1 // SPDX-License-Identifier: GPL-2.0-only
3 * GENEVE: Generic Network Virtualization Encapsulation
5 * Copyright (c) 2015 Red Hat, Inc.
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/ethtool.h>
11 #include <linux/kernel.h>
12 #include <linux/module.h>
13 #include <linux/etherdevice.h>
14 #include <linux/hash.h>
15 #include <net/ipv6_stubs.h>
16 #include <net/dst_metadata.h>
17 #include <net/gro_cells.h>
18 #include <net/rtnetlink.h>
19 #include <net/geneve.h>
21 #include <net/protocol.h>
23 #define GENEVE_NETDEV_VER "0.6"
25 #define GENEVE_N_VID (1u << 24)
26 #define GENEVE_VID_MASK (GENEVE_N_VID - 1)
28 #define VNI_HASH_BITS 10
29 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
31 static bool log_ecn_error = true;
32 module_param(log_ecn_error, bool, 0644);
33 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
36 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
37 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
38 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
40 /* per-network namespace private data for this module */
42 struct list_head geneve_list;
43 struct list_head sock_list;
46 static unsigned int geneve_net_id;
48 struct geneve_dev_node {
49 struct hlist_node hlist;
50 struct geneve_dev *geneve;
53 struct geneve_config {
54 struct ip_tunnel_info info;
56 bool use_udp6_rx_checksums;
58 enum ifla_geneve_df df;
59 bool inner_proto_inherit;
62 /* Pseudo network device */
64 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
65 #if IS_ENABLED(CONFIG_IPV6)
66 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
68 struct net *net; /* netns for packet i/o */
69 struct net_device *dev; /* netdev for geneve tunnel */
70 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
71 #if IS_ENABLED(CONFIG_IPV6)
72 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
74 struct list_head next; /* geneve's per namespace list */
75 struct gro_cells gro_cells;
76 struct geneve_config cfg;
81 struct list_head list;
85 struct hlist_head vni_list[VNI_HASH_SIZE];
88 static inline __u32 geneve_net_vni_hash(u8 vni[3])
92 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
93 return hash_32(vnid, VNI_HASH_BITS);
96 static __be64 vni_to_tunnel_id(const __u8 *vni)
99 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
101 return (__force __be64)(((__force u64)vni[0] << 40) |
102 ((__force u64)vni[1] << 48) |
103 ((__force u64)vni[2] << 56));
107 /* Convert 64 bit tunnel ID to 24 bit VNI. */
108 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
111 vni[0] = (__force __u8)(tun_id >> 16);
112 vni[1] = (__force __u8)(tun_id >> 8);
113 vni[2] = (__force __u8)tun_id;
115 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
116 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
117 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
121 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
123 return !memcmp(vni, &tun_id[5], 3);
126 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
128 return gs->sock->sk->sk_family;
131 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
132 __be32 addr, u8 vni[])
134 struct hlist_head *vni_list_head;
135 struct geneve_dev_node *node;
138 /* Find the device for this VNI */
139 hash = geneve_net_vni_hash(vni);
140 vni_list_head = &gs->vni_list[hash];
141 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
142 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
143 addr == node->geneve->cfg.info.key.u.ipv4.dst)
149 #if IS_ENABLED(CONFIG_IPV6)
150 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
151 struct in6_addr addr6, u8 vni[])
153 struct hlist_head *vni_list_head;
154 struct geneve_dev_node *node;
157 /* Find the device for this VNI */
158 hash = geneve_net_vni_hash(vni);
159 vni_list_head = &gs->vni_list[hash];
160 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
161 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
162 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
169 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
171 return (struct genevehdr *)(udp_hdr(skb) + 1);
174 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
177 static u8 zero_vni[3];
180 if (geneve_get_sk_family(gs) == AF_INET) {
184 iph = ip_hdr(skb); /* outer IP header... */
186 if (gs->collect_md) {
190 vni = geneve_hdr(skb)->vni;
194 return geneve_lookup(gs, addr, vni);
195 #if IS_ENABLED(CONFIG_IPV6)
196 } else if (geneve_get_sk_family(gs) == AF_INET6) {
197 static struct in6_addr zero_addr6;
198 struct ipv6hdr *ip6h;
199 struct in6_addr addr6;
201 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
203 if (gs->collect_md) {
207 vni = geneve_hdr(skb)->vni;
211 return geneve6_lookup(gs, addr6, vni);
217 /* geneve receive/decap routine */
218 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
221 struct genevehdr *gnvh = geneve_hdr(skb);
222 struct metadata_dst *tun_dst = NULL;
227 if (ip_tunnel_collect_metadata() || gs->collect_md) {
230 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
231 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
233 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
234 vni_to_tunnel_id(gnvh->vni),
237 DEV_STATS_INC(geneve->dev, rx_dropped);
240 /* Update tunnel dst according to Geneve options. */
241 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
242 gnvh->options, gnvh->opt_len * 4,
245 /* Drop packets w/ critical options,
246 * since we don't support any...
