1 // SPDX-License-Identifier: GPL-2.0-only
3 * GENEVE: Generic Network Virtualization Encapsulation
5 * Copyright (c) 2015 Red Hat, Inc.
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/kernel.h>
11 #include <linux/module.h>
12 #include <linux/etherdevice.h>
13 #include <linux/hash.h>
14 #include <net/ipv6_stubs.h>
15 #include <net/dst_metadata.h>
16 #include <net/gro_cells.h>
17 #include <net/rtnetlink.h>
18 #include <net/geneve.h>
19 #include <net/protocol.h>
21 #define GENEVE_NETDEV_VER "0.6"
23 #define GENEVE_N_VID (1u << 24)
24 #define GENEVE_VID_MASK (GENEVE_N_VID - 1)
26 #define VNI_HASH_BITS 10
27 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
29 static bool log_ecn_error = true;
30 module_param(log_ecn_error, bool, 0644);
31 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
34 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr))
35 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN)
36 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN)
38 /* per-network namespace private data for this module */
40 struct list_head geneve_list;
41 struct list_head sock_list;
44 static unsigned int geneve_net_id;
46 struct geneve_dev_node {
47 struct hlist_node hlist;
48 struct geneve_dev *geneve;
51 struct geneve_config {
52 struct ip_tunnel_info info;
54 bool use_udp6_rx_checksums;
56 enum ifla_geneve_df df;
59 /* Pseudo network device */
61 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */
62 #if IS_ENABLED(CONFIG_IPV6)
63 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */
65 struct net *net; /* netns for packet i/o */
66 struct net_device *dev; /* netdev for geneve tunnel */
67 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */
68 #if IS_ENABLED(CONFIG_IPV6)
69 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */
71 struct list_head next; /* geneve's per namespace list */
72 struct gro_cells gro_cells;
73 struct geneve_config cfg;
78 struct list_head list;
82 struct hlist_head vni_list[VNI_HASH_SIZE];
85 static inline __u32 geneve_net_vni_hash(u8 vni[3])
89 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2];
90 return hash_32(vnid, VNI_HASH_BITS);
93 static __be64 vni_to_tunnel_id(const __u8 *vni)
96 return (vni[0] << 16) | (vni[1] << 8) | vni[2];
98 return (__force __be64)(((__force u64)vni[0] << 40) |
99 ((__force u64)vni[1] << 48) |
100 ((__force u64)vni[2] << 56));
104 /* Convert 64 bit tunnel ID to 24 bit VNI. */
105 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni)
108 vni[0] = (__force __u8)(tun_id >> 16);
109 vni[1] = (__force __u8)(tun_id >> 8);
110 vni[2] = (__force __u8)tun_id;
112 vni[0] = (__force __u8)((__force u64)tun_id >> 40);
113 vni[1] = (__force __u8)((__force u64)tun_id >> 48);
114 vni[2] = (__force __u8)((__force u64)tun_id >> 56);
118 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni)
120 return !memcmp(vni, &tun_id[5], 3);
123 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs)
125 return gs->sock->sk->sk_family;
128 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs,
129 __be32 addr, u8 vni[])
131 struct hlist_head *vni_list_head;
132 struct geneve_dev_node *node;
135 /* Find the device for this VNI */
136 hash = geneve_net_vni_hash(vni);
137 vni_list_head = &gs->vni_list[hash];
138 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
139 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
140 addr == node->geneve->cfg.info.key.u.ipv4.dst)
146 #if IS_ENABLED(CONFIG_IPV6)
147 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs,
148 struct in6_addr addr6, u8 vni[])
150 struct hlist_head *vni_list_head;
151 struct geneve_dev_node *node;
154 /* Find the device for this VNI */
155 hash = geneve_net_vni_hash(vni);
156 vni_list_head = &gs->vni_list[hash];
157 hlist_for_each_entry_rcu(node, vni_list_head, hlist) {
158 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) &&
159 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst))
166 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb)
