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)
782 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
783 struct geneve_dev *geneve = netdev_priv(dev);
784 struct dst_cache *dst_cache;
785 struct rtable *rt = NULL;
789 return ERR_PTR(-EIO);
791 memset(fl4, 0, sizeof(*fl4));
792 fl4->flowi4_mark = skb->mark;
793 fl4->flowi4_proto = IPPROTO_UDP;
794 fl4->daddr = info->key.u.ipv4.dst;
795 fl4->saddr = info->key.u.ipv4.src;
798 if ((tos == 1) && !geneve->cfg.collect_md) {
799 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
802 fl4->flowi4_tos = RT_TOS(tos);
804 dst_cache = (struct dst_cache *)&info->dst_cache;
806 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
810 rt = ip_route_output_key(geneve->net, fl4);
812 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
813 return ERR_PTR(-ENETUNREACH);
815 if (rt->dst.dev == dev) { /* is this necessary? */
816 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
818 return ERR_PTR(-ELOOP);
821 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
825 #if IS_ENABLED(CONFIG_IPV6)
826 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
827 struct net_device *dev,
828 struct geneve_sock *gs6,
830 const struct ip_tunnel_info *info)
832 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
833 struct geneve_dev *geneve = netdev_priv(dev);
834 struct dst_entry *dst = NULL;
835 struct dst_cache *dst_cache;
839 return ERR_PTR(-EIO);
841 memset(fl6, 0, sizeof(*fl6));
842 fl6->flowi6_mark = skb->mark;
843 fl6->flowi6_proto = IPPROTO_UDP;
844 fl6->daddr = info->key.u.ipv6.dst;
845 fl6->saddr = info->key.u.ipv6.src;
846 prio = info->key.tos;
847 if ((prio == 1) && !geneve->cfg.collect_md) {
848 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
852 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
854 dst_cache = (struct dst_cache *)&info->dst_cache;
856 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
860 dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
863 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
864 return ERR_PTR(-ENETUNREACH);
866 if (dst->dev == dev) { /* is this necessary? */
867 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
869 return ERR_PTR(-ELOOP);
873 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
878 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
879 struct geneve_dev *geneve,
880 const struct ip_tunnel_info *info)
882 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
883 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
884 const struct ip_tunnel_key *key = &info->key;
892 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
896 skb_tunnel_check_pmtu(skb, &rt->dst,
897 GENEVE_IPV4_HLEN + info->options_len);
899 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
900 if (geneve->cfg.collect_md) {
901 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
904 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
906 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
907 if (geneve->cfg.ttl_inherit)
908 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
911 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
913 if (geneve->cfg.df == GENEVE_DF_SET) {
915 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
916 struct ethhdr *eth = eth_hdr(skb);
918 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
920 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
921 struct iphdr *iph = ip_hdr(skb);
923 if (iph->frag_off & htons(IP_DF))
929 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
933 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
934 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
935 !net_eq(geneve->net, dev_net(geneve->dev)),
936 !(info->key.tun_flags & TUNNEL_CSUM));
940 #if IS_ENABLED(CONFIG_IPV6)
941 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
942 struct geneve_dev *geneve,
943 const struct ip_tunnel_info *info)
945 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
946 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
947 const struct ip_tunnel_key *key = &info->key;
948 struct dst_entry *dst = NULL;
954 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
958 skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len);
960 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
961 if (geneve->cfg.collect_md) {
962 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
965 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
967 if (geneve->cfg.ttl_inherit)
