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))
39 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
40 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
42 /* per-network namespace private data for this module */
44 struct list_head geneve_list;
45 struct list_head sock_list;
48 static unsigned int geneve_net_id;
50 struct geneve_dev_node {
51 struct hlist_node hlist;
52 struct geneve_dev *geneve;
55 /* Pseudo network device */
57 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
58 #if IS_ENABLED(CONFIG_IPV6)
59 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
61 struct net *net; /* netns for packet i/o */
62 struct net_device *dev; /* netdev for geneve tunnel */
63 struct ip_tunnel_info info;
64 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
65 #if IS_ENABLED(CONFIG_IPV6)
66 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
68 struct list_head next; /* geneve's per namespace list */
69 struct gro_cells gro_cells;
71 bool use_udp6_rx_checksums;
77 struct list_head list;
81 struct hlist_head vni_list[VNI_HASH_SIZE];
84 static inline __u32 geneve_net_vni_hash(u8 vni[3])
88 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
89 return hash_32(vnid, VNI_HASH_BITS);
92 static __be64 vni_to_tunnel_id(const __u8 *vni)
95 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
97 return (__force __be64)(((__force u64)vni[0] << 40) |
98 ((__force u64)vni[1] << 48) |
99 ((__force u64)vni[2] << 56));
103 /* Convert 64 bit tunnel ID to 24 bit VNI. */
104 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
107 vni[0] = (__force __u8)(tun_id >> 16);
108 vni[1] = (__force __u8)(tun_id >> 8);
109 vni[2] = (__force __u8)tun_id;
111 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
112 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
113 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
117 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
119 return !memcmp(vni, &tun_id[5], 3);
122 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
124 return gs->sock->sk->sk_family;
127 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
128 __be32 addr, u8 vni[])
130 struct hlist_head *vni_list_head;
131 struct geneve_dev_node *node;
134 /* Find the device for this VNI */
135 hash = geneve_net_vni_hash(vni);
136 vni_list_head = &gs->vni_list[hash];
137 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
138 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
139 addr == node->geneve->info.key.u.ipv4.dst)
145 #if IS_ENABLED(CONFIG_IPV6)
146 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
147 struct in6_addr addr6, u8 vni[])
149 struct hlist_head *vni_list_head;
150 struct geneve_dev_node *node;
153 /* Find the device for this VNI */
154 hash = geneve_net_vni_hash(vni);
155 vni_list_head = &gs->vni_list[hash];
156 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
157 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) &&
158 ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst))
165 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
167 return (struct genevehdr *)(udp_hdr(skb) + 1);
170 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
173 static u8 zero_vni[3];
176 if (geneve_get_sk_family(gs) == AF_INET) {
180 iph = ip_hdr(skb); /* outer IP header... */
182 if (gs->collect_md) {
186 vni = geneve_hdr(skb)->vni;
190 return geneve_lookup(gs, addr, vni);
191 #if IS_ENABLED(CONFIG_IPV6)
192 } else if (geneve_get_sk_family(gs) == AF_INET6) {
193 static struct in6_addr zero_addr6;
194 struct ipv6hdr *ip6h;
195 struct in6_addr addr6;
197 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
199 if (gs->collect_md) {
203 vni = geneve_hdr(skb)->vni;
207 return geneve6_lookup(gs, addr6, vni);
213 /* geneve receive/decap routine */
214 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
217 struct genevehdr *gnvh = geneve_hdr(skb);
218 struct metadata_dst *tun_dst = NULL;
219 struct pcpu_sw_netstats *stats;
224 if (ip_tunnel_collect_metadata() || gs->collect_md) {
227 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
228 (gnvh->oam ? TUNNEL_OAM : 0) |
229 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
231 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
232 vni_to_tunnel_id(gnvh->vni),
235 geneve->dev->stats.rx_dropped++;
238 /* Update tunnel dst according to Geneve options. */
239 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
240 gnvh->options, gnvh->opt_len * 4,
243 /* Drop packets w/ critical options,
244 * since we don't support any...
