geneve: add initial netdev driver for GENEVE tunnels
authorJohn W. Linville <linville@tuxdriver.com>
Wed, 13 May 2015 16:57:30 +0000 (12:57 -0400)
committerDavid S. Miller <davem@davemloft.net>
Wed, 13 May 2015 19:59:13 +0000 (15:59 -0400)
This is an initial implementation of a netdev driver for GENEVE
tunnels.  This implementation uses a fixed UDP port, and only supports
point-to-point links with specific partner endpoints.  Only IPv4
links are supported at this time.

Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/Kconfig
drivers/net/Makefile
drivers/net/geneve.c [new file with mode: 0644]
include/uapi/linux/if_link.h

index df51d6025a9017413500046edb78401abe0dfdb3..019fceffc9e52980d1f3837b01a299b97782dfca 100644 (file)
@@ -179,6 +179,20 @@ config VXLAN
          To compile this driver as a module, choose M here: the module
          will be called vxlan.
 
+config GENEVE
+       tristate "Generic Network Virtualization Encapsulation netdev"
+       depends on INET && GENEVE_CORE
+       select NET_IP_TUNNEL
+       ---help---
+         This allows one to create geneve virtual interfaces that provide
+         Layer 2 Networks over Layer 3 Networks. GENEVE is often used
+         to tunnel virtual network infrastructure in virtualized environments.
+         For more information see:
+           http://tools.ietf.org/html/draft-gross-geneve-02
+
+         To compile this driver as a module, choose M here: the module
+         will be called geneve.
+
 config NETCONSOLE
        tristate "Network console logging support"
        ---help---
index e25fdd7d905e334dff7ccbcaba73b57436248f9f..c12cb22478a7daa39e2935f1c0aa22c26ab04bdc 100644 (file)
@@ -23,6 +23,7 @@ obj-$(CONFIG_TUN) += tun.o
 obj-$(CONFIG_VETH) += veth.o
 obj-$(CONFIG_VIRTIO_NET) += virtio_net.o
 obj-$(CONFIG_VXLAN) += vxlan.o
+obj-$(CONFIG_GENEVE) += geneve.o
 obj-$(CONFIG_NLMON) += nlmon.o
 
 #
diff --git a/drivers/net/geneve.c b/drivers/net/geneve.c
new file mode 100644 (file)
index 0000000..b7eafa4
--- /dev/null
@@ -0,0 +1,503 @@
+/*
+ * GENEVE: Generic Network Virtualization Encapsulation
+ *
+ * Copyright (c) 2015 Red Hat, Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/hash.h>
+#include <net/rtnetlink.h>
+#include <net/geneve.h>
+
+#define GENEVE_NETDEV_VER      "0.6"
+
+#define GENEVE_UDP_PORT                6081
+
+#define GENEVE_N_VID           (1u << 24)
+#define GENEVE_VID_MASK                (GENEVE_N_VID - 1)
+
+#define VNI_HASH_BITS          10
+#define VNI_HASH_SIZE          (1<<VNI_HASH_BITS)
+
+static bool log_ecn_error = true;
+module_param(log_ecn_error, bool, 0644);
+MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
+
+/* per-network namespace private data for this module */
+struct geneve_net {
+       struct list_head  geneve_list;
+       struct hlist_head vni_list[VNI_HASH_SIZE];
+};
+
+/* Pseudo network device */
+struct geneve_dev {
+       struct hlist_node  hlist;       /* vni hash table */
+       struct net         *net;        /* netns for packet i/o */
+       struct net_device  *dev;        /* netdev for geneve tunnel */
+       struct geneve_sock *sock;       /* socket used for geneve tunnel */
+       u8 vni[3];                      /* virtual network ID for tunnel */
+       struct sockaddr_in remote;      /* IPv4 address for link partner */
+       struct list_head   next;        /* geneve's per namespace list */
+};
+
+static int geneve_net_id;
+
+static inline __u32 geneve_net_vni_hash(u8 vni[3])
+{
+       __u32 vnid;
+
+       vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
+       return hash_32(vnid, VNI_HASH_BITS);
+}
+
+/* geneve receive/decap routine */
+static void geneve_rx(struct geneve_sock *gs, struct sk_buff *skb)
+{
+       struct genevehdr *gnvh = geneve_hdr(skb);
+       struct geneve_dev *dummy, *geneve = NULL;
+       struct geneve_net *gn;
+       struct iphdr *iph = NULL;
+       struct pcpu_sw_netstats *stats;
+       struct hlist_head *vni_list_head;
+       int err = 0;
+       __u32 hash;
+
+       iph = ip_hdr(skb); /* Still outer IP header... */
+
+       gn = gs->rcv_data;
+
+       /* Find the device for this VNI */
+       hash = geneve_net_vni_hash(gnvh->vni);
+       vni_list_head = &gn->vni_list[hash];
+       hlist_for_each_entry_rcu(dummy, vni_list_head, hlist) {
+               if (!memcmp(gnvh->vni, dummy->vni, sizeof(dummy->vni)) &&
+                   iph->saddr == dummy->remote.sin_addr.s_addr) {
+                       geneve = dummy;
+                       break;
+               }
+       }
+       if (!geneve)
+               goto drop;
+
+       /* Drop packets w/ critical options,
+        * since we don't support any...
