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/kernel.h>
11 #include <linux/module.h>
12 #include <linux/etherdevice.h>
13 #include <linux/hash.h>
14 #include <net/ipv6_stubs.h>
15 #include <net/dst_metadata.h>
16 #include <net/gro_cells.h>
17 #include <net/rtnetlink.h>
18 #include <net/geneve.h>
19 #include <net/protocol.h>
21 #define GENEVE_NETDEV_VER "0.6"
23 #define GENEVE_N_VID (1u << 24)
24 #define GENEVE_VID_MASK (GENEVE_N_VID - 1)
26 #define VNI_HASH_BITS 10
27 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
29 static bool log_ecn_error = true;
30 module_param(log_ecn_error, bool, 0644);
31 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
34 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
35 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
36 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
38 /* per-network namespace private data for this module */
40 struct list_head geneve_list;
41 struct list_head sock_list;
44 static unsigned int geneve_net_id;
46 struct geneve_dev_node {
47 struct hlist_node hlist;
48 struct geneve_dev *geneve;
51 struct geneve_config {
52 struct ip_tunnel_info info;
54 bool use_udp6_rx_checksums;
56 enum ifla_geneve_df df;
59 /* Pseudo network device */
61 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
62 #if IS_ENABLED(CONFIG_IPV6)
63 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
65 struct net *net; /* netns for packet i/o */
66 struct net_device *dev; /* netdev for geneve tunnel */
67 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
68 #if IS_ENABLED(CONFIG_IPV6)
69 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
71 struct list_head next; /* geneve's per namespace list */
72 struct gro_cells gro_cells;
73 struct geneve_config cfg;
78 struct list_head list;
82 struct hlist_head vni_list[VNI_HASH_SIZE];
85 static inline __u32 geneve_net_vni_hash(u8 vni[3])
89 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
90 return hash_32(vnid, VNI_HASH_BITS);
93 static __be64 vni_to_tunnel_id(const __u8 *vni)
96 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
98 return (__force __be64)(((__force u64)vni[0] << 40) |
99 ((__force u64)vni[1] << 48) |
100 ((__force u64)vni[2] << 56));
104 /* Convert 64 bit tunnel ID to 24 bit VNI. */
105 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
108 vni[0] = (__force __u8)(tun_id >> 16);
109 vni[1] = (__force __u8)(tun_id >> 8);
110 vni[2] = (__force __u8)tun_id;
112 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
113 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
114 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
118 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
120 return !memcmp(vni, &tun_id[5], 3);
123 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
125 return gs->sock->sk->sk_family;
128 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
129 __be32 addr, u8 vni[])
131 struct hlist_head *vni_list_head;
132 struct geneve_dev_node *node;
135 /* Find the device for this VNI */
136 hash = geneve_net_vni_hash(vni);
137 vni_list_head = &gs->vni_list[hash];
138 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
139 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
140 addr == node->geneve->cfg.info.key.u.ipv4.dst)
146 #if IS_ENABLED(CONFIG_IPV6)
147 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
148 struct in6_addr addr6, u8 vni[])
150 struct hlist_head *vni_list_head;
151 struct geneve_dev_node *node;
154 /* Find the device for this VNI */
155 hash = geneve_net_vni_hash(vni);
156 vni_list_head = &gs->vni_list[hash];
157 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
158 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
159 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
166 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
168 return (struct genevehdr *)(udp_hdr(skb) + 1);
171 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
174 static u8 zero_vni[3];
177 if (geneve_get_sk_family(gs) == AF_INET) {
181 iph = ip_hdr(skb); /* outer IP header... */
183 if (gs->collect_md) {
187 vni = geneve_hdr(skb)->vni;
191 return geneve_lookup(gs, addr, vni);
192 #if IS_ENABLED(CONFIG_IPV6)
193 } else if (geneve_get_sk_family(gs) == AF_INET6) {
194 static struct in6_addr zero_addr6;
195 struct ipv6hdr *ip6h;
196 struct in6_addr addr6;
198 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
200 if (gs->collect_md) {
204 vni = geneve_hdr(skb)->vni;
208 return geneve6_lookup(gs, addr6, vni);
214 /* geneve receive/decap routine */
215 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
218 struct genevehdr *gnvh = geneve_hdr(skb);
219 struct metadata_dst *tun_dst = NULL;
220 struct pcpu_sw_netstats *stats;
225 if (ip_tunnel_collect_metadata() || gs->collect_md) {
228 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT |
229 (gnvh->oam ? TUNNEL_OAM : 0) |
230 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
232 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
233 vni_to_tunnel_id(gnvh->vni),
236 geneve->dev->stats.rx_dropped++;
239 /* Update tunnel dst according to Geneve options. */
240 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
241 gnvh->options, gnvh->opt_len * 4,
244 /* Drop packets w/ critical options,
245 * since we don't support any...
