1 // SPDX-License-Identifier: GPL-2.0-only
3 * GENEVE: Generic Network Virtualization Encapsulation
5 * Copyright (c) 2015 Red Hat, Inc.
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/ethtool.h>
11 #include <linux/kernel.h>
12 #include <linux/module.h>
13 #include <linux/etherdevice.h>
14 #include <linux/hash.h>
15 #include <net/ipv6_stubs.h>
16 #include <net/dst_metadata.h>
17 #include <net/gro_cells.h>
18 #include <net/rtnetlink.h>
19 #include <net/geneve.h>
21 #include <net/protocol.h>
23 #define GENEVE_NETDEV_VER "0.6"
25 #define GENEVE_N_VID (1u << 24)
26 #define GENEVE_VID_MASK (GENEVE_N_VID - 1)
28 #define VNI_HASH_BITS 10
29 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
31 static bool log_ecn_error = true;
32 module_param(log_ecn_error, bool, 0644);
33 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
36 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
37 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
38 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
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 struct geneve_dev_node {
49 struct hlist_node hlist;
50 struct geneve_dev *geneve;
53 struct geneve_config {
54 struct ip_tunnel_info info;
56 bool use_udp6_rx_checksums;
58 enum ifla_geneve_df df;
59 bool inner_proto_inherit;
62 /* Pseudo network device */
64 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
65 #if IS_ENABLED(CONFIG_IPV6)
66 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
68 struct net *net; /* netns for packet i/o */
69 struct net_device *dev; /* netdev for geneve tunnel */
70 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
71 #if IS_ENABLED(CONFIG_IPV6)
72 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
74 struct list_head next; /* geneve's per namespace list */
75 struct gro_cells gro_cells;
76 struct geneve_config cfg;
81 struct list_head list;
85 struct hlist_head vni_list[VNI_HASH_SIZE];
88 static inline __u32 geneve_net_vni_hash(u8 vni[3])
92 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
93 return hash_32(vnid, VNI_HASH_BITS);
96 static __be64 vni_to_tunnel_id(const __u8 *vni)
99 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
101 return (__force __be64)(((__force u64)vni[0] << 40) |
102 ((__force u64)vni[1] << 48) |
103 ((__force u64)vni[2] << 56));
107 /* Convert 64 bit tunnel ID to 24 bit VNI. */
108 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
111 vni[0] = (__force __u8)(tun_id >> 16);
112 vni[1] = (__force __u8)(tun_id >> 8);
113 vni[2] = (__force __u8)tun_id;
115 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
116 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
117 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
121 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
123 return !memcmp(vni, &tun_id[5], 3);
126 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
128 return gs->sock->sk->sk_family;
131 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
132 __be32 addr, u8 vni[])
134 struct hlist_head *vni_list_head;
135 struct geneve_dev_node *node;
138 /* Find the device for this VNI */
139 hash = geneve_net_vni_hash(vni);
140 vni_list_head = &gs->vni_list[hash];
141 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
142 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
143 addr == node->geneve->cfg.info.key.u.ipv4.dst)
149 #if IS_ENABLED(CONFIG_IPV6)
150 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
151 struct in6_addr addr6, u8 vni[])
153 struct hlist_head *vni_list_head;
154 struct geneve_dev_node *node;
157 /* Find the device for this VNI */
158 hash = geneve_net_vni_hash(vni);
159 vni_list_head = &gs->vni_list[hash];
160 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
161 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
162 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
169 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
171 return (struct genevehdr *)(udp_hdr(skb) + 1);
174 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
177 static u8 zero_vni[3];
180 if (geneve_get_sk_family(gs) == AF_INET) {
184 iph = ip_hdr(skb); /* outer IP header... */
186 if (gs->collect_md) {
190 vni = geneve_hdr(skb)->vni;
194 return geneve_lookup(gs, addr, vni);
195 #if IS_ENABLED(CONFIG_IPV6)
196 } else if (geneve_get_sk_family(gs) == AF_INET6) {
197 static struct in6_addr zero_addr6;
198 struct ipv6hdr *ip6h;
199 struct in6_addr addr6;
201 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
203 if (gs->collect_md) {
207 vni = geneve_hdr(skb)->vni;
211 return geneve6_lookup(gs, addr6, vni);
217 /* geneve receive/decap routine */
218 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
221 struct genevehdr *gnvh = geneve_hdr(skb);
222 struct metadata_dst *tun_dst = NULL;
227 if (ip_tunnel_collect_metadata() || gs->collect_md) {
230 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
231 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
233 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
234 vni_to_tunnel_id(gnvh->vni),
237 geneve->dev->stats.rx_dropped++;
240 /* Update tunnel dst according to Geneve options. */
241 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
242 gnvh->options, gnvh->opt_len * 4,
245 /* Drop packets w/ critical options,
246 * since we don't support any...
