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 err = gro_cells_init(&geneve->gro_cells, dev);
340 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
342 gro_cells_destroy(&geneve->gro_cells);
345 netdev_lockdep_set_classes(dev);
349 static void geneve_uninit(struct net_device *dev)
351 struct geneve_dev *geneve = netdev_priv(dev);
353 dst_cache_destroy(&geneve->cfg.info.dst_cache);
354 gro_cells_destroy(&geneve->gro_cells);
357 /* Callback from net/ipv4/udp.c to receive packets */
358 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
360 struct genevehdr *geneveh;
361 struct geneve_dev *geneve;
362 struct geneve_sock *gs;
366 /* Need UDP and Geneve header to be present */
367 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
370 /* Return packets with reserved bits set */
371 geneveh = geneve_hdr(skb);
372 if (unlikely(geneveh->ver != GENEVE_VER))
375 gs = rcu_dereference_sk_user_data(sk);
379 geneve = geneve_lookup_skb(gs, skb);
383 inner_proto = geneveh->proto_type;
385 if (unlikely((!geneve->cfg.inner_proto_inherit &&
386 inner_proto != htons(ETH_P_TEB)))) {
387 DEV_STATS_INC(geneve->dev, rx_dropped);
391 opts_len = geneveh->opt_len * 4;
392 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
393 !net_eq(geneve->net, dev_net(geneve->dev)))) {
394 DEV_STATS_INC(geneve->dev, rx_dropped);
398 geneve_rx(geneve, gs, skb);
402 /* Consume bad packet */
407 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
408 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
410 struct genevehdr *geneveh;
411 struct geneve_sock *gs;
412 u8 zero_vni[3] = { 0 };
415 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
418 geneveh = geneve_hdr(skb);
419 if (geneveh->ver != GENEVE_VER)
422 if (geneveh->proto_type != htons(ETH_P_TEB))
425 gs = rcu_dereference_sk_user_data(sk);
429 if (geneve_get_sk_family(gs) == AF_INET) {
430 struct iphdr *iph = ip_hdr(skb);
433 if (!gs->collect_md) {
434 vni = geneve_hdr(skb)->vni;
438 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
441 #if IS_ENABLED(CONFIG_IPV6)
442 if (geneve_get_sk_family(gs) == AF_INET6) {
443 struct ipv6hdr *ip6h = ipv6_hdr(skb);
444 struct in6_addr addr6;
446 memset(&addr6, 0, sizeof(struct in6_addr));
448 if (!gs->collect_md) {
449 vni = geneve_hdr(skb)->vni;
453 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
457 return -EPFNOSUPPORT;
460 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
461 __be16 port, bool ipv6_rx_csum)
464 struct udp_port_cfg udp_conf;
467 memset(&udp_conf, 0, sizeof(udp_conf));
470 udp_conf.family = AF_INET6;
471 udp_conf.ipv6_v6only = 1;
472 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
474 udp_conf.family = AF_INET;
475 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
478 udp_conf.local_udp_port = port;
480 /* Open UDP socket */
481 err = udp_sock_create(net, &udp_conf, &sock);
485 udp_allow_gso(sock->sk);
489 static int geneve_hlen(struct genevehdr *gh)
491 return sizeof(*gh) + gh->opt_len * 4;
494 static struct sk_buff *geneve_gro_receive(struct sock *sk,
495 struct list_head *head,
498 struct sk_buff *pp = NULL;
500 struct genevehdr *gh, *gh2;
501 unsigned int hlen, gh_len, off_gnv;
502 const struct packet_offload *ptype;
506 off_gnv = skb_gro_offset(skb);
507 hlen = off_gnv + sizeof(*gh);
508 gh = skb_gro_header(skb, hlen, off_gnv);
512 if (gh->ver != GENEVE_VER || gh->oam)
514 gh_len = geneve_hlen(gh);
516 hlen = off_gnv + gh_len;
517 if (!skb_gro_may_pull(skb, hlen)) {
518 gh = skb_gro_header_slow(skb, hlen, off_gnv);
523 list_for_each_entry(p, head, list) {
524 if (!NAPI_GRO_CB(p)->same_flow)
527 gh2 = (struct genevehdr *)(p->data + off_gnv);
528 if (gh->opt_len != gh2->opt_len ||
529 memcmp(gh, gh2, gh_len)) {
530 NAPI_GRO_CB(p)->same_flow = 0;
535 skb_gro_pull(skb, gh_len);
536 skb_gro_postpull_rcsum(skb, gh, gh_len);
537 type = gh->proto_type;
538 if (likely(type == htons(ETH_P_TEB)))
539 return call_gro_receive(eth_gro_receive, head, skb);
541 ptype = gro_find_receive_by_type(type);
545 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
549 skb_gro_flush_final(skb, pp, flush);
554 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
557 struct genevehdr *gh;
558 struct packet_offload *ptype;
563 gh = (struct genevehdr *)(skb->data + nhoff);