247 if (gnvh->critical) {
248 geneve->dev->stats.rx_frame_errors++;
249 geneve->dev->stats.rx_errors++;
254 skb_reset_mac_header(skb);
255 skb->protocol = eth_type_trans(skb, geneve->dev);
256 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
259 skb_dst_set(skb, &tun_dst->dst);
261 /* Ignore packet loops (and multicast echo) */
262 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
263 geneve->dev->stats.rx_errors++;
267 oiph = skb_network_header(skb);
268 skb_reset_network_header(skb);
270 if (geneve_get_sk_family(gs) == AF_INET)
271 err = IP_ECN_decapsulate(oiph, skb);
272 #if IS_ENABLED(CONFIG_IPV6)
274 err = IP6_ECN_decapsulate(oiph, skb);
279 if (geneve_get_sk_family(gs) == AF_INET)
280 net_info_ratelimited("non-ECT from %pI4 "
282 &((struct iphdr *)oiph)->saddr,
283 ((struct iphdr *)oiph)->tos);
284 #if IS_ENABLED(CONFIG_IPV6)
286 net_info_ratelimited("non-ECT from %pI6\n",
287 &((struct ipv6hdr *)oiph)->saddr);
291 ++geneve->dev->stats.rx_frame_errors;
292 ++geneve->dev->stats.rx_errors;
298 err = gro_cells_receive(&geneve->gro_cells, skb);
299 if (likely(err == NET_RX_SUCCESS)) {
300 stats = this_cpu_ptr(geneve->dev->tstats);
301 u64_stats_update_begin(&stats->syncp);
303 stats->rx_bytes += len;
304 u64_stats_update_end(&stats->syncp);
308 /* Consume bad packet */
312 /* Setup stats when device is created */
313 static int geneve_init(struct net_device *dev)
315 struct geneve_dev *geneve = netdev_priv(dev);
318 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
322 err = gro_cells_init(&geneve->gro_cells, dev);
324 free_percpu(dev->tstats);
328 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
330 free_percpu(dev->tstats);
331 gro_cells_destroy(&geneve->gro_cells);
337 static void geneve_uninit(struct net_device *dev)
339 struct geneve_dev *geneve = netdev_priv(dev);
341 dst_cache_destroy(&geneve->cfg.info.dst_cache);
342 gro_cells_destroy(&geneve->gro_cells);
343 free_percpu(dev->tstats);
346 /* Callback from net/ipv4/udp.c to receive packets */
347 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
349 struct genevehdr *geneveh;
350 struct geneve_dev *geneve;
351 struct geneve_sock *gs;
354 /* Need UDP and Geneve header to be present */
355 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
358 /* Return packets with reserved bits set */
359 geneveh = geneve_hdr(skb);
360 if (unlikely(geneveh->ver != GENEVE_VER))
363 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
366 gs = rcu_dereference_sk_user_data(sk);
370 geneve = geneve_lookup_skb(gs, skb);
374 opts_len = geneveh->opt_len * 4;
375 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
377 !net_eq(geneve->net, dev_net(geneve->dev)))) {
378 geneve->dev->stats.rx_dropped++;
382 geneve_rx(geneve, gs, skb);
386 /* Consume bad packet */
391 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
392 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
394 struct genevehdr *geneveh;
395 struct geneve_sock *gs;
396 u8 zero_vni[3] = { 0 };
399 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
402 geneveh = geneve_hdr(skb);
403 if (geneveh->ver != GENEVE_VER)
406 if (geneveh->proto_type != htons(ETH_P_TEB))
409 gs = rcu_dereference_sk_user_data(sk);
413 if (geneve_get_sk_family(gs) == AF_INET) {
414 struct iphdr *iph = ip_hdr(skb);
417 if (!gs->collect_md) {
418 vni = geneve_hdr(skb)->vni;
422 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
425 #if IS_ENABLED(CONFIG_IPV6)
426 if (geneve_get_sk_family(gs) == AF_INET6) {
427 struct ipv6hdr *ip6h = ipv6_hdr(skb);
428 struct in6_addr addr6;
430 memset(&addr6, 0, sizeof(struct in6_addr));
432 if (!gs->collect_md) {
433 vni = geneve_hdr(skb)->vni;
437 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
441 return -EPFNOSUPPORT;
444 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
445 __be16 port, bool ipv6_rx_csum)
448 struct udp_port_cfg udp_conf;
451 memset(&udp_conf, 0, sizeof(udp_conf));
454 udp_conf.family = AF_INET6;
455 udp_conf.ipv6_v6only = 1;
456 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
458 udp_conf.family = AF_INET;
459 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
462 udp_conf.local_udp_port = port;
464 /* Open UDP socket */
465 err = udp_sock_create(net, &udp_conf, &sock);
472 static int geneve_hlen(struct genevehdr *gh)
474 return sizeof(*gh) + gh->opt_len * 4;
477 static struct sk_buff *geneve_gro_receive(struct sock *sk,
478 struct list_head *head,
481 struct sk_buff *pp = NULL;
483 struct genevehdr *gh, *gh2;
484 unsigned int hlen, gh_len, off_gnv;
485 const struct packet_offload *ptype;
489 off_gnv = skb_gro_offset(skb);
490 hlen = off_gnv + sizeof(*gh);
491 gh = skb_gro_header_fast(skb, off_gnv);
492 if (skb_gro_header_hard(skb, hlen)) {
493 gh = skb_gro_header_slow(skb, hlen, off_gnv);
498 if (gh->ver != GENEVE_VER || gh->oam)
500 gh_len = geneve_hlen(gh);
502 hlen = off_gnv + gh_len;
503 if (skb_gro_header_hard(skb, hlen)) {
504 gh = skb_gro_header_slow(skb, hlen, off_gnv);
509 list_for_each_entry(p, head, list) {
510 if (!NAPI_GRO_CB(p)->same_flow)
513 gh2 = (struct genevehdr *)(p->data + off_gnv);
514 if (gh->opt_len != gh2->opt_len ||