248 if (gnvh->critical) {
249 DEV_STATS_INC(geneve->dev, rx_frame_errors);
250 DEV_STATS_INC(geneve->dev, rx_errors);
256 skb_dst_set(skb, &tun_dst->dst);
258 if (gnvh->proto_type == htons(ETH_P_TEB)) {
259 skb_reset_mac_header(skb);
260 skb->protocol = eth_type_trans(skb, geneve->dev);
261 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
263 /* Ignore packet loops (and multicast echo) */
264 if (ether_addr_equal(eth_hdr(skb)->h_source,
265 geneve->dev->dev_addr)) {
266 DEV_STATS_INC(geneve->dev, rx_errors);
270 skb_reset_mac_header(skb);
271 skb->dev = geneve->dev;
272 skb->pkt_type = PACKET_HOST;
275 /* Save offset of outer header relative to skb->head,
276 * because we are going to reset the network header to the inner header
277 * and might change skb->head.
279 nh = skb_network_header(skb) - skb->head;
281 skb_reset_network_header(skb);
283 if (!pskb_inet_may_pull(skb)) {
284 DEV_STATS_INC(geneve->dev, rx_length_errors);
285 DEV_STATS_INC(geneve->dev, rx_errors);
289 /* Get the outer header. */
290 oiph = skb->head + nh;
292 if (geneve_get_sk_family(gs) == AF_INET)
293 err = IP_ECN_decapsulate(oiph, skb);
294 #if IS_ENABLED(CONFIG_IPV6)
296 err = IP6_ECN_decapsulate(oiph, skb);
301 if (geneve_get_sk_family(gs) == AF_INET)
302 net_info_ratelimited("non-ECT from %pI4 "
304 &((struct iphdr *)oiph)->saddr,
305 ((struct iphdr *)oiph)->tos);
306 #if IS_ENABLED(CONFIG_IPV6)
308 net_info_ratelimited("non-ECT from %pI6\n",
309 &((struct ipv6hdr *)oiph)->saddr);
313 DEV_STATS_INC(geneve->dev, rx_frame_errors);
314 DEV_STATS_INC(geneve->dev, rx_errors);
320 err = gro_cells_receive(&geneve->gro_cells, skb);
321 if (likely(err == NET_RX_SUCCESS))
322 dev_sw_netstats_rx_add(geneve->dev, len);
326 /* Consume bad packet */
330 /* Setup stats when device is created */
331 static int geneve_init(struct net_device *dev)
333 struct geneve_dev *geneve = netdev_priv(dev);
336 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
340 err = gro_cells_init(&geneve->gro_cells, dev);
342 free_percpu(dev->tstats);
346 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
348 free_percpu(dev->tstats);
349 gro_cells_destroy(&geneve->gro_cells);
352 netdev_lockdep_set_classes(dev);
356 static void geneve_uninit(struct net_device *dev)
358 struct geneve_dev *geneve = netdev_priv(dev);
360 dst_cache_destroy(&geneve->cfg.info.dst_cache);
361 gro_cells_destroy(&geneve->gro_cells);
362 free_percpu(dev->tstats);
365 /* Callback from net/ipv4/udp.c to receive packets */
366 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
368 struct genevehdr *geneveh;
369 struct geneve_dev *geneve;
370 struct geneve_sock *gs;
374 /* Need UDP and Geneve header to be present */
375 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
378 /* Return packets with reserved bits set */
379 geneveh = geneve_hdr(skb);
380 if (unlikely(geneveh->ver != GENEVE_VER))
383 gs = rcu_dereference_sk_user_data(sk);
387 geneve = geneve_lookup_skb(gs, skb);
391 inner_proto = geneveh->proto_type;
393 if (unlikely((!geneve->cfg.inner_proto_inherit &&
394 inner_proto != htons(ETH_P_TEB)))) {
395 DEV_STATS_INC(geneve->dev, rx_dropped);
399 opts_len = geneveh->opt_len * 4;
400 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
401 !net_eq(geneve->net, dev_net(geneve->dev)))) {
402 DEV_STATS_INC(geneve->dev, rx_dropped);
406 geneve_rx(geneve, gs, skb);
410 /* Consume bad packet */
415 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
416 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
418 struct genevehdr *geneveh;
419 struct geneve_sock *gs;
420 u8 zero_vni[3] = { 0 };
423 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
426 geneveh = geneve_hdr(skb);
427 if (geneveh->ver != GENEVE_VER)
430 if (geneveh->proto_type != htons(ETH_P_TEB))
433 gs = rcu_dereference_sk_user_data(sk);
437 if (geneve_get_sk_family(gs) == AF_INET) {
438 struct iphdr *iph = ip_hdr(skb);
441 if (!gs->collect_md) {
442 vni = geneve_hdr(skb)->vni;
446 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
449 #if IS_ENABLED(CONFIG_IPV6)
450 if (geneve_get_sk_family(gs) == AF_INET6) {
451 struct ipv6hdr *ip6h = ipv6_hdr(skb);
452 struct in6_addr addr6;
454 memset(&addr6, 0, sizeof(struct in6_addr));
456 if (!gs->collect_md) {
457 vni = geneve_hdr(skb)->vni;
461 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
465 return -EPFNOSUPPORT;
468 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
469 __be16 port, bool ipv6_rx_csum)
472 struct udp_port_cfg udp_conf;
475 memset(&udp_conf, 0, sizeof(udp_conf));
478 udp_conf.family = AF_INET6;
479 udp_conf.ipv6_v6only = 1;
480 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
482 udp_conf.family = AF_INET;
483 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
486 udp_conf.local_udp_port = port;
488 /* Open UDP socket */
489 err = udp_sock_create(net, &udp_conf, &sock);
493 udp_allow_gso(sock->sk);
497 static int geneve_hlen(struct genevehdr *gh)
499 return sizeof(*gh) + gh->opt_len * 4;
502 static struct sk_buff *geneve_gro_receive(struct sock *sk,
503 struct list_head *head,
506 struct sk_buff *pp = NULL;
