2 * GENEVE: Generic Network Virtualization Encapsulation
4 * Copyright (c) 2015 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/netdevice.h>
16 #include <linux/etherdevice.h>
17 #include <linux/hash.h>
18 #include <linux/if_link.h>
20 #include <net/dst_metadata.h>
21 #include <net/net_namespace.h>
22 #include <net/netns/generic.h>
23 #include <net/rtnetlink.h>
24 #include <net/geneve.h>
25 #include <net/protocol.h>
28 #include "vport-netdev.h"
31 #ifndef HAVE_METADATA_DST
32 #define GENEVE_NETDEV_VER "0.6"
34 #define GENEVE_UDP_PORT 6081
36 #define GENEVE_N_VID (1u << 24)
37 #define GENEVE_VID_MASK (GENEVE_N_VID - 1)
39 #define VNI_HASH_BITS 10
40 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
43 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
45 /* per-network namespace private data for this module */
47 struct list_head geneve_list;
48 struct list_head sock_list;
51 static int geneve_net_id;
53 /* Pseudo network device */
55 struct hlist_node hlist; /* vni hash table */
56 struct net *net; /* netns for packet i/o */
57 struct net_device *dev; /* netdev for geneve tunnel */
58 struct geneve_sock *sock; /* socket used for geneve tunnel */
59 u8 vni[3]; /* virtual network ID for tunnel */
60 u8 ttl; /* TTL override */
61 u8 tos; /* TOS override */
62 struct sockaddr_in remote; /* IPv4 address for link partner */
63 struct list_head next; /* geneve's per namespace list */
70 struct list_head list;
74 #ifdef HAVE_UDP_OFFLOAD
75 struct udp_offload udp_offloads;
77 struct hlist_head vni_list[VNI_HASH_SIZE];
80 static inline __u32 geneve_net_vni_hash(u8 vni[3])
84 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
85 return hash_32(vnid, VNI_HASH_BITS);
88 static __be64 vni_to_tunnel_id(const __u8 *vni)
91 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
93 return (__force __be64)(((__force u64)vni[0] << 40) |
94 ((__force u64)vni[1] << 48) |
95 ((__force u64)vni[2] << 56));
99 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
100 __be32 addr, u8 vni[])
102 struct hlist_head *vni_list_head;
103 struct geneve_dev *geneve;
106 /* Find the device for this VNI */
107 hash = geneve_net_vni_hash(vni);
108 vni_list_head = &gs->vni_list[hash];
109 hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
110 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
111 addr == geneve->remote.sin_addr.s_addr)
117 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
119 return (struct genevehdr *)(udp_hdr(skb) + 1);
122 /* geneve receive/decap routine */
123 static void geneve_rx(struct geneve_sock *gs, struct sk_buff *skb)
125 struct genevehdr *gnvh = geneve_hdr(skb);
126 struct metadata_dst *tun_dst;
127 struct geneve_dev *geneve = NULL;
128 #ifdef HAVE_DEV_TSTATS
129 struct pcpu_sw_netstats *stats;
136 struct metadata_dst dst;
137 char buf[sizeof(struct metadata_dst) + 256];
140 iph = ip_hdr(skb); /* outer IP header... */
142 if (gs->collect_md) {
143 static u8 zero_vni[3];
152 geneve = geneve_lookup(gs, addr, vni);
156 if (ip_tunnel_collect_metadata() || gs->collect_md) {
159 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
160 (gnvh->oam ? TUNNEL_OAM : 0) |
161 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
164 ovs_udp_tun_rx_dst(&tun_dst->u.tun_info, skb, AF_INET, flags,
165 vni_to_tunnel_id(gnvh->vni), gnvh->opt_len * 4);
166 /* Update tunnel dst according to Geneve options. */
167 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
168 gnvh->options, gnvh->opt_len * 4);
170 /* Drop packets w/ critical options,
171 * since we don't support any...