515 memcmp(gh, gh2, gh_len)) {
516 NAPI_GRO_CB(p)->same_flow = 0;
521 type = gh->proto_type;
524 ptype = gro_find_receive_by_type(type);
528 skb_gro_pull(skb, gh_len);
529 skb_gro_postpull_rcsum(skb, gh, gh_len);
530 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
536 skb_gro_flush_final(skb, pp, flush);
541 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
544 struct genevehdr *gh;
545 struct packet_offload *ptype;
550 gh = (struct genevehdr *)(skb->data + nhoff);
551 gh_len = geneve_hlen(gh);
552 type = gh->proto_type;
555 ptype = gro_find_complete_by_type(type);
557 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
561 skb_set_inner_mac_header(skb, nhoff + gh_len);
566 /* Create new listen socket if needed */
567 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
568 bool ipv6, bool ipv6_rx_csum)
570 struct geneve_net *gn = net_generic(net, geneve_net_id);
571 struct geneve_sock *gs;
573 struct udp_tunnel_sock_cfg tunnel_cfg;
576 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
578 return ERR_PTR(-ENOMEM);
580 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
583 return ERR_CAST(sock);
588 for (h = 0; h < VNI_HASH_SIZE; ++h)
589 INIT_HLIST_HEAD(&gs->vni_list[h]);
591 /* Initialize the geneve udp offloads structure */
592 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
594 /* Mark socket as an encapsulation socket */
595 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
596 tunnel_cfg.sk_user_data = gs;
597 tunnel_cfg.encap_type = 1;
598 tunnel_cfg.gro_receive = geneve_gro_receive;
599 tunnel_cfg.gro_complete = geneve_gro_complete;
600 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
601 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
602 tunnel_cfg.encap_destroy = NULL;
603 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
604 list_add(&gs->list, &gn->sock_list);
608 static void __geneve_sock_release(struct geneve_sock *gs)
610 if (!gs || --gs->refcnt)
614 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
615 udp_tunnel_sock_release(gs->sock);
619 static void geneve_sock_release(struct geneve_dev *geneve)
621 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
622 #if IS_ENABLED(CONFIG_IPV6)
623 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
625 rcu_assign_pointer(geneve->sock6, NULL);
628 rcu_assign_pointer(geneve->sock4, NULL);
631 __geneve_sock_release(gs4);
632 #if IS_ENABLED(CONFIG_IPV6)
633 __geneve_sock_release(gs6);
637 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
641 struct geneve_sock *gs;
643 list_for_each_entry(gs, &gn->sock_list, list) {
644 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
645 geneve_get_sk_family(gs) == family) {
652 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
654 struct net *net = geneve->net;
655 struct geneve_net *gn = net_generic(net, geneve_net_id);
656 struct geneve_dev_node *node;
657 struct geneve_sock *gs;
661 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
667 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
668 geneve->cfg.use_udp6_rx_checksums);
673 gs->collect_md = geneve->cfg.collect_md;
674 #if IS_ENABLED(CONFIG_IPV6)
676 rcu_assign_pointer(geneve->sock6, gs);
677 node = &geneve->hlist6;
681 rcu_assign_pointer(geneve->sock4, gs);
682 node = &geneve->hlist4;
684 node->geneve = geneve;
686 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
687 hash = geneve_net_vni_hash(vni);
688 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
692 static int geneve_open(struct net_device *dev)
694 struct geneve_dev *geneve = netdev_priv(dev);
695 bool metadata = geneve->cfg.collect_md;
699 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
700 ipv4 = !ipv6 || metadata;
701 #if IS_ENABLED(CONFIG_IPV6)
703 ret = geneve_sock_add(geneve, true);
704 if (ret < 0 && ret != -EAFNOSUPPORT)
709 ret = geneve_sock_add(geneve, false);
711 geneve_sock_release(geneve);
716 static int geneve_stop(struct net_device *dev)
718 struct geneve_dev *geneve = netdev_priv(dev);
720 hlist_del_init_rcu(&geneve->hlist4.hlist);
721 #if IS_ENABLED(CONFIG_IPV6)
722 hlist_del_init_rcu(&geneve->hlist6.hlist);
724 geneve_sock_release(geneve);
728 static void geneve_build_header(struct genevehdr *geneveh,
729 const struct ip_tunnel_info *info)
731 geneveh->ver = GENEVE_VER;
732 geneveh->opt_len = info->options_len / 4;
733 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
734 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
736 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
737 geneveh->proto_type = htons(ETH_P_TEB);
740 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
741 ip_tunnel_info_opts_get(geneveh->options, info);
744 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
745 const struct ip_tunnel_info *info,
746 bool xnet, int ip_hdr_len)
748 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
749 struct genevehdr *gnvh;
753 skb_reset_mac_header(skb);
754 skb_scrub_packet(skb, xnet);
756 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
757 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
758 err = skb_cow_head(skb, min_headroom);
762 err = udp_tunnel_handle_offloads(skb, udp_sum);
