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;
61 /* Pseudo network device */
63 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
64 #if IS_ENABLED(CONFIG_IPV6)
65 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
67 struct net *net; /* netns for packet i/o */
68 struct net_device *dev; /* netdev for geneve tunnel */
69 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
70 #if IS_ENABLED(CONFIG_IPV6)
71 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
73 struct list_head next; /* geneve's per namespace list */
74 struct gro_cells gro_cells;
75 struct geneve_config cfg;
80 struct list_head list;
84 struct hlist_head vni_list[VNI_HASH_SIZE];
87 static inline __u32 geneve_net_vni_hash(u8 vni[3])
91 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
92 return hash_32(vnid, VNI_HASH_BITS);
95 static __be64 vni_to_tunnel_id(const __u8 *vni)
98 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
100 return (__force __be64)(((__force u64)vni[0] << 40) |
101 ((__force u64)vni[1] << 48) |
102 ((__force u64)vni[2] << 56));
106 /* Convert 64 bit tunnel ID to 24 bit VNI. */
107 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
110 vni[0] = (__force __u8)(tun_id >> 16);
111 vni[1] = (__force __u8)(tun_id >> 8);
112 vni[2] = (__force __u8)tun_id;
114 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
115 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
116 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
120 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
122 return !memcmp(vni, &tun_id[5], 3);
125 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
127 return gs->sock->sk->sk_family;
130 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
131 __be32 addr, u8 vni[])
133 struct hlist_head *vni_list_head;
134 struct geneve_dev_node *node;
137 /* Find the device for this VNI */
138 hash = geneve_net_vni_hash(vni);
139 vni_list_head = &gs->vni_list[hash];
140 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
141 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
142 addr == node->geneve->cfg.info.key.u.ipv4.dst)
148 #if IS_ENABLED(CONFIG_IPV6)
149 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
150 struct in6_addr addr6, u8 vni[])
152 struct hlist_head *vni_list_head;
153 struct geneve_dev_node *node;
156 /* Find the device for this VNI */
157 hash = geneve_net_vni_hash(vni);
158 vni_list_head = &gs->vni_list[hash];
159 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
160 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
161 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
168 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
170 return (struct genevehdr *)(udp_hdr(skb) + 1);
173 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
176 static u8 zero_vni[3];
179 if (geneve_get_sk_family(gs) == AF_INET) {
183 iph = ip_hdr(skb); /* outer IP header... */
185 if (gs->collect_md) {
189 vni = geneve_hdr(skb)->vni;
193 return geneve_lookup(gs, addr, vni);
194 #if IS_ENABLED(CONFIG_IPV6)
195 } else if (geneve_get_sk_family(gs) == AF_INET6) {
196 static struct in6_addr zero_addr6;
197 struct ipv6hdr *ip6h;
198 struct in6_addr addr6;
200 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
202 if (gs->collect_md) {
206 vni = geneve_hdr(skb)->vni;
210 return geneve6_lookup(gs, addr6, vni);
216 /* geneve receive/decap routine */
217 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
220 struct genevehdr *gnvh = geneve_hdr(skb);
221 struct metadata_dst *tun_dst = NULL;
226 if (ip_tunnel_collect_metadata() || gs->collect_md) {
229 flags = TUNNEL_KEY | (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 dev_sw_netstats_rx_add(geneve->dev, len);
304 /* Consume bad packet */
308 /* Setup stats when device is created */
309 static int geneve_init(struct net_device *dev)
311 struct geneve_dev *geneve = netdev_priv(dev);
314 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
318 err = gro_cells_init(&geneve->gro_cells, dev);
320 free_percpu(dev->tstats);
324 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
326 free_percpu(dev->tstats);
327 gro_cells_destroy(&geneve->gro_cells);
333 static void geneve_uninit(struct net_device *dev)
335 struct geneve_dev *geneve = netdev_priv(dev);
337 dst_cache_destroy(&geneve->cfg.info.dst_cache);
338 gro_cells_destroy(&geneve->gro_cells);
339 free_percpu(dev->tstats);
342 /* Callback from net/ipv4/udp.c to receive packets */
343 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
345 struct genevehdr *geneveh;
346 struct geneve_dev *geneve;
347 struct geneve_sock *gs;
350 /* Need UDP and Geneve header to be present */
351 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