246 if (gnvh->critical) {
247 geneve->dev->stats.rx_frame_errors++;
248 geneve->dev->stats.rx_errors++;
253 skb_reset_mac_header(skb);
254 skb->protocol = eth_type_trans(skb, geneve->dev);
255 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
258 skb_dst_set(skb, &tun_dst->dst);
260 /* Ignore packet loops (and multicast echo) */
261 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
262 geneve->dev->stats.rx_errors++;
266 oiph = skb_network_header(skb);
267 skb_reset_network_header(skb);
269 if (geneve_get_sk_family(gs) == AF_INET)
270 err = IP_ECN_decapsulate(oiph, skb);
271 #if IS_ENABLED(CONFIG_IPV6)
273 err = IP6_ECN_decapsulate(oiph, skb);
278 if (geneve_get_sk_family(gs) == AF_INET)
279 net_info_ratelimited("non-ECT from %pI4 "
281 &((struct iphdr *)oiph)->saddr,
282 ((struct iphdr *)oiph)->tos);
283 #if IS_ENABLED(CONFIG_IPV6)
285 net_info_ratelimited("non-ECT from %pI6\n",
286 &((struct ipv6hdr *)oiph)->saddr);
290 ++geneve->dev->stats.rx_frame_errors;
291 ++geneve->dev->stats.rx_errors;
297 err = gro_cells_receive(&geneve->gro_cells, skb);
298 if (likely(err == NET_RX_SUCCESS)) {
299 stats = this_cpu_ptr(geneve->dev->tstats);
300 u64_stats_update_begin(&stats->syncp);
302 stats->rx_bytes += len;
303 u64_stats_update_end(&stats->syncp);
307 /* Consume bad packet */
311 /* Setup stats when device is created */
312 static int geneve_init(struct net_device *dev)
314 struct geneve_dev *geneve = netdev_priv(dev);
317 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
321 err = gro_cells_init(&geneve->gro_cells, dev);
323 free_percpu(dev->tstats);
327 err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL);
329 free_percpu(dev->tstats);
330 gro_cells_destroy(&geneve->gro_cells);
336 static void geneve_uninit(struct net_device *dev)
338 struct geneve_dev *geneve = netdev_priv(dev);
340 dst_cache_destroy(&geneve->info.dst_cache);
341 gro_cells_destroy(&geneve->gro_cells);
342 free_percpu(dev->tstats);
345 /* Callback from net/ipv4/udp.c to receive packets */
346 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
348 struct genevehdr *geneveh;
349 struct geneve_dev *geneve;
350 struct geneve_sock *gs;
353 /* Need UDP and Geneve header to be present */
354 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
357 /* Return packets with reserved bits set */
358 geneveh = geneve_hdr(skb);
359 if (unlikely(geneveh->ver != GENEVE_VER))
362 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
365 gs = rcu_dereference_sk_user_data(sk);
369 geneve = geneve_lookup_skb(gs, skb);
373 opts_len = geneveh->opt_len * 4;
374 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
376 !net_eq(geneve->net, dev_net(geneve->dev)))) {
377 geneve->dev->stats.rx_dropped++;
381 geneve_rx(geneve, gs, skb);
385 /* Consume bad packet */
390 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
391 __be16 port, bool ipv6_rx_csum)
394 struct udp_port_cfg udp_conf;
397 memset(&udp_conf, 0, sizeof(udp_conf));
400 udp_conf.family = AF_INET6;
401 udp_conf.ipv6_v6only = 1;
402 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
404 udp_conf.family = AF_INET;
405 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
408 udp_conf.local_udp_port = port;
410 /* Open UDP socket */
411 err = udp_sock_create(net, &udp_conf, &sock);
418 static int geneve_hlen(struct genevehdr *gh)
420 return sizeof(*gh) + gh->opt_len * 4;
423 static struct sk_buff *geneve_gro_receive(struct sock *sk,
424 struct list_head *head,
427 struct sk_buff *pp = NULL;
429 struct genevehdr *gh, *gh2;
430 unsigned int hlen, gh_len, off_gnv;
431 const struct packet_offload *ptype;
435 off_gnv = skb_gro_offset(skb);
436 hlen = off_gnv + sizeof(*gh);
437 gh = skb_gro_header_fast(skb, off_gnv);
438 if (skb_gro_header_hard(skb, hlen)) {
439 gh = skb_gro_header_slow(skb, hlen, off_gnv);
444 if (gh->ver != GENEVE_VER || gh->oam)
446 gh_len = geneve_hlen(gh);
448 hlen = off_gnv + gh_len;
449 if (skb_gro_header_hard(skb, hlen)) {
450 gh = skb_gro_header_slow(skb, hlen, off_gnv);
455 list_for_each_entry(p, head, list) {
456 if (!NAPI_GRO_CB(p)->same_flow)
459 gh2 = (struct genevehdr *)(p->data + off_gnv);
460 if (gh->opt_len != gh2->opt_len ||
461 memcmp(gh, gh2, gh_len)) {
462 NAPI_GRO_CB(p)->same_flow = 0;
467 type = gh->proto_type;
470 ptype = gro_find_receive_by_type(type);
474 skb_gro_pull(skb, gh_len);
475 skb_gro_postpull_rcsum(skb, gh, gh_len);
476 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
482 skb_gro_flush_final(skb, pp, flush);
487 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
490 struct genevehdr *gh;
491 struct packet_offload *ptype;
496 gh = (struct genevehdr *)(skb->data + nhoff);
497 gh_len = geneve_hlen(gh);
498 type = gh->proto_type;
501 ptype = gro_find_complete_by_type(type);
503 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
507 skb_set_inner_mac_header(skb, nhoff + gh_len);
512 /* Create new listen socket if needed */
513 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
514 bool ipv6, bool ipv6_rx_csum)
516 struct geneve_net *gn = net_generic(net, geneve_net_id);
517 struct geneve_sock *gs;
519 struct udp_tunnel_sock_cfg tunnel_cfg;
522 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
524 return ERR_PTR(-ENOMEM);
526 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
529 return ERR_CAST(sock);
534 for (h = 0; h < VNI_HASH_SIZE; ++h)
535 INIT_HLIST_HEAD(&gs->vni_list[h]);
537 /* Initialize the geneve udp offloads structure */
538 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
540 /* Mark socket as an encapsulation socket */