+        */
+       if (gnvh->critical)
+               goto drop;
+
+       skb_reset_mac_header(skb);
+       skb_scrub_packet(skb, !net_eq(geneve->net, dev_net(geneve->dev)));
+       skb->protocol = eth_type_trans(skb, geneve->dev);
+       skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
+
+       /* Ignore packet loops (and multicast echo) */
+       if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
+               goto drop;
+
+       skb_reset_network_header(skb);
+
+       iph = ip_hdr(skb); /* Now inner IP header... */
+       err = IP_ECN_decapsulate(iph, skb);
+
+       if (unlikely(err)) {
+               if (log_ecn_error)
+                       net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
+                                            &iph->saddr, iph->tos);
+               if (err > 1) {
+                       ++geneve->dev->stats.rx_frame_errors;
+                       ++geneve->dev->stats.rx_errors;
+                       goto drop;
+               }
+       }
+
+       stats = this_cpu_ptr(geneve->dev->tstats);
+       u64_stats_update_begin(&stats->syncp);
+       stats->rx_packets++;
+       stats->rx_bytes += skb->len;
+       u64_stats_update_end(&stats->syncp);
+
+       netif_rx(skb);
+
+       return;
+drop:
+       /* Consume bad packet */
+       kfree_skb(skb);
+}
+
+/* Setup stats when device is created */
+static int geneve_init(struct net_device *dev)
+{
+       dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
+       if (!dev->tstats)
+               return -ENOMEM;
+
+       return 0;
+}
+
+static void geneve_uninit(struct net_device *dev)
+{
+       free_percpu(dev->tstats);
+}
+
+static int geneve_open(struct net_device *dev)
+{
+       struct geneve_dev *geneve = netdev_priv(dev);
+       struct net *net = geneve->net;
+       struct geneve_net *gn = net_generic(geneve->net, geneve_net_id);
+       struct geneve_sock *gs;
+
+       gs = geneve_sock_add(net, htons(GENEVE_UDP_PORT), geneve_rx, gn,
+                            false, false);
+       if (IS_ERR(gs))
+               return PTR_ERR(gs);
+
+       geneve->sock = gs;
+
+       return 0;
+}
+
+static int geneve_stop(struct net_device *dev)
+{
+       struct geneve_dev *geneve = netdev_priv(dev);
+       struct geneve_sock *gs = geneve->sock;
+
+       geneve_sock_release(gs);
+
+       return 0;
+}
+
+static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct geneve_dev *geneve = netdev_priv(dev);
+       struct geneve_sock *gs = geneve->sock;
+       struct rtable *rt = NULL;
+       const struct iphdr *iip; /* interior IP header */
+       struct flowi4 fl4;
+       int err;
+       __be16 sport;
+       __u8 tos, ttl = 0;
+
+       iip = ip_hdr(skb);
+
+       skb_reset_mac_header(skb);
+
+       /* TODO: port min/max limits should be configurable */
+       sport = udp_flow_src_port(dev_net(dev), skb, 0, 0, true);
+
+       memset(&fl4, 0, sizeof(fl4));
+       fl4.daddr = geneve->remote.sin_addr.s_addr;
+       rt = ip_route_output_key(geneve->net, &fl4);
+       if (IS_ERR(rt)) {
+               netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
+               dev->stats.tx_carrier_errors++;
+               goto tx_error;
+       }
+       if (rt->dst.dev == dev) { /* is this necessary? */
+               netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr);
+               dev->stats.collisions++;
+               goto rt_tx_error;
+       }
+
+       /* TODO: tos and ttl should be configurable */
+
+       tos = ip_tunnel_ecn_encap(0, iip, skb);
+
+       if (IN_MULTICAST(ntohl(fl4.daddr)))
+               ttl = 1;
+
+       ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
+
+       /* no need to handle local destination and encap bypass...yet... */
+
+       err = geneve_xmit_skb(gs, rt, skb, fl4.saddr, fl4.daddr,
+                             tos, ttl, 0, sport, htons(GENEVE_UDP_PORT), 0,
+                             geneve->vni, 0, NULL, false,
+                             !net_eq(geneve->net, dev_net(geneve->dev)));
+       if (err < 0)
+               ip_rt_put(rt);
+
+       iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
+
+       return NETDEV_TX_OK;
+
+rt_tx_error:
+       ip_rt_put(rt);
+tx_error:
+       dev->stats.tx_errors++;
+       dev_kfree_skb(skb);
+       return NETDEV_TX_OK;
+}
+
+static const struct net_device_ops geneve_netdev_ops = {