354 /* Return packets with reserved bits set */
355 geneveh = geneve_hdr(skb);
356 if (unlikely(geneveh->ver != GENEVE_VER))
359 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
362 gs = rcu_dereference_sk_user_data(sk);
366 geneve = geneve_lookup_skb(gs, skb);
370 opts_len = geneveh->opt_len * 4;
371 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
373 !net_eq(geneve->net, dev_net(geneve->dev)))) {
374 geneve->dev->stats.rx_dropped++;
378 geneve_rx(geneve, gs, skb);
382 /* Consume bad packet */
387 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
388 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
390 struct genevehdr *geneveh;
391 struct geneve_sock *gs;
392 u8 zero_vni[3] = { 0 };
395 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
398 geneveh = geneve_hdr(skb);
399 if (geneveh->ver != GENEVE_VER)
402 if (geneveh->proto_type != htons(ETH_P_TEB))
405 gs = rcu_dereference_sk_user_data(sk);
409 if (geneve_get_sk_family(gs) == AF_INET) {
410 struct iphdr *iph = ip_hdr(skb);
413 if (!gs->collect_md) {
414 vni = geneve_hdr(skb)->vni;
418 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
421 #if IS_ENABLED(CONFIG_IPV6)
422 if (geneve_get_sk_family(gs) == AF_INET6) {
423 struct ipv6hdr *ip6h = ipv6_hdr(skb);
424 struct in6_addr addr6;
426 memset(&addr6, 0, sizeof(struct in6_addr));
428 if (!gs->collect_md) {
429 vni = geneve_hdr(skb)->vni;
433 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
437 return -EPFNOSUPPORT;
440 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
441 __be16 port, bool ipv6_rx_csum)
444 struct udp_port_cfg udp_conf;
447 memset(&udp_conf, 0, sizeof(udp_conf));
450 udp_conf.family = AF_INET6;
451 udp_conf.ipv6_v6only = 1;
452 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
454 udp_conf.family = AF_INET;
455 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
458 udp_conf.local_udp_port = port;
460 /* Open UDP socket */
461 err = udp_sock_create(net, &udp_conf, &sock);
465 udp_allow_gso(sock->sk);
469 static int geneve_hlen(struct genevehdr *gh)
471 return sizeof(*gh) + gh->opt_len * 4;
474 static struct sk_buff *geneve_gro_receive(struct sock *sk,
475 struct list_head *head,
478 struct sk_buff *pp = NULL;
480 struct genevehdr *gh, *gh2;
481 unsigned int hlen, gh_len, off_gnv;
482 const struct packet_offload *ptype;
486 off_gnv = skb_gro_offset(skb);
487 hlen = off_gnv + sizeof(*gh);
488 gh = skb_gro_header_fast(skb, off_gnv);
489 if (skb_gro_header_hard(skb, hlen)) {
490 gh = skb_gro_header_slow(skb, hlen, off_gnv);
495 if (gh->ver != GENEVE_VER || gh->oam)
497 gh_len = geneve_hlen(gh);
499 hlen = off_gnv + gh_len;
500 if (skb_gro_header_hard(skb, hlen)) {
501 gh = skb_gro_header_slow(skb, hlen, off_gnv);
506 list_for_each_entry(p, head, list) {
507 if (!NAPI_GRO_CB(p)->same_flow)
510 gh2 = (struct genevehdr *)(p->data + off_gnv);
511 if (gh->opt_len != gh2->opt_len ||
512 memcmp(gh, gh2, gh_len)) {
513 NAPI_GRO_CB(p)->same_flow = 0;
518 type = gh->proto_type;
520 ptype = gro_find_receive_by_type(type);
524 skb_gro_pull(skb, gh_len);
525 skb_gro_postpull_rcsum(skb, gh, gh_len);
526 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
530 skb_gro_flush_final(skb, pp, flush);
535 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
538 struct genevehdr *gh;
539 struct packet_offload *ptype;
544 gh = (struct genevehdr *)(skb->data + nhoff);
545 gh_len = geneve_hlen(gh);
546 type = gh->proto_type;
548 ptype = gro_find_complete_by_type(type);
550 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
552 skb_set_inner_mac_header(skb, nhoff + gh_len);
557 /* Create new listen socket if needed */
558 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
559 bool ipv6, bool ipv6_rx_csum)
561 struct geneve_net *gn = net_generic(net, geneve_net_id);
562 struct geneve_sock *gs;
564 struct udp_tunnel_sock_cfg tunnel_cfg;
567 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
569 return ERR_PTR(-ENOMEM);
571 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
574 return ERR_CAST(sock);
579 for (h = 0; h < VNI_HASH_SIZE; ++h)
580 INIT_HLIST_HEAD(&gs->vni_list[h]);
582 /* Initialize the geneve udp offloads structure */
583 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
585 /* Mark socket as an encapsulation socket */
586 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
587 tunnel_cfg.sk_user_data = gs;
588 tunnel_cfg.encap_type = 1;
589 tunnel_cfg.gro_receive = geneve_gro_receive;
590 tunnel_cfg.gro_complete = geneve_gro_complete;
591 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
592 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
593 tunnel_cfg.encap_destroy = NULL;