248 if (gnvh->critical) {
249 geneve->dev->stats.rx_frame_errors++;
250 geneve->dev->stats.rx_errors++;
256 skb_dst_set(skb, &tun_dst->dst);
258 if (gnvh->proto_type == htons(ETH_P_TEB)) {
259 skb_reset_mac_header(skb);
260 skb->protocol = eth_type_trans(skb, geneve->dev);
261 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
263 /* Ignore packet loops (and multicast echo) */
264 if (ether_addr_equal(eth_hdr(skb)->h_source,
265 geneve->dev->dev_addr)) {
266 geneve->dev->stats.rx_errors++;
270 skb_reset_mac_header(skb);
271 skb->dev = geneve->dev;
272 skb->pkt_type = PACKET_HOST;
275 oiph = skb_network_header(skb);
276 skb_reset_network_header(skb);
278 if (geneve_get_sk_family(gs) == AF_INET)
279 err = IP_ECN_decapsulate(oiph, skb);
280 #if IS_ENABLED(CONFIG_IPV6)
282 err = IP6_ECN_decapsulate(oiph, skb);
287 if (geneve_get_sk_family(gs) == AF_INET)
288 net_info_ratelimited("non-ECT from %pI4 "
290 &((struct iphdr *)oiph)->saddr,
291 ((struct iphdr *)oiph)->tos);
292 #if IS_ENABLED(CONFIG_IPV6)
294 net_info_ratelimited("non-ECT from %pI6\n",
295 &((struct ipv6hdr *)oiph)->saddr);
299 ++geneve->dev->stats.rx_frame_errors;
300 ++geneve->dev->stats.rx_errors;
306 err = gro_cells_receive(&geneve->gro_cells, skb);
307 if (likely(err == NET_RX_SUCCESS))
308 dev_sw_netstats_rx_add(geneve->dev, len);
312 /* Consume bad packet */
316 /* Setup stats when device is created */
317 static int geneve_init(struct net_device *dev)
319 struct geneve_dev *geneve = netdev_priv(dev);
322 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
326 err = gro_cells_init(&geneve->gro_cells, dev);
328 free_percpu(dev->tstats);
332 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
334 free_percpu(dev->tstats);
335 gro_cells_destroy(&geneve->gro_cells);
341 static void geneve_uninit(struct net_device *dev)
343 struct geneve_dev *geneve = netdev_priv(dev);
345 dst_cache_destroy(&geneve->cfg.info.dst_cache);
346 gro_cells_destroy(&geneve->gro_cells);
347 free_percpu(dev->tstats);
350 /* Callback from net/ipv4/udp.c to receive packets */
351 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
353 struct genevehdr *geneveh;
354 struct geneve_dev *geneve;
355 struct geneve_sock *gs;
359 /* Need UDP and Geneve header to be present */
360 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
363 /* Return packets with reserved bits set */
364 geneveh = geneve_hdr(skb);
365 if (unlikely(geneveh->ver != GENEVE_VER))
368 inner_proto = geneveh->proto_type;
370 if (unlikely((inner_proto != htons(ETH_P_TEB) &&
371 inner_proto != htons(ETH_P_IP) &&
372 inner_proto != htons(ETH_P_IPV6))))
375 gs = rcu_dereference_sk_user_data(sk);
379 geneve = geneve_lookup_skb(gs, skb);
383 if (unlikely((!geneve->cfg.inner_proto_inherit &&
384 inner_proto != htons(ETH_P_TEB)))) {
385 geneve->dev->stats.rx_dropped++;
389 opts_len = geneveh->opt_len * 4;
390 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
391 !net_eq(geneve->net, dev_net(geneve->dev)))) {
392 geneve->dev->stats.rx_dropped++;
396 geneve_rx(geneve, gs, skb);
400 /* Consume bad packet */
405 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
406 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
408 struct genevehdr *geneveh;
409 struct geneve_sock *gs;
410 u8 zero_vni[3] = { 0 };
413 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
416 geneveh = geneve_hdr(skb);
417 if (geneveh->ver != GENEVE_VER)
420 if (geneveh->proto_type != htons(ETH_P_TEB))
423 gs = rcu_dereference_sk_user_data(sk);
427 if (geneve_get_sk_family(gs) == AF_INET) {
428 struct iphdr *iph = ip_hdr(skb);
431 if (!gs->collect_md) {
432 vni = geneve_hdr(skb)->vni;
436 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
439 #if IS_ENABLED(CONFIG_IPV6)
440 if (geneve_get_sk_family(gs) == AF_INET6) {
441 struct ipv6hdr *ip6h = ipv6_hdr(skb);
442 struct in6_addr addr6;
444 memset(&addr6, 0, sizeof(struct in6_addr));
446 if (!gs->collect_md) {
447 vni = geneve_hdr(skb)->vni;
451 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
455 return -EPFNOSUPPORT;
458 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
459 __be16 port, bool ipv6_rx_csum)
462 struct udp_port_cfg udp_conf;
465 memset(&udp_conf, 0, sizeof(udp_conf));
468 udp_conf.family = AF_INET6;
469 udp_conf.ipv6_v6only = 1;
470 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
472 udp_conf.family = AF_INET;
473 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
476 udp_conf.local_udp_port = port;
478 /* Open UDP socket */
479 err = udp_sock_create(net, &udp_conf, &sock);
483 udp_allow_gso(sock->sk);
487 static int geneve_hlen(struct genevehdr *gh)
489 return sizeof(*gh) + gh->opt_len * 4;
492 static struct sk_buff *geneve_gro_receive(struct sock *sk,
493 struct list_head *head,
496 struct sk_buff *pp = NULL;
498 struct genevehdr *gh, *gh2;
499 unsigned int hlen, gh_len, off_gnv;
500 const struct packet_offload *ptype;
504 off_gnv = skb_gro_offset(skb);
505 hlen = off_gnv + sizeof(*gh);
506 gh = skb_gro_header(skb, hlen, off_gnv);
510 if (gh->ver != GENEVE_VER || gh->oam)
512 gh_len = geneve_hlen(gh);
514 hlen = off_gnv + gh_len;
515 if (skb_gro_header_hard(skb, hlen)) {
516 gh = skb_gro_header_slow(skb, hlen, off_gnv);
521 list_for_each_entry(p, head, list) {
522 if (!NAPI_GRO_CB(p)->same_flow)
525 gh2 = (struct genevehdr *)(p->data + off_gnv);
526 if (gh->opt_len != gh2->opt_len ||
527 memcmp(gh, gh2, gh_len)) {
528 NAPI_GRO_CB(p)->same_flow = 0;