968 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
971 ttl = ttl ? : ip6_dst_hoplimit(dst);
973 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
977 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
978 &fl6.saddr, &fl6.daddr, prio, ttl,
979 info->key.label, sport, geneve->cfg.info.key.tp_dst,
980 !(info->key.tun_flags & TUNNEL_CSUM));
985 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
987 struct geneve_dev *geneve = netdev_priv(dev);
988 struct ip_tunnel_info *info = NULL;
991 if (geneve->cfg.collect_md) {
992 info = skb_tunnel_info(skb);
993 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
994 netdev_dbg(dev, "no tunnel metadata\n");
996 dev->stats.tx_dropped++;
1000 info = &geneve->cfg.info;
1004 #if IS_ENABLED(CONFIG_IPV6)
1005 if (info->mode & IP_TUNNEL_INFO_IPV6)
1006 err = geneve6_xmit_skb(skb, dev, geneve, info);
1009 err = geneve_xmit_skb(skb, dev, geneve, info);
1013 return NETDEV_TX_OK;
1018 dev->stats.collisions++;
1019 else if (err == -ENETUNREACH)
1020 dev->stats.tx_carrier_errors++;
1022 dev->stats.tx_errors++;
1023 return NETDEV_TX_OK;
1026 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1028 if (new_mtu > dev->max_mtu)
1029 new_mtu = dev->max_mtu;
1030 else if (new_mtu < dev->min_mtu)
1031 new_mtu = dev->min_mtu;
1037 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1039 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1040 struct geneve_dev *geneve = netdev_priv(dev);
1042 if (ip_tunnel_info_af(info) == AF_INET) {
1045 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1047 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info);
1052 info->key.u.ipv4.src = fl4.saddr;
1053 #if IS_ENABLED(CONFIG_IPV6)
1054 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1055 struct dst_entry *dst;
1057 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1059 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info);
1061 return PTR_ERR(dst);
1064 info->key.u.ipv6.src = fl6.saddr;
1070 info->key.tp_src = udp_flow_src_port(geneve->net, skb,
1071 1, USHRT_MAX, true);
1072 info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1076 static const struct net_device_ops geneve_netdev_ops = {
1077 .ndo_init = geneve_init,
1078 .ndo_uninit = geneve_uninit,
1079 .ndo_open = geneve_open,
1080 .ndo_stop = geneve_stop,
1081 .ndo_start_xmit = geneve_xmit,
1082 .ndo_get_stats64 = ip_tunnel_get_stats64,
1083 .ndo_change_mtu = geneve_change_mtu,
1084 .ndo_validate_addr = eth_validate_addr,
1085 .ndo_set_mac_address = eth_mac_addr,
1086 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1089 static void geneve_get_drvinfo(struct net_device *dev,
1090 struct ethtool_drvinfo *drvinfo)
1092 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1093 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1096 static const struct ethtool_ops geneve_ethtool_ops = {
1097 .get_drvinfo = geneve_get_drvinfo,
1098 .get_link = ethtool_op_get_link,
1101 /* Info for udev, that this is a virtual tunnel endpoint */
1102 static struct device_type geneve_type = {
1106 /* Calls the ndo_udp_tunnel_add of the caller in order to
1107 * supply the listening GENEVE udp ports. Callers are expected
1108 * to implement the ndo_udp_tunnel_add.
1110 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1112 struct net *net = dev_net(dev);
1113 struct geneve_net *gn = net_generic(net, geneve_net_id);
1114 struct geneve_sock *gs;
1117 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1119 udp_tunnel_push_rx_port(dev, gs->sock,
1120 UDP_TUNNEL_TYPE_GENEVE);
1122 udp_tunnel_drop_rx_port(dev, gs->sock,
1123 UDP_TUNNEL_TYPE_GENEVE);
1129 /* Initialize the device structure. */
1130 static void geneve_setup(struct net_device *dev)
1134 dev->netdev_ops = &geneve_netdev_ops;
1135 dev->ethtool_ops = &geneve_ethtool_ops;
1136 dev->needs_free_netdev = true;
1138 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1140 dev->features |= NETIF_F_LLTX;
1141 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1142 dev->features |= NETIF_F_RXCSUM;
1143 dev->features |= NETIF_F_GSO_SOFTWARE;
1145 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1146 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1148 /* MTU range: 68 - (something less than 65535) */
1149 dev->min_mtu = ETH_MIN_MTU;
1150 /* The max_mtu calculation does not take account of GENEVE
1151 * options, to avoid excluding potentially valid
1152 * configurations. This will be further reduced by IPvX hdr size.