178 skb_reset_mac_header(skb);
179 skb_scrub_packet(skb, !net_eq(geneve->net, dev_net(geneve->dev)));
180 skb->protocol = eth_type_trans(skb, geneve->dev);
181 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
184 ovs_skb_dst_set(skb, &tun_dst->dst);
187 /* Ignore packet loops (and multicast echo) */
188 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr))
191 skb_reset_network_header(skb);
193 err = IP_ECN_decapsulate(iph, skb);
197 ++geneve->dev->stats.rx_frame_errors;
198 ++geneve->dev->stats.rx_errors;
203 #ifdef HAVE_DEV_TSTATS
204 stats = this_cpu_ptr((struct pcpu_sw_netstats __percpu *)geneve->dev->tstats);
205 u64_stats_update_begin(&stats->syncp);
207 stats->rx_bytes += skb->len;
208 u64_stats_update_end(&stats->syncp);
210 netdev_port_receive(skb, &tun_dst->u.tun_info);
213 /* Consume bad packet */
217 #ifdef HAVE_DEV_TSTATS
218 /* Setup stats when device is created */
219 static int geneve_init(struct net_device *dev)
221 dev->tstats = (typeof(dev->tstats)) netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
228 static void geneve_uninit(struct net_device *dev)
230 free_percpu(dev->tstats);
234 /* Callback from net/ipv4/udp.c to receive packets */
235 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
237 struct genevehdr *geneveh;
238 struct geneve_sock *gs;
241 /* Need Geneve and inner Ethernet header to be present */
242 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
245 /* Return packets with reserved bits set */
246 geneveh = geneve_hdr(skb);
247 if (unlikely(geneveh->ver != GENEVE_VER))
250 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
253 opts_len = geneveh->opt_len * 4;
254 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
258 gs = rcu_dereference_sk_user_data(sk);
266 /* Consume bad packet */
271 /* Let the UDP layer deal with the skb */
275 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
279 struct udp_port_cfg udp_conf;
282 memset(&udp_conf, 0, sizeof(udp_conf));
285 udp_conf.family = AF_INET6;
287 udp_conf.family = AF_INET;
288 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
291 udp_conf.local_udp_port = port;
293 /* Open UDP socket */
294 err = udp_sock_create(net, &udp_conf, &sock);
301 #ifdef HAVE_UDP_OFFLOAD
302 static void geneve_notify_add_rx_port(struct geneve_sock *gs)
304 struct sock *sk = gs->sock->sk;
305 sa_family_t sa_family = sk->sk_family;
308 if (sa_family == AF_INET) {
309 err = udp_add_offload(&gs->udp_offloads);
311 pr_warn("geneve: udp_add_offload failed with status %d\n",
316 static int geneve_hlen(struct genevehdr *gh)
318 return sizeof(*gh) + gh->opt_len * 4;
321 #ifndef HAVE_UDP_OFFLOAD_ARG_UOFF
322 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
325 static struct sk_buff **geneve_gro_receive(struct sk_buff **head,
327 struct udp_offload *uoff)
330 struct sk_buff *p, **pp = NULL;
331 struct genevehdr *gh, *gh2;
332 unsigned int hlen, gh_len, off_gnv;
333 const struct packet_offload *ptype;
337 off_gnv = skb_gro_offset(skb);
338 hlen = off_gnv + sizeof(*gh);
339 gh = skb_gro_header_fast(skb, off_gnv);
340 if (skb_gro_header_hard(skb, hlen)) {
341 gh = skb_gro_header_slow(skb, hlen, off_gnv);
346 if (gh->ver != GENEVE_VER || gh->oam)
348 gh_len = geneve_hlen(gh);
350 hlen = off_gnv + gh_len;
351 if (skb_gro_header_hard(skb, hlen)) {
352 gh = skb_gro_header_slow(skb, hlen, off_gnv);
359 for (p = *head; p; p = p->next) {
360 if (!NAPI_GRO_CB(p)->same_flow)
363 gh2 = (struct genevehdr *)(p->data + off_gnv);
364 if (gh->opt_len != gh2->opt_len ||
365 memcmp(gh, gh2, gh_len)) {
366 NAPI_GRO_CB(p)->same_flow = 0;
371 type = gh->proto_type;
374 ptype = gro_find_receive_by_type(type);
380 skb_gro_pull(skb, gh_len);