508 struct genevehdr *gh, *gh2;
509 unsigned int hlen, gh_len, off_gnv;
510 const struct packet_offload *ptype;
514 off_gnv = skb_gro_offset(skb);
515 hlen = off_gnv + sizeof(*gh);
516 gh = skb_gro_header(skb, hlen, off_gnv);
520 if (gh->ver != GENEVE_VER || gh->oam)
522 gh_len = geneve_hlen(gh);
524 hlen = off_gnv + gh_len;
525 if (skb_gro_header_hard(skb, hlen)) {
526 gh = skb_gro_header_slow(skb, hlen, off_gnv);
531 list_for_each_entry(p, head, list) {
532 if (!NAPI_GRO_CB(p)->same_flow)
535 gh2 = (struct genevehdr *)(p->data + off_gnv);
536 if (gh->opt_len != gh2->opt_len ||
537 memcmp(gh, gh2, gh_len)) {
538 NAPI_GRO_CB(p)->same_flow = 0;
543 skb_gro_pull(skb, gh_len);
544 skb_gro_postpull_rcsum(skb, gh, gh_len);
545 type = gh->proto_type;
546 if (likely(type == htons(ETH_P_TEB)))
547 return call_gro_receive(eth_gro_receive, head, skb);
549 ptype = gro_find_receive_by_type(type);
553 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
557 skb_gro_flush_final(skb, pp, flush);
562 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
565 struct genevehdr *gh;
566 struct packet_offload *ptype;
571 gh = (struct genevehdr *)(skb->data + nhoff);
572 gh_len = geneve_hlen(gh);
573 type = gh->proto_type;
575 /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */
576 if (likely(type == htons(ETH_P_TEB)))
577 return eth_gro_complete(skb, nhoff + gh_len);
579 ptype = gro_find_complete_by_type(type);
581 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
583 skb_set_inner_mac_header(skb, nhoff + gh_len);
588 /* Create new listen socket if needed */
589 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
590 bool ipv6, bool ipv6_rx_csum)
592 struct geneve_net *gn = net_generic(net, geneve_net_id);
593 struct geneve_sock *gs;
595 struct udp_tunnel_sock_cfg tunnel_cfg;
598 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
600 return ERR_PTR(-ENOMEM);
602 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
605 return ERR_CAST(sock);
610 for (h = 0; h < VNI_HASH_SIZE; ++h)
611 INIT_HLIST_HEAD(&gs->vni_list[h]);
613 /* Initialize the geneve udp offloads structure */
614 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
616 /* Mark socket as an encapsulation socket */
617 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
618 tunnel_cfg.sk_user_data = gs;
619 tunnel_cfg.encap_type = 1;
620 tunnel_cfg.gro_receive = geneve_gro_receive;
621 tunnel_cfg.gro_complete = geneve_gro_complete;
622 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
623 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
624 tunnel_cfg.encap_destroy = NULL;
625 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
626 list_add(&gs->list, &gn->sock_list);
630 static void __geneve_sock_release(struct geneve_sock *gs)
632 if (!gs || --gs->refcnt)
636 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
637 udp_tunnel_sock_release(gs->sock);
641 static void geneve_sock_release(struct geneve_dev *geneve)
643 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
644 #if IS_ENABLED(CONFIG_IPV6)
645 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
647 rcu_assign_pointer(geneve->sock6, NULL);
650 rcu_assign_pointer(geneve->sock4, NULL);
653 __geneve_sock_release(gs4);
654 #if IS_ENABLED(CONFIG_IPV6)
655 __geneve_sock_release(gs6);
659 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
663 struct geneve_sock *gs;
665 list_for_each_entry(gs, &gn->sock_list, list) {
666 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
667 geneve_get_sk_family(gs) == family) {
674 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
676 struct net *net = geneve->net;
677 struct geneve_net *gn = net_generic(net, geneve_net_id);
678 struct geneve_dev_node *node;
679 struct geneve_sock *gs;
683 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
689 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
690 geneve->cfg.use_udp6_rx_checksums);
695 gs->collect_md = geneve->cfg.collect_md;
696 #if IS_ENABLED(CONFIG_IPV6)
698 rcu_assign_pointer(geneve->sock6, gs);
699 node = &geneve->hlist6;
703 rcu_assign_pointer(geneve->sock4, gs);
704 node = &geneve->hlist4;
706 node->geneve = geneve;
708 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
709 hash = geneve_net_vni_hash(vni);
710 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
714 static int geneve_open(struct net_device *dev)
716 struct geneve_dev *geneve = netdev_priv(dev);
717 bool metadata = geneve->cfg.collect_md;
721 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
722 ipv4 = !ipv6 || metadata;
723 #if IS_ENABLED(CONFIG_IPV6)
725 ret = geneve_sock_add(geneve, true);
726 if (ret < 0 && ret != -EAFNOSUPPORT)
731 ret = geneve_sock_add(geneve, false);
733 geneve_sock_release(geneve);
738 static int geneve_stop(struct net_device *dev)
740 struct geneve_dev *geneve = netdev_priv(dev);
742 hlist_del_init_rcu(&geneve->hlist4.hlist);
743 #if IS_ENABLED(CONFIG_IPV6)
744 hlist_del_init_rcu(&geneve->hlist6.hlist);
746 geneve_sock_release(geneve);
750 static void geneve_build_header(struct genevehdr *geneveh,
751 const struct ip_tunnel_info *info,
