
        struct hlist_head vni_list[VNI_HASH_SIZE];
 };
 
+struct vxlan_rdst {
+       struct rcu_head          rcu;
+       __be32                   remote_ip;
+       __be16                   remote_port;
+       u32                      remote_vni;
+       u32                      remote_ifindex;
+       struct vxlan_rdst       *remote_next;
+};
+
 /* Forwarding table entry */
 struct vxlan_fdb {
        struct hlist_node hlist;        /* linked list of entries */
        struct rcu_head   rcu;
        unsigned long     updated;      /* jiffies */
        unsigned long     used;
-       __be32            remote_ip;
+       struct vxlan_rdst remote;
        u16               state;        /* see ndm_state */
        u8                eth_addr[ETH_ALEN];
 };
@@ -157,7 +166,8 @@ static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
 /* Fill in neighbour message in skbuff. */
 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
                           const struct vxlan_fdb *fdb,
-                          u32 portid, u32 seq, int type, unsigned int flags)
+                          u32 portid, u32 seq, int type, unsigned int flags,
+                          const struct vxlan_rdst *rdst)
 {
        unsigned long now = jiffies;
        struct nda_cacheinfo ci;
@@ -176,7 +186,7 @@ static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 
        if (type == RTM_GETNEIGH) {
                ndm->ndm_family = AF_INET;
-               send_ip = fdb->remote_ip != 0;
+               send_ip = rdst->remote_ip != htonl(INADDR_ANY);
                send_eth = !is_zero_ether_addr(fdb->eth_addr);
        } else
                ndm->ndm_family = AF_BRIDGE;
@@ -188,7 +198,17 @@ static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
        if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
                goto nla_put_failure;
 
-       if (send_ip && nla_put_be32(skb, NDA_DST, fdb->remote_ip))
+       if (send_ip && nla_put_be32(skb, NDA_DST, rdst->remote_ip))
+               goto nla_put_failure;
+
+       if (rdst->remote_port && rdst->remote_port != vxlan_port &&
+           nla_put_be16(skb, NDA_PORT, rdst->remote_port))
+               goto nla_put_failure;
+       if (rdst->remote_vni != vxlan->vni &&
+           nla_put_be32(skb, NDA_VNI, rdst->remote_vni))
+               goto nla_put_failure;
+       if (rdst->remote_ifindex &&
+           nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
                goto nla_put_failure;
 
        ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
@@ -211,6 +231,9 @@ static inline size_t vxlan_nlmsg_size(void)
        return NLMSG_ALIGN(sizeof(struct ndmsg))
                + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
                + nla_total_size(sizeof(__be32)) /* NDA_DST */
+               + nla_total_size(sizeof(__be32)) /* NDA_PORT */
+               + nla_total_size(sizeof(__be32)) /* NDA_VNI */
+               + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
                + nla_total_size(sizeof(struct nda_cacheinfo));
 }
 
@@ -225,7 +248,7 @@ static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
        if (skb == NULL)
                goto errout;
 
-       err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0);
+       err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, &fdb->remote);
        if (err < 0) {
                /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
                WARN_ON(err == -EMSGSIZE);
@@ -247,7 +270,8 @@ static void vxlan_ip_miss(struct net_device *dev, __be32 ipa)
 
        memset(&f, 0, sizeof f);
        f.state = NUD_STALE;
-       f.remote_ip = ipa; /* goes to NDA_DST */
+       f.remote.remote_ip = ipa; /* goes to NDA_DST */
+       f.remote.remote_vni = VXLAN_N_VID;
 
        vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
 }
@@ -300,10 +324,38 @@ static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
        return NULL;
 }
 