381 skb_gro_postpull_rcsum(skb, gh, gh_len);
382 pp = ptype->callbacks.gro_receive(head, skb);
387 NAPI_GRO_CB(skb)->flush |= flush;
392 #ifndef HAVE_UDP_OFFLOAD_ARG_UOFF
393 static int geneve_gro_complete(struct sk_buff *skb, int nhoff)
395 static int geneve_gro_complete(struct sk_buff *skb, int nhoff,
396 struct udp_offload *uoff)
399 struct genevehdr *gh;
400 struct packet_offload *ptype;
405 udp_tunnel_gro_complete(skb, nhoff);
407 gh = (struct genevehdr *)(skb->data + nhoff);
408 gh_len = geneve_hlen(gh);
409 type = gh->proto_type;
412 ptype = gro_find_complete_by_type(type);
414 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
421 /* Create new listen socket if needed */
422 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
425 struct geneve_net *gn = net_generic(net, geneve_net_id);
426 struct geneve_sock *gs;
428 struct udp_tunnel_sock_cfg tunnel_cfg;
431 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
433 return ERR_PTR(-ENOMEM);
435 sock = geneve_create_sock(net, ipv6, port);
438 return ERR_CAST(sock);
443 for (h = 0; h < VNI_HASH_SIZE; ++h)
444 INIT_HLIST_HEAD(&gs->vni_list[h]);
446 /* Initialize the geneve udp offloads structure */
447 #ifdef HAVE_UDP_OFFLOAD
448 gs->udp_offloads.port = port;
449 gs->udp_offloads.callbacks.gro_receive = geneve_gro_receive;
450 gs->udp_offloads.callbacks.gro_complete = geneve_gro_complete;
451 geneve_notify_add_rx_port(gs);
453 /* Mark socket as an encapsulation socket */
454 tunnel_cfg.sk_user_data = gs;
455 tunnel_cfg.encap_type = 1;
456 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
457 tunnel_cfg.encap_destroy = NULL;
458 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
459 list_add(&gs->list, &gn->sock_list);
463 static void geneve_notify_del_rx_port(struct geneve_sock *gs)
465 #ifdef HAVE_UDP_OFFLOAD
466 struct sock *sk = gs->sock->sk;
467 sa_family_t sa_family = sk->sk_family;
469 if (sa_family == AF_INET)
470 udp_del_offload(&gs->udp_offloads);
474 static void free_gs_rcu(struct rcu_head *rcu)
476 struct geneve_sock *gs = container_of(rcu, struct geneve_sock, rcu);
481 static void geneve_sock_release(struct geneve_sock *gs)
487 geneve_notify_del_rx_port(gs);
488 udp_tunnel_sock_release(gs->sock);
489 call_rcu(&gs->rcu, free_gs_rcu);
492 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
495 struct geneve_sock *gs;
497 list_for_each_entry(gs, &gn->sock_list, list) {
498 if (inet_sport(gs->sock->sk) == dst_port &&
499 inet_sk(gs->sock->sk)->sk.sk_family == AF_INET) {
506 static int geneve_open(struct net_device *dev)
508 struct geneve_dev *geneve = netdev_priv(dev);
509 struct net *net = geneve->net;
510 struct geneve_net *gn = net_generic(net, geneve_net_id);
511 struct geneve_sock *gs;
514 gs = geneve_find_sock(gn, geneve->dst_port);
520 gs = geneve_socket_create(net, geneve->dst_port, false);
525 gs->collect_md = geneve->collect_md;
528 hash = geneve_net_vni_hash(geneve->vni);
529 hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
533 static int geneve_stop(struct net_device *dev)
535 struct geneve_dev *geneve = netdev_priv(dev);
536 struct geneve_sock *gs = geneve->sock;
538 if (!hlist_unhashed(&geneve->hlist))
539 hlist_del_rcu(&geneve->hlist);
540 geneve_sock_release(gs);
544 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb,
545 __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt,
548 struct genevehdr *gnvh;
552 min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
553 + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr)
554 + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
555 err = skb_cow_head(skb, min_headroom);
561 skb = vlan_hwaccel_push_inside(skb);
567 skb = udp_tunnel_handle_offloads(skb, csum, 0, false);