168 return (struct genevehdr *)(udp_hdr(skb) + 1);
171 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs,
174 static u8 zero_vni[3];
177 if (geneve_get_sk_family(gs) == AF_INET) {
181 iph = ip_hdr(skb); /* outer IP header... */
183 if (gs->collect_md) {
187 vni = geneve_hdr(skb)->vni;
191 return geneve_lookup(gs, addr, vni);
192 #if IS_ENABLED(CONFIG_IPV6)
193 } else if (geneve_get_sk_family(gs) == AF_INET6) {
194 static struct in6_addr zero_addr6;
195 struct ipv6hdr *ip6h;
196 struct in6_addr addr6;
198 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */
200 if (gs->collect_md) {
204 vni = geneve_hdr(skb)->vni;
208 return geneve6_lookup(gs, addr6, vni);
214 /* geneve receive/decap routine */
215 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
218 struct genevehdr *gnvh = geneve_hdr(skb);
219 struct metadata_dst *tun_dst = NULL;
224 if (ip_tunnel_collect_metadata() || gs->collect_md) {
227 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) |
228 (gnvh->critical ? TUNNEL_CRIT_OPT : 0);
230 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags,
231 vni_to_tunnel_id(gnvh->vni),
234 geneve->dev->stats.rx_dropped++;
237 /* Update tunnel dst according to Geneve options. */
238 ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
239 gnvh->options, gnvh->opt_len * 4,
242 /* Drop packets w/ critical options,
243 * since we don't support any...
245 if (gnvh->critical) {
246 geneve->dev->stats.rx_frame_errors++;
247 geneve->dev->stats.rx_errors++;
252 skb_reset_mac_header(skb);
253 skb->protocol = eth_type_trans(skb, geneve->dev);
254 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
257 skb_dst_set(skb, &tun_dst->dst);
259 /* Ignore packet loops (and multicast echo) */
260 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
261 geneve->dev->stats.rx_errors++;
265 oiph = skb_network_header(skb);
266 skb_reset_network_header(skb);
268 if (geneve_get_sk_family(gs) == AF_INET)
269 err = IP_ECN_decapsulate(oiph, skb);
270 #if IS_ENABLED(CONFIG_IPV6)
272 err = IP6_ECN_decapsulate(oiph, skb);
277 if (geneve_get_sk_family(gs) == AF_INET)
278 net_info_ratelimited("non-ECT from %pI4 "
280 &((struct iphdr *)oiph)->saddr,
281 ((struct iphdr *)oiph)->tos);
282 #if IS_ENABLED(CONFIG_IPV6)
284 net_info_ratelimited("non-ECT from %pI6\n",
285 &((struct ipv6hdr *)oiph)->saddr);
289 ++geneve->dev->stats.rx_frame_errors;
290 ++geneve->dev->stats.rx_errors;
296 err = gro_cells_receive(&geneve->gro_cells, skb);
297 if (likely(err == NET_RX_SUCCESS))
298 dev_sw_netstats_rx_add(geneve->dev, len);
302 /* Consume bad packet */
306 /* Setup stats when device is created */
307 static int geneve_init(struct net_device *dev)
309 struct geneve_dev *geneve = netdev_priv(dev);
312 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
316 err = gro_cells_init(&geneve->gro_cells, dev);
318 free_percpu(dev->tstats);
322 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL);
324 free_percpu(dev->tstats);
325 gro_cells_destroy(&geneve->gro_cells);
331 static void geneve_uninit(struct net_device *dev)
333 struct geneve_dev *geneve = netdev_priv(dev);
335 dst_cache_destroy(&geneve->cfg.info.dst_cache);
336 gro_cells_destroy(&geneve->gro_cells);
337 free_percpu(dev->tstats);
340 /* Callback from net/ipv4/udp.c to receive packets */
341 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
343 struct genevehdr *geneveh;
344 struct geneve_dev *geneve;
345 struct geneve_sock *gs;
348 /* Need UDP and Geneve header to be present */
349 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN)))
352 /* Return packets with reserved bits set */
353 geneveh = geneve_hdr(skb);
354 if (unlikely(geneveh->ver != GENEVE_VER))
357 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
360 gs = rcu_dereference_sk_user_data(sk);
364 geneve = geneve_lookup_skb(gs, skb);
368 opts_len = geneveh->opt_len * 4;
369 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
371 !net_eq(geneve->net, dev_net(geneve->dev)))) {
372 geneve->dev->stats.rx_dropped++;