594 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
595 list_add(&gs->list, &gn->sock_list);
599 static void __geneve_sock_release(struct geneve_sock *gs)
601 if (!gs || --gs->refcnt)
605 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
606 udp_tunnel_sock_release(gs->sock);
610 static void geneve_sock_release(struct geneve_dev *geneve)
612 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
613 #if IS_ENABLED(CONFIG_IPV6)
614 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
616 rcu_assign_pointer(geneve->sock6, NULL);
619 rcu_assign_pointer(geneve->sock4, NULL);
622 __geneve_sock_release(gs4);
623 #if IS_ENABLED(CONFIG_IPV6)
624 __geneve_sock_release(gs6);
628 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
632 struct geneve_sock *gs;
634 list_for_each_entry(gs, &gn->sock_list, list) {
635 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
636 geneve_get_sk_family(gs) == family) {
643 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
645 struct net *net = geneve->net;
646 struct geneve_net *gn = net_generic(net, geneve_net_id);
647 struct geneve_dev_node *node;
648 struct geneve_sock *gs;
652 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
658 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
659 geneve->cfg.use_udp6_rx_checksums);
664 gs->collect_md = geneve->cfg.collect_md;
665 #if IS_ENABLED(CONFIG_IPV6)
667 rcu_assign_pointer(geneve->sock6, gs);
668 node = &geneve->hlist6;
672 rcu_assign_pointer(geneve->sock4, gs);
673 node = &geneve->hlist4;
675 node->geneve = geneve;
677 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
678 hash = geneve_net_vni_hash(vni);
679 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
683 static int geneve_open(struct net_device *dev)
685 struct geneve_dev *geneve = netdev_priv(dev);
686 bool metadata = geneve->cfg.collect_md;
690 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
691 ipv4 = !ipv6 || metadata;
692 #if IS_ENABLED(CONFIG_IPV6)
694 ret = geneve_sock_add(geneve, true);
695 if (ret < 0 && ret != -EAFNOSUPPORT)
700 ret = geneve_sock_add(geneve, false);
702 geneve_sock_release(geneve);
707 static int geneve_stop(struct net_device *dev)
709 struct geneve_dev *geneve = netdev_priv(dev);
711 hlist_del_init_rcu(&geneve->hlist4.hlist);
712 #if IS_ENABLED(CONFIG_IPV6)
713 hlist_del_init_rcu(&geneve->hlist6.hlist);
715 geneve_sock_release(geneve);
719 static void geneve_build_header(struct genevehdr *geneveh,
720 const struct ip_tunnel_info *info)
722 geneveh->ver = GENEVE_VER;
723 geneveh->opt_len = info->options_len / 4;
724 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
725 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
727 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
728 geneveh->proto_type = htons(ETH_P_TEB);
731 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
732 ip_tunnel_info_opts_get(geneveh->options, info);
735 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
736 const struct ip_tunnel_info *info,
737 bool xnet, int ip_hdr_len)
739 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
740 struct genevehdr *gnvh;
744 skb_reset_mac_header(skb);
745 skb_scrub_packet(skb, xnet);
747 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
748 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
749 err = skb_cow_head(skb, min_headroom);
753 err = udp_tunnel_handle_offloads(skb, udp_sum);
757 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
758 geneve_build_header(gnvh, info);
759 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
767 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
768 struct net_device *dev,
769 struct geneve_sock *gs4,
771 const struct ip_tunnel_info *info,
772 __be16 dport, __be16 sport)
774 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
775 struct geneve_dev *geneve = netdev_priv(dev);
776 struct dst_cache *dst_cache;
777 struct rtable *rt = NULL;
781 return ERR_PTR(-EIO);
783 memset(fl4, 0, sizeof(*fl4));
784 fl4->flowi4_mark = skb->mark;
785 fl4->flowi4_proto = IPPROTO_UDP;
786 fl4->daddr = info->key.u.ipv4.dst;
787 fl4->saddr = info->key.u.ipv4.src;
788 fl4->fl4_dport = dport;
789 fl4->fl4_sport = sport;
792 if ((tos == 1) && !geneve->cfg.collect_md) {
793 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
796 fl4->flowi4_tos = RT_TOS(tos);
798 dst_cache = (struct dst_cache *)&info->dst_cache;
800 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
804 rt = ip_route_output_key(geneve->net, fl4);
806 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
807 return ERR_PTR(-ENETUNREACH);
809 if (rt->dst.dev == dev) { /* is this necessary? */
810 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
812 return ERR_PTR(-ELOOP);
