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/etherdevice.h>
16 #include <linux/hash.h>
17 #include <net/dst_metadata.h>
18 #include <net/gro_cells.h>
19 #include <net/rtnetlink.h>
20 #include <net/geneve.h>
21 #include <net/protocol.h>
23 #define GENEVE_NETDEV_VER "0.6"
25 #define GENEVE_UDP_PORT 6081
27 #define GENEVE_N_VID (1u << 24)
28 #define GENEVE_VID_MASK (GENEVE_N_VID - 1)
30 #define VNI_HASH_BITS 10
31 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
33 static bool log_ecn_error = true;
34 module_param(log_ecn_error, bool, 0644);
35 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
38 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
40 /* per-network namespace private data for this module */
42 struct list_head geneve_list;
43 struct list_head sock_list;
46 static unsigned int geneve_net_id;
48 /* Pseudo network device */
50 struct hlist_node hlist; /* vni hash table */
51 struct net *net; /* netns for packet i/o */
52 struct net_device *dev; /* netdev for geneve tunnel */
53 struct ip_tunnel_info info;
54 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
55 #if IS_ENABLED(CONFIG_IPV6)
56 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
58 struct list_head next; /* geneve's per namespace list */
59 struct gro_cells gro_cells;
61 bool use_udp6_rx_checksums;
66 struct list_head list;
70 struct hlist_head vni_list[VNI_HASH_SIZE];
73 static inline __u32 geneve_net_vni_hash(u8 vni[3])
77 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
78 return hash_32(vnid, VNI_HASH_BITS);
81 static __be64 vni_to_tunnel_id(const __u8 *vni)
84 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
86 return (__force __be64)(((__force u64)vni[0] << 40) |
87 ((__force u64)vni[1] << 48) |
88 ((__force u64)vni[2] << 56));
92 /* Convert 64 bit tunnel ID to 24 bit VNI. */
93 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
96 vni[0] = (__force __u8)(tun_id >> 16);
97 vni[1] = (__force __u8)(tun_id >> 8);
98 vni[2] = (__force __u8)tun_id;
100 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
101 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
102 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
106 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
109 return (vni[0] == tun_id[2]) &&
110 (vni[1] == tun_id[1]) &&
111 (vni[2] == tun_id[0]);
113 return !memcmp(vni, &tun_id[5], 3);
117 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
119 return gs->sock->sk->sk_family;
122 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
123 __be32 addr, u8 vni[])
125 struct hlist_head *vni_list_head;
126 struct geneve_dev *geneve;
129 /* Find the device for this VNI */
130 hash = geneve_net_vni_hash(vni);
131 vni_list_head = &gs->vni_list[hash];
132 hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
133 if (eq_tun_id_and_vni((u8 *)&geneve->info.key.tun_id, vni) &&
134 addr == geneve->info.key.u.ipv4.dst)
140 #if IS_ENABLED(CONFIG_IPV6)
141 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
142 struct in6_addr addr6, u8 vni[])
144 struct hlist_head *vni_list_head;
145 struct geneve_dev *geneve;
148 /* Find the device for this VNI */
149 hash = geneve_net_vni_hash(vni);
150 vni_list_head = &gs->vni_list[hash];
151 hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) {
152 if (eq_tun_id_and_vni((u8 *)&geneve->info.key.tun_id, vni) &&
153 ipv6_addr_equal(&addr6, &geneve->info.key.u.ipv6.dst))
160 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
162 return (struct genevehdr *)(udp_hdr(skb) + 1);
165 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
168 static u8 zero_vni[3];
171 if (geneve_get_sk_family(gs) == AF_INET) {
175 iph = ip_hdr(skb); /* outer IP header... */
177 if (gs->collect_md) {
181 vni = geneve_hdr(skb)->vni;
185 return geneve_lookup(gs, addr, vni);
186 #if IS_ENABLED(CONFIG_IPV6)
187 } else if (geneve_get_sk_family(gs) == AF_INET6) {
188 static struct in6_addr zero_addr6;
189 struct ipv6hdr *ip6h;
190 struct in6_addr addr6;
192 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
194 if (gs->collect_md) {
198 vni = geneve_hdr(skb)->vni;
202 return geneve6_lookup(gs, addr6, vni);
208 /* geneve receive/decap routine */
209 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
212 struct genevehdr *gnvh = geneve_hdr(skb);
213 struct metadata_dst *tun_dst = NULL;
214 struct pcpu_sw_netstats *stats;
219 if (ip_tunnel_collect_metadata() || gs->collect_md) {
222 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
223 (gnvh->oam ? TUNNEL_OAM : 0) |
224 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
226 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
227 vni_to_tunnel_id(gnvh->vni),
230 geneve->dev->stats.rx_dropped++;
233 /* Update tunnel dst according to Geneve options. */
234 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
235 gnvh->options, gnvh->opt_len * 4);
237 /* Drop packets w/ critical options,
238 * since we don't support any...