376 geneve_rx(geneve, gs, skb);
380 /* Consume bad packet */
385 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */
386 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb)
388 struct genevehdr *geneveh;
389 struct geneve_sock *gs;
390 u8 zero_vni[3] = { 0 };
393 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN))
396 geneveh = geneve_hdr(skb);
397 if (geneveh->ver != GENEVE_VER)
400 if (geneveh->proto_type != htons(ETH_P_TEB))
403 gs = rcu_dereference_sk_user_data(sk);
407 if (geneve_get_sk_family(gs) == AF_INET) {
408 struct iphdr *iph = ip_hdr(skb);
411 if (!gs->collect_md) {
412 vni = geneve_hdr(skb)->vni;
416 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT;
419 #if IS_ENABLED(CONFIG_IPV6)
420 if (geneve_get_sk_family(gs) == AF_INET6) {
421 struct ipv6hdr *ip6h = ipv6_hdr(skb);
422 struct in6_addr addr6;
424 memset(&addr6, 0, sizeof(struct in6_addr));
426 if (!gs->collect_md) {
427 vni = geneve_hdr(skb)->vni;
431 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT;
435 return -EPFNOSUPPORT;
438 static struct socket *geneve_create_sock(struct net *net, bool ipv6,
439 __be16 port, bool ipv6_rx_csum)
442 struct udp_port_cfg udp_conf;
445 memset(&udp_conf, 0, sizeof(udp_conf));
448 udp_conf.family = AF_INET6;
449 udp_conf.ipv6_v6only = 1;
450 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum;
452 udp_conf.family = AF_INET;
453 udp_conf.local_ip.s_addr = htonl(INADDR_ANY);
456 udp_conf.local_udp_port = port;
458 /* Open UDP socket */
459 err = udp_sock_create(net, &udp_conf, &sock);
466 static int geneve_hlen(struct genevehdr *gh)
468 return sizeof(*gh) + gh->opt_len * 4;
471 static struct sk_buff *geneve_gro_receive(struct sock *sk,
472 struct list_head *head,
475 struct sk_buff *pp = NULL;
477 struct genevehdr *gh, *gh2;
478 unsigned int hlen, gh_len, off_gnv;
479 const struct packet_offload *ptype;
483 off_gnv = skb_gro_offset(skb);
484 hlen = off_gnv + sizeof(*gh);
485 gh = skb_gro_header_fast(skb, off_gnv);
486 if (skb_gro_header_hard(skb, hlen)) {
487 gh = skb_gro_header_slow(skb, hlen, off_gnv);
492 if (gh->ver != GENEVE_VER || gh->oam)
494 gh_len = geneve_hlen(gh);
496 hlen = off_gnv + gh_len;
497 if (skb_gro_header_hard(skb, hlen)) {
498 gh = skb_gro_header_slow(skb, hlen, off_gnv);
503 list_for_each_entry(p, head, list) {
504 if (!NAPI_GRO_CB(p)->same_flow)
507 gh2 = (struct genevehdr *)(p->data + off_gnv);
508 if (gh->opt_len != gh2->opt_len ||
509 memcmp(gh, gh2, gh_len)) {
510 NAPI_GRO_CB(p)->same_flow = 0;
515 type = gh->proto_type;
518 ptype = gro_find_receive_by_type(type);
522 skb_gro_pull(skb, gh_len);
523 skb_gro_postpull_rcsum(skb, gh, gh_len);
524 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
530 skb_gro_flush_final(skb, pp, flush);
535 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb,
538 struct genevehdr *gh;
539 struct packet_offload *ptype;
544 gh = (struct genevehdr *)(skb->data + nhoff);
545 gh_len = geneve_hlen(gh);
546 type = gh->proto_type;
549 ptype = gro_find_complete_by_type(type);
551 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len);
555 skb_set_inner_mac_header(skb, nhoff + gh_len);
560 /* Create new listen socket if needed */
561 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port,
562 bool ipv6, bool ipv6_rx_csum)
564 struct geneve_net *gn = net_generic(net, geneve_net_id);
565 struct geneve_sock *gs;
567 struct udp_tunnel_sock_cfg tunnel_cfg;
570 gs = kzalloc(sizeof(*gs), GFP_KERNEL);
572 return ERR_PTR(-ENOMEM);
574 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum);
577 return ERR_CAST(sock);
582 for (h = 0; h < VNI_HASH_SIZE; ++h)
583 INIT_HLIST_HEAD(&gs->vni_list[h]);
585 /* Initialize the geneve udp offloads structure */
586 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
588 /* Mark socket as an encapsulation socket */
589 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
590 tunnel_cfg.sk_user_data = gs;
591 tunnel_cfg.encap_type = 1;
592 tunnel_cfg.gro_receive = geneve_gro_receive;