815 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
819 #if IS_ENABLED(CONFIG_IPV6)
820 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
821 struct net_device *dev,
822 struct geneve_sock *gs6,
824 const struct ip_tunnel_info *info,
825 __be16 dport, __be16 sport)
827 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
828 struct geneve_dev *geneve = netdev_priv(dev);
829 struct dst_entry *dst = NULL;
830 struct dst_cache *dst_cache;
834 return ERR_PTR(-EIO);
836 memset(fl6, 0, sizeof(*fl6));
837 fl6->flowi6_mark = skb->mark;
838 fl6->flowi6_proto = IPPROTO_UDP;
839 fl6->daddr = info->key.u.ipv6.dst;
840 fl6->saddr = info->key.u.ipv6.src;
841 fl6->fl6_dport = dport;
842 fl6->fl6_sport = sport;
844 prio = info->key.tos;
845 if ((prio == 1) && !geneve->cfg.collect_md) {
846 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
850 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
852 dst_cache = (struct dst_cache *)&info->dst_cache;
854 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
858 dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
861 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
862 return ERR_PTR(-ENETUNREACH);
864 if (dst->dev == dev) { /* is this necessary? */
865 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
867 return ERR_PTR(-ELOOP);
871 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
876 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
877 struct geneve_dev *geneve,
878 const struct ip_tunnel_info *info)
880 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
881 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
882 const struct ip_tunnel_key *key = &info->key;
890 if (!pskb_inet_may_pull(skb))
893 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
894 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
895 geneve->cfg.info.key.tp_dst, sport);
899 err = skb_tunnel_check_pmtu(skb, &rt->dst,
900 GENEVE_IPV4_HLEN + info->options_len,
901 netif_is_any_bridge_port(dev));
903 dst_release(&rt->dst);
906 struct ip_tunnel_info *info;
908 info = skb_tunnel_info(skb);
910 struct ip_tunnel_info *unclone;
912 unclone = skb_tunnel_info_unclone(skb);
913 if (unlikely(!unclone)) {
914 dst_release(&rt->dst);
918 unclone->key.u.ipv4.dst = fl4.saddr;
919 unclone->key.u.ipv4.src = fl4.daddr;
922 if (!pskb_may_pull(skb, ETH_HLEN)) {
923 dst_release(&rt->dst);
927 skb->protocol = eth_type_trans(skb, geneve->dev);
929 dst_release(&rt->dst);
933 if (geneve->cfg.collect_md) {
934 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
937 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
939 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
940 if (geneve->cfg.ttl_inherit)
941 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
944 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
946 if (geneve->cfg.df == GENEVE_DF_SET) {
948 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
949 struct ethhdr *eth = eth_hdr(skb);
951 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
953 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
954 struct iphdr *iph = ip_hdr(skb);
956 if (iph->frag_off & htons(IP_DF))
962 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
966 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
967 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
968 !net_eq(geneve->net, dev_net(geneve->dev)),
969 !(info->key.tun_flags & TUNNEL_CSUM));
973 #if IS_ENABLED(CONFIG_IPV6)
974 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
975 struct geneve_dev *geneve,
976 const struct ip_tunnel_info *info)
978 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
979 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
980 const struct ip_tunnel_key *key = &info->key;
981 struct dst_entry *dst = NULL;
987 if (!pskb_inet_may_pull(skb))
990 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
991 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
992 geneve->cfg.info.key.tp_dst, sport);
996 err = skb_tunnel_check_pmtu(skb, dst,
997 GENEVE_IPV6_HLEN + info->options_len,
998 netif_is_any_bridge_port(dev));
1003 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1006 struct ip_tunnel_info *unclone;
1008 unclone = skb_tunnel_info_unclone(skb);
1009 if (unlikely(!unclone)) {
1014 unclone->key.u.ipv6.dst = fl6.saddr;
1015 unclone->key.u.ipv6.src = fl6.daddr;
1018 if (!pskb_may_pull(skb, ETH_HLEN)) {
1023 skb->protocol = eth_type_trans(skb, geneve->dev);
1029 if (geneve->cfg.collect_md) {
1030 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1033 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1035 if (geneve->cfg.ttl_inherit)
1036 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1039 ttl = ttl ? : ip6_dst_hoplimit(dst);