533 skb_gro_pull(skb, gh_len);
534 skb_gro_postpull_rcsum(skb, gh, gh_len);
535 type = gh->proto_type;
536 if (likely(type == htons(ETH_P_TEB)))
537 return call_gro_receive(eth_gro_receive, head, skb);
539 ptype = gro_find_receive_by_type(type);
543 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
547 skb_gro_flush_final(skb, pp, flush);
552 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
555 struct genevehdr *gh;
556 struct packet_offload *ptype;
561 gh = (struct genevehdr *)(skb->data + nhoff);
562 gh_len = geneve_hlen(gh);
563 type = gh->proto_type;
565 /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */
566 if (likely(type == htons(ETH_P_TEB)))
567 return eth_gro_complete(skb, nhoff + gh_len);
569 ptype = gro_find_complete_by_type(type);
571 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
573 skb_set_inner_mac_header(skb, nhoff + gh_len);
578 /* Create new listen socket if needed */
579 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
580 bool ipv6, bool ipv6_rx_csum)
582 struct geneve_net *gn = net_generic(net, geneve_net_id);
583 struct geneve_sock *gs;
585 struct udp_tunnel_sock_cfg tunnel_cfg;
588 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
590 return ERR_PTR(-ENOMEM);
592 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
595 return ERR_CAST(sock);
600 for (h = 0; h < VNI_HASH_SIZE; ++h)
601 INIT_HLIST_HEAD(&gs->vni_list[h]);
603 /* Initialize the geneve udp offloads structure */
604 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
606 /* Mark socket as an encapsulation socket */
607 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
608 tunnel_cfg.sk_user_data = gs;
609 tunnel_cfg.encap_type = 1;
610 tunnel_cfg.gro_receive = geneve_gro_receive;
611 tunnel_cfg.gro_complete = geneve_gro_complete;
612 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
613 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
614 tunnel_cfg.encap_destroy = NULL;
615 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
616 list_add(&gs->list, &gn->sock_list);
620 static void __geneve_sock_release(struct geneve_sock *gs)
622 if (!gs || --gs->refcnt)
626 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
627 udp_tunnel_sock_release(gs->sock);
631 static void geneve_sock_release(struct geneve_dev *geneve)
633 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
634 #if IS_ENABLED(CONFIG_IPV6)
635 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
637 rcu_assign_pointer(geneve->sock6, NULL);
640 rcu_assign_pointer(geneve->sock4, NULL);
643 __geneve_sock_release(gs4);
644 #if IS_ENABLED(CONFIG_IPV6)
645 __geneve_sock_release(gs6);
649 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
653 struct geneve_sock *gs;
655 list_for_each_entry(gs, &gn->sock_list, list) {
656 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
657 geneve_get_sk_family(gs) == family) {
664 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
666 struct net *net = geneve->net;
667 struct geneve_net *gn = net_generic(net, geneve_net_id);
668 struct geneve_dev_node *node;
669 struct geneve_sock *gs;
673 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
679 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
680 geneve->cfg.use_udp6_rx_checksums);
685 gs->collect_md = geneve->cfg.collect_md;
686 #if IS_ENABLED(CONFIG_IPV6)
688 rcu_assign_pointer(geneve->sock6, gs);
689 node = &geneve->hlist6;
693 rcu_assign_pointer(geneve->sock4, gs);
694 node = &geneve->hlist4;
696 node->geneve = geneve;
698 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
699 hash = geneve_net_vni_hash(vni);
700 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
704 static int geneve_open(struct net_device *dev)
706 struct geneve_dev *geneve = netdev_priv(dev);
707 bool metadata = geneve->cfg.collect_md;
711 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
712 ipv4 = !ipv6 || metadata;
713 #if IS_ENABLED(CONFIG_IPV6)
715 ret = geneve_sock_add(geneve, true);
716 if (ret < 0 && ret != -EAFNOSUPPORT)
721 ret = geneve_sock_add(geneve, false);
723 geneve_sock_release(geneve);
728 static int geneve_stop(struct net_device *dev)
730 struct geneve_dev *geneve = netdev_priv(dev);
732 hlist_del_init_rcu(&geneve->hlist4.hlist);
733 #if IS_ENABLED(CONFIG_IPV6)
734 hlist_del_init_rcu(&geneve->hlist6.hlist);
736 geneve_sock_release(geneve);
740 static void geneve_build_header(struct genevehdr *geneveh,
741 const struct ip_tunnel_info *info,
744 geneveh->ver = GENEVE_VER;
745 geneveh->opt_len = info->options_len / 4;
746 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
747 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
749 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
750 geneveh->proto_type = inner_proto;
753 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
754 ip_tunnel_info_opts_get(geneveh->options, info);
757 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
758 const struct ip_tunnel_info *info,
759 bool xnet, int ip_hdr_len,
760 bool inner_proto_inherit)
762 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
763 struct genevehdr *gnvh;
768 skb_reset_mac_header(skb);
769 skb_scrub_packet(skb, xnet);
771 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
772 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
773 err = skb_cow_head(skb, min_headroom);
777 err = udp_tunnel_handle_offloads(skb, udp_sum);
781 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