240 if (gnvh->critical) {
241 geneve->dev->stats.rx_frame_errors++;
242 geneve->dev->stats.rx_errors++;
247 skb_reset_mac_header(skb);
248 skb->protocol = eth_type_trans(skb, geneve->dev);
249 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
252 skb_dst_set(skb, &tun_dst->dst);
254 /* Ignore packet loops (and multicast echo) */
255 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
256 geneve->dev->stats.rx_errors++;
260 oiph = skb_network_header(skb);
261 skb_reset_network_header(skb);
263 if (geneve_get_sk_family(gs) == AF_INET)
264 err = IP_ECN_decapsulate(oiph, skb);
265 #if IS_ENABLED(CONFIG_IPV6)
267 err = IP6_ECN_decapsulate(oiph, skb);
272 if (geneve_get_sk_family(gs) == AF_INET)
273 net_info_ratelimited("non-ECT from %pI4 "
275 &((struct iphdr *)oiph)->saddr,
276 ((struct iphdr *)oiph)->tos);
277 #if IS_ENABLED(CONFIG_IPV6)
279 net_info_ratelimited("non-ECT from %pI6\n",
280 &((struct ipv6hdr *)oiph)->saddr);
284 ++geneve->dev->stats.rx_frame_errors;
285 ++geneve->dev->stats.rx_errors;
291 err = gro_cells_receive(&geneve->gro_cells, skb);
292 if (likely(err == NET_RX_SUCCESS)) {
293 stats = this_cpu_ptr(geneve->dev->tstats);
294 u64_stats_update_begin(&stats->syncp);
296 stats->rx_bytes += len;
297 u64_stats_update_end(&stats->syncp);
301 /* Consume bad packet */
305 /* Setup stats when device is created */
306 static int geneve_init(struct net_device *dev)
308 struct geneve_dev *geneve = netdev_priv(dev);
311 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
315 err = gro_cells_init(&geneve->gro_cells, dev);
317 free_percpu(dev->tstats);
321 err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
323 free_percpu(dev->tstats);
324 gro_cells_destroy(&geneve->gro_cells);
330 static void geneve_uninit(struct net_device *dev)
332 struct geneve_dev *geneve = netdev_priv(dev);
334 dst_cache_destroy(&geneve->info.dst_cache);
335 gro_cells_destroy(&geneve->gro_cells);
336 free_percpu(dev->tstats);
339 /* Callback from net/ipv4/udp.c to receive packets */
340 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
342 struct genevehdr *geneveh;
343 struct geneve_dev *geneve;
344 struct geneve_sock *gs;
347 /* Need UDP and Geneve header to be present */
348 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
351 /* Return packets with reserved bits set */
352 geneveh = geneve_hdr(skb);
353 if (unlikely(geneveh->ver != GENEVE_VER))
356 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
359 gs = rcu_dereference_sk_user_data(sk);
363 geneve = geneve_lookup_skb(gs, skb);
367 opts_len = geneveh->opt_len * 4;
368 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
370 !net_eq(geneve->net, dev_net(geneve->dev)))) {
371 geneve->dev->stats.rx_dropped++;
375 geneve_rx(geneve, gs, skb);
379 /* Consume bad packet */
384 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
385 __be16 port, bool ipv6_rx_csum)
388 struct udp_port_cfg udp_conf;
391 memset(&udp_conf, 0, sizeof(udp_conf));
394 udp_conf.family = AF_INET6;
395 udp_conf.ipv6_v6only = 1;
396 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
398 udp_conf.family = AF_INET;
399 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
402 udp_conf.local_udp_port = port;
404 /* Open UDP socket */
405 err = udp_sock_create(net, &udp_conf, &sock);
412 static int geneve_hlen(struct genevehdr *gh)
414 return sizeof(*gh) + gh->opt_len * 4;
417 static struct sk_buff **geneve_gro_receive(struct sock *sk,
418 struct sk_buff **head,
421 struct sk_buff *p, **pp = NULL;
422 struct genevehdr *gh, *gh2;
423 unsigned int hlen, gh_len, off_gnv;
424 const struct packet_offload *ptype;
428 off_gnv = skb_gro_offset(skb);
429 hlen = off_gnv + sizeof(*gh);
430 gh = skb_gro_header_fast(skb, off_gnv);
431 if (skb_gro_header_hard(skb, hlen)) {
432 gh = skb_gro_header_slow(skb, hlen, off_gnv);
437 if (gh->ver != GENEVE_VER || gh->oam)
439 gh_len = geneve_hlen(gh);
441 hlen = off_gnv + gh_len;
442 if (skb_gro_header_hard(skb, hlen)) {
443 gh = skb_gro_header_slow(skb, hlen, off_gnv);