593 tunnel_cfg.gro_complete = geneve_gro_complete;
594 tunnel_cfg.encap_rcv = geneve_udp_encap_recv;
595 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup;
596 tunnel_cfg.encap_destroy = NULL;
597 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
598 list_add(&gs->list, &gn->sock_list);
602 static void __geneve_sock_release(struct geneve_sock *gs)
604 if (!gs || --gs->refcnt)
608 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE);
609 udp_tunnel_sock_release(gs->sock);
613 static void geneve_sock_release(struct geneve_dev *geneve)
615 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4);
616 #if IS_ENABLED(CONFIG_IPV6)
617 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6);
619 rcu_assign_pointer(geneve->sock6, NULL);
622 rcu_assign_pointer(geneve->sock4, NULL);
625 __geneve_sock_release(gs4);
626 #if IS_ENABLED(CONFIG_IPV6)
627 __geneve_sock_release(gs6);
631 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn,
635 struct geneve_sock *gs;
637 list_for_each_entry(gs, &gn->sock_list, list) {
638 if (inet_sk(gs->sock->sk)->inet_sport == dst_port &&
639 geneve_get_sk_family(gs) == family) {
646 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6)
648 struct net *net = geneve->net;
649 struct geneve_net *gn = net_generic(net, geneve_net_id);
650 struct geneve_dev_node *node;
651 struct geneve_sock *gs;
655 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst);
661 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6,
662 geneve->cfg.use_udp6_rx_checksums);
667 gs->collect_md = geneve->cfg.collect_md;
668 #if IS_ENABLED(CONFIG_IPV6)
670 rcu_assign_pointer(geneve->sock6, gs);
671 node = &geneve->hlist6;
675 rcu_assign_pointer(geneve->sock4, gs);
676 node = &geneve->hlist4;
678 node->geneve = geneve;
680 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni);
681 hash = geneve_net_vni_hash(vni);
682 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]);
686 static int geneve_open(struct net_device *dev)
688 struct geneve_dev *geneve = netdev_priv(dev);
689 bool metadata = geneve->cfg.collect_md;
693 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata;
694 ipv4 = !ipv6 || metadata;
695 #if IS_ENABLED(CONFIG_IPV6)
697 ret = geneve_sock_add(geneve, true);
698 if (ret < 0 && ret != -EAFNOSUPPORT)
703 ret = geneve_sock_add(geneve, false);
705 geneve_sock_release(geneve);
710 static int geneve_stop(struct net_device *dev)
712 struct geneve_dev *geneve = netdev_priv(dev);
714 hlist_del_init_rcu(&geneve->hlist4.hlist);
715 #if IS_ENABLED(CONFIG_IPV6)
716 hlist_del_init_rcu(&geneve->hlist6.hlist);
718 geneve_sock_release(geneve);
722 static void geneve_build_header(struct genevehdr *geneveh,
723 const struct ip_tunnel_info *info)
725 geneveh->ver = GENEVE_VER;
726 geneveh->opt_len = info->options_len / 4;
727 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM);
728 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
730 tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
731 geneveh->proto_type = htons(ETH_P_TEB);
734 if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
735 ip_tunnel_info_opts_get(geneveh->options, info);
738 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
739 const struct ip_tunnel_info *info,
740 bool xnet, int ip_hdr_len)
742 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
743 struct genevehdr *gnvh;
747 skb_reset_mac_header(skb);
748 skb_scrub_packet(skb, xnet);
750 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len +
751 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len;
752 err = skb_cow_head(skb, min_headroom);
756 err = udp_tunnel_handle_offloads(skb, udp_sum);
760 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
761 geneve_build_header(gnvh, info);
762 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
770 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb,
771 struct net_device *dev,
772 struct geneve_sock *gs4,
774 const struct ip_tunnel_info *info,