782 inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
783 geneve_build_header(gnvh, info, inner_proto);
784 skb_set_inner_protocol(skb, inner_proto);
792 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
793 struct net_device *dev,
794 struct geneve_sock *gs4,
796 const struct ip_tunnel_info *info,
797 __be16 dport, __be16 sport,
800 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
801 struct geneve_dev *geneve = netdev_priv(dev);
802 struct dst_cache *dst_cache;
803 struct rtable *rt = NULL;
807 return ERR_PTR(-EIO);
809 memset(fl4, 0, sizeof(*fl4));
810 fl4->flowi4_mark = skb->mark;
811 fl4->flowi4_proto = IPPROTO_UDP;
812 fl4->daddr = info->key.u.ipv4.dst;
813 fl4->saddr = info->key.u.ipv4.src;
814 fl4->fl4_dport = dport;
815 fl4->fl4_sport = sport;
816 fl4->flowi4_flags = info->key.flow_flags;
819 if ((tos == 1) && !geneve->cfg.collect_md) {
820 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
823 fl4->flowi4_tos = RT_TOS(tos);
827 dst_cache = (struct dst_cache *)&info->dst_cache;
829 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
833 rt = ip_route_output_key(geneve->net, fl4);
835 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
836 return ERR_PTR(-ENETUNREACH);
838 if (rt->dst.dev == dev) { /* is this necessary? */
839 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
841 return ERR_PTR(-ELOOP);
844 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
848 #if IS_ENABLED(CONFIG_IPV6)
849 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
850 struct net_device *dev,
851 struct geneve_sock *gs6,
853 const struct ip_tunnel_info *info,
854 __be16 dport, __be16 sport)
856 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
857 struct geneve_dev *geneve = netdev_priv(dev);
858 struct dst_entry *dst = NULL;
859 struct dst_cache *dst_cache;
863 return ERR_PTR(-EIO);
865 memset(fl6, 0, sizeof(*fl6));
866 fl6->flowi6_mark = skb->mark;
867 fl6->flowi6_proto = IPPROTO_UDP;
868 fl6->daddr = info->key.u.ipv6.dst;
869 fl6->saddr = info->key.u.ipv6.src;
870 fl6->fl6_dport = dport;
871 fl6->fl6_sport = sport;
873 prio = info->key.tos;
874 if ((prio == 1) && !geneve->cfg.collect_md) {
875 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
879 fl6->flowlabel = ip6_make_flowinfo(prio, info->key.label);
880 dst_cache = (struct dst_cache *)&info->dst_cache;
882 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
886 dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
889 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
890 return ERR_PTR(-ENETUNREACH);
892 if (dst->dev == dev) { /* is this necessary? */
893 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
895 return ERR_PTR(-ELOOP);
899 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
904 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
905 struct geneve_dev *geneve,
906 const struct ip_tunnel_info *info)
908 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
909 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
910 const struct ip_tunnel_key *key = &info->key;
919 if (!pskb_inet_may_pull(skb))
922 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
923 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
924 geneve->cfg.info.key.tp_dst, sport, &full_tos);
928 err = skb_tunnel_check_pmtu(skb, &rt->dst,
929 GENEVE_IPV4_HLEN + info->options_len,
930 netif_is_any_bridge_port(dev));
932 dst_release(&rt->dst);
935 struct ip_tunnel_info *info;
937 info = skb_tunnel_info(skb);
939 struct ip_tunnel_info *unclone;
941 unclone = skb_tunnel_info_unclone(skb);
942 if (unlikely(!unclone)) {
943 dst_release(&rt->dst);
947 unclone->key.u.ipv4.dst = fl4.saddr;
948 unclone->key.u.ipv4.src = fl4.daddr;
951 if (!pskb_may_pull(skb, ETH_HLEN)) {
952 dst_release(&rt->dst);
956 skb->protocol = eth_type_trans(skb, geneve->dev);
958 dst_release(&rt->dst);
962 if (geneve->cfg.collect_md) {
963 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
966 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
968 tos = ip_tunnel_ecn_encap(full_tos, ip_hdr(skb), skb);
969 if (geneve->cfg.ttl_inherit)
970 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
973 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
975 if (geneve->cfg.df == GENEVE_DF_SET) {
977 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
978 struct ethhdr *eth = eth_hdr(skb);
980 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
982 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
983 struct iphdr *iph = ip_hdr(skb);
985 if (iph->frag_off & htons(IP_DF))
991 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
992 geneve->cfg.inner_proto_inherit);
996 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
997 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
998 !net_eq(geneve->net, dev_net(geneve->dev)),
999 !(info->key.tun_flags & TUNNEL_CSUM));
1003 #if IS_ENABLED(CONFIG_IPV6)
1004 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
1005 struct geneve_dev *geneve,
1006 const struct ip_tunnel_info *info)
1008 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
1009 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1010 const struct ip_tunnel_key *key = &info->key;
1011 struct dst_entry *dst = NULL;
1017 if (!pskb_inet_may_pull(skb))
1020 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
1021 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1022 geneve->cfg.info.key.tp_dst, sport);
1024 return PTR_ERR(dst);