541 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
542 tunnel_cfg.sk_user_data = gs;
543 tunnel_cfg.encap_type = 1;
544 tunnel_cfg.gro_receive = geneve_gro_receive;
545 tunnel_cfg.gro_complete = geneve_gro_complete;
546 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
547 tunnel_cfg.encap_destroy = NULL;
548 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
549 list_add(&gs->list, &gn->sock_list);
553 static void __geneve_sock_release(struct geneve_sock *gs)
555 if (!gs || --gs->refcnt)
559 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
560 udp_tunnel_sock_release(gs->sock);
564 static void geneve_sock_release(struct geneve_dev *geneve)
566 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
567 #if IS_ENABLED(CONFIG_IPV6)
568 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
570 rcu_assign_pointer(geneve->sock6, NULL);
573 rcu_assign_pointer(geneve->sock4, NULL);
576 __geneve_sock_release(gs4);
577 #if IS_ENABLED(CONFIG_IPV6)
578 __geneve_sock_release(gs6);
582 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
586 struct geneve_sock *gs;
588 list_for_each_entry(gs, &gn->sock_list, list) {
589 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
590 geneve_get_sk_family(gs) == family) {
597 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
599 struct net *net = geneve->net;
600 struct geneve_net *gn = net_generic(net, geneve_net_id);
601 struct geneve_dev_node *node;
602 struct geneve_sock *gs;
606 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst);
612 gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6,
613 geneve->use_udp6_rx_checksums);
618 gs->collect_md = geneve->collect_md;
619 #if IS_ENABLED(CONFIG_IPV6)
621 rcu_assign_pointer(geneve->sock6, gs);
622 node = &geneve->hlist6;
626 rcu_assign_pointer(geneve->sock4, gs);
627 node = &geneve->hlist4;
629 node->geneve = geneve;
631 tunnel_id_to_vni(geneve->info.key.tun_id, vni);
632 hash = geneve_net_vni_hash(vni);
633 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
637 static int geneve_open(struct net_device *dev)
639 struct geneve_dev *geneve = netdev_priv(dev);
640 bool ipv6 = !!(geneve->info.mode & IP_TUNNEL_INFO_IPV6);
641 bool metadata = geneve->collect_md;
644 #if IS_ENABLED(CONFIG_IPV6)
645 if (ipv6 || metadata)
646 ret = geneve_sock_add(geneve, true);
648 if (!ret && (!ipv6 || metadata))
649 ret = geneve_sock_add(geneve, false);
651 geneve_sock_release(geneve);
656 static int geneve_stop(struct net_device *dev)
658 struct geneve_dev *geneve = netdev_priv(dev);
660 hlist_del_init_rcu(&geneve->hlist4.hlist);
661 #if IS_ENABLED(CONFIG_IPV6)
662 hlist_del_init_rcu(&geneve->hlist6.hlist);
664 geneve_sock_release(geneve);
668 static void geneve_build_header(struct genevehdr *geneveh,
669 const struct ip_tunnel_info *info)
671 geneveh->ver = GENEVE_VER;
672 geneveh->opt_len = info->options_len / 4;
673 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
674 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
676 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
677 geneveh->proto_type = htons(ETH_P_TEB);
680 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
681 ip_tunnel_info_opts_get(geneveh->options, info);
684 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
685 const struct ip_tunnel_info *info,
686 bool xnet, int ip_hdr_len)
688 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
689 struct genevehdr *gnvh;
693 skb_reset_mac_header(skb);
694 skb_scrub_packet(skb, xnet);
696 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
697 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
698 err = skb_cow_head(skb, min_headroom);
702 err = udp_tunnel_handle_offloads(skb, udp_sum);
706 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
707 geneve_build_header(gnvh, info);
708 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
716 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
717 struct net_device *dev,
718 struct geneve_sock *gs4,
720 const struct ip_tunnel_info *info)
722 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
723 struct geneve_dev *geneve = netdev_priv(dev);
724 struct dst_cache *dst_cache;
725 struct rtable *rt = NULL;
729 return ERR_PTR(-EIO);
731 memset(fl4, 0, sizeof(*fl4));
732 fl4->flowi4_mark = skb->mark;
733 fl4->flowi4_proto = IPPROTO_UDP;
734 fl4->daddr = info->key.u.ipv4.dst;
735 fl4->saddr = info->key.u.ipv4.src;
738 if ((tos == 1) && !geneve->collect_md) {
739 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
742 fl4->flowi4_tos = RT_TOS(tos);
744 dst_cache = (struct dst_cache *)&info->dst_cache;
746 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
750 rt = ip_route_output_key(geneve->net, fl4);
752 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
753 return ERR_PTR(-ENETUNREACH);
755 if (rt->dst.dev == dev) { /* is this necessary? */
756 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
758 return ERR_PTR(-ELOOP);
761 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
765 #if IS_ENABLED(CONFIG_IPV6)
766 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
767 struct net_device *dev,
768 struct geneve_sock *gs6,
770 const struct ip_tunnel_info *info)
772 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
773 struct geneve_dev *geneve = netdev_priv(dev);
774 struct dst_entry *dst = NULL;
775 struct dst_cache *dst_cache;
779 return ERR_PTR(-EIO);
781 memset(fl6, 0, sizeof(*fl6));
782 fl6->flowi6_mark = skb->mark;
783 fl6->flowi6_proto = IPPROTO_UDP;
784 fl6->daddr = info->key.u.ipv6.dst;
785 fl6->saddr = info->key.u.ipv6.src;