+/* Add/update destinations for multicast */
+static int vxlan_fdb_append(struct vxlan_fdb *f,
+                           __be32 ip, __u32 port, __u32 vni, __u32 ifindex)
+{
+       struct vxlan_rdst *rd_prev, *rd;
+
+       rd_prev = NULL;
+       for (rd = &f->remote; rd; rd = rd->remote_next) {
+               if (rd->remote_ip == ip &&
+                   rd->remote_port == port &&
+                   rd->remote_vni == vni &&
+                   rd->remote_ifindex == ifindex)
+                       return 0;
+               rd_prev = rd;
+       }
+       rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
+       if (rd == NULL)
+               return -ENOBUFS;
+       rd->remote_ip = ip;
+       rd->remote_port = port;
+       rd->remote_vni = vni;
+       rd->remote_ifindex = ifindex;
+       rd->remote_next = NULL;
+       rd_prev->remote_next = rd;
+       return 1;
+}
+
 /* Add new entry to forwarding table -- assumes lock held */
 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
                            const u8 *mac, __be32 ip,
-                           __u16 state, __u16 flags)
+                           __u16 state, __u16 flags,
+                           __u32 port, __u32 vni, __u32 ifindex)
 {
        struct vxlan_fdb *f;
        int notify = 0;
@@ -320,6 +372,14 @@ static int vxlan_fdb_create(struct vxlan_dev *vxlan,
                        f->updated = jiffies;
                        notify = 1;
                }
+               if ((flags & NLM_F_APPEND) &&
+                   is_multicast_ether_addr(f->eth_addr)) {
+                       int rc = vxlan_fdb_append(f, ip, port, vni, ifindex);
+
+                       if (rc < 0)
+                               return rc;
+                       notify |= rc;
+               }
        } else {
                if (!(flags & NLM_F_CREATE))
                        return -ENOENT;
@@ -333,7 +393,11 @@ static int vxlan_fdb_create(struct vxlan_dev *vxlan,
                        return -ENOMEM;
 
                notify = 1;
-               f->remote_ip = ip;
+               f->remote.remote_ip = ip;
+               f->remote.remote_port = port;
+               f->remote.remote_vni = vni;
+               f->remote.remote_ifindex = ifindex;
+               f->remote.remote_next = NULL;
                f->state = state;
                f->updated = f->used = jiffies;
                memcpy(f->eth_addr, mac, ETH_ALEN);
@@ -349,6 +413,19 @@ static int vxlan_fdb_create(struct vxlan_dev *vxlan,
        return 0;
 }
 
+void vxlan_fdb_free(struct rcu_head *head)
+{
+       struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
+
+       while (f->remote.remote_next) {
+               struct vxlan_rdst *rd = f->remote.remote_next;
+
+               f->remote.remote_next = rd->remote_next;
+               kfree(rd);
+       }
+       kfree(f);
+}
+
 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 {
        netdev_dbg(vxlan->dev,
@@ -358,7 +435,7 @@ static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
        vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
 
        hlist_del_rcu(&f->hlist);
-       kfree_rcu(f, rcu);
+       call_rcu(&f->rcu, vxlan_fdb_free);
 }
 
 /* Add static entry (via netlink) */
@@ -367,7 +444,9 @@ static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
                         const unsigned char *addr, u16 flags)
 {
        struct vxlan_dev *vxlan = netdev_priv(dev);
+       struct net *net = dev_net(vxlan->dev);
        __be32 ip;
+       u32 port, vni, ifindex;
        int err;
 
        if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
@@ -384,8 +463,36 @@ static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 
        ip = nla_get_be32(tb[NDA_DST]);
 
+       if (tb[NDA_PORT]) {
+               if (nla_len(tb[NDA_PORT]) != sizeof(u32))
+                       return -EINVAL;
+               port = nla_get_u32(tb[NDA_PORT]);
+       } else
+               port = vxlan_port;
+
+       if (tb[NDA_VNI]) {
+               if (nla_len(tb[NDA_VNI]) != sizeof(u32))
+                       return -EINVAL;
+               vni = nla_get_u32(tb[NDA_VNI]);
+       } else
+               vni = vxlan->vni;
+
+       if (tb[NDA_IFINDEX]) {
+               struct net_device *dev;
+
+               if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
+                       return -EINVAL;
+               ifindex = nla_get_u32(tb[NDA_IFINDEX]);
+               dev = dev_get_by_index(net, ifindex);
+               if (!dev)
+                       return -EADDRNOTAVAIL;
+               dev_put(dev);
+       } else
+               ifindex = 0;
+
        spin_lock_bh(&vxlan->hash_lock);
-       err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags);
+       err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags, port,
+               vni, ifindex);
        spin_unlock_bh(&vxlan->hash_lock);
 
        return err;
@@ -423,18 +530,21 @@ static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
                int err;
 
                hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
-                       if (idx < cb->args[0])
-                               goto skip;
-
-                       err = vxlan_fdb_info(skb, vxlan, f,
-                                            NETLINK_CB(cb->skb).portid,
-                                            cb->nlh->nlmsg_seq,
-                                            RTM_NEWNEIGH,
-                                            NLM_F_MULTI);
-                       if (err < 0)
-                               break;
+                       struct vxlan_rdst *rd;
+                       for (rd = &f->remote; rd; rd = rd->remote_next) {
+                               if (idx < cb->args[0])
+                                       goto skip;
+
+                               err = vxlan_fdb_info(skb, vxlan, f,
+                                                    NETLINK_CB(cb->skb).portid,
+                                                    cb->nlh->nlmsg_seq,
+                                                    RTM_NEWNEIGH,
+                                                    NLM_F_MULTI, rd);
+                               if (err < 0)
+                                       break;
 skip:
-                       ++idx;
+                               ++idx;
+                       }
                }
        }
 