766 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
767 geneve_build_header(gnvh, info);
768 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
776 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
777 struct net_device *dev,
778 struct geneve_sock *gs4,
780 const struct ip_tunnel_info *info,
781 __be16 dport, __be16 sport)
783 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
784 struct geneve_dev *geneve = netdev_priv(dev);
785 struct dst_cache *dst_cache;
786 struct rtable *rt = NULL;
790 return ERR_PTR(-EIO);
792 memset(fl4, 0, sizeof(*fl4));
793 fl4->flowi4_mark = skb->mark;
794 fl4->flowi4_proto = IPPROTO_UDP;
795 fl4->daddr = info->key.u.ipv4.dst;
796 fl4->saddr = info->key.u.ipv4.src;
797 fl4->fl4_dport = dport;
798 fl4->fl4_sport = sport;
801 if ((tos == 1) && !geneve->cfg.collect_md) {
802 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
805 fl4->flowi4_tos = RT_TOS(tos);
807 dst_cache = (struct dst_cache *)&info->dst_cache;
809 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
813 rt = ip_route_output_key(geneve->net, fl4);
815 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
816 return ERR_PTR(-ENETUNREACH);
818 if (rt->dst.dev == dev) { /* is this necessary? */
819 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
821 return ERR_PTR(-ELOOP);
824 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
828 #if IS_ENABLED(CONFIG_IPV6)
829 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
830 struct net_device *dev,
831 struct geneve_sock *gs6,
833 const struct ip_tunnel_info *info,
834 __be16 dport, __be16 sport)
836 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
837 struct geneve_dev *geneve = netdev_priv(dev);
838 struct dst_entry *dst = NULL;
839 struct dst_cache *dst_cache;
843 return ERR_PTR(-EIO);
845 memset(fl6, 0, sizeof(*fl6));
846 fl6->flowi6_mark = skb->mark;
847 fl6->flowi6_proto = IPPROTO_UDP;
848 fl6->daddr = info->key.u.ipv6.dst;
849 fl6->saddr = info->key.u.ipv6.src;
850 fl6->fl6_dport = dport;
851 fl6->fl6_sport = sport;
853 prio = info->key.tos;
854 if ((prio == 1) && !geneve->cfg.collect_md) {
855 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
859 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
861 dst_cache = (struct dst_cache *)&info->dst_cache;
863 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
867 dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
870 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
871 return ERR_PTR(-ENETUNREACH);
873 if (dst->dev == dev) { /* is this necessary? */
874 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
876 return ERR_PTR(-ELOOP);
880 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
885 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
886 struct geneve_dev *geneve,
887 const struct ip_tunnel_info *info)
889 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
890 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
891 const struct ip_tunnel_key *key = &info->key;
899 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
900 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
901 geneve->cfg.info.key.tp_dst, sport);
905 err = skb_tunnel_check_pmtu(skb, &rt->dst,
906 GENEVE_IPV4_HLEN + info->options_len,
907 netif_is_any_bridge_port(dev));
909 dst_release(&rt->dst);
912 struct ip_tunnel_info *info;
914 info = skb_tunnel_info(skb);
916 info->key.u.ipv4.dst = fl4.saddr;
917 info->key.u.ipv4.src = fl4.daddr;
920 if (!pskb_may_pull(skb, ETH_HLEN)) {
921 dst_release(&rt->dst);
925 skb->protocol = eth_type_trans(skb, geneve->dev);
927 dst_release(&rt->dst);
931 if (geneve->cfg.collect_md) {
932 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
935 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
937 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
938 if (geneve->cfg.ttl_inherit)
939 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
942 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
944 if (geneve->cfg.df == GENEVE_DF_SET) {
946 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
947 struct ethhdr *eth = eth_hdr(skb);
949 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
951 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
952 struct iphdr *iph = ip_hdr(skb);
954 if (iph->frag_off & htons(IP_DF))
960 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
964 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
965 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
966 !net_eq(geneve->net, dev_net(geneve->dev)),
967 !(info->key.tun_flags & TUNNEL_CSUM));
971 #if IS_ENABLED(CONFIG_IPV6)
972 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
973 struct geneve_dev *geneve,
974 const struct ip_tunnel_info *info)
976 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
977 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
978 const struct ip_tunnel_key *key = &info->key;
979 struct dst_entry *dst = NULL;
985 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
986 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
987 geneve->cfg.info.key.tp_dst, sport);
991 err = skb_tunnel_check_pmtu(skb, dst,
992 GENEVE_IPV6_HLEN + info->options_len,