786 prio = info->key.tos;
787 if ((prio == 1) && !geneve->collect_md) {
788 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
792 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
794 dst_cache = (struct dst_cache *)&info->dst_cache;
796 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
800 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) {
801 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
802 return ERR_PTR(-ENETUNREACH);
804 if (dst->dev == dev) { /* is this necessary? */
805 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
807 return ERR_PTR(-ELOOP);
811 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
816 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
817 struct geneve_dev *geneve,
818 const struct ip_tunnel_info *info)
820 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
821 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
822 const struct ip_tunnel_key *key = &info->key;
830 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
835 int mtu = dst_mtu(&rt->dst) - GENEVE_IPV4_HLEN -
838 skb_dst_update_pmtu(skb, mtu);
841 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
842 if (geneve->collect_md) {
843 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
846 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
847 if (geneve->ttl_inherit)
848 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
851 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
853 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
855 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
859 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
860 tos, ttl, df, sport, geneve->info.key.tp_dst,
861 !net_eq(geneve->net, dev_net(geneve->dev)),
862 !(info->key.tun_flags & TUNNEL_CSUM));
866 #if IS_ENABLED(CONFIG_IPV6)
867 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
868 struct geneve_dev *geneve,
869 const struct ip_tunnel_info *info)
871 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
872 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
873 const struct ip_tunnel_key *key = &info->key;
874 struct dst_entry *dst = NULL;
880 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
885 int mtu = dst_mtu(dst) - GENEVE_IPV6_HLEN - info->options_len;
887 skb_dst_update_pmtu(skb, mtu);
890 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
891 if (geneve->collect_md) {
892 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
895 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
897 if (geneve->ttl_inherit)
898 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
901 ttl = ttl ? : ip6_dst_hoplimit(dst);
903 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
907 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
908 &fl6.saddr, &fl6.daddr, prio, ttl,
909 info->key.label, sport, geneve->info.key.tp_dst,
910 !(info->key.tun_flags & TUNNEL_CSUM));
915 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
917 struct geneve_dev *geneve = netdev_priv(dev);
918 struct ip_tunnel_info *info = NULL;
921 if (geneve->collect_md) {
922 info = skb_tunnel_info(skb);
923 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
925 netdev_dbg(dev, "no tunnel metadata\n");
929 info = &geneve->info;
933 #if IS_ENABLED(CONFIG_IPV6)
934 if (info->mode & IP_TUNNEL_INFO_IPV6)
935 err = geneve6_xmit_skb(skb, dev, geneve, info);
938 err = geneve_xmit_skb(skb, dev, geneve, info);
947 dev->stats.collisions++;
948 else if (err == -ENETUNREACH)
949 dev->stats.tx_carrier_errors++;
951 dev->stats.tx_errors++;
955 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
957 if (new_mtu > dev->max_mtu)
958 new_mtu = dev->max_mtu;
959 else if (new_mtu < dev->min_mtu)
960 new_mtu = dev->min_mtu;
966 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
968 struct ip_tunnel_info *info = skb_tunnel_info(skb);
969 struct geneve_dev *geneve = netdev_priv(dev);
971 if (ip_tunnel_info_af(info) == AF_INET) {
974 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
976 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
981 info->key.u.ipv4.src = fl4.saddr;
982 #if IS_ENABLED(CONFIG_IPV6)
983 } else if (ip_tunnel_info_af(info) == AF_INET6) {
984 struct dst_entry *dst;
986 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
988 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
993 info->key.u.ipv6.src = fl6.saddr;
999 info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1000 1, USHRT_MAX, true);
1001 info->key.tp_dst = geneve->info.key.tp_dst;
1005 static const struct net_device_ops geneve_netdev_ops = {
1006 .ndo_init = geneve_init,
1007 .ndo_uninit = geneve_uninit,
1008 .ndo_open = geneve_open,
1009 .ndo_stop = geneve_stop,
1010 .ndo_start_xmit = geneve_xmit,
1011 .ndo_get_stats64 = ip_tunnel_get_stats64,
1012 .ndo_change_mtu = geneve_change_mtu,
1013 .ndo_validate_addr = eth_validate_addr,
1014 .ndo_set_mac_address = eth_mac_addr,
1015 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1018 static void geneve_get_drvinfo(struct net_device *dev,
1019 struct ethtool_drvinfo *drvinfo)
1021 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1022 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1025 static const struct ethtool_ops geneve_ethtool_ops = {
1026 .get_drvinfo = geneve_get_drvinfo,
1027 .get_link = ethtool_op_get_link,
1030 /* Info for udev, that this is a virtual tunnel endpoint */
1031 static struct device_type geneve_type = {
1035 /* Calls the ndo_udp_tunnel_add of the caller in order to
1036 * supply the listening GENEVE udp ports. Callers are expected
1037 * to implement the ndo_udp_tunnel_add.