564 gh_len = geneve_hlen(gh);
565 type = gh->proto_type;
567 /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */
568 if (likely(type == htons(ETH_P_TEB)))
569 return eth_gro_complete(skb, nhoff + gh_len);
571 ptype = gro_find_complete_by_type(type);
573 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
575 skb_set_inner_mac_header(skb, nhoff + gh_len);
580 /* Create new listen socket if needed */
581 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
582 bool ipv6, bool ipv6_rx_csum)
584 struct geneve_net *gn = net_generic(net, geneve_net_id);
585 struct geneve_sock *gs;
587 struct udp_tunnel_sock_cfg tunnel_cfg;
590 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
592 return ERR_PTR(-ENOMEM);
594 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
597 return ERR_CAST(sock);
602 for (h = 0; h < VNI_HASH_SIZE; ++h)
603 INIT_HLIST_HEAD(&gs->vni_list[h]);
605 /* Initialize the geneve udp offloads structure */
606 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
608 /* Mark socket as an encapsulation socket */
609 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
610 tunnel_cfg.sk_user_data = gs;
611 tunnel_cfg.encap_type = 1;
612 tunnel_cfg.gro_receive = geneve_gro_receive;
613 tunnel_cfg.gro_complete = geneve_gro_complete;
614 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
615 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
616 tunnel_cfg.encap_destroy = NULL;
617 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
618 list_add(&gs->list, &gn->sock_list);
622 static void __geneve_sock_release(struct geneve_sock *gs)
624 if (!gs || --gs->refcnt)
628 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
629 udp_tunnel_sock_release(gs->sock);
633 static void geneve_sock_release(struct geneve_dev *geneve)
635 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
636 #if IS_ENABLED(CONFIG_IPV6)
637 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
639 rcu_assign_pointer(geneve->sock6, NULL);
642 rcu_assign_pointer(geneve->sock4, NULL);
645 __geneve_sock_release(gs4);
646 #if IS_ENABLED(CONFIG_IPV6)
647 __geneve_sock_release(gs6);
651 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
655 struct geneve_sock *gs;
657 list_for_each_entry(gs, &gn->sock_list, list) {
658 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
659 geneve_get_sk_family(gs) == family) {
666 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
668 struct net *net = geneve->net;
669 struct geneve_net *gn = net_generic(net, geneve_net_id);
670 struct geneve_dev_node *node;
671 struct geneve_sock *gs;
675 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
681 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
682 geneve->cfg.use_udp6_rx_checksums);
687 gs->collect_md = geneve->cfg.collect_md;
688 #if IS_ENABLED(CONFIG_IPV6)
690 rcu_assign_pointer(geneve->sock6, gs);
691 node = &geneve->hlist6;
695 rcu_assign_pointer(geneve->sock4, gs);
696 node = &geneve->hlist4;
698 node->geneve = geneve;
700 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
701 hash = geneve_net_vni_hash(vni);
702 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
706 static int geneve_open(struct net_device *dev)
708 struct geneve_dev *geneve = netdev_priv(dev);
709 bool metadata = geneve->cfg.collect_md;
713 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
714 ipv4 = !ipv6 || metadata;
715 #if IS_ENABLED(CONFIG_IPV6)
717 ret = geneve_sock_add(geneve, true);
718 if (ret < 0 && ret != -EAFNOSUPPORT)
723 ret = geneve_sock_add(geneve, false);
725 geneve_sock_release(geneve);
730 static int geneve_stop(struct net_device *dev)
732 struct geneve_dev *geneve = netdev_priv(dev);
734 hlist_del_init_rcu(&geneve->hlist4.hlist);
735 #if IS_ENABLED(CONFIG_IPV6)
736 hlist_del_init_rcu(&geneve->hlist6.hlist);
738 geneve_sock_release(geneve);
742 static void geneve_build_header(struct genevehdr *geneveh,
743 const struct ip_tunnel_info *info,
746 geneveh->ver = GENEVE_VER;
747 geneveh->opt_len = info->options_len / 4;
748 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
749 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
751 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
752 geneveh->proto_type = inner_proto;
755 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