1154 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1156 netif_keep_dst(dev);
1157 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1158 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1159 eth_hw_addr_random(dev);
1162 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1163 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1164 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1165 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1166 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1167 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1168 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1169 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1170 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1171 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1172 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1173 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1174 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1175 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1178 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1179 struct netlink_ext_ack *extack)
1181 if (tb[IFLA_ADDRESS]) {
1182 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1183 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1184 "Provided link layer address is not Ethernet");
1188 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1189 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1190 "Provided Ethernet address is not unicast");
1191 return -EADDRNOTAVAIL;
1196 NL_SET_ERR_MSG(extack,
1197 "Not enough attributes provided to perform the operation");
1201 if (data[IFLA_GENEVE_ID]) {
1202 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1204 if (vni >= GENEVE_N_VID) {
1205 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1206 "Geneve ID must be lower than 16777216");
1211 if (data[IFLA_GENEVE_DF]) {
1212 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1214 if (df < 0 || df > GENEVE_DF_MAX) {
1215 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1216 "Invalid DF attribute");
1224 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1225 const struct ip_tunnel_info *info,
1226 bool *tun_on_same_port,
1227 bool *tun_collect_md)
1229 struct geneve_dev *geneve, *t = NULL;
1231 *tun_on_same_port = false;
1232 *tun_collect_md = false;
1233 list_for_each_entry(geneve, &gn->geneve_list, next) {
1234 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1235 *tun_collect_md = geneve->cfg.collect_md;
1236 *tun_on_same_port = true;
1238 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1239 info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1240 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1246 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1248 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1249 info->key.ttl || info->key.label || info->key.tp_src ||
1250 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1253 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1254 struct ip_tunnel_info *b)
1256 if (ip_tunnel_info_af(a) == AF_INET)
1257 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1259 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1262 static int geneve_configure(struct net *net, struct net_device *dev,
1263 struct netlink_ext_ack *extack,
1264 const struct geneve_config *cfg)
1266 struct geneve_net *gn = net_generic(net, geneve_net_id);
1267 struct geneve_dev *t, *geneve = netdev_priv(dev);
1268 const struct ip_tunnel_info *info = &cfg->info;
1269 bool tun_collect_md, tun_on_same_port;
1272 if (cfg->collect_md && !is_tnl_info_zero(info)) {
1273 NL_SET_ERR_MSG(extack,
1274 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1281 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1285 /* make enough headroom for basic scenario */
1286 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1287 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1288 encap_len += sizeof(struct iphdr);
1289 dev->max_mtu -= sizeof(struct iphdr);
1291 encap_len += sizeof(struct ipv6hdr);
1292 dev->max_mtu -= sizeof(struct ipv6hdr);
1294 dev->needed_headroom = encap_len + ETH_HLEN;
1296 if (cfg->collect_md) {
1297 if (tun_on_same_port) {
1298 NL_SET_ERR_MSG(extack,
1299 "There can be only one externally controlled device on a destination port");
1303 if (tun_collect_md) {
1304 NL_SET_ERR_MSG(extack,
1305 "There already exists an externally controlled device on this destination port");
1310 dst_cache_reset(&geneve->cfg.info.dst_cache);
1311 memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1313 err = register_netdevice(dev);
1317 list_add(&geneve->next, &gn->geneve_list);
1321 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1323 memset(info, 0, sizeof(*info));
1324 info->key.tp_dst = htons(dst_port);
1327 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1328 struct netlink_ext_ack *extack,
1329 struct geneve_config *cfg, bool changelink)
1332 struct ip_tunnel_info *info = &cfg->info;
1334 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1335 NL_SET_ERR_MSG(extack,
1336 "Cannot specify both IPv4 and IPv6 Remote addresses");
1340 if (data[IFLA_GENEVE_REMOTE]) {
1341 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1342 attrtype = IFLA_GENEVE_REMOTE;
1346 info->key.u.ipv4.dst =