1026 err = skb_tunnel_check_pmtu(skb, dst,
1027 GENEVE_IPV6_HLEN + info->options_len,
1028 netif_is_any_bridge_port(dev));
1033 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1036 struct ip_tunnel_info *unclone;
1038 unclone = skb_tunnel_info_unclone(skb);
1039 if (unlikely(!unclone)) {
1044 unclone->key.u.ipv6.dst = fl6.saddr;
1045 unclone->key.u.ipv6.src = fl6.daddr;
1048 if (!pskb_may_pull(skb, ETH_HLEN)) {
1053 skb->protocol = eth_type_trans(skb, geneve->dev);
1059 if (geneve->cfg.collect_md) {
1060 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1063 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1065 if (geneve->cfg.ttl_inherit)
1066 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1069 ttl = ttl ? : ip6_dst_hoplimit(dst);
1071 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
1072 geneve->cfg.inner_proto_inherit);
1076 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1077 &fl6.saddr, &fl6.daddr, prio, ttl,
1078 info->key.label, sport, geneve->cfg.info.key.tp_dst,
1079 !(info->key.tun_flags & TUNNEL_CSUM));
1084 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1086 struct geneve_dev *geneve = netdev_priv(dev);
1087 struct ip_tunnel_info *info = NULL;
1090 if (geneve->cfg.collect_md) {
1091 info = skb_tunnel_info(skb);
1092 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1093 netdev_dbg(dev, "no tunnel metadata\n");
1095 dev->stats.tx_dropped++;
1096 return NETDEV_TX_OK;
1099 info = &geneve->cfg.info;
1103 #if IS_ENABLED(CONFIG_IPV6)
1104 if (info->mode & IP_TUNNEL_INFO_IPV6)
1105 err = geneve6_xmit_skb(skb, dev, geneve, info);
1108 err = geneve_xmit_skb(skb, dev, geneve, info);
1112 return NETDEV_TX_OK;
1114 if (err != -EMSGSIZE)
1118 dev->stats.collisions++;
1119 else if (err == -ENETUNREACH)
1120 dev->stats.tx_carrier_errors++;
1122 dev->stats.tx_errors++;
1123 return NETDEV_TX_OK;
1126 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1128 if (new_mtu > dev->max_mtu)
1129 new_mtu = dev->max_mtu;
1130 else if (new_mtu < dev->min_mtu)
1131 new_mtu = dev->min_mtu;
1137 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1139 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1140 struct geneve_dev *geneve = netdev_priv(dev);
1143 if (ip_tunnel_info_af(info) == AF_INET) {
1147 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1148 sport = udp_flow_src_port(geneve->net, skb,
1149 1, USHRT_MAX, true);
1151 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1152 geneve->cfg.info.key.tp_dst, sport, NULL);
1157 info->key.u.ipv4.src = fl4.saddr;
1158 #if IS_ENABLED(CONFIG_IPV6)
1159 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1160 struct dst_entry *dst;
1163 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1164 sport = udp_flow_src_port(geneve->net, skb,
1165 1, USHRT_MAX, true);
1167 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1168 geneve->cfg.info.key.tp_dst, sport);
1170 return PTR_ERR(dst);
1173 info->key.u.ipv6.src = fl6.saddr;
1179 info->key.tp_src = sport;
1180 info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1184 static const struct net_device_ops geneve_netdev_ops = {
1185 .ndo_init = geneve_init,
1186 .ndo_uninit = geneve_uninit,
1187 .ndo_open = geneve_open,
1188 .ndo_stop = geneve_stop,
1189 .ndo_start_xmit = geneve_xmit,
1190 .ndo_get_stats64 = dev_get_tstats64,
1191 .ndo_change_mtu = geneve_change_mtu,
1192 .ndo_validate_addr = eth_validate_addr,
1193 .ndo_set_mac_address = eth_mac_addr,
1194 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1197 static void geneve_get_drvinfo(struct net_device *dev,
1198 struct ethtool_drvinfo *drvinfo)
1200 strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1201 strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1204 static const struct ethtool_ops geneve_ethtool_ops = {
1205 .get_drvinfo = geneve_get_drvinfo,
1206 .get_link = ethtool_op_get_link,
1209 /* Info for udev, that this is a virtual tunnel endpoint */
1210 static struct device_type geneve_type = {
1214 /* Calls the ndo_udp_tunnel_add of the caller in order to
1215 * supply the listening GENEVE udp ports. Callers are expected
1216 * to implement the ndo_udp_tunnel_add.
1218 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1220 struct net *net = dev_net(dev);
1221 struct geneve_net *gn = net_generic(net, geneve_net_id);
1222 struct geneve_sock *gs;
1225 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1227 udp_tunnel_push_rx_port(dev, gs->sock,
1228 UDP_TUNNEL_TYPE_GENEVE);
1230 udp_tunnel_drop_rx_port(dev, gs->sock,
1231 UDP_TUNNEL_TYPE_GENEVE);
1237 /* Initialize the device structure. */
1238 static void geneve_setup(struct net_device *dev)
1242 dev->netdev_ops = &geneve_netdev_ops;
1243 dev->ethtool_ops = &geneve_ethtool_ops;
1244 dev->needs_free_netdev = true;
1246 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1248 dev->features |= NETIF_F_LLTX;
1249 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1250 dev->features |= NETIF_F_RXCSUM;
1251 dev->features |= NETIF_F_GSO_SOFTWARE;
1253 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1254 dev->hw_features |= NETIF_F_RXCSUM;
1255 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1257 /* MTU range: 68 - (something less than 65535) */
1258 dev->min_mtu = ETH_MIN_MTU;
1259 /* The max_mtu calculation does not take account of GENEVE
1260 * options, to avoid excluding potentially valid
1261 * configurations. This will be further reduced by IPvX hdr size.