756 ip_tunnel_info_opts_get(geneveh->options, info);
759 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
760 const struct ip_tunnel_info *info,
761 bool xnet, int ip_hdr_len,
762 bool inner_proto_inherit)
764 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
765 struct genevehdr *gnvh;
770 skb_reset_mac_header(skb);
771 skb_scrub_packet(skb, xnet);
773 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
774 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
775 err = skb_cow_head(skb, min_headroom);
779 err = udp_tunnel_handle_offloads(skb, udp_sum);
783 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
784 inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB);
785 geneve_build_header(gnvh, info, inner_proto);
786 skb_set_inner_protocol(skb, inner_proto);
794 static u8 geneve_get_dsfield(struct sk_buff *skb, struct net_device *dev,
795 const struct ip_tunnel_info *info,
798 struct geneve_dev *geneve = netdev_priv(dev);
801 dsfield = info->key.tos;
802 if (dsfield == 1 && !geneve->cfg.collect_md) {
803 dsfield = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
810 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
811 struct geneve_dev *geneve,
812 const struct ip_tunnel_info *info)
814 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
815 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
816 const struct ip_tunnel_key *key = &info->key;
825 if (!pskb_inet_may_pull(skb))
831 use_cache = ip_tunnel_dst_cache_usable(skb, info);
832 tos = geneve_get_dsfield(skb, dev, info, &use_cache);
833 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
835 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
837 sport, geneve->cfg.info.key.tp_dst, tos,
839 (struct dst_cache *)&info->dst_cache : NULL);
843 err = skb_tunnel_check_pmtu(skb, &rt->dst,
844 GENEVE_IPV4_HLEN + info->options_len,
845 netif_is_any_bridge_port(dev));
847 dst_release(&rt->dst);
850 struct ip_tunnel_info *info;
852 info = skb_tunnel_info(skb);
854 struct ip_tunnel_info *unclone;
856 unclone = skb_tunnel_info_unclone(skb);
857 if (unlikely(!unclone)) {
858 dst_release(&rt->dst);
862 unclone->key.u.ipv4.dst = saddr;
863 unclone->key.u.ipv4.src = info->key.u.ipv4.dst;
866 if (!pskb_may_pull(skb, ETH_HLEN)) {
867 dst_release(&rt->dst);
871 skb->protocol = eth_type_trans(skb, geneve->dev);
873 dst_release(&rt->dst);
877 tos = ip_tunnel_ecn_encap(tos, ip_hdr(skb), skb);
878 if (geneve->cfg.collect_md) {
881 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
883 if (geneve->cfg.ttl_inherit)
884 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
887 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
889 if (geneve->cfg.df == GENEVE_DF_SET) {
891 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
892 struct ethhdr *eth = eth_hdr(skb);
894 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
896 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
897 struct iphdr *iph = ip_hdr(skb);
899 if (iph->frag_off & htons(IP_DF))
905 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
906 geneve->cfg.inner_proto_inherit);
910 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, saddr, info->key.u.ipv4.dst,
911 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
912 !net_eq(geneve->net, dev_net(geneve->dev)),
913 !(info->key.tun_flags & TUNNEL_CSUM));
917 #if IS_ENABLED(CONFIG_IPV6)
918 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
919 struct geneve_dev *geneve,
920 const struct ip_tunnel_info *info)
922 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
923 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
924 const struct ip_tunnel_key *key = &info->key;
925 struct dst_entry *dst = NULL;
926 struct in6_addr saddr;
932 if (!pskb_inet_may_pull(skb))
938 use_cache = ip_tunnel_dst_cache_usable(skb, info);
939 prio = geneve_get_dsfield(skb, dev, info, &use_cache);
940 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
942 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
944 geneve->cfg.info.key.tp_dst, prio,
946 (struct dst_cache *)&info->dst_cache : NULL);
950 err = skb_tunnel_check_pmtu(skb, dst,
951 GENEVE_IPV6_HLEN + info->options_len,
952 netif_is_any_bridge_port(dev));
957 struct ip_tunnel_info *info = skb_tunnel_info(skb);