754 geneveh->ver = GENEVE_VER;
755 geneveh->opt_len = info->options_len / 4;
756 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
757 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
759 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
760 geneveh->proto_type = inner_proto;
763 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
764 ip_tunnel_info_opts_get(geneveh->options, info);
767 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
768 const struct ip_tunnel_info *info,
769 bool xnet, int ip_hdr_len,
770 bool inner_proto_inherit)
772 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
773 struct genevehdr *gnvh;
778 skb_reset_mac_header(skb);
779 skb_scrub_packet(skb, xnet);
781 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
782 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
783 err = skb_cow_head(skb, min_headroom);
787 err = udp_tunnel_handle_offloads(skb, udp_sum);
791 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
792 inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
793 geneve_build_header(gnvh, info, inner_proto);
794 skb_set_inner_protocol(skb, inner_proto);
802 static u8 geneve_get_dsfield(struct sk_buff *skb, struct net_device *dev,
803 const struct ip_tunnel_info *info,
806 struct geneve_dev *geneve = netdev_priv(dev);
809 dsfield = info->key.tos;
810 if (dsfield == 1 && !geneve->cfg.collect_md) {
811 dsfield = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
818 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
819 struct geneve_dev *geneve,
820 const struct ip_tunnel_info *info)
822 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
823 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
824 const struct ip_tunnel_key *key = &info->key;
833 if (!skb_vlan_inet_prepare(skb))
839 use_cache = ip_tunnel_dst_cache_usable(skb, info);
840 tos = geneve_get_dsfield(skb, dev, info, &use_cache);
841 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
843 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
845 sport, geneve->cfg.info.key.tp_dst, tos,
847 (struct dst_cache *)&info->dst_cache : NULL);
851 err = skb_tunnel_check_pmtu(skb, &rt->dst,
852 GENEVE_IPV4_HLEN + info->options_len,
853 netif_is_any_bridge_port(dev));
855 dst_release(&rt->dst);
858 struct ip_tunnel_info *info;
860 info = skb_tunnel_info(skb);
862 struct ip_tunnel_info *unclone;
864 unclone = skb_tunnel_info_unclone(skb);
865 if (unlikely(!unclone)) {
866 dst_release(&rt->dst);
870 unclone->key.u.ipv4.dst = saddr;
871 unclone->key.u.ipv4.src = info->key.u.ipv4.dst;
874 if (!pskb_may_pull(skb, ETH_HLEN)) {
875 dst_release(&rt->dst);
879 skb->protocol = eth_type_trans(skb, geneve->dev);
881 dst_release(&rt->dst);
885 tos = ip_tunnel_ecn_encap(tos, ip_hdr(skb), skb);
886 if (geneve->cfg.collect_md) {
889 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
891 if (geneve->cfg.ttl_inherit)
892 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
895 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
897 if (geneve->cfg.df == GENEVE_DF_SET) {
899 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
900 struct ethhdr *eth = eth_hdr(skb);
902 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
904 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
905 struct iphdr *iph = ip_hdr(skb);
907 if (iph->frag_off & htons(IP_DF))
913 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
914 geneve->cfg.inner_proto_inherit);
918 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, saddr, info->key.u.ipv4.dst,
919 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
920 !net_eq(geneve->net, dev_net(geneve->dev)),
921 !(info->key.tun_flags & TUNNEL_CSUM));
925 #if IS_ENABLED(CONFIG_IPV6)
926 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
927 struct geneve_dev *geneve,
928 const struct ip_tunnel_info *info)
930 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
931 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
932 const struct ip_tunnel_key *key = &info->key;
933 struct dst_entry *dst = NULL;
934 struct in6_addr saddr;
940 if (!skb_vlan_inet_prepare(skb))
946 use_cache = ip_tunnel_dst_cache_usable(skb, info);
947 prio = geneve_get_dsfield(skb, dev, info, &use_cache);
948 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
950 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
952 geneve->cfg.info.key.tp_dst, prio,
954 (struct dst_cache *)&info->dst_cache : NULL);
958 err = skb_tunnel_check_pmtu(skb, dst,
959 GENEVE_IPV6_HLEN + info->options_len,
960 netif_is_any_bridge_port(dev));
965 struct ip_tunnel_info *info = skb_tunnel_info(skb);
968 struct ip_tunnel_info *unclone;
970 unclone = skb_tunnel_info_unclone(skb);
971 if (unlikely(!unclone)) {
976 unclone->key.u.ipv6.dst = saddr;
977 unclone->key.u.ipv6.src = info->key.u.ipv6.dst;
980 if (!pskb_may_pull(skb, ETH_HLEN)) {
985 skb->protocol = eth_type_trans(skb, geneve->dev);
991 prio = ip_tunnel_ecn_encap(prio, ip_hdr(skb), skb);
992 if (geneve->cfg.collect_md) {
995 if (geneve->cfg.ttl_inherit)
996 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
999 ttl = ttl ? : ip6_dst_hoplimit(dst);