448 for (p = *head; p; p = p->next) {
449 if (!NAPI_GRO_CB(p)->same_flow)
452 gh2 = (struct genevehdr *)(p->data + off_gnv);
453 if (gh->opt_len != gh2->opt_len ||
454 memcmp(gh, gh2, gh_len)) {
455 NAPI_GRO_CB(p)->same_flow = 0;
460 type = gh->proto_type;
463 ptype = gro_find_receive_by_type(type);
467 skb_gro_pull(skb, gh_len);
468 skb_gro_postpull_rcsum(skb, gh, gh_len);
469 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
475 NAPI_GRO_CB(skb)->flush |= flush;
480 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
483 struct genevehdr *gh;
484 struct packet_offload *ptype;
489 gh = (struct genevehdr *)(skb->data + nhoff);
490 gh_len = geneve_hlen(gh);
491 type = gh->proto_type;
494 ptype = gro_find_complete_by_type(type);
496 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
500 skb_set_inner_mac_header(skb, nhoff + gh_len);
505 /* Create new listen socket if needed */
506 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
507 bool ipv6, bool ipv6_rx_csum)
509 struct geneve_net *gn = net_generic(net, geneve_net_id);
510 struct geneve_sock *gs;
512 struct udp_tunnel_sock_cfg tunnel_cfg;
515 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
517 return ERR_PTR(-ENOMEM);
519 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
522 return ERR_CAST(sock);
527 for (h = 0; h < VNI_HASH_SIZE; ++h)
528 INIT_HLIST_HEAD(&gs->vni_list[h]);
530 /* Initialize the geneve udp offloads structure */
531 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
533 /* Mark socket as an encapsulation socket */
534 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
535 tunnel_cfg.sk_user_data = gs;
536 tunnel_cfg.encap_type = 1;
537 tunnel_cfg.gro_receive = geneve_gro_receive;
538 tunnel_cfg.gro_complete = geneve_gro_complete;
539 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
540 tunnel_cfg.encap_destroy = NULL;
541 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
542 list_add(&gs->list, &gn->sock_list);
546 static void __geneve_sock_release(struct geneve_sock *gs)
548 if (!gs || --gs->refcnt)
552 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
553 udp_tunnel_sock_release(gs->sock);
557 static void geneve_sock_release(struct geneve_dev *geneve)
559 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
560 #if IS_ENABLED(CONFIG_IPV6)
561 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
563 rcu_assign_pointer(geneve->sock6, NULL);
566 rcu_assign_pointer(geneve->sock4, NULL);
569 __geneve_sock_release(gs4);
570 #if IS_ENABLED(CONFIG_IPV6)
571 __geneve_sock_release(gs6);
575 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
579 struct geneve_sock *gs;
581 list_for_each_entry(gs, &gn->sock_list, list) {
582 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
583 geneve_get_sk_family(gs) == family) {
590 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
592 struct net *net = geneve->net;
593 struct geneve_net *gn = net_generic(net, geneve_net_id);
594 struct geneve_sock *gs;
598 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
604 gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
605 geneve->use_udp6_rx_checksums);
610 gs->collect_md = geneve->collect_md;
611 #if IS_ENABLED(CONFIG_IPV6)
613 rcu_assign_pointer(geneve->sock6, gs);
616 rcu_assign_pointer(geneve->sock4, gs);
618 tunnel_id_to_vni(geneve->info.key.tun_id, vni);
619 hash = geneve_net_vni_hash(vni);
620 hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]);
624 static int geneve_open(struct net_device *dev)
626 struct geneve_dev *geneve = netdev_priv(dev);
627 bool ipv6 = !!(geneve->info.mode & IP_TUNNEL_INFO_IPV6);
628 bool metadata = geneve->collect_md;
631 #if IS_ENABLED(CONFIG_IPV6)
632 if (ipv6 || metadata)
633 ret = geneve_sock_add(geneve, true);
635 if (!ret && (!ipv6 || metadata))
636 ret = geneve_sock_add(geneve, false);
638 geneve_sock_release(geneve);
643 static int geneve_stop(struct net_device *dev)
645 struct geneve_dev *geneve = netdev_priv(dev);
647 if (!hlist_unhashed(&geneve->hlist))