775 __be16 dport, __be16 sport)
777 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
778 struct geneve_dev *geneve = netdev_priv(dev);
779 struct dst_cache *dst_cache;
780 struct rtable *rt = NULL;
784 return ERR_PTR(-EIO);
786 memset(fl4, 0, sizeof(*fl4));
787 fl4->flowi4_mark = skb->mark;
788 fl4->flowi4_proto = IPPROTO_UDP;
789 fl4->daddr = info->key.u.ipv4.dst;
790 fl4->saddr = info->key.u.ipv4.src;
791 fl4->fl4_dport = dport;
792 fl4->fl4_sport = sport;
795 if ((tos == 1) && !geneve->cfg.collect_md) {
796 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
799 fl4->flowi4_tos = RT_TOS(tos);
801 dst_cache = (struct dst_cache *)&info->dst_cache;
803 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr);
807 rt = ip_route_output_key(geneve->net, fl4);
809 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr);
810 return ERR_PTR(-ENETUNREACH);
812 if (rt->dst.dev == dev) { /* is this necessary? */
813 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr);
815 return ERR_PTR(-ELOOP);
818 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr);
822 #if IS_ENABLED(CONFIG_IPV6)
823 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb,
824 struct net_device *dev,
825 struct geneve_sock *gs6,
827 const struct ip_tunnel_info *info,
828 __be16 dport, __be16 sport)
830 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
831 struct geneve_dev *geneve = netdev_priv(dev);
832 struct dst_entry *dst = NULL;
833 struct dst_cache *dst_cache;
837 return ERR_PTR(-EIO);
839 memset(fl6, 0, sizeof(*fl6));
840 fl6->flowi6_mark = skb->mark;
841 fl6->flowi6_proto = IPPROTO_UDP;
842 fl6->daddr = info->key.u.ipv6.dst;
843 fl6->saddr = info->key.u.ipv6.src;
844 fl6->fl6_dport = dport;
845 fl6->fl6_sport = sport;
847 prio = info->key.tos;
848 if ((prio == 1) && !geneve->cfg.collect_md) {
849 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
853 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio),
855 dst_cache = (struct dst_cache *)&info->dst_cache;
857 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr);
861 dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6,
864 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr);
865 return ERR_PTR(-ENETUNREACH);
867 if (dst->dev == dev) { /* is this necessary? */
868 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr);
870 return ERR_PTR(-ELOOP);
874 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr);
879 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
880 struct geneve_dev *geneve,
881 const struct ip_tunnel_info *info)
883 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
884 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
885 const struct ip_tunnel_key *key = &info->key;
893 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
894 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
895 geneve->cfg.info.key.tp_dst, sport);
899 err = skb_tunnel_check_pmtu(skb, &rt->dst,
900 GENEVE_IPV4_HLEN + info->options_len,
901 netif_is_any_bridge_port(dev));
903 dst_release(&rt->dst);
906 struct ip_tunnel_info *info;
908 info = skb_tunnel_info(skb);
910 info->key.u.ipv4.dst = fl4.saddr;
911 info->key.u.ipv4.src = fl4.daddr;
914 if (!pskb_may_pull(skb, ETH_HLEN)) {
915 dst_release(&rt->dst);
919 skb->protocol = eth_type_trans(skb, geneve->dev);
921 dst_release(&rt->dst);
925 if (geneve->cfg.collect_md) {
926 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
929 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
931 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb);
932 if (geneve->cfg.ttl_inherit)
933 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
936 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
938 if (geneve->cfg.df == GENEVE_DF_SET) {
940 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) {
941 struct ethhdr *eth = eth_hdr(skb);
943 if (ntohs(eth->h_proto) == ETH_P_IPV6) {
945 } else if (ntohs(eth->h_proto) == ETH_P_IP) {
946 struct iphdr *iph = ip_hdr(skb);
948 if (iph->frag_off & htons(IP_DF))