1041 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
1045 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1046 &fl6.saddr, &fl6.daddr, prio, ttl,
1047 info->key.label, sport, geneve->cfg.info.key.tp_dst,
1048 !(info->key.tun_flags & TUNNEL_CSUM));
1053 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1055 struct geneve_dev *geneve = netdev_priv(dev);
1056 struct ip_tunnel_info *info = NULL;
1059 if (geneve->cfg.collect_md) {
1060 info = skb_tunnel_info(skb);
1061 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1062 netdev_dbg(dev, "no tunnel metadata\n");
1064 dev->stats.tx_dropped++;
1065 return NETDEV_TX_OK;
1068 info = &geneve->cfg.info;
1072 #if IS_ENABLED(CONFIG_IPV6)
1073 if (info->mode & IP_TUNNEL_INFO_IPV6)
1074 err = geneve6_xmit_skb(skb, dev, geneve, info);
1077 err = geneve_xmit_skb(skb, dev, geneve, info);
1081 return NETDEV_TX_OK;
1083 if (err != -EMSGSIZE)
1087 dev->stats.collisions++;
1088 else if (err == -ENETUNREACH)
1089 dev->stats.tx_carrier_errors++;
1091 dev->stats.tx_errors++;
1092 return NETDEV_TX_OK;
1095 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1097 if (new_mtu > dev->max_mtu)
1098 new_mtu = dev->max_mtu;
1099 else if (new_mtu < dev->min_mtu)
1100 new_mtu = dev->min_mtu;
1106 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1108 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1109 struct geneve_dev *geneve = netdev_priv(dev);
1112 if (ip_tunnel_info_af(info) == AF_INET) {
1116 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1117 sport = udp_flow_src_port(geneve->net, skb,
1118 1, USHRT_MAX, true);
1120 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1121 geneve->cfg.info.key.tp_dst, sport);
1126 info->key.u.ipv4.src = fl4.saddr;
1127 #if IS_ENABLED(CONFIG_IPV6)
1128 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1129 struct dst_entry *dst;
1132 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1133 sport = udp_flow_src_port(geneve->net, skb,
1134 1, USHRT_MAX, true);
1136 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1137 geneve->cfg.info.key.tp_dst, sport);
1139 return PTR_ERR(dst);
1142 info->key.u.ipv6.src = fl6.saddr;
1148 info->key.tp_src = sport;
1149 info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1153 static const struct net_device_ops geneve_netdev_ops = {
1154 .ndo_init = geneve_init,
1155 .ndo_uninit = geneve_uninit,
1156 .ndo_open = geneve_open,
1157 .ndo_stop = geneve_stop,
1158 .ndo_start_xmit = geneve_xmit,
1159 .ndo_get_stats64 = dev_get_tstats64,
1160 .ndo_change_mtu = geneve_change_mtu,
1161 .ndo_validate_addr = eth_validate_addr,
1162 .ndo_set_mac_address = eth_mac_addr,
1163 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1166 static void geneve_get_drvinfo(struct net_device *dev,
1167 struct ethtool_drvinfo *drvinfo)
1169 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1170 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1173 static const struct ethtool_ops geneve_ethtool_ops = {
1174 .get_drvinfo = geneve_get_drvinfo,
1175 .get_link = ethtool_op_get_link,
1178 /* Info for udev, that this is a virtual tunnel endpoint */
1179 static struct device_type geneve_type = {
1183 /* Calls the ndo_udp_tunnel_add of the caller in order to
1184 * supply the listening GENEVE udp ports. Callers are expected
1185 * to implement the ndo_udp_tunnel_add.
1187 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1189 struct net *net = dev_net(dev);
1190 struct geneve_net *gn = net_generic(net, geneve_net_id);
1191 struct geneve_sock *gs;
1194 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1196 udp_tunnel_push_rx_port(dev, gs->sock,
1197 UDP_TUNNEL_TYPE_GENEVE);
1199 udp_tunnel_drop_rx_port(dev, gs->sock,
1200 UDP_TUNNEL_TYPE_GENEVE);
1206 /* Initialize the device structure. */
1207 static void geneve_setup(struct net_device *dev)
1211 dev->netdev_ops = &geneve_netdev_ops;
1212 dev->ethtool_ops = &geneve_ethtool_ops;
1213 dev->needs_free_netdev = true;
1215 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1217 dev->features |= NETIF_F_LLTX;
1218 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1219 dev->features |= NETIF_F_RXCSUM;
1220 dev->features |= NETIF_F_GSO_SOFTWARE;
1222 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1223 dev->hw_features |= NETIF_F_RXCSUM;
1224 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1226 /* MTU range: 68 - (something less than 65535) */
1227 dev->min_mtu = ETH_MIN_MTU;
1228 /* The max_mtu calculation does not take account of GENEVE
1229 * options, to avoid excluding potentially valid
1230 * configurations. This will be further reduced by IPvX hdr size.