648 hlist_del_rcu(&geneve->hlist);
649 geneve_sock_release(geneve);
653 static void geneve_build_header(struct genevehdr *geneveh,
654 const struct ip_tunnel_info *info)
656 geneveh->ver = GENEVE_VER;
657 geneveh->opt_len = info->options_len / 4;
658 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
659 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
661 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
662 geneveh->proto_type = htons(ETH_P_TEB);
665 ip_tunnel_info_opts_get(geneveh->options, info);
668 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
669 const struct ip_tunnel_info *info,
670 bool xnet, int ip_hdr_len)
672 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
673 struct genevehdr *gnvh;
677 skb_reset_mac_header(skb);
678 skb_scrub_packet(skb, xnet);
680 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
681 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
682 err = skb_cow_head(skb, min_headroom);
686 err = udp_tunnel_handle_offloads(skb, udp_sum);
690 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
691 geneve_build_header(gnvh, info);
692 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
700 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
701 struct net_device *dev,
703 const struct ip_tunnel_info *info)
705 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
706 struct geneve_dev *geneve = netdev_priv(dev);
707 struct dst_cache *dst_cache;
708 struct rtable *rt = NULL;
711 if (!rcu_dereference(geneve->sock4))
712 return ERR_PTR(-EIO);
714 memset(fl4, 0, sizeof(*fl4));
715 fl4->flowi4_mark = skb->mark;
716 fl4->flowi4_proto = IPPROTO_UDP;
717 fl4->daddr = info->key.u.ipv4.dst;
718 fl4->saddr = info->key.u.ipv4.src;
721 if ((tos == 1) && !geneve->collect_md) {
722 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
725 fl4->flowi4_tos = RT_TOS(tos);
727 dst_cache = (struct dst_cache *)&info->dst_cache;
729 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
733 rt = ip_route_output_key(geneve->net, fl4);
735 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
736 return ERR_PTR(-ENETUNREACH);
738 if (rt->dst.dev == dev) { /* is this necessary? */
739 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
741 return ERR_PTR(-ELOOP);
744 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
748 #if IS_ENABLED(CONFIG_IPV6)
749 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
750 struct net_device *dev,
752 const struct ip_tunnel_info *info)
754 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
755 struct geneve_dev *geneve = netdev_priv(dev);
756 struct dst_entry *dst = NULL;
757 struct dst_cache *dst_cache;
758 struct geneve_sock *gs6;
761 gs6 = rcu_dereference(geneve->sock6);
763 return ERR_PTR(-EIO);
765 memset(fl6, 0, sizeof(*fl6));
766 fl6->flowi6_mark = skb->mark;
767 fl6->flowi6_proto = IPPROTO_UDP;
768 fl6->daddr = info->key.u.ipv6.dst;
769 fl6->saddr = info->key.u.ipv6.src;
770 prio = info->key.tos;
771 if ((prio == 1) && !geneve->collect_md) {
772 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
776 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
778 dst_cache = (struct dst_cache *)&info->dst_cache;
780 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
784 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
785 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
786 return ERR_PTR(-ENETUNREACH);
788 if (dst->dev == dev) { /* is this necessary? */
789 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
791 return ERR_PTR(-ELOOP);
795 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
800 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
801 struct geneve_dev *geneve,
802 const struct ip_tunnel_info *info)
804 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
805 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
806 const struct ip_tunnel_key *key = &info->key;
814 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
818 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
819 if (geneve->collect_md) {