1347 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1349 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1350 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1351 "Remote IPv4 address cannot be Multicast");
1356 if (data[IFLA_GENEVE_REMOTE6]) {
1357 #if IS_ENABLED(CONFIG_IPV6)
1358 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1359 attrtype = IFLA_GENEVE_REMOTE6;
1363 info->mode = IP_TUNNEL_INFO_IPV6;
1364 info->key.u.ipv6.dst =
1365 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1367 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1368 IPV6_ADDR_LINKLOCAL) {
1369 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1370 "Remote IPv6 address cannot be link-local");
1373 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1374 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1375 "Remote IPv6 address cannot be Multicast");
1378 info->key.tun_flags |= TUNNEL_CSUM;
1379 cfg->use_udp6_rx_checksums = true;
1381 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1382 "IPv6 support not enabled in the kernel");
1383 return -EPFNOSUPPORT;
1387 if (data[IFLA_GENEVE_ID]) {
1392 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1393 tvni[0] = (vni & 0x00ff0000) >> 16;
1394 tvni[1] = (vni & 0x0000ff00) >> 8;
1395 tvni[2] = vni & 0x000000ff;
1397 tunid = vni_to_tunnel_id(tvni);
1398 if (changelink && (tunid != info->key.tun_id)) {
1399 attrtype = IFLA_GENEVE_ID;
1402 info->key.tun_id = tunid;
1405 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1406 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1407 cfg->ttl_inherit = true;
1409 cfg->ttl_inherit = false;
1410 } else if (data[IFLA_GENEVE_TTL]) {
1411 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1412 cfg->ttl_inherit = false;
1415 if (data[IFLA_GENEVE_TOS])
1416 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1418 if (data[IFLA_GENEVE_DF])
1419 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1421 if (data[IFLA_GENEVE_LABEL]) {
1422 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1423 IPV6_FLOWLABEL_MASK;
1424 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1425 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1426 "Label attribute only applies for IPv6 Geneve devices");
1431 if (data[IFLA_GENEVE_PORT]) {
1433 attrtype = IFLA_GENEVE_PORT;
1436 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1439 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1441 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1444 cfg->collect_md = true;
1447 if (data[IFLA_GENEVE_UDP_CSUM]) {
1449 attrtype = IFLA_GENEVE_UDP_CSUM;
1452 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1453 info->key.tun_flags |= TUNNEL_CSUM;
1456 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1457 #if IS_ENABLED(CONFIG_IPV6)
1459 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1462 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1463 info->key.tun_flags &= ~TUNNEL_CSUM;
1465 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1466 "IPv6 support not enabled in the kernel");
1467 return -EPFNOSUPPORT;
1471 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1472 #if IS_ENABLED(CONFIG_IPV6)
1474 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1477 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1478 cfg->use_udp6_rx_checksums = false;
1480 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1481 "IPv6 support not enabled in the kernel");
1482 return -EPFNOSUPPORT;
1488 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1489 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1493 static void geneve_link_config(struct net_device *dev,
1494 struct ip_tunnel_info *info, struct nlattr *tb[])
1496 struct geneve_dev *geneve = netdev_priv(dev);
1500 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1504 switch (ip_tunnel_info_af(info)) {
1506 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1507 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1509 if (!IS_ERR(rt) && rt->dst.dev) {
1510 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1515 #if IS_ENABLED(CONFIG_IPV6)
1517 struct rt6_info *rt;
1519 if (!__in6_dev_get(dev))
1522 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1525 if (rt && rt->dst.dev)
1526 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1536 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1539 static int geneve_newlink(struct net *net, struct net_device *dev,
1540 struct nlattr *tb[], struct nlattr *data[],
1541 struct netlink_ext_ack *extack)
1543 struct geneve_config cfg = {
1544 .df = GENEVE_DF_UNSET,
1545 .use_udp6_rx_checksums = false,
1546 .ttl_inherit = false,
1547 .collect_md = false,
1551 init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1552 err = geneve_nl2info(tb, data, extack, &cfg, false);
1556 err = geneve_configure(net, dev, extack, &cfg);
1560 geneve_link_config(dev, &cfg.info, tb);
1565 /* Quiesces the geneve device data path for both TX and RX.
1567 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1568 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1569 * to complete for the existing set of in-flight packets to be transmitted,
1570 * then we would have quiesced the transmit data path. All the future packets
1571 * will get dropped until we unquiesce the data path.