1232 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1234 netif_keep_dst(dev);
1235 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1236 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1237 eth_hw_addr_random(dev);
1240 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1241 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1242 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1243 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1244 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1245 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1246 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1247 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1248 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1249 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1250 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1251 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1252 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1253 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1256 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1257 struct netlink_ext_ack *extack)
1259 if (tb[IFLA_ADDRESS]) {
1260 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1261 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1262 "Provided link layer address is not Ethernet");
1266 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1267 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1268 "Provided Ethernet address is not unicast");
1269 return -EADDRNOTAVAIL;
1274 NL_SET_ERR_MSG(extack,
1275 "Not enough attributes provided to perform the operation");
1279 if (data[IFLA_GENEVE_ID]) {
1280 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1282 if (vni >= GENEVE_N_VID) {
1283 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1284 "Geneve ID must be lower than 16777216");
1289 if (data[IFLA_GENEVE_DF]) {
1290 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1292 if (df < 0 || df > GENEVE_DF_MAX) {
1293 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1294 "Invalid DF attribute");
1302 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1303 const struct ip_tunnel_info *info,
1304 bool *tun_on_same_port,
1305 bool *tun_collect_md)
1307 struct geneve_dev *geneve, *t = NULL;
1309 *tun_on_same_port = false;
1310 *tun_collect_md = false;
1311 list_for_each_entry(geneve, &gn->geneve_list, next) {
1312 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1313 *tun_collect_md = geneve->cfg.collect_md;
1314 *tun_on_same_port = true;
1316 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1317 info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1318 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1324 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1326 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1327 info->key.ttl || info->key.label || info->key.tp_src ||
1328 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1331 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1332 struct ip_tunnel_info *b)
1334 if (ip_tunnel_info_af(a) == AF_INET)
1335 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1337 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1340 static int geneve_configure(struct net *net, struct net_device *dev,
1341 struct netlink_ext_ack *extack,
1342 const struct geneve_config *cfg)
1344 struct geneve_net *gn = net_generic(net, geneve_net_id);
1345 struct geneve_dev *t, *geneve = netdev_priv(dev);
1346 const struct ip_tunnel_info *info = &cfg->info;
1347 bool tun_collect_md, tun_on_same_port;
1350 if (cfg->collect_md && !is_tnl_info_zero(info)) {
1351 NL_SET_ERR_MSG(extack,
1352 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1359 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1363 /* make enough headroom for basic scenario */
1364 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1365 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1366 encap_len += sizeof(struct iphdr);
1367 dev->max_mtu -= sizeof(struct iphdr);
1369 encap_len += sizeof(struct ipv6hdr);
1370 dev->max_mtu -= sizeof(struct ipv6hdr);
1372 dev->needed_headroom = encap_len + ETH_HLEN;
1374 if (cfg->collect_md) {
1375 if (tun_on_same_port) {
1376 NL_SET_ERR_MSG(extack,
1377 "There can be only one externally controlled device on a destination port");
1381 if (tun_collect_md) {
1382 NL_SET_ERR_MSG(extack,
1383 "There already exists an externally controlled device on this destination port");
1388 dst_cache_reset(&geneve->cfg.info.dst_cache);
1389 memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1391 err = register_netdevice(dev);
1395 list_add(&geneve->next, &gn->geneve_list);
1399 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1401 memset(info, 0, sizeof(*info));
1402 info->key.tp_dst = htons(dst_port);
1405 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1406 struct netlink_ext_ack *extack,
1407 struct geneve_config *cfg, bool changelink)
1409 struct ip_tunnel_info *info = &cfg->info;
1412 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1413 NL_SET_ERR_MSG(extack,
1414 "Cannot specify both IPv4 and IPv6 Remote addresses");
1418 if (data[IFLA_GENEVE_REMOTE]) {
1419 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1420 attrtype = IFLA_GENEVE_REMOTE;
1424 info->key.u.ipv4.dst =
1425 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1427 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1428 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1429 "Remote IPv4 address cannot be Multicast");
1434 if (data[IFLA_GENEVE_REMOTE6]) {
1435 #if IS_ENABLED(CONFIG_IPV6)
1436 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1437 attrtype = IFLA_GENEVE_REMOTE6;
1441 info->mode = IP_TUNNEL_INFO_IPV6;
1442 info->key.u.ipv6.dst =
1443 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1445 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1446 IPV6_ADDR_LINKLOCAL) {
1447 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1448 "Remote IPv6 address cannot be link-local");
1451 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1452 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1453 "Remote IPv6 address cannot be Multicast");
1456 info->key.tun_flags |= TUNNEL_CSUM;
1457 cfg->use_udp6_rx_checksums = true;
1459 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1460 "IPv6 support not enabled in the kernel");
1461 return -EPFNOSUPPORT;
1465 if (data[IFLA_GENEVE_ID]) {