1263 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1265 netif_keep_dst(dev);
1266 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1267 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1268 eth_hw_addr_random(dev);
1271 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1272 [IFLA_GENEVE_UNSPEC] = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1273 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1274 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1275 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1276 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1277 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1278 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1279 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1280 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1281 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1282 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1283 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1284 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1285 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1286 [IFLA_GENEVE_INNER_PROTO_INHERIT] = { .type = NLA_FLAG },
1289 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1290 struct netlink_ext_ack *extack)
1292 if (tb[IFLA_ADDRESS]) {
1293 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1294 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1295 "Provided link layer address is not Ethernet");
1299 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1300 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1301 "Provided Ethernet address is not unicast");
1302 return -EADDRNOTAVAIL;
1307 NL_SET_ERR_MSG(extack,
1308 "Not enough attributes provided to perform the operation");
1312 if (data[IFLA_GENEVE_ID]) {
1313 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1315 if (vni >= GENEVE_N_VID) {
1316 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1317 "Geneve ID must be lower than 16777216");
1322 if (data[IFLA_GENEVE_DF]) {
1323 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1325 if (df < 0 || df > GENEVE_DF_MAX) {
1326 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1327 "Invalid DF attribute");
1335 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1336 const struct ip_tunnel_info *info,
1337 bool *tun_on_same_port,
1338 bool *tun_collect_md)
1340 struct geneve_dev *geneve, *t = NULL;
1342 *tun_on_same_port = false;
1343 *tun_collect_md = false;
1344 list_for_each_entry(geneve, &gn->geneve_list, next) {
1345 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1346 *tun_collect_md = geneve->cfg.collect_md;
1347 *tun_on_same_port = true;
1349 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1350 info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1351 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1357 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1359 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1360 info->key.ttl || info->key.label || info->key.tp_src ||
1361 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1364 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1365 struct ip_tunnel_info *b)
1367 if (ip_tunnel_info_af(a) == AF_INET)
1368 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1370 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1373 static int geneve_configure(struct net *net, struct net_device *dev,
1374 struct netlink_ext_ack *extack,
1375 const struct geneve_config *cfg)
1377 struct geneve_net *gn = net_generic(net, geneve_net_id);
1378 struct geneve_dev *t, *geneve = netdev_priv(dev);
1379 const struct ip_tunnel_info *info = &cfg->info;
1380 bool tun_collect_md, tun_on_same_port;
1383 if (cfg->collect_md && !is_tnl_info_zero(info)) {
1384 NL_SET_ERR_MSG(extack,
1385 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1392 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1396 /* make enough headroom for basic scenario */
1397 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1398 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1399 encap_len += sizeof(struct iphdr);
1400 dev->max_mtu -= sizeof(struct iphdr);
1402 encap_len += sizeof(struct ipv6hdr);
1403 dev->max_mtu -= sizeof(struct ipv6hdr);
1405 dev->needed_headroom = encap_len + ETH_HLEN;
1407 if (cfg->collect_md) {
1408 if (tun_on_same_port) {
1409 NL_SET_ERR_MSG(extack,
1410 "There can be only one externally controlled device on a destination port");
1414 if (tun_collect_md) {
1415 NL_SET_ERR_MSG(extack,
1416 "There already exists an externally controlled device on this destination port");
1421 dst_cache_reset(&geneve->cfg.info.dst_cache);
1422 memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1424 if (geneve->cfg.inner_proto_inherit) {
1425 dev->header_ops = NULL;
1426 dev->type = ARPHRD_NONE;
1427 dev->hard_header_len = 0;
1429 dev->flags = IFF_NOARP;
1432 err = register_netdevice(dev);
1436 list_add(&geneve->next, &gn->geneve_list);
1440 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1442 memset(info, 0, sizeof(*info));
1443 info->key.tp_dst = htons(dst_port);
1446 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1447 struct netlink_ext_ack *extack,
1448 struct geneve_config *cfg, bool changelink)
1450 struct ip_tunnel_info *info = &cfg->info;
1453 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1454 NL_SET_ERR_MSG(extack,
1455 "Cannot specify both IPv4 and IPv6 Remote addresses");
1459 if (data[IFLA_GENEVE_REMOTE]) {
1460 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1461 attrtype = IFLA_GENEVE_REMOTE;
1465 info->key.u.ipv4.dst =
1466 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1468 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1469 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1470 "Remote IPv4 address cannot be Multicast");
1475 if (data[IFLA_GENEVE_REMOTE6]) {
1476 #if IS_ENABLED(CONFIG_IPV6)
1477 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1478 attrtype = IFLA_GENEVE_REMOTE6;
1482 info->mode = IP_TUNNEL_INFO_IPV6;
1483 info->key.u.ipv6.dst =
1484 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1486 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1487 IPV6_ADDR_LINKLOCAL) {