960 struct ip_tunnel_info *unclone;
962 unclone = skb_tunnel_info_unclone(skb);
963 if (unlikely(!unclone)) {
968 unclone->key.u.ipv6.dst = saddr;
969 unclone->key.u.ipv6.src = info->key.u.ipv6.dst;
972 if (!pskb_may_pull(skb, ETH_HLEN)) {
977 skb->protocol = eth_type_trans(skb, geneve->dev);
983 prio = ip_tunnel_ecn_encap(prio, ip_hdr(skb), skb);
984 if (geneve->cfg.collect_md) {
987 if (geneve->cfg.ttl_inherit)
988 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
991 ttl = ttl ? : ip6_dst_hoplimit(dst);
993 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
994 geneve->cfg.inner_proto_inherit);
998 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
999 &saddr, &key->u.ipv6.dst, prio, ttl,
1000 info->key.label, sport, geneve->cfg.info.key.tp_dst,
1001 !(info->key.tun_flags & TUNNEL_CSUM));
1006 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1008 struct geneve_dev *geneve = netdev_priv(dev);
1009 struct ip_tunnel_info *info = NULL;
1012 if (geneve->cfg.collect_md) {
1013 info = skb_tunnel_info(skb);
1014 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1015 netdev_dbg(dev, "no tunnel metadata\n");
1017 DEV_STATS_INC(dev, tx_dropped);
1018 return NETDEV_TX_OK;
1021 info = &geneve->cfg.info;
1025 #if IS_ENABLED(CONFIG_IPV6)
1026 if (info->mode & IP_TUNNEL_INFO_IPV6)
1027 err = geneve6_xmit_skb(skb, dev, geneve, info);
1030 err = geneve_xmit_skb(skb, dev, geneve, info);
1034 return NETDEV_TX_OK;
1036 if (err != -EMSGSIZE)
1040 DEV_STATS_INC(dev, collisions);
1041 else if (err == -ENETUNREACH)
1042 DEV_STATS_INC(dev, tx_carrier_errors);
1044 DEV_STATS_INC(dev, tx_errors);
1045 return NETDEV_TX_OK;
1048 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1050 if (new_mtu > dev->max_mtu)
1051 new_mtu = dev->max_mtu;
1052 else if (new_mtu < dev->min_mtu)
1053 new_mtu = dev->min_mtu;
1059 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1061 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1062 struct geneve_dev *geneve = netdev_priv(dev);
1065 if (ip_tunnel_info_af(info) == AF_INET) {
1067 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1075 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1076 tos = geneve_get_dsfield(skb, dev, info, &use_cache);
1077 sport = udp_flow_src_port(geneve->net, skb,
1078 1, USHRT_MAX, true);
1080 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr,
1082 sport, geneve->cfg.info.key.tp_dst,
1084 use_cache ? &info->dst_cache : NULL);
1089 info->key.u.ipv4.src = saddr;
1090 #if IS_ENABLED(CONFIG_IPV6)
1091 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1092 struct dst_entry *dst;
1093 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1094 struct in6_addr saddr;
1101 use_cache = ip_tunnel_dst_cache_usable(skb, info);
1102 prio = geneve_get_dsfield(skb, dev, info, &use_cache);
1103 sport = udp_flow_src_port(geneve->net, skb,
1104 1, USHRT_MAX, true);
1106 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0,
1107 &saddr, &info->key, sport,
1108 geneve->cfg.info.key.tp_dst, prio,
1109 use_cache ? &info->dst_cache : NULL);
1111 return PTR_ERR(dst);
1114 info->key.u.ipv6.src = saddr;
1120 info->key.tp_src = sport;
1121 info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1125 static const struct net_device_ops geneve_netdev_ops = {
1126 .ndo_init = geneve_init,
1127 .ndo_uninit = geneve_uninit,
1128 .ndo_open = geneve_open,
1129 .ndo_stop = geneve_stop,
1130 .ndo_start_xmit = geneve_xmit,
1131 .ndo_change_mtu = geneve_change_mtu,
1132 .ndo_validate_addr = eth_validate_addr,
1133 .ndo_set_mac_address = eth_mac_addr,
1134 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1137 static void geneve_get_drvinfo(struct net_device *dev,
1138 struct ethtool_drvinfo *drvinfo)
1140 strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1141 strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1144 static const struct ethtool_ops geneve_ethtool_ops = {
1145 .get_drvinfo = geneve_get_drvinfo,
1146 .get_link = ethtool_op_get_link,
1149 /* Info for udev, that this is a virtual tunnel endpoint */
1150 static const struct device_type geneve_type = {
1154 /* Calls the ndo_udp_tunnel_add of the caller in order to
1155 * supply the listening GENEVE udp ports. Callers are expected
1156 * to implement the ndo_udp_tunnel_add.