1470 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1471 tvni[0] = (vni & 0x00ff0000) >> 16;
1472 tvni[1] = (vni & 0x0000ff00) >> 8;
1473 tvni[2] = vni & 0x000000ff;
1475 tunid = vni_to_tunnel_id(tvni);
1476 if (changelink && (tunid != info->key.tun_id)) {
1477 attrtype = IFLA_GENEVE_ID;
1480 info->key.tun_id = tunid;
1483 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1484 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1485 cfg->ttl_inherit = true;
1487 cfg->ttl_inherit = false;
1488 } else if (data[IFLA_GENEVE_TTL]) {
1489 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1490 cfg->ttl_inherit = false;
1493 if (data[IFLA_GENEVE_TOS])
1494 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1496 if (data[IFLA_GENEVE_DF])
1497 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1499 if (data[IFLA_GENEVE_LABEL]) {
1500 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1501 IPV6_FLOWLABEL_MASK;
1502 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1503 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1504 "Label attribute only applies for IPv6 Geneve devices");
1509 if (data[IFLA_GENEVE_PORT]) {
1511 attrtype = IFLA_GENEVE_PORT;
1514 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1517 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1519 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1522 cfg->collect_md = true;
1525 if (data[IFLA_GENEVE_UDP_CSUM]) {
1527 attrtype = IFLA_GENEVE_UDP_CSUM;
1530 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1531 info->key.tun_flags |= TUNNEL_CSUM;
1534 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1535 #if IS_ENABLED(CONFIG_IPV6)
1537 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1540 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1541 info->key.tun_flags &= ~TUNNEL_CSUM;
1543 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1544 "IPv6 support not enabled in the kernel");
1545 return -EPFNOSUPPORT;
1549 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1550 #if IS_ENABLED(CONFIG_IPV6)
1552 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1555 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1556 cfg->use_udp6_rx_checksums = false;
1558 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1559 "IPv6 support not enabled in the kernel");
1560 return -EPFNOSUPPORT;
1566 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1567 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1571 static void geneve_link_config(struct net_device *dev,
1572 struct ip_tunnel_info *info, struct nlattr *tb[])
1574 struct geneve_dev *geneve = netdev_priv(dev);
1578 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1582 switch (ip_tunnel_info_af(info)) {
1584 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1585 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1587 if (!IS_ERR(rt) && rt->dst.dev) {
1588 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1593 #if IS_ENABLED(CONFIG_IPV6)
1595 struct rt6_info *rt;
1597 if (!__in6_dev_get(dev))
1600 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1603 if (rt && rt->dst.dev)
1604 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1614 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1617 static int geneve_newlink(struct net *net, struct net_device *dev,
1618 struct nlattr *tb[], struct nlattr *data[],
1619 struct netlink_ext_ack *extack)
1621 struct geneve_config cfg = {
1622 .df = GENEVE_DF_UNSET,
1623 .use_udp6_rx_checksums = false,
1624 .ttl_inherit = false,
1625 .collect_md = false,
1629 init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1630 err = geneve_nl2info(tb, data, extack, &cfg, false);
1634 err = geneve_configure(net, dev, extack, &cfg);
1638 geneve_link_config(dev, &cfg.info, tb);
1643 /* Quiesces the geneve device data path for both TX and RX.
1645 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1646 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1647 * to complete for the existing set of in-flight packets to be transmitted,
1648 * then we would have quiesced the transmit data path. All the future packets
1649 * will get dropped until we unquiesce the data path.
1651 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1652 * if we set that to NULL under RCU and wait for synchronize_net() to
1653 * complete, then we would have quiesced the receive data path.
1655 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1656 struct geneve_sock **gs6)
1658 *gs4 = rtnl_dereference(geneve->sock4);
1659 rcu_assign_pointer(geneve->sock4, NULL);
1661 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1662 #if IS_ENABLED(CONFIG_IPV6)
1663 *gs6 = rtnl_dereference(geneve->sock6);
1664 rcu_assign_pointer(geneve->sock6, NULL);
1666 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1673 /* Resumes the geneve device data path for both TX and RX. */
1674 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1675 struct geneve_sock __maybe_unused *gs6)
1677 rcu_assign_pointer(geneve->sock4, gs4);
1679 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1680 #if IS_ENABLED(CONFIG_IPV6)
1681 rcu_assign_pointer(geneve->sock6, gs6);
1683 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1688 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1689 struct nlattr *data[],
1690 struct netlink_ext_ack *extack)
1692 struct geneve_dev *geneve = netdev_priv(dev);
1693 struct geneve_sock *gs4, *gs6;
1694 struct geneve_config cfg;
1697 /* If the geneve device is configured for metadata (or externally
1698 * controlled, for example, OVS), then nothing can be changed.
1700 if (geneve->cfg.collect_md)
1703 /* Start with the existing info. */
1704 memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1705 err = geneve_nl2info(tb, data, extack, &cfg, true);
1709 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1710 dst_cache_reset(&cfg.info.dst_cache);
1711 geneve_link_config(dev, &cfg.info, tb);
1714 geneve_quiesce(geneve, &gs4, &gs6);
1715 memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1716 geneve_unquiesce(geneve, gs4, gs6);
1721 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1723 struct geneve_dev *geneve = netdev_priv(dev);
1725 list_del(&geneve->next);
1726 unregister_netdevice_queue(dev, head);
1729 static size_t geneve_get_size(const struct net_device *dev)