993 netif_is_any_bridge_port(dev));
998 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1001 info->key.u.ipv6.dst = fl6.saddr;
1002 info->key.u.ipv6.src = fl6.daddr;
1005 if (!pskb_may_pull(skb, ETH_HLEN)) {
1010 skb->protocol = eth_type_trans(skb, geneve->dev);
1016 if (geneve->cfg.collect_md) {
1017 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1020 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1022 if (geneve->cfg.ttl_inherit)
1023 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1026 ttl = ttl ? : ip6_dst_hoplimit(dst);
1028 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
1032 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1033 &fl6.saddr, &fl6.daddr, prio, ttl,
1034 info->key.label, sport, geneve->cfg.info.key.tp_dst,
1035 !(info->key.tun_flags & TUNNEL_CSUM));
1040 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1042 struct geneve_dev *geneve = netdev_priv(dev);
1043 struct ip_tunnel_info *info = NULL;
1046 if (geneve->cfg.collect_md) {
1047 info = skb_tunnel_info(skb);
1048 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1049 netdev_dbg(dev, "no tunnel metadata\n");
1051 dev->stats.tx_dropped++;
1052 return NETDEV_TX_OK;
1055 info = &geneve->cfg.info;
1059 #if IS_ENABLED(CONFIG_IPV6)
1060 if (info->mode & IP_TUNNEL_INFO_IPV6)
1061 err = geneve6_xmit_skb(skb, dev, geneve, info);
1064 err = geneve_xmit_skb(skb, dev, geneve, info);
1068 return NETDEV_TX_OK;
1070 if (err != -EMSGSIZE)
1074 dev->stats.collisions++;
1075 else if (err == -ENETUNREACH)
1076 dev->stats.tx_carrier_errors++;
1078 dev->stats.tx_errors++;
1079 return NETDEV_TX_OK;
1082 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1084 if (new_mtu > dev->max_mtu)
1085 new_mtu = dev->max_mtu;
1086 else if (new_mtu < dev->min_mtu)
1087 new_mtu = dev->min_mtu;
1093 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1095 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1096 struct geneve_dev *geneve = netdev_priv(dev);
1099 if (ip_tunnel_info_af(info) == AF_INET) {
1103 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1104 sport = udp_flow_src_port(geneve->net, skb,
1105 1, USHRT_MAX, true);
1107 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1108 geneve->cfg.info.key.tp_dst, sport);
1113 info->key.u.ipv4.src = fl4.saddr;
1114 #if IS_ENABLED(CONFIG_IPV6)
1115 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1116 struct dst_entry *dst;
1119 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1120 sport = udp_flow_src_port(geneve->net, skb,
1121 1, USHRT_MAX, true);
1123 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1124 geneve->cfg.info.key.tp_dst, sport);
1126 return PTR_ERR(dst);
1129 info->key.u.ipv6.src = fl6.saddr;
1135 info->key.tp_src = sport;
1136 info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1140 static const struct net_device_ops geneve_netdev_ops = {
1141 .ndo_init = geneve_init,
1142 .ndo_uninit = geneve_uninit,
1143 .ndo_open = geneve_open,
1144 .ndo_stop = geneve_stop,
1145 .ndo_start_xmit = geneve_xmit,
1146 .ndo_get_stats64 = ip_tunnel_get_stats64,
1147 .ndo_change_mtu = geneve_change_mtu,
1148 .ndo_validate_addr = eth_validate_addr,
1149 .ndo_set_mac_address = eth_mac_addr,
1150 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1153 static void geneve_get_drvinfo(struct net_device *dev,
1154 struct ethtool_drvinfo *drvinfo)
1156 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1157 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1160 static const struct ethtool_ops geneve_ethtool_ops = {
1161 .get_drvinfo = geneve_get_drvinfo,
1162 .get_link = ethtool_op_get_link,
1165 /* Info for udev, that this is a virtual tunnel endpoint */
1166 static struct device_type geneve_type = {
1170 /* Calls the ndo_udp_tunnel_add of the caller in order to
1171 * supply the listening GENEVE udp ports. Callers are expected
1172 * to implement the ndo_udp_tunnel_add.
1174 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1176 struct net *net = dev_net(dev);
1177 struct geneve_net *gn = net_generic(net, geneve_net_id);
1178 struct geneve_sock *gs;
1181 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1183 udp_tunnel_push_rx_port(dev, gs->sock,
1184 UDP_TUNNEL_TYPE_GENEVE);
1186 udp_tunnel_drop_rx_port(dev, gs->sock,
1187 UDP_TUNNEL_TYPE_GENEVE);
1193 /* Initialize the device structure. */
1194 static void geneve_setup(struct net_device *dev)
1198 dev->netdev_ops = &geneve_netdev_ops;
1199 dev->ethtool_ops = &geneve_ethtool_ops;
1200 dev->needs_free_netdev = true;
1202 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1204 dev->features |= NETIF_F_LLTX;
1205 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1206 dev->features |= NETIF_F_RXCSUM;
1207 dev->features |= NETIF_F_GSO_SOFTWARE;
1209 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1210 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1212 /* MTU range: 68 - (something less than 65535) */
1213 dev->min_mtu = ETH_MIN_MTU;
1214 /* The max_mtu calculation does not take account of GENEVE
1215 * options, to avoid excluding potentially valid
1216 * configurations. This will be further reduced by IPvX hdr size.