1001 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
1002 geneve->cfg.inner_proto_inherit);
1006 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1007 &saddr, &key->u.ipv6.dst, prio, ttl,
1008 info->key.label, sport, geneve->cfg.info.key.tp_dst,
1009 !(info->key.tun_flags & TUNNEL_CSUM));
1014 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1016 struct geneve_dev *geneve = netdev_priv(dev);
1017 struct ip_tunnel_info *info = NULL;
1020 if (geneve->cfg.collect_md) {
1021 info = skb_tunnel_info(skb);
1022 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1023 netdev_dbg(dev, "no tunnel metadata\n");
1025 DEV_STATS_INC(dev, tx_dropped);
1026 return NETDEV_TX_OK;
1029 info = &geneve->cfg.info;
1033 #if IS_ENABLED(CONFIG_IPV6)
1034 if (info->mode & IP_TUNNEL_INFO_IPV6)
1035 err = geneve6_xmit_skb(skb, dev, geneve, info);
1038 err = geneve_xmit_skb(skb, dev, geneve, info);
1042 return NETDEV_TX_OK;
1044 if (err != -EMSGSIZE)
1048 DEV_STATS_INC(dev, collisions);
1049 else if (err == -ENETUNREACH)
1050 DEV_STATS_INC(dev, tx_carrier_errors);
1052 DEV_STATS_INC(dev, tx_errors);
1053 return NETDEV_TX_OK;
1056 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1058 if (new_mtu > dev->max_mtu)
1059 new_mtu = dev->max_mtu;
1060 else if (new_mtu < dev->min_mtu)
1061 new_mtu = dev->min_mtu;
1067 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1069 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1070 struct geneve_dev *geneve = netdev_priv(dev);
1073 if (ip_tunnel_info_af(info) == AF_INET) {
1075 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1083 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1084 tos = geneve_get_dsfield(skb, dev, info, &use_cache);
1085 sport = udp_flow_src_port(geneve->net, skb,
1086 1, USHRT_MAX, true);
1088 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
1090 sport, geneve->cfg.info.key.tp_dst,
1092 use_cache ? &info->dst_cache : NULL);
1097 info->key.u.ipv4.src = saddr;
1098 #if IS_ENABLED(CONFIG_IPV6)
1099 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1100 struct dst_entry *dst;
1101 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1102 struct in6_addr saddr;
1109 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1110 prio = geneve_get_dsfield(skb, dev, info, &use_cache);
1111 sport = udp_flow_src_port(geneve->net, skb,
1112 1, USHRT_MAX, true);
1114 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
1115 &saddr, &info->key, sport,
1116 geneve->cfg.info.key.tp_dst, prio,
1117 use_cache ? &info->dst_cache : NULL);
1119 return PTR_ERR(dst);
1122 info->key.u.ipv6.src = saddr;
1128 info->key.tp_src = sport;
1129 info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1133 static const struct net_device_ops geneve_netdev_ops = {
1134 .ndo_init = geneve_init,
1135 .ndo_uninit = geneve_uninit,
1136 .ndo_open = geneve_open,
1137 .ndo_stop = geneve_stop,
1138 .ndo_start_xmit = geneve_xmit,
1139 .ndo_get_stats64 = dev_get_tstats64,
1140 .ndo_change_mtu = geneve_change_mtu,
1141 .ndo_validate_addr = eth_validate_addr,
1142 .ndo_set_mac_address = eth_mac_addr,
1143 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1146 static void geneve_get_drvinfo(struct net_device *dev,
1147 struct ethtool_drvinfo *drvinfo)
1149 strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1150 strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1153 static const struct ethtool_ops geneve_ethtool_ops = {
1154 .get_drvinfo = geneve_get_drvinfo,
1155 .get_link = ethtool_op_get_link,
1158 /* Info for udev, that this is a virtual tunnel endpoint */
1159 static struct device_type geneve_type = {
1163 /* Calls the ndo_udp_tunnel_add of the caller in order to
1164 * supply the listening GENEVE udp ports. Callers are expected
1165 * to implement the ndo_udp_tunnel_add.
1167 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1169 struct net *net = dev_net(dev);
1170 struct geneve_net *gn = net_generic(net, geneve_net_id);
1171 struct geneve_sock *gs;
1174 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1176 udp_tunnel_push_rx_port(dev, gs->sock,
1177 UDP_TUNNEL_TYPE_GENEVE);
1179 udp_tunnel_drop_rx_port(dev, gs->sock,
1180 UDP_TUNNEL_TYPE_GENEVE);
1186 /* Initialize the device structure. */
1187 static void geneve_setup(struct net_device *dev)
1191 dev->netdev_ops = &geneve_netdev_ops;
1192 dev->ethtool_ops = &geneve_ethtool_ops;
1193 dev->needs_free_netdev = true;
1195 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1197 dev->features |= NETIF_F_LLTX;
1198 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1199 dev->features |= NETIF_F_RXCSUM;
1200 dev->features |= NETIF_F_GSO_SOFTWARE;
1202 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1203 dev->hw_features |= NETIF_F_RXCSUM;
1204 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1206 /* MTU range: 68 - (something less than 65535) */
1207 dev->min_mtu = ETH_MIN_MTU;
1208 /* The max_mtu calculation does not take account of GENEVE
1209 * options, to avoid excluding potentially valid
1210 * configurations. This will be further reduced by IPvX hdr size.