820 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
823 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
824 ttl = key->ttl ? : ip4_dst_hoplimit(&rt->dst);
826 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
828 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
832 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
833 tos, ttl, df, sport, geneve->info.key.tp_dst,
834 !net_eq(geneve->net, dev_net(geneve->dev)),
835 !(info->key.tun_flags & TUNNEL_CSUM));
839 #if IS_ENABLED(CONFIG_IPV6)
840 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
841 struct geneve_dev *geneve,
842 const struct ip_tunnel_info *info)
844 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
845 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
846 const struct ip_tunnel_key *key = &info->key;
847 struct dst_entry *dst = NULL;
853 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
857 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
858 if (geneve->collect_md) {
859 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
862 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
864 ttl = key->ttl ? : ip6_dst_hoplimit(dst);
866 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
870 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
871 &fl6.saddr, &fl6.daddr, prio, ttl,
872 info->key.label, sport, geneve->info.key.tp_dst,
873 !(info->key.tun_flags & TUNNEL_CSUM));
878 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
880 struct geneve_dev *geneve = netdev_priv(dev);
881 struct ip_tunnel_info *info = NULL;
884 if (geneve->collect_md) {
885 info = skb_tunnel_info(skb);
886 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
888 netdev_dbg(dev, "no tunnel metadata\n");
892 info = &geneve->info;
896 #if IS_ENABLED(CONFIG_IPV6)
897 if (info->mode & IP_TUNNEL_INFO_IPV6)
898 err = geneve6_xmit_skb(skb, dev, geneve, info);
901 err = geneve_xmit_skb(skb, dev, geneve, info);
910 dev->stats.collisions++;
911 else if (err == -ENETUNREACH)
912 dev->stats.tx_carrier_errors++;
914 dev->stats.tx_errors++;
918 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
920 /* Only possible if called internally, ndo_change_mtu path's new_mtu
921 * is guaranteed to be between dev->min_mtu and dev->max_mtu.
923 if (new_mtu > dev->max_mtu)
924 new_mtu = dev->max_mtu;
930 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
932 struct ip_tunnel_info *info = skb_tunnel_info(skb);
933 struct geneve_dev *geneve = netdev_priv(dev);
935 if (ip_tunnel_info_af(info) == AF_INET) {
939 rt = geneve_get_v4_rt(skb, dev, &fl4, info);
944 info->key.u.ipv4.src = fl4.saddr;
945 #if IS_ENABLED(CONFIG_IPV6)
946 } else if (ip_tunnel_info_af(info) == AF_INET6) {
947 struct dst_entry *dst;
950 dst = geneve_get_v6_dst(skb, dev, &fl6, info);
955 info->key.u.ipv6.src = fl6.saddr;
961 info->key.tp_src = udp_flow_src_port(geneve->net, skb,
963 info->key.tp_dst = geneve->info.key.tp_dst;
967 static const struct net_device_ops geneve_netdev_ops = {
968 .ndo_init = geneve_init,
969 .ndo_uninit = geneve_uninit,
970 .ndo_open = geneve_open,
971 .ndo_stop = geneve_stop,
972 .ndo_start_xmit = geneve_xmit,
973 .ndo_get_stats64 = ip_tunnel_get_stats64,
974 .ndo_change_mtu = geneve_change_mtu,
975 .ndo_validate_addr = eth_validate_addr,
976 .ndo_set_mac_address = eth_mac_addr,
977 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
980 static void geneve_get_drvinfo(struct net_device *dev,
981 struct ethtool_drvinfo *drvinfo)
983 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
984 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
987 static const struct ethtool_ops geneve_ethtool_ops = {
988 .get_drvinfo = geneve_get_drvinfo,
989 .get_link = ethtool_op_get_link,
992 /* Info for udev, that this is a virtual tunnel endpoint */
993 static struct device_type geneve_type = {
997 /* Calls the ndo_udp_tunnel_add of the caller in order to
998 * supply the listening GENEVE udp ports. Callers are expected
999 * to implement the ndo_udp_tunnel_add.