572 gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len);
573 gnvh->ver = GENEVE_VER;
574 gnvh->opt_len = opt_len / 4;
575 gnvh->oam = !!(tun_flags & TUNNEL_OAM);
576 gnvh->critical = !!(tun_flags & TUNNEL_CRIT_OPT);
578 memcpy(gnvh->vni, vni, 3);
579 gnvh->proto_type = htons(ETH_P_TEB);
581 memcpy(gnvh->options, opt, opt_len);
583 ovs_skb_set_inner_protocol(skb, htons(ETH_P_TEB));
591 static struct rtable *geneve_get_rt(struct sk_buff *skb,
592 struct net_device *dev,
594 struct ip_tunnel_info *info)
596 struct geneve_dev *geneve = netdev_priv(dev);
597 struct rtable *rt = NULL;
600 memset(fl4, 0, sizeof(*fl4));
601 fl4->flowi4_mark = skb->mark;
602 fl4->flowi4_proto = IPPROTO_UDP;
605 fl4->daddr = info->key.u.ipv4.dst;
606 fl4->saddr = info->key.u.ipv4.src;
607 fl4->flowi4_tos = RT_TOS(info->key.tos);
611 const struct iphdr *iip = ip_hdr(skb);
613 tos = ip_tunnel_get_dsfield(iip, skb);
616 fl4->flowi4_tos = RT_TOS(tos);
617 fl4->daddr = geneve->remote.sin_addr.s_addr;
620 rt = ip_route_output_key(geneve->net, fl4);
622 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
623 dev->stats.tx_carrier_errors++;
626 if (rt_dst(rt).dev == dev) { /* is this necessary? */
627 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
628 dev->stats.collisions++;
630 return ERR_PTR(-EINVAL);
635 /* Convert 64 bit tunnel ID to 24 bit VNI. */
636 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
639 vni[0] = (__force __u8)(tun_id >> 16);
640 vni[1] = (__force __u8)(tun_id >> 8);
641 vni[2] = (__force __u8)tun_id;
643 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
644 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
645 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
649 netdev_tx_t rpl_geneve_xmit(struct sk_buff *skb)
651 struct net_device *dev = skb->dev;
652 struct geneve_dev *geneve = netdev_priv(dev);
653 struct geneve_sock *gs = geneve->sock;
654 struct ip_tunnel_info *info = NULL;
655 struct rtable *rt = NULL;
656 const struct iphdr *iip; /* interior IP header */
664 if (geneve->collect_md) {
665 info = skb_tunnel_info(skb);
666 if (unlikely(info && !(info->mode & IP_TUNNEL_INFO_TX))) {
667 netdev_dbg(dev, "no tunnel metadata\n");
670 if (info && ip_tunnel_info_af(info) != AF_INET)
674 rt = geneve_get_rt(skb, dev, &fl4, info);
676 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr);
677 dev->stats.tx_carrier_errors++;
681 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
682 skb_reset_mac_header(skb);
687 const struct ip_tunnel_key *key = &info->key;
691 tunnel_id_to_vni(key->tun_id, vni);
692 if (key->tun_flags & TUNNEL_GENEVE_OPT)
693 opts = ip_tunnel_info_opts(info);
695 udp_csum = !!(key->tun_flags & TUNNEL_CSUM);
696 err = geneve_build_skb(rt, skb, key->tun_flags, vni,
697 info->options_len, opts, udp_csum);
701 tos = ip_tunnel_ecn_encap(key->tos, iip, skb);
703 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
706 err = geneve_build_skb(rt, skb, 0, geneve->vni,
711 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, iip, skb);
713 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr)))
715 ttl = ttl ? : ip4_dst_hoplimit(&rt_dst(rt));
718 err = udp_tunnel_xmit_skb(rt, gs->sock->sk, skb, fl4.saddr, fl4.daddr,
719 tos, ttl, df, sport, geneve->dst_port,
720 !net_eq(geneve->net, dev_net(geneve->dev)),
723 iptunnel_xmit_stats(err, &dev->stats, (struct pcpu_sw_netstats __percpu *) dev->tstats);
729 dev->stats.tx_errors++;
732 EXPORT_SYMBOL(rpl_geneve_xmit);
734 static netdev_tx_t geneve_dev_xmit(struct sk_buff *skb, struct net_device *dev)
736 /* Drop All packets coming from networking stack. OVS-CB is
737 * not initialized for these packets.