1158 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1160 struct net *net = dev_net(dev);
1161 struct geneve_net *gn = net_generic(net, geneve_net_id);
1162 struct geneve_sock *gs;
1165 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1167 udp_tunnel_push_rx_port(dev, gs->sock,
1168 UDP_TUNNEL_TYPE_GENEVE);
1170 udp_tunnel_drop_rx_port(dev, gs->sock,
1171 UDP_TUNNEL_TYPE_GENEVE);
1177 /* Initialize the device structure. */
1178 static void geneve_setup(struct net_device *dev)
1182 dev->netdev_ops = &geneve_netdev_ops;
1183 dev->ethtool_ops = &geneve_ethtool_ops;
1184 dev->needs_free_netdev = true;
1186 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1188 dev->features |= NETIF_F_LLTX;
1189 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1190 dev->features |= NETIF_F_RXCSUM;
1191 dev->features |= NETIF_F_GSO_SOFTWARE;
1193 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
1194 dev->hw_features |= NETIF_F_RXCSUM;
1195 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1197 dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
1198 /* MTU range: 68 - (something less than 65535) */
1199 dev->min_mtu = ETH_MIN_MTU;
1200 /* The max_mtu calculation does not take account of GENEVE
1201 * options, to avoid excluding potentially valid
1202 * configurations. This will be further reduced by IPvX hdr size.
1204 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len;
1206 netif_keep_dst(dev);
1207 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1208 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1209 eth_hw_addr_random(dev);
1212 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = {
1213 [IFLA_GENEVE_UNSPEC] = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT },
1214 [IFLA_GENEVE_ID] = { .type = NLA_U32 },
1215 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1216 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1217 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1218 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1219 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1220 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1221 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1222 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1223 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1224 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1225 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1226 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1227 [IFLA_GENEVE_INNER_PROTO_INHERIT] = { .type = NLA_FLAG },
1230 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1231 struct netlink_ext_ack *extack)
1233 if (tb[IFLA_ADDRESS]) {
1234 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1235 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1236 "Provided link layer address is not Ethernet");
1240 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1241 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1242 "Provided Ethernet address is not unicast");
1243 return -EADDRNOTAVAIL;
1248 NL_SET_ERR_MSG(extack,
1249 "Not enough attributes provided to perform the operation");
1253 if (data[IFLA_GENEVE_ID]) {
1254 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1256 if (vni >= GENEVE_N_VID) {
1257 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1258 "Geneve ID must be lower than 16777216");
1263 if (data[IFLA_GENEVE_DF]) {
1264 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1266 if (df < 0 || df > GENEVE_DF_MAX) {
1267 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1268 "Invalid DF attribute");
1276 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1277 const struct ip_tunnel_info *info,
1278 bool *tun_on_same_port,
1279 bool *tun_collect_md)
1281 struct geneve_dev *geneve, *t = NULL;
1283 *tun_on_same_port = false;
1284 *tun_collect_md = false;
1285 list_for_each_entry(geneve, &gn->geneve_list, next) {
1286 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1287 *tun_collect_md = geneve->cfg.collect_md;
1288 *tun_on_same_port = true;
1290 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1291 info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1292 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1298 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1300 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1301 info->key.ttl || info->key.label || info->key.tp_src ||
1302 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1305 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1306 struct ip_tunnel_info *b)
1308 if (ip_tunnel_info_af(a) == AF_INET)
1309 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1311 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1314 static int geneve_configure(struct net *net, struct net_device *dev,
1315 struct netlink_ext_ack *extack,
1316 const struct geneve_config *cfg)
1318 struct geneve_net *gn = net_generic(net, geneve_net_id);
1319 struct geneve_dev *t, *geneve = netdev_priv(dev);
1320 const struct ip_tunnel_info *info = &cfg->info;
1321 bool tun_collect_md, tun_on_same_port;
1324 if (cfg->collect_md && !is_tnl_info_zero(info)) {
1325 NL_SET_ERR_MSG(extack,
1326 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1333 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1337 /* make enough headroom for basic scenario */
1338 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1339 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1340 encap_len += sizeof(struct iphdr);
1341 dev->max_mtu -= sizeof(struct iphdr);
1343 encap_len += sizeof(struct ipv6hdr);
1344 dev->max_mtu -= sizeof(struct ipv6hdr);
1346 dev->needed_headroom = encap_len + ETH_HLEN;
1348 if (cfg->collect_md) {
1349 if (tun_on_same_port) {
1350 NL_SET_ERR_MSG(extack,
1351 "There can be only one externally controlled device on a destination port");
1355 if (tun_collect_md) {
1356 NL_SET_ERR_MSG(extack,
1357 "There already exists an externally controlled device on this destination port");
1362 dst_cache_reset(&geneve->cfg.info.dst_cache);