954 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
958 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr,
959 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst,
960 !net_eq(geneve->net, dev_net(geneve->dev)),
961 !(info->key.tun_flags & TUNNEL_CSUM));
965 #if IS_ENABLED(CONFIG_IPV6)
966 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
967 struct geneve_dev *geneve,
968 const struct ip_tunnel_info *info)
970 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev));
971 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
972 const struct ip_tunnel_key *key = &info->key;
973 struct dst_entry *dst = NULL;
979 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
980 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
981 geneve->cfg.info.key.tp_dst, sport);
985 err = skb_tunnel_check_pmtu(skb, dst,
986 GENEVE_IPV6_HLEN + info->options_len,
987 netif_is_any_bridge_port(dev));
992 struct ip_tunnel_info *info = skb_tunnel_info(skb);
995 info->key.u.ipv6.dst = fl6.saddr;
996 info->key.u.ipv6.src = fl6.daddr;
999 if (!pskb_may_pull(skb, ETH_HLEN)) {
1004 skb->protocol = eth_type_trans(skb, geneve->dev);
1010 if (geneve->cfg.collect_md) {
1011 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb);
1014 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel),
1016 if (geneve->cfg.ttl_inherit)
1017 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb);
1020 ttl = ttl ? : ip6_dst_hoplimit(dst);
1022 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
1026 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev,
1027 &fl6.saddr, &fl6.daddr, prio, ttl,
1028 info->key.label, sport, geneve->cfg.info.key.tp_dst,
1029 !(info->key.tun_flags & TUNNEL_CSUM));
1034 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
1036 struct geneve_dev *geneve = netdev_priv(dev);
1037 struct ip_tunnel_info *info = NULL;
1040 if (geneve->cfg.collect_md) {
1041 info = skb_tunnel_info(skb);
1042 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) {
1043 netdev_dbg(dev, "no tunnel metadata\n");
1045 dev->stats.tx_dropped++;
1046 return NETDEV_TX_OK;
1049 info = &geneve->cfg.info;
1053 #if IS_ENABLED(CONFIG_IPV6)
1054 if (info->mode & IP_TUNNEL_INFO_IPV6)
1055 err = geneve6_xmit_skb(skb, dev, geneve, info);
1058 err = geneve_xmit_skb(skb, dev, geneve, info);
1062 return NETDEV_TX_OK;
1064 if (err != -EMSGSIZE)
1068 dev->stats.collisions++;
1069 else if (err == -ENETUNREACH)
1070 dev->stats.tx_carrier_errors++;
1072 dev->stats.tx_errors++;
1073 return NETDEV_TX_OK;
1076 static int geneve_change_mtu(struct net_device *dev, int new_mtu)
1078 if (new_mtu > dev->max_mtu)
1079 new_mtu = dev->max_mtu;
1080 else if (new_mtu < dev->min_mtu)
1081 new_mtu = dev->min_mtu;
1087 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
1089 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1090 struct geneve_dev *geneve = netdev_priv(dev);
1093 if (ip_tunnel_info_af(info) == AF_INET) {
1097 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4);
1098 sport = udp_flow_src_port(geneve->net, skb,
1099 1, USHRT_MAX, true);
1101 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info,
1102 geneve->cfg.info.key.tp_dst, sport);
1107 info->key.u.ipv4.src = fl4.saddr;
1108 #if IS_ENABLED(CONFIG_IPV6)
1109 } else if (ip_tunnel_info_af(info) == AF_INET6) {
1110 struct dst_entry *dst;
1113 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6);
1114 sport = udp_flow_src_port(geneve->net, skb,
1115 1, USHRT_MAX, true);
1117 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info,
1118 geneve->cfg.info.key.tp_dst, sport);
1120 return PTR_ERR(dst);
1123 info->key.u.ipv6.src = fl6.saddr;
1129 info->key.tp_src = sport;
1130 info->key.tp_dst = geneve->cfg.info.key.tp_dst;
1134 static const struct net_device_ops geneve_netdev_ops = {
1135 .ndo_init = geneve_init,
1136 .ndo_uninit = geneve_uninit,
1137 .ndo_open = geneve_open,
1138 .ndo_stop = geneve_stop,
1139 .ndo_start_xmit = geneve_xmit,
1140 .ndo_get_stats64 = ip_tunnel_get_stats64,