1039 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1041 struct net *net = dev_net(dev);
1042 struct geneve_net *gn = net_generic(net, geneve_net_id);
1043 struct geneve_sock *gs;
1046 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1048 udp_tunnel_push_rx_port(dev, gs->sock,
1049 UDP_TUNNEL_TYPE_GENEVE);
1051 udp_tunnel_drop_rx_port(dev, gs->sock,
1052 UDP_TUNNEL_TYPE_GENEVE);
1058 /* Initialize the device structure. */
1059 static void geneve_setup(struct net_device *dev)
1063 dev->netdev_ops = &geneve_netdev_ops;
1064 dev->ethtool_ops = &geneve_ethtool_ops;
1065 dev->needs_free_netdev = true;
1067 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1069 dev->features |= NETIF_F_LLTX;
1070 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1071 dev->features |= NETIF_F_RXCSUM;
1072 dev->features |= NETIF_F_GSO_SOFTWARE;
1074 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1075 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1077 /* MTU range: 68 - (something less than 65535) */
1078 dev->min_mtu = ETH_MIN_MTU;
1079 /* The max_mtu calculation does not take account of GENEVE
1080 * options, to avoid excluding potentially valid
1081 * configurations. This will be further reduced by IPvX hdr size.
1083 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1085 netif_keep_dst(dev);
1086 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1087 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1088 eth_hw_addr_random(dev);
1091 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1092 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1093 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1094 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1095 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1096 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1097 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1098 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1099 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1100 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1101 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1102 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1103 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1106 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1107 struct netlink_ext_ack *extack)
1109 if (tb[IFLA_ADDRESS]) {
1110 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1111 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1112 "Provided link layer address is not Ethernet");
1116 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1117 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1118 "Provided Ethernet address is not unicast");
1119 return -EADDRNOTAVAIL;
1124 NL_SET_ERR_MSG(extack,
1125 "Not enough attributes provided to perform the operation");
1129 if (data[IFLA_GENEVE_ID]) {
1130 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1132 if (vni >= GENEVE_N_VID) {
1133 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1134 "Geneve ID must be lower than 16777216");
1142 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1143 const struct ip_tunnel_info *info,
1144 bool *tun_on_same_port,
1145 bool *tun_collect_md)
1147 struct geneve_dev *geneve, *t = NULL;
1149 *tun_on_same_port = false;
1150 *tun_collect_md = false;
1151 list_for_each_entry(geneve, &gn->geneve_list, next) {
1152 if (info->key.tp_dst == geneve->info.key.tp_dst) {
1153 *tun_collect_md = geneve->collect_md;
1154 *tun_on_same_port = true;
1156 if (info->key.tun_id == geneve->info.key.tun_id &&
1157 info->key.tp_dst == geneve->info.key.tp_dst &&
1158 !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u)))
1164 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1166 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1167 info->key.ttl || info->key.label || info->key.tp_src ||
1168 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1171 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1172 struct ip_tunnel_info *b)
1174 if (ip_tunnel_info_af(a) == AF_INET)
1175 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1177 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1180 static int geneve_configure(struct net *net, struct net_device *dev,
1181 struct netlink_ext_ack *extack,
1182 const struct ip_tunnel_info *info,
1183 bool metadata, bool ipv6_rx_csum,
1186 struct geneve_net *gn = net_generic(net, geneve_net_id);
1187 struct geneve_dev *t, *geneve = netdev_priv(dev);
1188 bool tun_collect_md, tun_on_same_port;
1191 if (metadata && !is_tnl_info_zero(info)) {
1192 NL_SET_ERR_MSG(extack,
1193 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1200 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1204 /* make enough headroom for basic scenario */
1205 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1206 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1207 encap_len += sizeof(struct iphdr);
1208 dev->max_mtu -= sizeof(struct iphdr);
1210 encap_len += sizeof(struct ipv6hdr);
1211 dev->max_mtu -= sizeof(struct ipv6hdr);
1213 dev->needed_headroom = encap_len + ETH_HLEN;
1216 if (tun_on_same_port) {
1217 NL_SET_ERR_MSG(extack,
1218 "There can be only one externally controlled device on a destination port");
1222 if (tun_collect_md) {
1223 NL_SET_ERR_MSG(extack,
1224 "There already exists an externally controlled device on this destination port");
1229 dst_cache_reset(&geneve->info.dst_cache);
1230 geneve->info = *info;
1231 geneve->collect_md = metadata;