@@ -454,22 +564,23 @@ static void vxlan_snoop(struct net_device *dev,
        f = vxlan_find_mac(vxlan, src_mac);
        if (likely(f)) {
                f->used = jiffies;
-               if (likely(f->remote_ip == src_ip))
+               if (likely(f->remote.remote_ip == src_ip))
                        return;
 
                if (net_ratelimit())
                        netdev_info(dev,
                                    "%pM migrated from %pI4 to %pI4\n",
-                                   src_mac, &f->remote_ip, &src_ip);
+                                   src_mac, &f->remote.remote_ip, &src_ip);
 
-               f->remote_ip = src_ip;
+               f->remote.remote_ip = src_ip;
                f->updated = jiffies;
        } else {
                /* learned new entry */
                spin_lock(&vxlan->hash_lock);
                err = vxlan_fdb_create(vxlan, src_mac, src_ip,
                                       NUD_REACHABLE,
-                                      NLM_F_EXCL|NLM_F_CREATE);
+                                      NLM_F_EXCL|NLM_F_CREATE,
+                                      vxlan_port, vxlan->vni, 0);
                spin_unlock(&vxlan->hash_lock);
        }
 }
@@ -701,7 +812,7 @@ static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
                }
 
                f = vxlan_find_mac(vxlan, n->ha);
-               if (f && f->remote_ip == 0) {
+               if (f && f->remote.remote_ip == htonl(INADDR_ANY)) {
                        /* bridge-local neighbor */
                        neigh_release(n);
                        goto out;
@@ -834,47 +945,26 @@ static int handle_offloads(struct sk_buff *skb)
        return 0;
 }
 
-/* Transmit local packets over Vxlan
- *
- * Outer IP header inherits ECN and DF from inner header.
- * Outer UDP destination is the VXLAN assigned port.
- *           source port is based on hash of flow
- */
-static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
+static netdev_tx_t vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
+                                 struct vxlan_rdst *rdst, bool did_rsc)
 {
        struct vxlan_dev *vxlan = netdev_priv(dev);
        struct rtable *rt;
        const struct iphdr *old_iph;
-       struct ethhdr *eth;
        struct iphdr *iph;
        struct vxlanhdr *vxh;
        struct udphdr *uh;
        struct flowi4 fl4;
        unsigned int pkt_len = skb->len;
        __be32 dst;
-       __u16 src_port;
+       __u16 src_port, dst_port;
+        u32 vni;
        __be16 df = 0;
        __u8 tos, ttl;
-       bool did_rsc = false;
-       const struct vxlan_fdb *f;
-
-       skb_reset_mac_header(skb);
-       eth = eth_hdr(skb);
-
-       if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
-               return arp_reduce(dev, skb);
-       else if ((vxlan->flags&VXLAN_F_RSC) && ntohs(eth->h_proto) == ETH_P_IP)
-               did_rsc = route_shortcircuit(dev, skb);
 
-       f = vxlan_find_mac(vxlan, eth->h_dest);
-       if (f == NULL) {
-               did_rsc = false;
-               dst = vxlan->gaddr;
-               if (!dst && (vxlan->flags & VXLAN_F_L2MISS) &&
-                   !is_multicast_ether_addr(eth->h_dest))
-                       vxlan_fdb_miss(vxlan, eth->h_dest);
-       } else
-               dst = f->remote_ip;
+       dst_port = rdst->remote_port ? rdst->remote_port : vxlan_port;
+       vni = rdst->remote_vni;
+       dst = rdst->remote_ip;
 
        if (!dst) {
                if (did_rsc) {
@@ -922,7 +1012,7 @@ static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
        src_port = vxlan_src_port(vxlan, skb);
 
        memset(&fl4, 0, sizeof(fl4));
-       fl4.flowi4_oif = vxlan->link;
+       fl4.flowi4_oif = rdst->remote_ifindex;
        fl4.flowi4_tos = RT_TOS(tos);
        fl4.daddr = dst;
        fl4.saddr = vxlan->saddr;
@@ -949,13 +1039,13 @@ static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
 
        vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
        vxh->vx_flags = htonl(VXLAN_FLAGS);
-       vxh->vx_vni = htonl(vxlan->vni << 8);
+       vxh->vx_vni = htonl(vni << 8);
 