1363 memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1365 if (geneve->cfg.inner_proto_inherit) {
1366 dev->header_ops = NULL;
1367 dev->type = ARPHRD_NONE;
1368 dev->hard_header_len = 0;
1370 dev->flags = IFF_POINTOPOINT | IFF_NOARP;
1373 err = register_netdevice(dev);
1377 list_add(&geneve->next, &gn->geneve_list);
1381 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1383 memset(info, 0, sizeof(*info));
1384 info->key.tp_dst = htons(dst_port);
1387 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1388 struct netlink_ext_ack *extack,
1389 struct geneve_config *cfg, bool changelink)
1391 struct ip_tunnel_info *info = &cfg->info;
1394 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1395 NL_SET_ERR_MSG(extack,
1396 "Cannot specify both IPv4 and IPv6 Remote addresses");
1400 if (data[IFLA_GENEVE_REMOTE]) {
1401 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1402 attrtype = IFLA_GENEVE_REMOTE;
1406 info->key.u.ipv4.dst =
1407 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1409 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1410 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1411 "Remote IPv4 address cannot be Multicast");
1416 if (data[IFLA_GENEVE_REMOTE6]) {
1417 #if IS_ENABLED(CONFIG_IPV6)
1418 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1419 attrtype = IFLA_GENEVE_REMOTE6;
1423 info->mode = IP_TUNNEL_INFO_IPV6;
1424 info->key.u.ipv6.dst =
1425 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1427 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1428 IPV6_ADDR_LINKLOCAL) {
1429 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1430 "Remote IPv6 address cannot be link-local");
1433 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1434 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1435 "Remote IPv6 address cannot be Multicast");
1438 info->key.tun_flags |= TUNNEL_CSUM;
1439 cfg->use_udp6_rx_checksums = true;
1441 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1442 "IPv6 support not enabled in the kernel");
1443 return -EPFNOSUPPORT;
1447 if (data[IFLA_GENEVE_ID]) {
1452 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1453 tvni[0] = (vni & 0x00ff0000) >> 16;
1454 tvni[1] = (vni & 0x0000ff00) >> 8;
1455 tvni[2] = vni & 0x000000ff;
1457 tunid = vni_to_tunnel_id(tvni);
1458 if (changelink && (tunid != info->key.tun_id)) {
1459 attrtype = IFLA_GENEVE_ID;
1462 info->key.tun_id = tunid;
1465 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1466 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1467 cfg->ttl_inherit = true;
1469 cfg->ttl_inherit = false;
1470 } else if (data[IFLA_GENEVE_TTL]) {
1471 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1472 cfg->ttl_inherit = false;
1475 if (data[IFLA_GENEVE_TOS])
1476 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1478 if (data[IFLA_GENEVE_DF])
1479 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1481 if (data[IFLA_GENEVE_LABEL]) {
1482 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1483 IPV6_FLOWLABEL_MASK;
1484 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1485 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1486 "Label attribute only applies for IPv6 Geneve devices");
1491 if (data[IFLA_GENEVE_PORT]) {
1493 attrtype = IFLA_GENEVE_PORT;
1496 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1499 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1501 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1504 cfg->collect_md = true;
1507 if (data[IFLA_GENEVE_UDP_CSUM]) {
1509 attrtype = IFLA_GENEVE_UDP_CSUM;
1512 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1513 info->key.tun_flags |= TUNNEL_CSUM;
1516 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1517 #if IS_ENABLED(CONFIG_IPV6)
1519 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1522 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1523 info->key.tun_flags &= ~TUNNEL_CSUM;
1525 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1526 "IPv6 support not enabled in the kernel");
1527 return -EPFNOSUPPORT;
1531 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1532 #if IS_ENABLED(CONFIG_IPV6)
1534 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1537 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1538 cfg->use_udp6_rx_checksums = false;
1540 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1541 "IPv6 support not enabled in the kernel");
1542 return -EPFNOSUPPORT;
1546 if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) {
1548 attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT;
1551 cfg->inner_proto_inherit = true;
1556 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1557 "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported");
1561 static void geneve_link_config(struct net_device *dev,
1562 struct ip_tunnel_info *info, struct nlattr *tb[])
1564 struct geneve_dev *geneve = netdev_priv(dev);
1568 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1572 switch (ip_tunnel_info_af(info)) {
1574 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1575 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1577 if (!IS_ERR(rt) && rt->dst.dev) {
1578 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1583 #if IS_ENABLED(CONFIG_IPV6)
1585 struct rt6_info *rt;
1587 if (!__in6_dev_get(dev))
1590 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1593 if (rt && rt->dst.dev)
1594 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1604 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1607 static int geneve_newlink(struct net *net, struct net_device *dev,
1608 struct nlattr *tb[], struct nlattr *data[],
1609 struct netlink_ext_ack *extack)
1611 struct geneve_config cfg = {
1612 .df = GENEVE_DF_UNSET,
1613 .use_udp6_rx_checksums = false,
1614 .ttl_inherit = false,
1615 .collect_md = false,
1619 init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1620 err = geneve_nl2info(tb, data, extack, &cfg, false);
1624 err = geneve_configure(net, dev, extack, &cfg);
1628 geneve_link_config(dev, &cfg.info, tb);
1633 /* Quiesces the geneve device data path for both TX and RX.