1573 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1574 * if we set that to NULL under RCU and wait for synchronize_net() to
1575 * complete, then we would have quiesced the receive data path.
1577 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1578 struct geneve_sock **gs6)
1580 *gs4 = rtnl_dereference(geneve->sock4);
1581 rcu_assign_pointer(geneve->sock4, NULL);
1583 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1584 #if IS_ENABLED(CONFIG_IPV6)
1585 *gs6 = rtnl_dereference(geneve->sock6);
1586 rcu_assign_pointer(geneve->sock6, NULL);
1588 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1595 /* Resumes the geneve device data path for both TX and RX. */
1596 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1597 struct geneve_sock __maybe_unused *gs6)
1599 rcu_assign_pointer(geneve->sock4, gs4);
1601 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1602 #if IS_ENABLED(CONFIG_IPV6)
1603 rcu_assign_pointer(geneve->sock6, gs6);
1605 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1610 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1611 struct nlattr *data[],
1612 struct netlink_ext_ack *extack)
1614 struct geneve_dev *geneve = netdev_priv(dev);
1615 struct geneve_sock *gs4, *gs6;
1616 struct geneve_config cfg;
1619 /* If the geneve device is configured for metadata (or externally
1620 * controlled, for example, OVS), then nothing can be changed.
1622 if (geneve->cfg.collect_md)
1625 /* Start with the existing info. */
1626 memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1627 err = geneve_nl2info(tb, data, extack, &cfg, true);
1631 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1632 dst_cache_reset(&cfg.info.dst_cache);
1633 geneve_link_config(dev, &cfg.info, tb);
1636 geneve_quiesce(geneve, &gs4, &gs6);
1637 memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1638 geneve_unquiesce(geneve, gs4, gs6);
1643 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1645 struct geneve_dev *geneve = netdev_priv(dev);
1647 list_del(&geneve->next);
1648 unregister_netdevice_queue(dev, head);
1651 static size_t geneve_get_size(const struct net_device *dev)
1653 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1654 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1655 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1656 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1657 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1658 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1659 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1660 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1661 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1662 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1663 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1664 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1668 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1670 struct geneve_dev *geneve = netdev_priv(dev);
1671 struct ip_tunnel_info *info = &geneve->cfg.info;
1672 bool ttl_inherit = geneve->cfg.ttl_inherit;
1673 bool metadata = geneve->cfg.collect_md;
1677 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1678 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1679 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1680 goto nla_put_failure;
1682 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1683 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1684 info->key.u.ipv4.dst))
1685 goto nla_put_failure;
1686 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1687 !!(info->key.tun_flags & TUNNEL_CSUM)))
1688 goto nla_put_failure;
1690 #if IS_ENABLED(CONFIG_IPV6)
1691 } else if (!metadata) {
1692 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1693 &info->key.u.ipv6.dst))
1694 goto nla_put_failure;
1695 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1696 !(info->key.tun_flags & TUNNEL_CSUM)))
1697 goto nla_put_failure;
1701 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1702 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1703 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1704 goto nla_put_failure;
1706 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1707 goto nla_put_failure;
1709 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1710 goto nla_put_failure;
1712 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1713 goto nla_put_failure;
1715 #if IS_ENABLED(CONFIG_IPV6)
1716 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1717 !geneve->cfg.use_udp6_rx_checksums))
1718 goto nla_put_failure;
1721 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1722 goto nla_put_failure;
1730 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1732 .maxtype = IFLA_GENEVE_MAX,
1733 .policy = geneve_policy,
1734 .priv_size = sizeof(struct geneve_dev),
1735 .setup = geneve_setup,
1736 .validate = geneve_validate,
1737 .newlink = geneve_newlink,
1738 .changelink = geneve_changelink,
1739 .dellink = geneve_dellink,
1740 .get_size = geneve_get_size,
1741 .fill_info = geneve_fill_info,
1744 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1745 u8 name_assign_type, u16 dst_port)
1747 struct nlattr *tb[IFLA_MAX + 1];
1748 struct net_device *dev;
1749 LIST_HEAD(list_kill);
1751 struct geneve_config cfg = {
1752 .df = GENEVE_DF_UNSET,
1753 .use_udp6_rx_checksums = true,
1754 .ttl_inherit = false,
1758 memset(tb, 0, sizeof(tb));
1759 dev = rtnl_create_link(net, name, name_assign_type,
1760 &geneve_link_ops, tb, NULL);
1764 init_tnl_info(&cfg.info, dst_port);
1765 err = geneve_configure(net, dev, NULL, &cfg);
1768 return ERR_PTR(err);
1771 /* openvswitch users expect packet sizes to be unrestricted,
1772 * so set the largest MTU we can.