1001 static void geneve_push_rx_ports(struct net_device *dev)
1003 struct net *net = dev_net(dev);
1004 struct geneve_net *gn = net_generic(net, geneve_net_id);
1005 struct geneve_sock *gs;
1008 list_for_each_entry_rcu(gs, &gn->sock_list, list)
1009 udp_tunnel_push_rx_port(dev, gs->sock,
1010 UDP_TUNNEL_TYPE_GENEVE);
1014 /* Initialize the device structure. */
1015 static void geneve_setup(struct net_device *dev)
1019 dev->netdev_ops = &geneve_netdev_ops;
1020 dev->ethtool_ops = &geneve_ethtool_ops;
1021 dev->needs_free_netdev = true;
1023 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1025 dev->features |= NETIF_F_LLTX;
1026 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1027 dev->features |= NETIF_F_RXCSUM;
1028 dev->features |= NETIF_F_GSO_SOFTWARE;
1030 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1031 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1033 /* MTU range: 68 - (something less than 65535) */
1034 dev->min_mtu = ETH_MIN_MTU;
1035 /* The max_mtu calculation does not take account of GENEVE
1036 * options, to avoid excluding potentially valid
1037 * configurations. This will be further reduced by IPvX hdr size.
1039 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1041 netif_keep_dst(dev);
1042 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1043 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1044 eth_hw_addr_random(dev);
1047 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1048 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1049 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1050 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1051 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1052 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1053 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1054 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1055 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1056 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1057 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1058 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1061 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[])
1063 if (tb[IFLA_ADDRESS]) {
1064 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1067 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1068 return -EADDRNOTAVAIL;
1074 if (data[IFLA_GENEVE_ID]) {
1075 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1077 if (vni >= GENEVE_VID_MASK)
1084 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1085 const struct ip_tunnel_info *info,
1086 bool *tun_on_same_port,
1087 bool *tun_collect_md)
1089 struct geneve_dev *geneve, *t = NULL;
1091 *tun_on_same_port = false;
1092 *tun_collect_md = false;
1093 list_for_each_entry(geneve, &gn->geneve_list, next) {
1094 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1095 *tun_collect_md = geneve->collect_md;
1096 *tun_on_same_port = true;
1098 if (info->key.tun_id == geneve->info.key.tun_id &&
1099 info->key.tp_dst == geneve->info.key.tp_dst &&
1100 !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1106 static bool is_all_zero(const u8 *fp, size_t size)
1110 for (i = 0; i < size; i++)
1116 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1118 if (info->key.tun_id || info->key.tun_flags || info->key.tos ||
1119 info->key.ttl || info->key.label || info->key.tp_src ||
1120 !is_all_zero((const u8 *)&info->key.u, sizeof(info->key.u)))
1126 static int geneve_configure(struct net *net, struct net_device *dev,
1127 const struct ip_tunnel_info *info,
1128 bool metadata, bool ipv6_rx_csum)
1130 struct geneve_net *gn = net_generic(net, geneve_net_id);
1131 struct geneve_dev *t, *geneve = netdev_priv(dev);
1132 bool tun_collect_md, tun_on_same_port;
1135 if (metadata && !is_tnl_info_zero(info))
1141 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1145 /* make enough headroom for basic scenario */
1146 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1147 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1148 encap_len += sizeof(struct iphdr);
1149 dev->max_mtu -= sizeof(struct iphdr);
1151 encap_len += sizeof(struct ipv6hdr);
1152 dev->max_mtu -= sizeof(struct ipv6hdr);
1154 dev->needed_headroom = encap_len + ETH_HLEN;
1157 if (tun_on_same_port)
1164 dst_cache_reset(&geneve->info.dst_cache);
1165 geneve->info = *info;
1166 geneve->collect_md = metadata;