741 dev->stats.tx_dropped++;
745 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict)
747 /* The max_mtu calculation does not take account of GENEVE
748 * options, to avoid excluding potentially valid
751 int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - sizeof(struct iphdr)
752 - dev->hard_header_len;
757 if (new_mtu > max_mtu) {
768 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
770 return __geneve_change_mtu(dev, new_mtu, true);
773 static const struct net_device_ops geneve_netdev_ops = {
774 #ifdef HAVE_DEV_TSTATS
775 .ndo_init = geneve_init,
776 .ndo_uninit = geneve_uninit,
777 .ndo_get_stats64 = ip_tunnel_get_stats64,
779 .ndo_open = geneve_open,
780 .ndo_stop = geneve_stop,
781 .ndo_start_xmit = geneve_dev_xmit,
782 .ndo_change_mtu = geneve_change_mtu,
783 .ndo_validate_addr = eth_validate_addr,
784 .ndo_set_mac_address = eth_mac_addr,
787 static void geneve_get_drvinfo(struct net_device *dev,
788 struct ethtool_drvinfo *drvinfo)
790 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
791 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
794 static const struct ethtool_ops geneve_ethtool_ops = {
795 .get_drvinfo = geneve_get_drvinfo,
796 .get_link = ethtool_op_get_link,
799 /* Info for udev, that this is a virtual tunnel endpoint */
800 static struct device_type geneve_type = {
804 /* Initialize the device structure. */
805 static void geneve_setup(struct net_device *dev)
809 dev->netdev_ops = &geneve_netdev_ops;
810 dev->ethtool_ops = &geneve_ethtool_ops;
811 dev->destructor = free_netdev;
813 SET_NETDEV_DEVTYPE(dev, &geneve_type);
815 dev->features |= NETIF_F_LLTX;
816 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
817 dev->features |= NETIF_F_RXCSUM;
818 dev->features |= NETIF_F_GSO_SOFTWARE;
820 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
821 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
822 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
828 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
829 eth_hw_addr_random(dev);
832 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
833 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
834 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
835 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
836 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
837 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
838 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
841 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
843 if (tb[IFLA_ADDRESS]) {
844 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
847 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
848 return -EADDRNOTAVAIL;
854 if (data[IFLA_GENEVE_ID]) {
855 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
857 if (vni >= GENEVE_VID_MASK)
864 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
868 bool *tun_on_same_port,
869 bool *tun_collect_md)
871 struct geneve_dev *geneve, *t;
873 *tun_on_same_port = false;
874 *tun_collect_md = false;
876 list_for_each_entry(geneve, &gn->geneve_list, next) {
877 if (geneve->dst_port == dst_port) {
878 *tun_collect_md = geneve->collect_md;
879 *tun_on_same_port = true;
881 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) &&
882 rem_addr == geneve->remote.sin_addr.s_addr &&
883 dst_port == geneve->dst_port)
889 static int geneve_configure(struct net *net, struct net_device *dev,
890 __be32 rem_addr, __u32 vni, __u8 ttl, __u8 tos,
891 __be16 dst_port, bool metadata)
893 struct geneve_net *gn = net_generic(net, geneve_net_id);
894 struct geneve_dev *t, *geneve = netdev_priv(dev);
895 bool tun_collect_md, tun_on_same_port;
899 if (rem_addr || vni || tos || ttl)
906 geneve->vni[0] = (vni & 0x00ff0000) >> 16;
907 geneve->vni[1] = (vni & 0x0000ff00) >> 8;
908 geneve->vni[2] = vni & 0x000000ff;
910 geneve->remote.sin_addr.s_addr = rem_addr;
911 if (IN_MULTICAST(ntohl(geneve->remote.sin_addr.s_addr)))
916 geneve->dst_port = dst_port;
917 geneve->collect_md = metadata;
919 t = geneve_find_dev(gn, dst_port, rem_addr, geneve->vni,
920 &tun_on_same_port, &tun_collect_md);
925 if (tun_on_same_port)
932 err = register_netdevice(dev);
936 list_add(&geneve->next, &gn->geneve_list);
940 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
941 static int geneve_newlink(struct net_device *dev,
942 struct nlattr *tb[], struct nlattr *data[])
944 struct net *net = &init_net;
946 static int geneve_newlink(struct net *net, struct net_device *dev,
947 struct nlattr *tb[], struct nlattr *data[])
950 __be16 dst_port = htons(GENEVE_UDP_PORT);
951 __u8 ttl = 0, tos = 0;
952 bool metadata = false;
956 if (!data[IFLA_GENEVE_ID] || !data[IFLA_GENEVE_REMOTE])