1635 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1636 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1637 * to complete for the existing set of in-flight packets to be transmitted,
1638 * then we would have quiesced the transmit data path. All the future packets
1639 * will get dropped until we unquiesce the data path.
1641 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1642 * if we set that to NULL under RCU and wait for synchronize_net() to
1643 * complete, then we would have quiesced the receive data path.
1645 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1646 struct geneve_sock **gs6)
1648 *gs4 = rtnl_dereference(geneve->sock4);
1649 rcu_assign_pointer(geneve->sock4, NULL);
1651 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1652 #if IS_ENABLED(CONFIG_IPV6)
1653 *gs6 = rtnl_dereference(geneve->sock6);
1654 rcu_assign_pointer(geneve->sock6, NULL);
1656 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1663 /* Resumes the geneve device data path for both TX and RX. */
1664 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1665 struct geneve_sock __maybe_unused *gs6)
1667 rcu_assign_pointer(geneve->sock4, gs4);
1669 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1670 #if IS_ENABLED(CONFIG_IPV6)
1671 rcu_assign_pointer(geneve->sock6, gs6);
1673 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1678 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1679 struct nlattr *data[],
1680 struct netlink_ext_ack *extack)
1682 struct geneve_dev *geneve = netdev_priv(dev);
1683 struct geneve_sock *gs4, *gs6;
1684 struct geneve_config cfg;
1687 /* If the geneve device is configured for metadata (or externally
1688 * controlled, for example, OVS), then nothing can be changed.
1690 if (geneve->cfg.collect_md)
1693 /* Start with the existing info. */
1694 memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1695 err = geneve_nl2info(tb, data, extack, &cfg, true);
1699 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1700 dst_cache_reset(&cfg.info.dst_cache);
1701 geneve_link_config(dev, &cfg.info, tb);
1704 geneve_quiesce(geneve, &gs4, &gs6);
1705 memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1706 geneve_unquiesce(geneve, gs4, gs6);
1711 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1713 struct geneve_dev *geneve = netdev_priv(dev);
1715 list_del(&geneve->next);
1716 unregister_netdevice_queue(dev, head);
1719 static size_t geneve_get_size(const struct net_device *dev)
1721 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1722 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1723 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1724 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1725 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1726 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1727 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1728 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1729 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1730 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1731 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1732 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1733 nla_total_size(0) + /* IFLA_GENEVE_INNER_PROTO_INHERIT */
1737 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1739 struct geneve_dev *geneve = netdev_priv(dev);
1740 struct ip_tunnel_info *info = &geneve->cfg.info;
1741 bool ttl_inherit = geneve->cfg.ttl_inherit;
1742 bool metadata = geneve->cfg.collect_md;
1746 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1747 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1748 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1749 goto nla_put_failure;
1751 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1752 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1753 info->key.u.ipv4.dst))
1754 goto nla_put_failure;
1755 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1756 !!(info->key.tun_flags & TUNNEL_CSUM)))
1757 goto nla_put_failure;
1759 #if IS_ENABLED(CONFIG_IPV6)
1760 } else if (!metadata) {
1761 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1762 &info->key.u.ipv6.dst))
1763 goto nla_put_failure;
1764 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1765 !(info->key.tun_flags & TUNNEL_CSUM)))
1766 goto nla_put_failure;
1770 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1771 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1772 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1773 goto nla_put_failure;
1775 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1776 goto nla_put_failure;
1778 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1779 goto nla_put_failure;
1781 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1782 goto nla_put_failure;
1784 #if IS_ENABLED(CONFIG_IPV6)
1785 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1786 !geneve->cfg.use_udp6_rx_checksums))
1787 goto nla_put_failure;
1790 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1791 goto nla_put_failure;
1793 if (geneve->cfg.inner_proto_inherit &&
1794 nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT))
1795 goto nla_put_failure;
1803 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1805 .maxtype = IFLA_GENEVE_MAX,
1806 .policy = geneve_policy,
1807 .priv_size = sizeof(struct geneve_dev),
1808 .setup = geneve_setup,
1809 .validate = geneve_validate,
1810 .newlink = geneve_newlink,
1811 .changelink = geneve_changelink,
1812 .dellink = geneve_dellink,
1813 .get_size = geneve_get_size,
1814 .fill_info = geneve_fill_info,
1817 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1818 u8 name_assign_type, u16 dst_port)
1820 struct nlattr *tb[IFLA_MAX + 1];
1821 struct net_device *dev;
1822 LIST_HEAD(list_kill);
1824 struct geneve_config cfg = {
1825 .df = GENEVE_DF_UNSET,
1826 .use_udp6_rx_checksums = true,
1827 .ttl_inherit = false,
1831 memset(tb, 0, sizeof(tb));
1832 dev = rtnl_create_link(net, name, name_assign_type,
1833 &geneve_link_ops, tb, NULL);
1837 init_tnl_info(&cfg.info, dst_port);
1838 err = geneve_configure(net, dev, NULL, &cfg);
1841 return ERR_PTR(err);
1844 /* openvswitch users expect packet sizes to be unrestricted,
1845 * so set the largest MTU we can.