1167 geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1169 err = register_netdevice(dev);
1173 list_add(&geneve->next, &gn->geneve_list);
1177 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1179 memset(info, 0, sizeof(*info));
1180 info->key.tp_dst = htons(dst_port);
1183 static int geneve_newlink(struct net *net, struct net_device *dev,
1184 struct nlattr *tb[], struct nlattr *data[],
1185 struct netlink_ext_ack *extack)
1187 bool use_udp6_rx_checksums = false;
1188 struct ip_tunnel_info info;
1189 bool metadata = false;
1191 init_tnl_info(&info, GENEVE_UDP_PORT);
1193 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6])
1196 if (data[IFLA_GENEVE_REMOTE]) {
1197 info.key.u.ipv4.dst =
1198 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1200 if (IN_MULTICAST(ntohl(info.key.u.ipv4.dst))) {
1201 netdev_dbg(dev, "multicast remote is unsupported\n");
1206 if (data[IFLA_GENEVE_REMOTE6]) {
1207 #if IS_ENABLED(CONFIG_IPV6)
1208 info.mode = IP_TUNNEL_INFO_IPV6;
1209 info.key.u.ipv6.dst =
1210 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1212 if (ipv6_addr_type(&info.key.u.ipv6.dst) &
1213 IPV6_ADDR_LINKLOCAL) {
1214 netdev_dbg(dev, "link-local remote is unsupported\n");
1217 if (ipv6_addr_is_multicast(&info.key.u.ipv6.dst)) {
1218 netdev_dbg(dev, "multicast remote is unsupported\n");
1221 info.key.tun_flags |= TUNNEL_CSUM;
1222 use_udp6_rx_checksums = true;
1224 return -EPFNOSUPPORT;
1228 if (data[IFLA_GENEVE_ID]) {
1232 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1233 tvni[0] = (vni & 0x00ff0000) >> 16;
1234 tvni[1] = (vni & 0x0000ff00) >> 8;
1235 tvni[2] = vni & 0x000000ff;
1237 info.key.tun_id = vni_to_tunnel_id(tvni);
1239 if (data[IFLA_GENEVE_TTL])
1240 info.key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1242 if (data[IFLA_GENEVE_TOS])
1243 info.key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1245 if (data[IFLA_GENEVE_LABEL]) {
1246 info.key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1247 IPV6_FLOWLABEL_MASK;
1248 if (info.key.label && (!(info.mode & IP_TUNNEL_INFO_IPV6)))
1252 if (data[IFLA_GENEVE_PORT])
1253 info.key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1255 if (data[IFLA_GENEVE_COLLECT_METADATA])
1258 if (data[IFLA_GENEVE_UDP_CSUM] &&
1259 nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1260 info.key.tun_flags |= TUNNEL_CSUM;
1262 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] &&
1263 nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1264 info.key.tun_flags &= ~TUNNEL_CSUM;
1266 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] &&
1267 nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1268 use_udp6_rx_checksums = false;
1270 return geneve_configure(net, dev, &info, metadata, use_udp6_rx_checksums);
1273 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1275 struct geneve_dev *geneve = netdev_priv(dev);
1277 list_del(&geneve->next);
1278 unregister_netdevice_queue(dev, head);
1281 static size_t geneve_get_size(const struct net_device *dev)
1283 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1284 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1285 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1286 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1287 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1288 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1289 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1290 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1291 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1292 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1296 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1298 struct geneve_dev *geneve = netdev_priv(dev);
1299 struct ip_tunnel_info *info = &geneve->info;
1303 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1304 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1305 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1306 goto nla_put_failure;
1308 if (rtnl_dereference(geneve->sock4)) {
1309 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1310 info->key.u.ipv4.dst))
1311 goto nla_put_failure;
1313 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1314 !!(info->key.tun_flags & TUNNEL_CSUM)))
1315 goto nla_put_failure;
1319 #if IS_ENABLED(CONFIG_IPV6)
1320 if (rtnl_dereference(geneve->sock6)) {