959 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
960 rem_addr = nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
962 if (data[IFLA_GENEVE_TTL])
963 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
965 if (data[IFLA_GENEVE_TOS])
966 tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
968 if (data[IFLA_GENEVE_PORT])
969 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]);
971 if (data[IFLA_GENEVE_COLLECT_METADATA])
974 return geneve_configure(net, dev, rem_addr, vni,
975 ttl, tos, dst_port, metadata);
978 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
979 static void geneve_dellink(struct net_device *dev)
981 static void geneve_dellink(struct net_device *dev, struct list_head *head)
984 struct geneve_dev *geneve = netdev_priv(dev);
986 list_del(&geneve->next);
987 unregister_netdevice_queue(dev, head);
990 static size_t geneve_get_size(const struct net_device *dev)
992 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
993 nla_total_size(sizeof(struct in_addr)) + /* IFLA_GENEVE_REMOTE */
994 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
995 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
996 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
997 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1001 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1003 struct geneve_dev *geneve = netdev_priv(dev);
1006 vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2];
1007 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1008 goto nla_put_failure;
1010 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1011 geneve->remote.sin_addr.s_addr))
1012 goto nla_put_failure;
1014 if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) ||
1015 nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos))
1016 goto nla_put_failure;
1018 if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port))
1019 goto nla_put_failure;
1021 if (geneve->collect_md) {
1022 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1023 goto nla_put_failure;
1032 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1033 .kind = "ovs_geneve",
1034 .maxtype = IFLA_GENEVE_MAX,
1035 .policy = geneve_policy,
1036 .priv_size = sizeof(struct geneve_dev),
1037 .setup = geneve_setup,
1038 .validate = geneve_validate,
1039 .newlink = geneve_newlink,
1040 .dellink = geneve_dellink,
1041 .get_size = geneve_get_size,
1042 .fill_info = geneve_fill_info,
1045 struct net_device *rpl_geneve_dev_create_fb(struct net *net, const char *name,
1046 u8 name_assign_type, u16 dst_port)
1048 struct nlattr *tb[IFLA_MAX + 1];
1049 struct net_device *dev;
1052 memset(tb, 0, sizeof(tb));
1053 dev = rtnl_create_link(net, (char *) name, name_assign_type,
1054 &geneve_link_ops, tb);
1058 err = geneve_configure(net, dev, 0, 0, 0, 0, htons(dst_port), true);
1062 /* openvswitch users expect packet sizes to be unrestricted,
1063 * so set the largest MTU we can.
1065 err = __geneve_change_mtu(dev, IP_MAX_MTU, false);
1073 return ERR_PTR(err);
1075 EXPORT_SYMBOL_GPL(rpl_geneve_dev_create_fb);
1077 static __net_init int geneve_init_net(struct net *net)
1079 struct geneve_net *gn = net_generic(net, geneve_net_id);
1081 INIT_LIST_HEAD(&gn->geneve_list);
1082 INIT_LIST_HEAD(&gn->sock_list);
1086 static void __net_exit geneve_exit_net(struct net *net)
1088 struct geneve_net *gn = net_generic(net, geneve_net_id);
1089 struct geneve_dev *geneve, *next;
1090 struct net_device *dev, *aux;
1095 /* gather any geneve devices that were moved into this ns */
1096 for_each_netdev_safe(net, dev, aux)
1097 if (dev->rtnl_link_ops == &geneve_link_ops)
1098 unregister_netdevice_queue(dev, &list);
1100 /* now gather any other geneve devices that were created in this ns */
1101 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1102 /* If geneve->dev is in the same netns, it was already added
1103 * to the list by the previous loop.
1105 if (!net_eq(dev_net(geneve->dev), net))
1106 unregister_netdevice_queue(geneve->dev, &list);
1109 /* unregister the devices gathered above */
1110 unregister_netdevice_many(&list);
1114 static struct pernet_operations geneve_net_ops = {
1115 .init = geneve_init_net,
1116 .exit = geneve_exit_net,
1117 .id = &geneve_net_id,
1118 .size = sizeof(struct geneve_net),
1121 DEFINE_COMPAT_PNET_REG_FUNC(device)
1122 int rpl_geneve_init_module(void)
1126 rc = register_pernet_subsys(&geneve_net_ops);
1130 rc = rtnl_link_register(&geneve_link_ops);
1134 pr_info("Geneve tunneling driver\n");
1137 unregister_pernet_subsys(&geneve_net_ops);
1142 void rpl_geneve_cleanup_module(void)
1144 rtnl_link_unregister(&geneve_link_ops);
1145 unregister_pernet_subsys(&geneve_net_ops);