1218 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1220 netif_keep_dst(dev);
1221 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1222 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1223 eth_hw_addr_random(dev);
1226 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1227 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1228 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1229 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1230 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1231 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1232 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1233 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1234 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1235 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1236 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1237 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1238 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1239 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1242 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1243 struct netlink_ext_ack *extack)
1245 if (tb[IFLA_ADDRESS]) {
1246 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1247 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1248 "Provided link layer address is not Ethernet");
1252 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1253 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1254 "Provided Ethernet address is not unicast");
1255 return -EADDRNOTAVAIL;
1260 NL_SET_ERR_MSG(extack,
1261 "Not enough attributes provided to perform the operation");
1265 if (data[IFLA_GENEVE_ID]) {
1266 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1268 if (vni >= GENEVE_N_VID) {
1269 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1270 "Geneve ID must be lower than 16777216");
1275 if (data[IFLA_GENEVE_DF]) {
1276 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1278 if (df < 0 || df > GENEVE_DF_MAX) {
1279 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1280 "Invalid DF attribute");
1288 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1289 const struct ip_tunnel_info *info,
1290 bool *tun_on_same_port,
1291 bool *tun_collect_md)
1293 struct geneve_dev *geneve, *t = NULL;
1295 *tun_on_same_port = false;
1296 *tun_collect_md = false;
1297 list_for_each_entry(geneve, &gn->geneve_list, next) {
1298 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1299 *tun_collect_md = geneve->cfg.collect_md;
1300 *tun_on_same_port = true;
1302 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1303 info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1304 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1310 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1312 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1313 info->key.ttl || info->key.label || info->key.tp_src ||
1314 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1317 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1318 struct ip_tunnel_info *b)
1320 if (ip_tunnel_info_af(a) == AF_INET)
1321 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1323 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1326 static int geneve_configure(struct net *net, struct net_device *dev,
1327 struct netlink_ext_ack *extack,
1328 const struct geneve_config *cfg)
1330 struct geneve_net *gn = net_generic(net, geneve_net_id);
1331 struct geneve_dev *t, *geneve = netdev_priv(dev);
1332 const struct ip_tunnel_info *info = &cfg->info;
1333 bool tun_collect_md, tun_on_same_port;
1336 if (cfg->collect_md && !is_tnl_info_zero(info)) {
1337 NL_SET_ERR_MSG(extack,
1338 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1345 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1349 /* make enough headroom for basic scenario */
1350 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1351 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1352 encap_len += sizeof(struct iphdr);
1353 dev->max_mtu -= sizeof(struct iphdr);
1355 encap_len += sizeof(struct ipv6hdr);
1356 dev->max_mtu -= sizeof(struct ipv6hdr);
1358 dev->needed_headroom = encap_len + ETH_HLEN;
1360 if (cfg->collect_md) {
1361 if (tun_on_same_port) {
1362 NL_SET_ERR_MSG(extack,
1363 "There can be only one externally controlled device on a destination port");
1367 if (tun_collect_md) {
1368 NL_SET_ERR_MSG(extack,
1369 "There already exists an externally controlled device on this destination port");
1374 dst_cache_reset(&geneve->cfg.info.dst_cache);
1375 memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1377 err = register_netdevice(dev);
1381 list_add(&geneve->next, &gn->geneve_list);
1385 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1387 memset(info, 0, sizeof(*info));
1388 info->key.tp_dst = htons(dst_port);
1391 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1392 struct netlink_ext_ack *extack,
1393 struct geneve_config *cfg, bool changelink)
1395 struct ip_tunnel_info *info = &cfg->info;
1398 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1399 NL_SET_ERR_MSG(extack,
1400 "Cannot specify both IPv4 and IPv6 Remote addresses");
1404 if (data[IFLA_GENEVE_REMOTE]) {
1405 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1406 attrtype = IFLA_GENEVE_REMOTE;
1410 info->key.u.ipv4.dst =
1411 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1413 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1414 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1415 "Remote IPv4 address cannot be Multicast");
1420 if (data[IFLA_GENEVE_REMOTE6]) {
1421 #if IS_ENABLED(CONFIG_IPV6)
1422 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1423 attrtype = IFLA_GENEVE_REMOTE6;
1427 info->mode = IP_TUNNEL_INFO_IPV6;
1428 info->key.u.ipv6.dst =
1429 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1431 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1432 IPV6_ADDR_LINKLOCAL) {
1433 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1434 "Remote IPv6 address cannot be link-local");