1212 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1214 netif_keep_dst(dev);
1215 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1216 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1217 eth_hw_addr_random(dev);
1220 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1221 [IFLA_GENEVE_UNSPEC] = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1222 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1223 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1224 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1225 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1226 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1227 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1228 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1229 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1230 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1231 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1232 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1233 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1234 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1235 [IFLA_GENEVE_INNER_PROTO_INHERIT] = { .type = NLA_FLAG },
1238 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1239 struct netlink_ext_ack *extack)
1241 if (tb[IFLA_ADDRESS]) {
1242 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1243 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1244 "Provided link layer address is not Ethernet");
1248 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1249 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1250 "Provided Ethernet address is not unicast");
1251 return -EADDRNOTAVAIL;
1256 NL_SET_ERR_MSG(extack,
1257 "Not enough attributes provided to perform the operation");
1261 if (data[IFLA_GENEVE_ID]) {
1262 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1264 if (vni >= GENEVE_N_VID) {
1265 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1266 "Geneve ID must be lower than 16777216");
1271 if (data[IFLA_GENEVE_DF]) {
1272 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1274 if (df < 0 || df > GENEVE_DF_MAX) {
1275 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1276 "Invalid DF attribute");
1284 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1285 const struct ip_tunnel_info *info,
1286 bool *tun_on_same_port,
1287 bool *tun_collect_md)
1289 struct geneve_dev *geneve, *t = NULL;
1291 *tun_on_same_port = false;
1292 *tun_collect_md = false;
1293 list_for_each_entry(geneve, &gn->geneve_list, next) {
1294 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1295 *tun_collect_md = geneve->cfg.collect_md;
1296 *tun_on_same_port = true;
1298 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1299 info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1300 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1306 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1308 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1309 info->key.ttl || info->key.label || info->key.tp_src ||
1310 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1313 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1314 struct ip_tunnel_info *b)
1316 if (ip_tunnel_info_af(a) == AF_INET)
1317 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1319 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1322 static int geneve_configure(struct net *net, struct net_device *dev,
1323 struct netlink_ext_ack *extack,
1324 const struct geneve_config *cfg)
1326 struct geneve_net *gn = net_generic(net, geneve_net_id);
1327 struct geneve_dev *t, *geneve = netdev_priv(dev);
1328 const struct ip_tunnel_info *info = &cfg->info;
1329 bool tun_collect_md, tun_on_same_port;
1332 if (cfg->collect_md && !is_tnl_info_zero(info)) {
1333 NL_SET_ERR_MSG(extack,
1334 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1341 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1345 /* make enough headroom for basic scenario */
1346 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1347 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1348 encap_len += sizeof(struct iphdr);
1349 dev->max_mtu -= sizeof(struct iphdr);
1351 encap_len += sizeof(struct ipv6hdr);
1352 dev->max_mtu -= sizeof(struct ipv6hdr);
1354 dev->needed_headroom = encap_len + ETH_HLEN;
1356 if (cfg->collect_md) {
1357 if (tun_on_same_port) {
1358 NL_SET_ERR_MSG(extack,
1359 "There can be only one externally controlled device on a destination port");
1363 if (tun_collect_md) {
1364 NL_SET_ERR_MSG(extack,
1365 "There already exists an externally controlled device on this destination port");
1370 dst_cache_reset(&geneve->cfg.info.dst_cache);
1371 memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1373 if (geneve->cfg.inner_proto_inherit) {
1374 dev->header_ops = NULL;
1375 dev->type = ARPHRD_NONE;
1376 dev->hard_header_len = 0;
1378 dev->flags = IFF_POINTOPOINT | IFF_NOARP;
1381 err = register_netdevice(dev);
1385 list_add(&geneve->next, &gn->geneve_list);
1389 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1391 memset(info, 0, sizeof(*info));
1392 info->key.tp_dst = htons(dst_port);
1395 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1396 struct netlink_ext_ack *extack,
1397 struct geneve_config *cfg, bool changelink)
1399 struct ip_tunnel_info *info = &cfg->info;
1402 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1403 NL_SET_ERR_MSG(extack,
1404 "Cannot specify both IPv4 and IPv6 Remote addresses");
1408 if (data[IFLA_GENEVE_REMOTE]) {
1409 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1410 attrtype = IFLA_GENEVE_REMOTE;
1414 info->key.u.ipv4.dst =
1415 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1417 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1418 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1419 "Remote IPv4 address cannot be Multicast");
1424 if (data[IFLA_GENEVE_REMOTE6]) {
1425 #if IS_ENABLED(CONFIG_IPV6)
1426 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1427 attrtype = IFLA_GENEVE_REMOTE6;
1431 info->mode = IP_TUNNEL_INFO_IPV6;
1432 info->key.u.ipv6.dst =
1433 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1435 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1436 IPV6_ADDR_LINKLOCAL) {