1847 err = geneve_change_mtu(dev, IP_MAX_MTU);
1851 err = rtnl_configure_link(dev, NULL, 0, NULL);
1857 geneve_dellink(dev, &list_kill);
1858 unregister_netdevice_many(&list_kill);
1859 return ERR_PTR(err);
1861 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1863 static int geneve_netdevice_event(struct notifier_block *unused,
1864 unsigned long event, void *ptr)
1866 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1868 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
1869 geneve_offload_rx_ports(dev, true);
1870 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
1871 geneve_offload_rx_ports(dev, false);
1876 static struct notifier_block geneve_notifier_block __read_mostly = {
1877 .notifier_call = geneve_netdevice_event,
1880 static __net_init int geneve_init_net(struct net *net)
1882 struct geneve_net *gn = net_generic(net, geneve_net_id);
1884 INIT_LIST_HEAD(&gn->geneve_list);
1885 INIT_LIST_HEAD(&gn->sock_list);
1889 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1891 struct geneve_net *gn = net_generic(net, geneve_net_id);
1892 struct geneve_dev *geneve, *next;
1893 struct net_device *dev, *aux;
1895 /* gather any geneve devices that were moved into this ns */
1896 for_each_netdev_safe(net, dev, aux)
1897 if (dev->rtnl_link_ops == &geneve_link_ops)
1898 unregister_netdevice_queue(dev, head);
1900 /* now gather any other geneve devices that were created in this ns */
1901 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1902 /* If geneve->dev is in the same netns, it was already added
1903 * to the list by the previous loop.
1905 if (!net_eq(dev_net(geneve->dev), net))
1906 unregister_netdevice_queue(geneve->dev, head);
1910 static void __net_exit geneve_exit_batch_rtnl(struct list_head *net_list,
1911 struct list_head *dev_to_kill)
1915 list_for_each_entry(net, net_list, exit_list)
1916 geneve_destroy_tunnels(net, dev_to_kill);
1919 static void __net_exit geneve_exit_net(struct net *net)
1921 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1923 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1926 static struct pernet_operations geneve_net_ops = {
1927 .init = geneve_init_net,
1928 .exit_batch_rtnl = geneve_exit_batch_rtnl,
1929 .exit = geneve_exit_net,
1930 .id = &geneve_net_id,
1931 .size = sizeof(struct geneve_net),
1934 static int __init geneve_init_module(void)
1938 rc = register_pernet_subsys(&geneve_net_ops);
1942 rc = register_netdevice_notifier(&geneve_notifier_block);
1946 rc = rtnl_link_register(&geneve_link_ops);
1952 unregister_netdevice_notifier(&geneve_notifier_block);
1954 unregister_pernet_subsys(&geneve_net_ops);
1958 late_initcall(geneve_init_module);
1960 static void __exit geneve_cleanup_module(void)
1962 rtnl_link_unregister(&geneve_link_ops);
1963 unregister_netdevice_notifier(&geneve_notifier_block);
1964 unregister_pernet_subsys(&geneve_net_ops);
1966 module_exit(geneve_cleanup_module);
1968 MODULE_LICENSE("GPL");
1969 MODULE_VERSION(GENEVE_NETDEV_VER);
1970 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1971 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1972 MODULE_ALIAS_RTNL_LINK("geneve");