1437 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1438 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1439 "Remote IPv6 address cannot be Multicast");
1442 info->key.tun_flags |= TUNNEL_CSUM;
1443 cfg->use_udp6_rx_checksums = true;
1445 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1446 "IPv6 support not enabled in the kernel");
1447 return -EPFNOSUPPORT;
1451 if (data[IFLA_GENEVE_ID]) {
1456 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1457 tvni[0] = (vni & 0x00ff0000) >> 16;
1458 tvni[1] = (vni & 0x0000ff00) >> 8;
1459 tvni[2] = vni & 0x000000ff;
1461 tunid = vni_to_tunnel_id(tvni);
1462 if (changelink && (tunid != info->key.tun_id)) {
1463 attrtype = IFLA_GENEVE_ID;
1466 info->key.tun_id = tunid;
1469 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1470 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1471 cfg->ttl_inherit = true;
1473 cfg->ttl_inherit = false;
1474 } else if (data[IFLA_GENEVE_TTL]) {
1475 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1476 cfg->ttl_inherit = false;
1479 if (data[IFLA_GENEVE_TOS])
1480 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1482 if (data[IFLA_GENEVE_DF])
1483 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1485 if (data[IFLA_GENEVE_LABEL]) {
1486 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1487 IPV6_FLOWLABEL_MASK;
1488 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1489 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1490 "Label attribute only applies for IPv6 Geneve devices");
1495 if (data[IFLA_GENEVE_PORT]) {
1497 attrtype = IFLA_GENEVE_PORT;
1500 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1503 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1505 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1508 cfg->collect_md = true;
1511 if (data[IFLA_GENEVE_UDP_CSUM]) {
1513 attrtype = IFLA_GENEVE_UDP_CSUM;
1516 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1517 info->key.tun_flags |= TUNNEL_CSUM;
1520 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1521 #if IS_ENABLED(CONFIG_IPV6)
1523 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1526 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1527 info->key.tun_flags &= ~TUNNEL_CSUM;
1529 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1530 "IPv6 support not enabled in the kernel");
1531 return -EPFNOSUPPORT;
1535 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1536 #if IS_ENABLED(CONFIG_IPV6)
1538 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1541 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1542 cfg->use_udp6_rx_checksums = false;
1544 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1545 "IPv6 support not enabled in the kernel");
1546 return -EPFNOSUPPORT;
1552 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1553 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1557 static void geneve_link_config(struct net_device *dev,
1558 struct ip_tunnel_info *info, struct nlattr *tb[])
1560 struct geneve_dev *geneve = netdev_priv(dev);
1564 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1568 switch (ip_tunnel_info_af(info)) {
1570 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1571 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1573 if (!IS_ERR(rt) && rt->dst.dev) {
1574 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1579 #if IS_ENABLED(CONFIG_IPV6)
1581 struct rt6_info *rt;
1583 if (!__in6_dev_get(dev))
1586 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1589 if (rt && rt->dst.dev)
1590 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1600 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1603 static int geneve_newlink(struct net *net, struct net_device *dev,
1604 struct nlattr *tb[], struct nlattr *data[],
1605 struct netlink_ext_ack *extack)
1607 struct geneve_config cfg = {
1608 .df = GENEVE_DF_UNSET,
1609 .use_udp6_rx_checksums = false,
1610 .ttl_inherit = false,
1611 .collect_md = false,
1615 init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1616 err = geneve_nl2info(tb, data, extack, &cfg, false);
1620 err = geneve_configure(net, dev, extack, &cfg);
1624 geneve_link_config(dev, &cfg.info, tb);
1629 /* Quiesces the geneve device data path for both TX and RX.
1631 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1632 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1633 * to complete for the existing set of in-flight packets to be transmitted,
1634 * then we would have quiesced the transmit data path. All the future packets
1635 * will get dropped until we unquiesce the data path.
1637 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1638 * if we set that to NULL under RCU and wait for synchronize_net() to
1639 * complete, then we would have quiesced the receive data path.
1641 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1642 struct geneve_sock **gs6)
1644 *gs4 = rtnl_dereference(geneve->sock4);
1645 rcu_assign_pointer(geneve->sock4, NULL);
1647 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1648 #if IS_ENABLED(CONFIG_IPV6)
1649 *gs6 = rtnl_dereference(geneve->sock6);
1650 rcu_assign_pointer(geneve->sock6, NULL);
1652 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1659 /* Resumes the geneve device data path for both TX and RX. */
1660 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1661 struct geneve_sock __maybe_unused *gs6)
1663 rcu_assign_pointer(geneve->sock4, gs4);
1665 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1666 #if IS_ENABLED(CONFIG_IPV6)
1667 rcu_assign_pointer(geneve->sock6, gs6);
1669 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1674 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1675 struct nlattr *data[],
1676 struct netlink_ext_ack *extack)
1678 struct geneve_dev *geneve = netdev_priv(dev);
1679 struct geneve_sock *gs4, *gs6;
1680 struct geneve_config cfg;
1683 /* If the geneve device is configured for metadata (or externally
1684 * controlled, for example, OVS), then nothing can be changed.