1232 geneve->use_udp6_rx_checksums = ipv6_rx_csum;
1233 geneve->ttl_inherit = ttl_inherit;
1235 err = register_netdevice(dev);
1239 list_add(&geneve->next, &gn->geneve_list);
1243 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1245 memset(info, 0, sizeof(*info));
1246 info->key.tp_dst = htons(dst_port);
1249 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1250 struct netlink_ext_ack *extack,
1251 struct ip_tunnel_info *info, bool *metadata,
1252 bool *use_udp6_rx_checksums, bool *ttl_inherit,
1257 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1258 NL_SET_ERR_MSG(extack,
1259 "Cannot specify both IPv4 and IPv6 Remote addresses");
1263 if (data[IFLA_GENEVE_REMOTE]) {
1264 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1265 attrtype = IFLA_GENEVE_REMOTE;
1269 info->key.u.ipv4.dst =
1270 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1272 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) {
1273 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1274 "Remote IPv4 address cannot be Multicast");
1279 if (data[IFLA_GENEVE_REMOTE6]) {
1280 #if IS_ENABLED(CONFIG_IPV6)
1281 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1282 attrtype = IFLA_GENEVE_REMOTE6;
1286 info->mode = IP_TUNNEL_INFO_IPV6;
1287 info->key.u.ipv6.dst =
1288 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1290 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1291 IPV6_ADDR_LINKLOCAL) {
1292 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1293 "Remote IPv6 address cannot be link-local");
1296 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1297 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1298 "Remote IPv6 address cannot be Multicast");
1301 info->key.tun_flags |= TUNNEL_CSUM;
1302 *use_udp6_rx_checksums = true;
1304 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1305 "IPv6 support not enabled in the kernel");
1306 return -EPFNOSUPPORT;
1310 if (data[IFLA_GENEVE_ID]) {
1315 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1316 tvni[0] = (vni & 0x00ff0000) >> 16;
1317 tvni[1] = (vni & 0x0000ff00) >> 8;
1318 tvni[2] = vni & 0x000000ff;
1320 tunid = vni_to_tunnel_id(tvni);
1321 if (changelink && (tunid != info->key.tun_id)) {
1322 attrtype = IFLA_GENEVE_ID;
1325 info->key.tun_id = tunid;
1328 if (data[IFLA_GENEVE_TTL])
1329 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1331 if (data[IFLA_GENEVE_TTL_INHERIT])
1332 *ttl_inherit = true;
1334 if (data[IFLA_GENEVE_TOS])
1335 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1337 if (data[IFLA_GENEVE_LABEL]) {
1338 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1339 IPV6_FLOWLABEL_MASK;
1340 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1341 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1342 "Label attribute only applies for IPv6 Geneve devices");
1347 if (data[IFLA_GENEVE_PORT]) {
1349 attrtype = IFLA_GENEVE_PORT;
1352 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1355 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1357 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1363 if (data[IFLA_GENEVE_UDP_CSUM]) {
1365 attrtype = IFLA_GENEVE_UDP_CSUM;
1368 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1369 info->key.tun_flags |= TUNNEL_CSUM;
1372 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1373 #if IS_ENABLED(CONFIG_IPV6)
1375 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1378 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1379 info->key.tun_flags &= ~TUNNEL_CSUM;
1381 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1382 "IPv6 support not enabled in the kernel");
1383 return -EPFNOSUPPORT;
1387 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1388 #if IS_ENABLED(CONFIG_IPV6)
1390 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1393 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1394 *use_udp6_rx_checksums = false;
1396 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1397 "IPv6 support not enabled in the kernel");
1398 return -EPFNOSUPPORT;
1404 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1405 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1409 static void geneve_link_config(struct net_device *dev,
1410 struct ip_tunnel_info *info, struct nlattr *tb[])
1412 struct geneve_dev *geneve = netdev_priv(dev);
1416 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1420 switch (ip_tunnel_info_af(info)) {
1422 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1423 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1425 if (!IS_ERR(rt) && rt->dst.dev) {
1426 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1431 #if IS_ENABLED(CONFIG_IPV6)
1433 struct rt6_info *rt = rt6_lookup(geneve->net,
1434 &info->key.u.ipv6.dst, NULL, 0,
1437 if (rt && rt->dst.dev)
1438 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1448 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1451 static int geneve_newlink(struct net *net, struct net_device *dev,
1452 struct nlattr *tb[], struct nlattr *data[],
1453 struct netlink_ext_ack *extack)
1455 bool use_udp6_rx_checksums = false;
1456 struct ip_tunnel_info info;
1457 bool ttl_inherit = false;
1458 bool metadata = false;
1461 init_tnl_info(&info, GENEVE_UDP_PORT);
1462 err = geneve_nl2info(tb, data, extack, &info, &metadata,
1463 &use_udp6_rx_checksums, &ttl_inherit, false);
1467 err = geneve_configure(net, dev, extack, &info, metadata,
1468 use_udp6_rx_checksums, ttl_inherit);
1472 geneve_link_config(dev, &info, tb);
1477 /* Quiesces the geneve device data path for both TX and RX.