1488 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1489 "Remote IPv6 address cannot be link-local");
1492 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1493 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1494 "Remote IPv6 address cannot be Multicast");
1497 info->key.tun_flags |= TUNNEL_CSUM;
1498 cfg->use_udp6_rx_checksums = true;
1500 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1501 "IPv6 support not enabled in the kernel");
1502 return -EPFNOSUPPORT;
1506 if (data[IFLA_GENEVE_ID]) {
1511 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1512 tvni[0] = (vni & 0x00ff0000) >> 16;
1513 tvni[1] = (vni & 0x0000ff00) >> 8;
1514 tvni[2] = vni & 0x000000ff;
1516 tunid = vni_to_tunnel_id(tvni);
1517 if (changelink && (tunid != info->key.tun_id)) {
1518 attrtype = IFLA_GENEVE_ID;
1521 info->key.tun_id = tunid;
1524 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1525 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1526 cfg->ttl_inherit = true;
1528 cfg->ttl_inherit = false;
1529 } else if (data[IFLA_GENEVE_TTL]) {
1530 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1531 cfg->ttl_inherit = false;
1534 if (data[IFLA_GENEVE_TOS])
1535 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1537 if (data[IFLA_GENEVE_DF])
1538 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1540 if (data[IFLA_GENEVE_LABEL]) {
1541 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1542 IPV6_FLOWLABEL_MASK;
1543 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1544 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1545 "Label attribute only applies for IPv6 Geneve devices");
1550 if (data[IFLA_GENEVE_PORT]) {
1552 attrtype = IFLA_GENEVE_PORT;
1555 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1558 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1560 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1563 cfg->collect_md = true;
1566 if (data[IFLA_GENEVE_UDP_CSUM]) {
1568 attrtype = IFLA_GENEVE_UDP_CSUM;
1571 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1572 info->key.tun_flags |= TUNNEL_CSUM;
1575 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1576 #if IS_ENABLED(CONFIG_IPV6)
1578 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1581 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1582 info->key.tun_flags &= ~TUNNEL_CSUM;
1584 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1585 "IPv6 support not enabled in the kernel");
1586 return -EPFNOSUPPORT;
1590 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1591 #if IS_ENABLED(CONFIG_IPV6)
1593 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1596 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1597 cfg->use_udp6_rx_checksums = false;
1599 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1600 "IPv6 support not enabled in the kernel");
1601 return -EPFNOSUPPORT;
1605 if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1607 attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1610 cfg->inner_proto_inherit = true;
1615 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1616 "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1620 static void geneve_link_config(struct net_device *dev,
1621 struct ip_tunnel_info *info, struct nlattr *tb[])
1623 struct geneve_dev *geneve = netdev_priv(dev);
1627 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1631 switch (ip_tunnel_info_af(info)) {
1633 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1634 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1636 if (!IS_ERR(rt) && rt->dst.dev) {
1637 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1642 #if IS_ENABLED(CONFIG_IPV6)
1644 struct rt6_info *rt;
1646 if (!__in6_dev_get(dev))
1649 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1652 if (rt && rt->dst.dev)
1653 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1663 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1666 static int geneve_newlink(struct net *net, struct net_device *dev,
1667 struct nlattr *tb[], struct nlattr *data[],
1668 struct netlink_ext_ack *extack)
1670 struct geneve_config cfg = {
1671 .df = GENEVE_DF_UNSET,
1672 .use_udp6_rx_checksums = false,
1673 .ttl_inherit = false,
1674 .collect_md = false,
1678 init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1679 err = geneve_nl2info(tb, data, extack, &cfg, false);
1683 err = geneve_configure(net, dev, extack, &cfg);
1687 geneve_link_config(dev, &cfg.info, tb);
1692 /* Quiesces the geneve device data path for both TX and RX.
1694 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1695 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1696 * to complete for the existing set of in-flight packets to be transmitted,
1697 * then we would have quiesced the transmit data path. All the future packets
1698 * will get dropped until we unquiesce the data path.
1700 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1701 * if we set that to NULL under RCU and wait for synchronize_net() to
1702 * complete, then we would have quiesced the receive data path.
1704 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1705 struct geneve_sock **gs6)
1707 *gs4 = rtnl_dereference(geneve->sock4);
1708 rcu_assign_pointer(geneve->sock4, NULL);
1710 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1711 #if IS_ENABLED(CONFIG_IPV6)
1712 *gs6 = rtnl_dereference(geneve->sock6);
1713 rcu_assign_pointer(geneve->sock6, NULL);
1715 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1722 /* Resumes the geneve device data path for both TX and RX. */
1723 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1724 struct geneve_sock __maybe_unused *gs6)
1726 rcu_assign_pointer(geneve->sock4, gs4);
1728 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1729 #if IS_ENABLED(CONFIG_IPV6)
1730 rcu_assign_pointer(geneve->sock6, gs6);
1732 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1737 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1738 struct nlattr *data[],
1739 struct netlink_ext_ack *extack)
1741 struct geneve_dev *geneve = netdev_priv(dev);
1742 struct geneve_sock *gs4, *gs6;
1743 struct geneve_config cfg;
1746 /* If the geneve device is configured for metadata (or externally
1747 * controlled, for example, OVS), then nothing can be changed.