1686 if (geneve->cfg.collect_md)
1689 /* Start with the existing info. */
1690 memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1691 err = geneve_nl2info(tb, data, extack, &cfg, true);
1695 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1696 dst_cache_reset(&cfg.info.dst_cache);
1697 geneve_link_config(dev, &cfg.info, tb);
1700 geneve_quiesce(geneve, &gs4, &gs6);
1701 memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1702 geneve_unquiesce(geneve, gs4, gs6);
1707 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1709 struct geneve_dev *geneve = netdev_priv(dev);
1711 list_del(&geneve->next);
1712 unregister_netdevice_queue(dev, head);
1715 static size_t geneve_get_size(const struct net_device *dev)
1717 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1718 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1719 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1720 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1721 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1722 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1723 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1724 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1725 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1726 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1727 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1728 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1732 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1734 struct geneve_dev *geneve = netdev_priv(dev);
1735 struct ip_tunnel_info *info = &geneve->cfg.info;
1736 bool ttl_inherit = geneve->cfg.ttl_inherit;
1737 bool metadata = geneve->cfg.collect_md;
1741 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1742 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1743 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1744 goto nla_put_failure;
1746 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1747 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1748 info->key.u.ipv4.dst))
1749 goto nla_put_failure;
1750 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1751 !!(info->key.tun_flags & TUNNEL_CSUM)))
1752 goto nla_put_failure;
1754 #if IS_ENABLED(CONFIG_IPV6)
1755 } else if (!metadata) {
1756 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1757 &info->key.u.ipv6.dst))
1758 goto nla_put_failure;
1759 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1760 !(info->key.tun_flags & TUNNEL_CSUM)))
1761 goto nla_put_failure;
1765 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1766 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1767 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1768 goto nla_put_failure;
1770 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1771 goto nla_put_failure;
1773 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1774 goto nla_put_failure;
1776 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1777 goto nla_put_failure;
1779 #if IS_ENABLED(CONFIG_IPV6)
1780 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1781 !geneve->cfg.use_udp6_rx_checksums))
1782 goto nla_put_failure;
1785 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1786 goto nla_put_failure;
1794 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1796 .maxtype = IFLA_GENEVE_MAX,
1797 .policy = geneve_policy,
1798 .priv_size = sizeof(struct geneve_dev),
1799 .setup = geneve_setup,
1800 .validate = geneve_validate,
1801 .newlink = geneve_newlink,
1802 .changelink = geneve_changelink,
1803 .dellink = geneve_dellink,
1804 .get_size = geneve_get_size,
1805 .fill_info = geneve_fill_info,
1808 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1809 u8 name_assign_type, u16 dst_port)
1811 struct nlattr *tb[IFLA_MAX + 1];
1812 struct net_device *dev;
1813 LIST_HEAD(list_kill);
1815 struct geneve_config cfg = {
1816 .df = GENEVE_DF_UNSET,
1817 .use_udp6_rx_checksums = true,
1818 .ttl_inherit = false,
1822 memset(tb, 0, sizeof(tb));
1823 dev = rtnl_create_link(net, name, name_assign_type,
1824 &geneve_link_ops, tb, NULL);
1828 init_tnl_info(&cfg.info, dst_port);
1829 err = geneve_configure(net, dev, NULL, &cfg);
1832 return ERR_PTR(err);
1835 /* openvswitch users expect packet sizes to be unrestricted,
1836 * so set the largest MTU we can.
1838 err = geneve_change_mtu(dev, IP_MAX_MTU);
1842 err = rtnl_configure_link(dev, NULL);
1848 geneve_dellink(dev, &list_kill);
1849 unregister_netdevice_many(&list_kill);
1850 return ERR_PTR(err);
1852 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1854 static int geneve_netdevice_event(struct notifier_block *unused,
1855 unsigned long event, void *ptr)
1857 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1859 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1860 event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1861 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1862 } else if (event == NETDEV_UNREGISTER) {
1863 if (!dev->udp_tunnel_nic_info)
1864 geneve_offload_rx_ports(dev, false);
1865 } else if (event == NETDEV_REGISTER) {
1866 if (!dev->udp_tunnel_nic_info)
1867 geneve_offload_rx_ports(dev, true);
1873 static struct notifier_block geneve_notifier_block __read_mostly = {
1874 .notifier_call = geneve_netdevice_event,
1877 static __net_init int geneve_init_net(struct net *net)
1879 struct geneve_net *gn = net_generic(net, geneve_net_id);
1881 INIT_LIST_HEAD(&gn->geneve_list);
1882 INIT_LIST_HEAD(&gn->sock_list);
1886 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1888 struct geneve_net *gn = net_generic(net, geneve_net_id);
1889 struct geneve_dev *geneve, *next;
1890 struct net_device *dev, *aux;
1892 /* gather any geneve devices that were moved into this ns */
1893 for_each_netdev_safe(net, dev, aux)
1894 if (dev->rtnl_link_ops == &geneve_link_ops)
1895 unregister_netdevice_queue(dev, head);
1897 /* now gather any other geneve devices that were created in this ns */
1898 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1899 /* If geneve->dev is in the same netns, it was already added
1900 * to the list by the previous loop.
1902 if (!net_eq(dev_net(geneve->dev), net))
1903 unregister_netdevice_queue(geneve->dev, head);
1907 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1913 list_for_each_entry(net, net_list, exit_list)
1914 geneve_destroy_tunnels(net, &list);
1916 /* unregister the devices gathered above */
1917 unregister_netdevice_many(&list);
1920 list_for_each_entry(net, net_list, exit_list) {
1921 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1923 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1927 static struct pernet_operations geneve_net_ops = {
1928 .init = geneve_init_net,
1929 .exit_batch = geneve_exit_batch_net,
1930 .id = &geneve_net_id,
1931 .size = sizeof(struct geneve_net),
1934 static int __init geneve_init_module(void)
1938 rc = register_pernet_subsys(&geneve_net_ops);
1942 rc = register_netdevice_notifier(&geneve_notifier_block);
1946 rc = rtnl_link_register(&geneve_link_ops);
1952 unregister_netdevice_notifier(&geneve_notifier_block);
1954 unregister_pernet_subsys(&geneve_net_ops);
1958 late_initcall(geneve_init_module);
1960 static void __exit geneve_cleanup_module(void)
1962 rtnl_link_unregister(&geneve_link_ops);
1963 unregister_netdevice_notifier(&geneve_notifier_block);
1964 unregister_pernet_subsys(&geneve_net_ops);
1966 module_exit(geneve_cleanup_module);
1968 MODULE_LICENSE("GPL");
1969 MODULE_VERSION(GENEVE_NETDEV_VER);
1970 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1971 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1972 MODULE_ALIAS_RTNL_LINK("geneve");