1437 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1438 "Remote IPv6 address cannot be link-local");
1441 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1442 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1443 "Remote IPv6 address cannot be Multicast");
1446 info->key.tun_flags |= TUNNEL_CSUM;
1447 cfg->use_udp6_rx_checksums = true;
1449 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1450 "IPv6 support not enabled in the kernel");
1451 return -EPFNOSUPPORT;
1455 if (data[IFLA_GENEVE_ID]) {
1460 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1461 tvni[0] = (vni & 0x00ff0000) >> 16;
1462 tvni[1] = (vni & 0x0000ff00) >> 8;
1463 tvni[2] = vni & 0x000000ff;
1465 tunid = vni_to_tunnel_id(tvni);
1466 if (changelink && (tunid != info->key.tun_id)) {
1467 attrtype = IFLA_GENEVE_ID;
1470 info->key.tun_id = tunid;
1473 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1474 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1475 cfg->ttl_inherit = true;
1477 cfg->ttl_inherit = false;
1478 } else if (data[IFLA_GENEVE_TTL]) {
1479 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1480 cfg->ttl_inherit = false;
1483 if (data[IFLA_GENEVE_TOS])
1484 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1486 if (data[IFLA_GENEVE_DF])
1487 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1489 if (data[IFLA_GENEVE_LABEL]) {
1490 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1491 IPV6_FLOWLABEL_MASK;
1492 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1493 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1494 "Label attribute only applies for IPv6 Geneve devices");
1499 if (data[IFLA_GENEVE_PORT]) {
1501 attrtype = IFLA_GENEVE_PORT;
1504 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1507 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1509 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1512 cfg->collect_md = true;
1515 if (data[IFLA_GENEVE_UDP_CSUM]) {
1517 attrtype = IFLA_GENEVE_UDP_CSUM;
1520 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1521 info->key.tun_flags |= TUNNEL_CSUM;
1524 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1525 #if IS_ENABLED(CONFIG_IPV6)
1527 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1530 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1531 info->key.tun_flags &= ~TUNNEL_CSUM;
1533 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1534 "IPv6 support not enabled in the kernel");
1535 return -EPFNOSUPPORT;
1539 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1540 #if IS_ENABLED(CONFIG_IPV6)
1542 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1545 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1546 cfg->use_udp6_rx_checksums = false;
1548 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1549 "IPv6 support not enabled in the kernel");
1550 return -EPFNOSUPPORT;
1554 if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1556 attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1559 cfg->inner_proto_inherit = true;
1564 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1565 "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1569 static void geneve_link_config(struct net_device *dev,
1570 struct ip_tunnel_info *info, struct nlattr *tb[])
1572 struct geneve_dev *geneve = netdev_priv(dev);
1576 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1580 switch (ip_tunnel_info_af(info)) {
1582 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1583 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1585 if (!IS_ERR(rt) && rt->dst.dev) {
1586 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1591 #if IS_ENABLED(CONFIG_IPV6)
1593 struct rt6_info *rt;
1595 if (!__in6_dev_get(dev))
1598 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1601 if (rt && rt->dst.dev)
1602 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1612 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1615 static int geneve_newlink(struct net *net, struct net_device *dev,
1616 struct nlattr *tb[], struct nlattr *data[],
1617 struct netlink_ext_ack *extack)
1619 struct geneve_config cfg = {
1620 .df = GENEVE_DF_UNSET,
1621 .use_udp6_rx_checksums = false,
1622 .ttl_inherit = false,
1623 .collect_md = false,
1627 init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1628 err = geneve_nl2info(tb, data, extack, &cfg, false);
1632 err = geneve_configure(net, dev, extack, &cfg);
1636 geneve_link_config(dev, &cfg.info, tb);
1641 /* Quiesces the geneve device data path for both TX and RX.
1643 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1644 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1645 * to complete for the existing set of in-flight packets to be transmitted,
1646 * then we would have quiesced the transmit data path. All the future packets
1647 * will get dropped until we unquiesce the data path.
1649 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1650 * if we set that to NULL under RCU and wait for synchronize_net() to
1651 * complete, then we would have quiesced the receive data path.
1653 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1654 struct geneve_sock **gs6)
1656 *gs4 = rtnl_dereference(geneve->sock4);
1657 rcu_assign_pointer(geneve->sock4, NULL);
1659 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1660 #if IS_ENABLED(CONFIG_IPV6)
1661 *gs6 = rtnl_dereference(geneve->sock6);
1662 rcu_assign_pointer(geneve->sock6, NULL);
1664 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1671 /* Resumes the geneve device data path for both TX and RX. */
1672 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1673 struct geneve_sock __maybe_unused *gs6)
1675 rcu_assign_pointer(geneve->sock4, gs4);
1677 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1678 #if IS_ENABLED(CONFIG_IPV6)
1679 rcu_assign_pointer(geneve->sock6, gs6);
1681 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1686 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1687 struct nlattr *data[],
1688 struct netlink_ext_ack *extack)
1690 struct geneve_dev *geneve = netdev_priv(dev);
1691 struct geneve_sock *gs4, *gs6;
1692 struct geneve_config cfg;
1695 /* If the geneve device is configured for metadata (or externally
1696 * controlled, for example, OVS), then nothing can be changed.