1479 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1480 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1481 * to complete for the existing set of in-flight packets to be transmitted,
1482 * then we would have quiesced the transmit data path. All the future packets
1483 * will get dropped until we unquiesce the data path.
1485 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1486 * if we set that to NULL under RCU and wait for synchronize_net() to
1487 * complete, then we would have quiesced the receive data path.
1489 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1490 struct geneve_sock **gs6)
1492 *gs4 = rtnl_dereference(geneve->sock4);
1493 rcu_assign_pointer(geneve->sock4, NULL);
1495 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1496 #if IS_ENABLED(CONFIG_IPV6)
1497 *gs6 = rtnl_dereference(geneve->sock6);
1498 rcu_assign_pointer(geneve->sock6, NULL);
1500 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1507 /* Resumes the geneve device data path for both TX and RX. */
1508 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1509 struct geneve_sock __maybe_unused *gs6)
1511 rcu_assign_pointer(geneve->sock4, gs4);
1513 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1514 #if IS_ENABLED(CONFIG_IPV6)
1515 rcu_assign_pointer(geneve->sock6, gs6);
1517 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1522 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1523 struct nlattr *data[],
1524 struct netlink_ext_ack *extack)
1526 struct geneve_dev *geneve = netdev_priv(dev);
1527 struct geneve_sock *gs4, *gs6;
1528 struct ip_tunnel_info info;
1530 bool use_udp6_rx_checksums;
1534 /* If the geneve device is configured for metadata (or externally
1535 * controlled, for example, OVS), then nothing can be changed.
1537 if (geneve->collect_md)
1540 /* Start with the existing info. */
1541 memcpy(&info, &geneve->info, sizeof(info));
1542 metadata = geneve->collect_md;
1543 use_udp6_rx_checksums = geneve->use_udp6_rx_checksums;
1544 ttl_inherit = geneve->ttl_inherit;
1545 err = geneve_nl2info(tb, data, extack, &info, &metadata,
1546 &use_udp6_rx_checksums, &ttl_inherit, true);
1550 if (!geneve_dst_addr_equal(&geneve->info, &info)) {
1551 dst_cache_reset(&info.dst_cache);
1552 geneve_link_config(dev, &info, tb);
1555 geneve_quiesce(geneve, &gs4, &gs6);
1556 geneve->info = info;
1557 geneve->collect_md = metadata;
1558 geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
1559 geneve->ttl_inherit = ttl_inherit;
1560 geneve_unquiesce(geneve, gs4, gs6);
1565 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1567 struct geneve_dev *geneve = netdev_priv(dev);
1569 list_del(&geneve->next);
1570 unregister_netdevice_queue(dev, head);
1573 static size_t geneve_get_size(const struct net_device *dev)
1575 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1576 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1577 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1578 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1579 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1580 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1581 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1582 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1583 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1584 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1585 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1589 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1591 struct geneve_dev *geneve = netdev_priv(dev);
1592 struct ip_tunnel_info *info = &geneve->info;
1593 bool ttl_inherit = geneve->ttl_inherit;
1594 bool metadata = geneve->collect_md;
1598 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1599 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1600 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1601 goto nla_put_failure;
1603 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1604 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1605 info->key.u.ipv4.dst))
1606 goto nla_put_failure;
1607 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1608 !!(info->key.tun_flags & TUNNEL_CSUM)))
1609 goto nla_put_failure;
1611 #if IS_ENABLED(CONFIG_IPV6)
1612 } else if (!metadata) {
1613 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1614 &info->key.u.ipv6.dst))
1615 goto nla_put_failure;
1616 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1617 !(info->key.tun_flags & TUNNEL_CSUM)))
1618 goto nla_put_failure;
1622 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1623 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1624 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1625 goto nla_put_failure;