1749 if (geneve->cfg.collect_md)
1752 /* Start with the existing info. */
1753 memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1754 err = geneve_nl2info(tb, data, extack, &cfg, true);
1758 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1759 dst_cache_reset(&cfg.info.dst_cache);
1760 geneve_link_config(dev, &cfg.info, tb);
1763 geneve_quiesce(geneve, &gs4, &gs6);
1764 memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1765 geneve_unquiesce(geneve, gs4, gs6);
1770 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1772 struct geneve_dev *geneve = netdev_priv(dev);
1774 list_del(&geneve->next);
1775 unregister_netdevice_queue(dev, head);
1778 static size_t geneve_get_size(const struct net_device *dev)
1780 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1781 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1782 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1783 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1784 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1785 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1786 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1787 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1788 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1789 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1790 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1791 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1792 nla_total_size(0) + /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1796 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1798 struct geneve_dev *geneve = netdev_priv(dev);
1799 struct ip_tunnel_info *info = &geneve->cfg.info;
1800 bool ttl_inherit = geneve->cfg.ttl_inherit;
1801 bool metadata = geneve->cfg.collect_md;
1805 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1806 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1807 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1808 goto nla_put_failure;
1810 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1811 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1812 info->key.u.ipv4.dst))
1813 goto nla_put_failure;
1814 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1815 !!(info->key.tun_flags & TUNNEL_CSUM)))
1816 goto nla_put_failure;
1818 #if IS_ENABLED(CONFIG_IPV6)
1819 } else if (!metadata) {
1820 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1821 &info->key.u.ipv6.dst))
1822 goto nla_put_failure;
1823 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1824 !(info->key.tun_flags & TUNNEL_CSUM)))
1825 goto nla_put_failure;
1829 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1830 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1831 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1832 goto nla_put_failure;
1834 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1835 goto nla_put_failure;
1837 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1838 goto nla_put_failure;
1840 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1841 goto nla_put_failure;
1843 #if IS_ENABLED(CONFIG_IPV6)
1844 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1845 !geneve->cfg.use_udp6_rx_checksums))
1846 goto nla_put_failure;
1849 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1850 goto nla_put_failure;
1852 if (geneve->cfg.inner_proto_inherit &&
1853 nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1854 goto nla_put_failure;
1862 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1864 .maxtype = IFLA_GENEVE_MAX,
1865 .policy = geneve_policy,
1866 .priv_size = sizeof(struct geneve_dev),
1867 .setup = geneve_setup,
1868 .validate = geneve_validate,
1869 .newlink = geneve_newlink,
1870 .changelink = geneve_changelink,
1871 .dellink = geneve_dellink,
1872 .get_size = geneve_get_size,
1873 .fill_info = geneve_fill_info,
1876 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1877 u8 name_assign_type, u16 dst_port)
1879 struct nlattr *tb[IFLA_MAX + 1];
1880 struct net_device *dev;
1881 LIST_HEAD(list_kill);
1883 struct geneve_config cfg = {
1884 .df = GENEVE_DF_UNSET,
1885 .use_udp6_rx_checksums = true,
1886 .ttl_inherit = false,
1890 memset(tb, 0, sizeof(tb));
1891 dev = rtnl_create_link(net, name, name_assign_type,
1892 &geneve_link_ops, tb, NULL);
1896 init_tnl_info(&cfg.info, dst_port);
1897 err = geneve_configure(net, dev, NULL, &cfg);
1900 return ERR_PTR(err);
1903 /* openvswitch users expect packet sizes to be unrestricted,
1904 * so set the largest MTU we can.
1906 err = geneve_change_mtu(dev, IP_MAX_MTU);
1910 err = rtnl_configure_link(dev, NULL);
1916 geneve_dellink(dev, &list_kill);
1917 unregister_netdevice_many(&list_kill);
1918 return ERR_PTR(err);
1920 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1922 static int geneve_netdevice_event(struct notifier_block *unused,
1923 unsigned long event, void *ptr)
1925 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1927 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1928 geneve_offload_rx_ports(dev, true);
1929 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1930 geneve_offload_rx_ports(dev, false);
1935 static struct notifier_block geneve_notifier_block __read_mostly = {
1936 .notifier_call = geneve_netdevice_event,
1939 static __net_init int geneve_init_net(struct net *net)
1941 struct geneve_net *gn = net_generic(net, geneve_net_id);
1943 INIT_LIST_HEAD(&gn->geneve_list);
1944 INIT_LIST_HEAD(&gn->sock_list);
1948 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1950 struct geneve_net *gn = net_generic(net, geneve_net_id);
1951 struct geneve_dev *geneve, *next;
1952 struct net_device *dev, *aux;
1954 /* gather any geneve devices that were moved into this ns */
1955 for_each_netdev_safe(net, dev, aux)
1956 if (dev->rtnl_link_ops == &geneve_link_ops)
1957 unregister_netdevice_queue(dev, head);
1959 /* now gather any other geneve devices that were created in this ns */
1960 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1961 /* If geneve->dev is in the same netns, it was already added
1962 * to the list by the previous loop.
1964 if (!net_eq(dev_net(geneve->dev), net))
1965 unregister_netdevice_queue(geneve->dev, head);
1969 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1975 list_for_each_entry(net, net_list, exit_list)
1976 geneve_destroy_tunnels(net, &list);
1978 /* unregister the devices gathered above */
1979 unregister_netdevice_many(&list);
1982 list_for_each_entry(net, net_list, exit_list) {
1983 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1985 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1989 static struct pernet_operations geneve_net_ops = {
1990 .init = geneve_init_net,
1991 .exit_batch = geneve_exit_batch_net,
1992 .id = &geneve_net_id,
1993 .size = sizeof(struct geneve_net),
1996 static int __init geneve_init_module(void)
2000 rc = register_pernet_subsys(&geneve_net_ops);
2004 rc = register_netdevice_notifier(&geneve_notifier_block);
2008 rc = rtnl_link_register(&geneve_link_ops);
2014 unregister_netdevice_notifier(&geneve_notifier_block);
2016 unregister_pernet_subsys(&geneve_net_ops);
2020 late_initcall(geneve_init_module);
2022 static void __exit geneve_cleanup_module(void)
2024 rtnl_link_unregister(&geneve_link_ops);
2025 unregister_netdevice_notifier(&geneve_notifier_block);
2026 unregister_pernet_subsys(&geneve_net_ops);
2028 module_exit(geneve_cleanup_module);
2030 MODULE_LICENSE("GPL");
2031 MODULE_VERSION(GENEVE_NETDEV_VER);
2032 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
2033 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
2034 MODULE_ALIAS_RTNL_LINK("geneve");