1698 if (geneve->cfg.collect_md)
1701 /* Start with the existing info. */
1702 memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1703 err = geneve_nl2info(tb, data, extack, &cfg, true);
1707 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1708 dst_cache_reset(&cfg.info.dst_cache);
1709 geneve_link_config(dev, &cfg.info, tb);
1712 geneve_quiesce(geneve, &gs4, &gs6);
1713 memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1714 geneve_unquiesce(geneve, gs4, gs6);
1719 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1721 struct geneve_dev *geneve = netdev_priv(dev);
1723 list_del(&geneve->next);
1724 unregister_netdevice_queue(dev, head);
1727 static size_t geneve_get_size(const struct net_device *dev)
1729 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1730 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1731 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1732 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1733 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1734 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1735 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1736 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1737 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1738 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1739 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1740 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1741 nla_total_size(0) + /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1745 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1747 struct geneve_dev *geneve = netdev_priv(dev);
1748 struct ip_tunnel_info *info = &geneve->cfg.info;
1749 bool ttl_inherit = geneve->cfg.ttl_inherit;
1750 bool metadata = geneve->cfg.collect_md;
1754 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1755 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1756 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1757 goto nla_put_failure;
1759 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1760 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1761 info->key.u.ipv4.dst))
1762 goto nla_put_failure;
1763 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1764 !!(info->key.tun_flags & TUNNEL_CSUM)))
1765 goto nla_put_failure;
1767 #if IS_ENABLED(CONFIG_IPV6)
1768 } else if (!metadata) {
1769 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1770 &info->key.u.ipv6.dst))
1771 goto nla_put_failure;
1772 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1773 !(info->key.tun_flags & TUNNEL_CSUM)))
1774 goto nla_put_failure;
1778 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1779 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1780 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1781 goto nla_put_failure;
1783 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1784 goto nla_put_failure;
1786 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1787 goto nla_put_failure;
1789 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1790 goto nla_put_failure;
1792 #if IS_ENABLED(CONFIG_IPV6)
1793 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1794 !geneve->cfg.use_udp6_rx_checksums))
1795 goto nla_put_failure;
1798 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1799 goto nla_put_failure;
1801 if (geneve->cfg.inner_proto_inherit &&
1802 nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1803 goto nla_put_failure;
1811 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1813 .maxtype = IFLA_GENEVE_MAX,
1814 .policy = geneve_policy,
1815 .priv_size = sizeof(struct geneve_dev),
1816 .setup = geneve_setup,
1817 .validate = geneve_validate,
1818 .newlink = geneve_newlink,
1819 .changelink = geneve_changelink,
1820 .dellink = geneve_dellink,
1821 .get_size = geneve_get_size,
1822 .fill_info = geneve_fill_info,
1825 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1826 u8 name_assign_type, u16 dst_port)
1828 struct nlattr *tb[IFLA_MAX + 1];
1829 struct net_device *dev;
1830 LIST_HEAD(list_kill);
1832 struct geneve_config cfg = {
1833 .df = GENEVE_DF_UNSET,
1834 .use_udp6_rx_checksums = true,
1835 .ttl_inherit = false,
1839 memset(tb, 0, sizeof(tb));
1840 dev = rtnl_create_link(net, name, name_assign_type,
1841 &geneve_link_ops, tb, NULL);
1845 init_tnl_info(&cfg.info, dst_port);
1846 err = geneve_configure(net, dev, NULL, &cfg);
1849 return ERR_PTR(err);
1852 /* openvswitch users expect packet sizes to be unrestricted,
1853 * so set the largest MTU we can.
1855 err = geneve_change_mtu(dev, IP_MAX_MTU);
1859 err = rtnl_configure_link(dev, NULL, 0, NULL);
1865 geneve_dellink(dev, &list_kill);
1866 unregister_netdevice_many(&list_kill);
1867 return ERR_PTR(err);
1869 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1871 static int geneve_netdevice_event(struct notifier_block *unused,
1872 unsigned long event, void *ptr)
1874 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1876 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1877 geneve_offload_rx_ports(dev, true);
1878 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1879 geneve_offload_rx_ports(dev, false);
1884 static struct notifier_block geneve_notifier_block __read_mostly = {
1885 .notifier_call = geneve_netdevice_event,
1888 static __net_init int geneve_init_net(struct net *net)
1890 struct geneve_net *gn = net_generic(net, geneve_net_id);
1892 INIT_LIST_HEAD(&gn->geneve_list);
1893 INIT_LIST_HEAD(&gn->sock_list);
1897 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1899 struct geneve_net *gn = net_generic(net, geneve_net_id);
1900 struct geneve_dev *geneve, *next;
1901 struct net_device *dev, *aux;
1903 /* gather any geneve devices that were moved into this ns */
1904 for_each_netdev_safe(net, dev, aux)
1905 if (dev->rtnl_link_ops == &geneve_link_ops)
1906 unregister_netdevice_queue(dev, head);
1908 /* now gather any other geneve devices that were created in this ns */
1909 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1910 /* If geneve->dev is in the same netns, it was already added
1911 * to the list by the previous loop.
1913 if (!net_eq(dev_net(geneve->dev), net))
1914 unregister_netdevice_queue(geneve->dev, head);
1918 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1924 list_for_each_entry(net, net_list, exit_list)
1925 geneve_destroy_tunnels(net, &list);
1927 /* unregister the devices gathered above */
1928 unregister_netdevice_many(&list);
1931 list_for_each_entry(net, net_list, exit_list) {
1932 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1934 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1938 static struct pernet_operations geneve_net_ops = {
1939 .init = geneve_init_net,
1940 .exit_batch = geneve_exit_batch_net,
1941 .id = &geneve_net_id,
1942 .size = sizeof(struct geneve_net),
1945 static int __init geneve_init_module(void)
1949 rc = register_pernet_subsys(&geneve_net_ops);
1953 rc = register_netdevice_notifier(&geneve_notifier_block);
1957 rc = rtnl_link_register(&geneve_link_ops);
1963 unregister_netdevice_notifier(&geneve_notifier_block);
1965 unregister_pernet_subsys(&geneve_net_ops);
1969 late_initcall(geneve_init_module);
1971 static void __exit geneve_cleanup_module(void)
1973 rtnl_link_unregister(&geneve_link_ops);
1974 unregister_netdevice_notifier(&geneve_notifier_block);
1975 unregister_pernet_subsys(&geneve_net_ops);
1977 module_exit(geneve_cleanup_module);
1979 MODULE_LICENSE("GPL");
1980 MODULE_VERSION(GENEVE_NETDEV_VER);
1981 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1982 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1983 MODULE_ALIAS_RTNL_LINK("geneve");