+       .ndo_init               = geneve_init,
+       .ndo_uninit             = geneve_uninit,
+       .ndo_open               = geneve_open,
+       .ndo_stop               = geneve_stop,
+       .ndo_start_xmit         = geneve_xmit,
+       .ndo_get_stats64        = ip_tunnel_get_stats64,
+       .ndo_change_mtu         = eth_change_mtu,
+       .ndo_validate_addr      = eth_validate_addr,
+       .ndo_set_mac_address    = eth_mac_addr,
+};
+
+static void geneve_get_drvinfo(struct net_device *dev,
+                              struct ethtool_drvinfo *drvinfo)
+{
+       strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
+       strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
+}
+
+static const struct ethtool_ops geneve_ethtool_ops = {
+       .get_drvinfo    = geneve_get_drvinfo,
+       .get_link       = ethtool_op_get_link,
+};
+
+/* Info for udev, that this is a virtual tunnel endpoint */
+static struct device_type geneve_type = {
+       .name = "geneve",
+};
+
+/* Initialize the device structure. */
+static void geneve_setup(struct net_device *dev)
+{
+       ether_setup(dev);
+
+       dev->netdev_ops = &geneve_netdev_ops;
+       dev->ethtool_ops = &geneve_ethtool_ops;
+       dev->destructor = free_netdev;
+
+       SET_NETDEV_DEVTYPE(dev, &geneve_type);
+
+       dev->tx_queue_len = 0;
+       dev->features    |= NETIF_F_LLTX;
+       dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
+       dev->features    |= NETIF_F_RXCSUM;
+       dev->features    |= NETIF_F_GSO_SOFTWARE;
+
+       dev->vlan_features = dev->features;
+       dev->features    |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
+
+       dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
+       dev->hw_features |= NETIF_F_GSO_SOFTWARE;
+       dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
+
+       netif_keep_dst(dev);
+       dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
+}
+
+static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
+       [IFLA_GENEVE_ID]                = { .type = NLA_U32 },
+       [IFLA_GENEVE_REMOTE]            = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
+};
+
+static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
+{
+       if (tb[IFLA_ADDRESS]) {
+               if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
+                       return -EINVAL;
+
+               if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
+                       return -EADDRNOTAVAIL;
+       }
+
+       if (!data)
+               return -EINVAL;
+
+       if (data[IFLA_GENEVE_ID]) {
+               __u32 vni =  nla_get_u32(data[IFLA_GENEVE_ID]);
+
+               if (vni >= GENEVE_VID_MASK)
+                       return -ERANGE;
+       }
+
+       return 0;
+}
+
+static int geneve_newlink(struct net *net, struct net_device *dev,
+                        struct nlattr *tb[], struct nlattr *data[])
+{
+       struct geneve_net *gn = net_generic(net, geneve_net_id);
+       struct geneve_dev *dummy, *geneve = netdev_priv(dev);
+       struct hlist_head *vni_list_head;
+       struct sockaddr_in remote;      /* IPv4 address for link partner */
+       __u32 vni, hash;
+       int err;
+
+       if (!data[IFLA_GENEVE_ID] || !data[IFLA_GENEVE_REMOTE])
+               return -EINVAL;
+
+       geneve->net = net;
+       geneve->dev = dev;
+
+       vni = nla_get_u32(data[IFLA_GENEVE_ID]);
+       geneve->vni[0] = (vni & 0x00ff0000) >> 16;
+       geneve->vni[1] = (vni & 0x0000ff00) >> 8;
+       geneve->vni[2] =  vni & 0x000000ff;
+
+       geneve->remote.sin_addr.s_addr =
+               nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
+       if (IN_MULTICAST(ntohl(geneve->remote.sin_addr.s_addr)))
+               return -EINVAL;
+
+       remote = geneve->remote;
+       hash = geneve_net_vni_hash(geneve->vni);
+       vni_list_head = &gn->vni_list[hash];
+       hlist_for_each_entry_rcu(dummy, vni_list_head, hlist) {
+               if (!memcmp(geneve->vni, dummy->vni, sizeof(dummy->vni)) &&
+                   !memcmp(&remote, &dummy->remote, sizeof(dummy->remote)))
+                       return -EBUSY;
+       }
+
+       if (tb[IFLA_ADDRESS] == NULL)
+               eth_hw_addr_random(dev);
+
+       err = register_netdevice(dev);
+       if (err)
+               return err;
+
+       list_add(&geneve->next, &gn->geneve_list);
+
+       hlist_add_head_rcu(&geneve->hlist, &gn->vni_list[hash]);