        __skb_push(skb, sizeof(*uh));
        skb_reset_transport_header(skb);
        uh = udp_hdr(skb);
 
-       uh->dest = htons(vxlan_port);
+       uh->dest = htons(dst_port);
        uh->source = htons(src_port);
 
        uh->len = htons(skb->len);
@@ -995,6 +1085,64 @@ tx_free:
        return NETDEV_TX_OK;
 }
 
+/* Transmit local packets over Vxlan
+ *
+ * Outer IP header inherits ECN and DF from inner header.
+ * Outer UDP destination is the VXLAN assigned port.
+ *           source port is based on hash of flow
+ */
+static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct vxlan_dev *vxlan = netdev_priv(dev);
+       struct ethhdr *eth;
+       bool did_rsc = false;
+       struct vxlan_rdst group, *rdst0, *rdst;
+       struct vxlan_fdb *f;
+       int rc1, rc;
+
+       skb_reset_mac_header(skb);
+       eth = eth_hdr(skb);
+
+       if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
+               return arp_reduce(dev, skb);
+       else if ((vxlan->flags&VXLAN_F_RSC) && ntohs(eth->h_proto) == ETH_P_IP)
+               did_rsc = route_shortcircuit(dev, skb);
+
+       f = vxlan_find_mac(vxlan, eth->h_dest);
+       if (f == NULL) {
+               did_rsc = false;
+               group.remote_port = vxlan_port;
+               group.remote_vni = vxlan->vni;
+               group.remote_ip = vxlan->gaddr;
+               group.remote_ifindex = vxlan->link;
+               group.remote_next = 0;
+               rdst0 = &group;
+
+               if (group.remote_ip == htonl(INADDR_ANY) &&
+                   (vxlan->flags & VXLAN_F_L2MISS) &&
+                   !is_multicast_ether_addr(eth->h_dest))
+                       vxlan_fdb_miss(vxlan, eth->h_dest);
+       } else
+               rdst0 = &f->remote;
+
+       rc = NETDEV_TX_OK;
+
+       /* if there are multiple destinations, send copies */
+       for (rdst = rdst0->remote_next; rdst; rdst = rdst->remote_next) {
+               struct sk_buff *skb1;
+
+               skb1 = skb_clone(skb, GFP_ATOMIC);
+               rc1 = vxlan_xmit_one(skb1, dev, rdst, did_rsc);
+               if (rc == NETDEV_TX_OK)
+                       rc = rc1;
+       }
+
+       rc1 = vxlan_xmit_one(skb, dev, rdst0, did_rsc);
+       if (rc == NETDEV_TX_OK)
+               rc = rc1;
+       return rc;
+}
+
 /* Walk the forwarding table and purge stale entries */
 static void vxlan_cleanup(unsigned long arg)
 {
@@ -1558,6 +1706,7 @@ static void __exit vxlan_cleanup_module(void)
 {
        rtnl_link_unregister(&vxlan_link_ops);
        unregister_pernet_device(&vxlan_net_ops);
+       rcu_barrier();
 }
 module_exit(vxlan_cleanup_module);
 
index adb068c53c4e964fc1ea950a61f188b9159bc178..f175212420ab0bd1f4ffbdae7cac60457b77bb3c 100644 (file)
@@ -21,6 +21,9 @@ enum {
        NDA_CACHEINFO,
        NDA_PROBES,
        NDA_VLAN,
+       NDA_PORT,
+       NDA_VNI,
+       NDA_IFINDEX,
        __NDA_MAX
 };
 
index 55b5624a4b00f3b1466a757c1194b2dacca1bd09..0e86baf8f809043bc1db0412194aa8edc2f67c53 100644 (file)
@@ -2112,7 +2112,7 @@ static int rtnl_fdb_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
        }
 
        addr = nla_data(tb[NDA_LLADDR]);
-       if (!is_valid_ether_addr(addr)) {
+       if (is_zero_ether_addr(addr)) {
                pr_info("PF_BRIDGE: RTM_NEWNEIGH with invalid ether address\n");
                return -EINVAL;
        }