1774 err = geneve_change_mtu(dev, IP_MAX_MTU);
1778 err = rtnl_configure_link(dev, NULL);
1784 geneve_dellink(dev, &list_kill);
1785 unregister_netdevice_many(&list_kill);
1786 return ERR_PTR(err);
1788 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1790 static int geneve_netdevice_event(struct notifier_block *unused,
1791 unsigned long event, void *ptr)
1793 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1795 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1796 event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1797 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1798 } else if (event == NETDEV_UNREGISTER) {
1799 if (!dev->udp_tunnel_nic_info)
1800 geneve_offload_rx_ports(dev, false);
1801 } else if (event == NETDEV_REGISTER) {
1802 if (!dev->udp_tunnel_nic_info)
1803 geneve_offload_rx_ports(dev, true);
1809 static struct notifier_block geneve_notifier_block __read_mostly = {
1810 .notifier_call = geneve_netdevice_event,
1813 static __net_init int geneve_init_net(struct net *net)
1815 struct geneve_net *gn = net_generic(net, geneve_net_id);
1817 INIT_LIST_HEAD(&gn->geneve_list);
1818 INIT_LIST_HEAD(&gn->sock_list);
1822 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1824 struct geneve_net *gn = net_generic(net, geneve_net_id);
1825 struct geneve_dev *geneve, *next;
1826 struct net_device *dev, *aux;
1828 /* gather any geneve devices that were moved into this ns */
1829 for_each_netdev_safe(net, dev, aux)
1830 if (dev->rtnl_link_ops == &geneve_link_ops)
1831 unregister_netdevice_queue(dev, head);
1833 /* now gather any other geneve devices that were created in this ns */
1834 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1835 /* If geneve->dev is in the same netns, it was already added
1836 * to the list by the previous loop.
1838 if (!net_eq(dev_net(geneve->dev), net))
1839 unregister_netdevice_queue(geneve->dev, head);
1843 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1849 list_for_each_entry(net, net_list, exit_list)
1850 geneve_destroy_tunnels(net, &list);
1852 /* unregister the devices gathered above */
1853 unregister_netdevice_many(&list);
1856 list_for_each_entry(net, net_list, exit_list) {
1857 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1859 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1863 static struct pernet_operations geneve_net_ops = {
1864 .init = geneve_init_net,
1865 .exit_batch = geneve_exit_batch_net,
1866 .id = &geneve_net_id,
1867 .size = sizeof(struct geneve_net),
1870 static int __init geneve_init_module(void)
1874 rc = register_pernet_subsys(&geneve_net_ops);
1878 rc = register_netdevice_notifier(&geneve_notifier_block);
1882 rc = rtnl_link_register(&geneve_link_ops);
1888 unregister_netdevice_notifier(&geneve_notifier_block);
1890 unregister_pernet_subsys(&geneve_net_ops);
1894 late_initcall(geneve_init_module);
1896 static void __exit geneve_cleanup_module(void)
1898 rtnl_link_unregister(&geneve_link_ops);
1899 unregister_netdevice_notifier(&geneve_notifier_block);
1900 unregister_pernet_subsys(&geneve_net_ops);
1902 module_exit(geneve_cleanup_module);
1904 MODULE_LICENSE("GPL");
1905 MODULE_VERSION(GENEVE_NETDEV_VER);
1906 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1907 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1908 MODULE_ALIAS_RTNL_LINK("geneve");