+
+       return 0;
+}
+
+static void geneve_dellink(struct net_device *dev, struct list_head *head)
+{
+       struct geneve_dev *geneve = netdev_priv(dev);
+
+       if (!hlist_unhashed(&geneve->hlist))
+               hlist_del_rcu(&geneve->hlist);
+
+       list_del(&geneve->next);
+       unregister_netdevice_queue(dev, head);
+}
+
+static size_t geneve_get_size(const struct net_device *dev)
+{
+       return nla_total_size(sizeof(__u32)) +  /* IFLA_GENEVE_ID */
+               nla_total_size(sizeof(struct in_addr)) + /* IFLA_GENEVE_REMOTE */
+               0;
+}
+
+static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
+{
+       struct geneve_dev *geneve = netdev_priv(dev);
+       __u32 vni;
+
+       vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
+       if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
+               goto nla_put_failure;
+
+       if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
+                           geneve->remote.sin_addr.s_addr))
+               goto nla_put_failure;
+
+       return 0;
+
+nla_put_failure:
+       return -EMSGSIZE;
+}
+
+static struct rtnl_link_ops geneve_link_ops __read_mostly = {
+       .kind           = "geneve",
+       .maxtype        = IFLA_GENEVE_MAX,
+       .policy         = geneve_policy,
+       .priv_size      = sizeof(struct geneve_dev),
+       .setup          = geneve_setup,
+       .validate       = geneve_validate,
+       .newlink        = geneve_newlink,
+       .dellink        = geneve_dellink,
+       .get_size       = geneve_get_size,
+       .fill_info      = geneve_fill_info,
+};
+
+static __net_init int geneve_init_net(struct net *net)
+{
+       struct geneve_net *gn = net_generic(net, geneve_net_id);
+       unsigned int h;
+
+       INIT_LIST_HEAD(&gn->geneve_list);
+
+       for (h = 0; h < VNI_HASH_SIZE; ++h)
+               INIT_HLIST_HEAD(&gn->vni_list[h]);
+
+       return 0;
+}
+
+static void __net_exit geneve_exit_net(struct net *net)
+{
+       struct geneve_net *gn = net_generic(net, geneve_net_id);
+       struct geneve_dev *geneve, *next;
+       struct net_device *dev, *aux;
+       LIST_HEAD(list);
+
+       rtnl_lock();
+
+       /* gather any geneve devices that were moved into this ns */
+       for_each_netdev_safe(net, dev, aux)
+               if (dev->rtnl_link_ops == &geneve_link_ops)
+                       unregister_netdevice_queue(dev, &list);
+
+       /* now gather any other geneve devices that were created in this ns */
+       list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
+               /* If geneve->dev is in the same netns, it was already added
+                * to the list by the previous loop.
+                */
+               if (!net_eq(dev_net(geneve->dev), net))
+                       unregister_netdevice_queue(geneve->dev, &list);
+       }
+
+       /* unregister the devices gathered above */
+       unregister_netdevice_many(&list);
+       rtnl_unlock();
+}
+
+static struct pernet_operations geneve_net_ops = {
+       .init = geneve_init_net,
+       .exit = geneve_exit_net,
+       .id   = &geneve_net_id,
+       .size = sizeof(struct geneve_net),
+};
+
+static int __init geneve_init_module(void)
+{
+       int rc;
+
+       rc = register_pernet_subsys(&geneve_net_ops);
+       if (rc)
+               goto out1;
+
+       rc = rtnl_link_register(&geneve_link_ops);
+       if (rc)
+               goto out2;
+
+       return 0;
+out2:
+       unregister_pernet_subsys(&geneve_net_ops);
+out1:
+       return rc;
+}
+late_initcall(geneve_init_module);
+
+static void __exit geneve_cleanup_module(void)
+{
+       rtnl_link_unregister(&geneve_link_ops);
+       unregister_pernet_subsys(&geneve_net_ops);
+}
+module_exit(geneve_cleanup_module);
+
+MODULE_LICENSE("GPL");
+MODULE_VERSION(GENEVE_NETDEV_VER);
+MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
+MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
+MODULE_ALIAS_RTNL_LINK("geneve");
index 6d6e502e1051309ec38f3fe55788ca7cbe3a5425..afccc9393fefbc987477fc1cbe6ec483b208707d 100644 (file)
@@ -390,6 +390,15 @@ struct ifla_vxlan_port_range {
        __be16  high;
 };
 
+/* GENEVE section */
+enum {
+       IFLA_GENEVE_UNSPEC,
+       IFLA_GENEVE_ID,
+       IFLA_GENEVE_REMOTE,
+       __IFLA_GENEVE_MAX
+};
+#define IFLA_GENEVE_MAX        (__IFLA_GENEVE_MAX - 1)
+
 /* Bonding section */
 
 enum {