1321 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1322 &info->key.u.ipv6.dst))
1323 goto nla_put_failure;
1325 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1326 !(info->key.tun_flags & TUNNEL_CSUM)))
1327 goto nla_put_failure;
1329 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1330 !geneve->use_udp6_rx_checksums))
1331 goto nla_put_failure;
1335 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1336 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1337 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1338 goto nla_put_failure;
1340 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1341 goto nla_put_failure;
1343 if (geneve->collect_md) {
1344 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1345 goto nla_put_failure;
1353 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1355 .maxtype = IFLA_GENEVE_MAX,
1356 .policy = geneve_policy,
1357 .priv_size = sizeof(struct geneve_dev),
1358 .setup = geneve_setup,
1359 .validate = geneve_validate,
1360 .newlink = geneve_newlink,
1361 .dellink = geneve_dellink,
1362 .get_size = geneve_get_size,
1363 .fill_info = geneve_fill_info,
1366 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1367 u8 name_assign_type, u16 dst_port)
1369 struct nlattr *tb[IFLA_MAX + 1];
1370 struct ip_tunnel_info info;
1371 struct net_device *dev;
1372 LIST_HEAD(list_kill);
1375 memset(tb, 0, sizeof(tb));
1376 dev = rtnl_create_link(net, name, name_assign_type,
1377 &geneve_link_ops, tb);
1381 init_tnl_info(&info, dst_port);
1382 err = geneve_configure(net, dev, &info, true, true);
1385 return ERR_PTR(err);
1388 /* openvswitch users expect packet sizes to be unrestricted,
1389 * so set the largest MTU we can.
1391 err = geneve_change_mtu(dev, IP_MAX_MTU);
1395 err = rtnl_configure_link(dev, NULL);
1401 geneve_dellink(dev, &list_kill);
1402 unregister_netdevice_many(&list_kill);
1403 return ERR_PTR(err);
1405 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1407 static int geneve_netdevice_event(struct notifier_block *unused,
1408 unsigned long event, void *ptr)
1410 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1412 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1413 geneve_push_rx_ports(dev);
1418 static struct notifier_block geneve_notifier_block __read_mostly = {
1419 .notifier_call = geneve_netdevice_event,
1422 static __net_init int geneve_init_net(struct net *net)
1424 struct geneve_net *gn = net_generic(net, geneve_net_id);
1426 INIT_LIST_HEAD(&gn->geneve_list);
1427 INIT_LIST_HEAD(&gn->sock_list);
1431 static void __net_exit geneve_exit_net(struct net *net)
1433 struct geneve_net *gn = net_generic(net, geneve_net_id);
1434 struct geneve_dev *geneve, *next;
1435 struct net_device *dev, *aux;
1440 /* gather any geneve devices that were moved into this ns */
1441 for_each_netdev_safe(net, dev, aux)
1442 if (dev->rtnl_link_ops == &geneve_link_ops)
1443 unregister_netdevice_queue(dev, &list);
1445 /* now gather any other geneve devices that were created in this ns */
1446 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1447 /* If geneve->dev is in the same netns, it was already added
1448 * to the list by the previous loop.
1450 if (!net_eq(dev_net(geneve->dev), net))
1451 unregister_netdevice_queue(geneve->dev, &list);
1454 /* unregister the devices gathered above */
1455 unregister_netdevice_many(&list);
1459 static struct pernet_operations geneve_net_ops = {
1460 .init = geneve_init_net,
1461 .exit = geneve_exit_net,
1462 .id = &geneve_net_id,
1463 .size = sizeof(struct geneve_net),
1466 static int __init geneve_init_module(void)
1470 rc = register_pernet_subsys(&geneve_net_ops);
1474 rc = register_netdevice_notifier(&geneve_notifier_block);
1478 rc = rtnl_link_register(&geneve_link_ops);
1484 unregister_netdevice_notifier(&geneve_notifier_block);
1486 unregister_pernet_subsys(&geneve_net_ops);
1490 late_initcall(geneve_init_module);
1492 static void __exit geneve_cleanup_module(void)
1494 rtnl_link_unregister(&geneve_link_ops);
1495 unregister_netdevice_notifier(&geneve_notifier_block);
1496 unregister_pernet_subsys(&geneve_net_ops);
1498 module_exit(geneve_cleanup_module);
1500 MODULE_LICENSE("GPL");
1501 MODULE_VERSION(GENEVE_NETDEV_VER);
1502 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1503 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1504 MODULE_ALIAS_RTNL_LINK("geneve");