1627 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1628 goto nla_put_failure;
1630 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1631 goto nla_put_failure;
1633 #if IS_ENABLED(CONFIG_IPV6)
1634 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1635 !geneve->use_udp6_rx_checksums))
1636 goto nla_put_failure;
1639 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1640 goto nla_put_failure;
1648 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1650 .maxtype = IFLA_GENEVE_MAX,
1651 .policy = geneve_policy,
1652 .priv_size = sizeof(struct geneve_dev),
1653 .setup = geneve_setup,
1654 .validate = geneve_validate,
1655 .newlink = geneve_newlink,
1656 .changelink = geneve_changelink,
1657 .dellink = geneve_dellink,
1658 .get_size = geneve_get_size,
1659 .fill_info = geneve_fill_info,
1662 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1663 u8 name_assign_type, u16 dst_port)
1665 struct nlattr *tb[IFLA_MAX + 1];
1666 struct ip_tunnel_info info;
1667 struct net_device *dev;
1668 LIST_HEAD(list_kill);
1671 memset(tb, 0, sizeof(tb));
1672 dev = rtnl_create_link(net, name, name_assign_type,
1673 &geneve_link_ops, tb);
1677 init_tnl_info(&info, dst_port);
1678 err = geneve_configure(net, dev, NULL, &info, true, true, false);
1681 return ERR_PTR(err);
1684 /* openvswitch users expect packet sizes to be unrestricted,
1685 * so set the largest MTU we can.
1687 err = geneve_change_mtu(dev, IP_MAX_MTU);
1691 err = rtnl_configure_link(dev, NULL);
1697 geneve_dellink(dev, &list_kill);
1698 unregister_netdevice_many(&list_kill);
1699 return ERR_PTR(err);
1701 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1703 static int geneve_netdevice_event(struct notifier_block *unused,
1704 unsigned long event, void *ptr)
1706 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1708 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1709 event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1710 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1711 } else if (event == NETDEV_UNREGISTER) {
1712 geneve_offload_rx_ports(dev, false);
1713 } else if (event == NETDEV_REGISTER) {
1714 geneve_offload_rx_ports(dev, true);
1720 static struct notifier_block geneve_notifier_block __read_mostly = {
1721 .notifier_call = geneve_netdevice_event,
1724 static __net_init int geneve_init_net(struct net *net)
1726 struct geneve_net *gn = net_generic(net, geneve_net_id);
1728 INIT_LIST_HEAD(&gn->geneve_list);
1729 INIT_LIST_HEAD(&gn->sock_list);
1733 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1735 struct geneve_net *gn = net_generic(net, geneve_net_id);
1736 struct geneve_dev *geneve, *next;
1737 struct net_device *dev, *aux;
1739 /* gather any geneve devices that were moved into this ns */
1740 for_each_netdev_safe(net, dev, aux)
1741 if (dev->rtnl_link_ops == &geneve_link_ops)
1742 unregister_netdevice_queue(dev, head);
1744 /* now gather any other geneve devices that were created in this ns */
1745 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1746 /* If geneve->dev is in the same netns, it was already added
1747 * to the list by the previous loop.
1749 if (!net_eq(dev_net(geneve->dev), net))
1750 unregister_netdevice_queue(geneve->dev, head);
1753 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1756 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1762 list_for_each_entry(net, net_list, exit_list)
1763 geneve_destroy_tunnels(net, &list);
1765 /* unregister the devices gathered above */
1766 unregister_netdevice_many(&list);
1770 static struct pernet_operations geneve_net_ops = {
1771 .init = geneve_init_net,
1772 .exit_batch = geneve_exit_batch_net,
1773 .id = &geneve_net_id,
1774 .size = sizeof(struct geneve_net),
1777 static int __init geneve_init_module(void)
1781 rc = register_pernet_subsys(&geneve_net_ops);
1785 rc = register_netdevice_notifier(&geneve_notifier_block);
1789 rc = rtnl_link_register(&geneve_link_ops);
1795 unregister_netdevice_notifier(&geneve_notifier_block);
1797 unregister_pernet_subsys(&geneve_net_ops);
1801 late_initcall(geneve_init_module);
1803 static void __exit geneve_cleanup_module(void)
1805 rtnl_link_unregister(&geneve_link_ops);
1806 unregister_netdevice_notifier(&geneve_notifier_block);
1807 unregister_pernet_subsys(&geneve_net_ops);
1809 module_exit(geneve_cleanup_module);
1811 MODULE_LICENSE("GPL");
1812 MODULE_VERSION(GENEVE_NETDEV_VER);
1813 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1814 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1815 MODULE_ALIAS_RTNL_LINK("geneve");