1731 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1732 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1733 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1734 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1735 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1736 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1737 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1738 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1739 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1740 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1741 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1742 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1746 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1748 struct geneve_dev *geneve = netdev_priv(dev);
1749 struct ip_tunnel_info *info = &geneve->cfg.info;
1750 bool ttl_inherit = geneve->cfg.ttl_inherit;
1751 bool metadata = geneve->cfg.collect_md;
1755 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1756 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1757 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1758 goto nla_put_failure;
1760 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1761 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1762 info->key.u.ipv4.dst))
1763 goto nla_put_failure;
1764 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1765 !!(info->key.tun_flags & TUNNEL_CSUM)))
1766 goto nla_put_failure;
1768 #if IS_ENABLED(CONFIG_IPV6)
1769 } else if (!metadata) {
1770 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1771 &info->key.u.ipv6.dst))
1772 goto nla_put_failure;
1773 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1774 !(info->key.tun_flags & TUNNEL_CSUM)))
1775 goto nla_put_failure;
1779 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1780 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1781 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1782 goto nla_put_failure;
1784 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1785 goto nla_put_failure;
1787 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1788 goto nla_put_failure;
1790 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1791 goto nla_put_failure;
1793 #if IS_ENABLED(CONFIG_IPV6)
1794 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1795 !geneve->cfg.use_udp6_rx_checksums))
1796 goto nla_put_failure;
1799 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1800 goto nla_put_failure;
1808 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1810 .maxtype = IFLA_GENEVE_MAX,
1811 .policy = geneve_policy,
1812 .priv_size = sizeof(struct geneve_dev),
1813 .setup = geneve_setup,
1814 .validate = geneve_validate,
1815 .newlink = geneve_newlink,
1816 .changelink = geneve_changelink,
1817 .dellink = geneve_dellink,
1818 .get_size = geneve_get_size,
1819 .fill_info = geneve_fill_info,
1822 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1823 u8 name_assign_type, u16 dst_port)
1825 struct nlattr *tb[IFLA_MAX + 1];
1826 struct net_device *dev;
1827 LIST_HEAD(list_kill);
1829 struct geneve_config cfg = {
1830 .df = GENEVE_DF_UNSET,
1831 .use_udp6_rx_checksums = true,
1832 .ttl_inherit = false,
1836 memset(tb, 0, sizeof(tb));
1837 dev = rtnl_create_link(net, name, name_assign_type,
1838 &geneve_link_ops, tb, NULL);
1842 init_tnl_info(&cfg.info, dst_port);
1843 err = geneve_configure(net, dev, NULL, &cfg);
1846 return ERR_PTR(err);
1849 /* openvswitch users expect packet sizes to be unrestricted,
1850 * so set the largest MTU we can.
1852 err = geneve_change_mtu(dev, IP_MAX_MTU);
1856 err = rtnl_configure_link(dev, NULL);
1862 geneve_dellink(dev, &list_kill);
1863 unregister_netdevice_many(&list_kill);
1864 return ERR_PTR(err);
1866 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1868 static int geneve_netdevice_event(struct notifier_block *unused,
1869 unsigned long event, void *ptr)
1871 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1873 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1874 geneve_offload_rx_ports(dev, true);
1875 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1876 geneve_offload_rx_ports(dev, false);
1881 static struct notifier_block geneve_notifier_block __read_mostly = {
1882 .notifier_call = geneve_netdevice_event,
1885 static __net_init int geneve_init_net(struct net *net)
1887 struct geneve_net *gn = net_generic(net, geneve_net_id);
1889 INIT_LIST_HEAD(&gn->geneve_list);
1890 INIT_LIST_HEAD(&gn->sock_list);
1894 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1896 struct geneve_net *gn = net_generic(net, geneve_net_id);
1897 struct geneve_dev *geneve, *next;
1898 struct net_device *dev, *aux;
1900 /* gather any geneve devices that were moved into this ns */
1901 for_each_netdev_safe(net, dev, aux)
1902 if (dev->rtnl_link_ops == &geneve_link_ops)
1903 unregister_netdevice_queue(dev, head);
1905 /* now gather any other geneve devices that were created in this ns */
1906 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1907 /* If geneve->dev is in the same netns, it was already added
1908 * to the list by the previous loop.
1910 if (!net_eq(dev_net(geneve->dev), net))
1911 unregister_netdevice_queue(geneve->dev, head);
1915 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1921 list_for_each_entry(net, net_list, exit_list)
1922 geneve_destroy_tunnels(net, &list);
1924 /* unregister the devices gathered above */
1925 unregister_netdevice_many(&list);
1928 list_for_each_entry(net, net_list, exit_list) {
1929 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1931 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1935 static struct pernet_operations geneve_net_ops = {
1936 .init = geneve_init_net,
1937 .exit_batch = geneve_exit_batch_net,
1938 .id = &geneve_net_id,
1939 .size = sizeof(struct geneve_net),
1942 static int __init geneve_init_module(void)
1946 rc = register_pernet_subsys(&geneve_net_ops);
1950 rc = register_netdevice_notifier(&geneve_notifier_block);
1954 rc = rtnl_link_register(&geneve_link_ops);
1960 unregister_netdevice_notifier(&geneve_notifier_block);
1962 unregister_pernet_subsys(&geneve_net_ops);
1966 late_initcall(geneve_init_module);
1968 static void __exit geneve_cleanup_module(void)
1970 rtnl_link_unregister(&geneve_link_ops);
1971 unregister_netdevice_notifier(&geneve_notifier_block);
1972 unregister_pernet_subsys(&geneve_net_ops);
1974 module_exit(geneve_cleanup_module);
1976 MODULE_LICENSE("GPL");
1977 MODULE_VERSION(GENEVE_NETDEV_VER);
1978 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1979 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1980 MODULE_ALIAS_RTNL_LINK("geneve");