1141 .ndo_change_mtu = geneve_change_mtu,
1142 .ndo_validate_addr = eth_validate_addr,
1143 .ndo_set_mac_address = eth_mac_addr,
1144 .ndo_fill_metadata_dst = geneve_fill_metadata_dst,
1147 static void geneve_get_drvinfo(struct net_device *dev,
1148 struct ethtool_drvinfo *drvinfo)
1150 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version));
1151 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver));
1154 static const struct ethtool_ops geneve_ethtool_ops = {
1155 .get_drvinfo = geneve_get_drvinfo,
1156 .get_link = ethtool_op_get_link,
1159 /* Info for udev, that this is a virtual tunnel endpoint */
1160 static struct device_type geneve_type = {
1164 /* Calls the ndo_udp_tunnel_add of the caller in order to
1165 * supply the listening GENEVE udp ports. Callers are expected
1166 * to implement the ndo_udp_tunnel_add.
1168 static void geneve_offload_rx_ports(struct net_device *dev, bool push)
1170 struct net *net = dev_net(dev);
1171 struct geneve_net *gn = net_generic(net, geneve_net_id);
1172 struct geneve_sock *gs;
1175 list_for_each_entry_rcu(gs, &gn->sock_list, list) {
1177 udp_tunnel_push_rx_port(dev, gs->sock,
1178 UDP_TUNNEL_TYPE_GENEVE);
1180 udp_tunnel_drop_rx_port(dev, gs->sock,
1181 UDP_TUNNEL_TYPE_GENEVE);
1187 /* Initialize the device structure. */
1188 static void geneve_setup(struct net_device *dev)
1192 dev->netdev_ops = &geneve_netdev_ops;
1193 dev->ethtool_ops = &geneve_ethtool_ops;
1194 dev->needs_free_netdev = true;
1196 SET_NETDEV_DEVTYPE(dev, &geneve_type);
1198 dev->features |= NETIF_F_LLTX;
1199 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1200 dev->features |= NETIF_F_RXCSUM;
1201 dev->features |= NETIF_F_GSO_SOFTWARE;
1203 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | 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_ID] = { .type = NLA_U32 },
1222 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) },
1223 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) },
1224 [IFLA_GENEVE_TTL] = { .type = NLA_U8 },
1225 [IFLA_GENEVE_TOS] = { .type = NLA_U8 },
1226 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 },
1227 [IFLA_GENEVE_PORT] = { .type = NLA_U16 },
1228 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG },
1229 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 },
1230 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
1231 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
1232 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 },
1233 [IFLA_GENEVE_DF] = { .type = NLA_U8 },
1236 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[],
1237 struct netlink_ext_ack *extack)
1239 if (tb[IFLA_ADDRESS]) {
1240 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1241 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1242 "Provided link layer address is not Ethernet");
1246 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1247 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
1248 "Provided Ethernet address is not unicast");
1249 return -EADDRNOTAVAIL;
1254 NL_SET_ERR_MSG(extack,
1255 "Not enough attributes provided to perform the operation");
1259 if (data[IFLA_GENEVE_ID]) {
1260 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1262 if (vni >= GENEVE_N_VID) {
1263 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID],
1264 "Geneve ID must be lower than 16777216");
1269 if (data[IFLA_GENEVE_DF]) {
1270 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]);
1272 if (df < 0 || df > GENEVE_DF_MAX) {
1273 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF],
1274 "Invalid DF attribute");
1282 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn,
1283 const struct ip_tunnel_info *info,
1284 bool *tun_on_same_port,
1285 bool *tun_collect_md)
1287 struct geneve_dev *geneve, *t = NULL;
1289 *tun_on_same_port = false;
1290 *tun_collect_md = false;
1291 list_for_each_entry(geneve, &gn->geneve_list, next) {
1292 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) {
1293 *tun_collect_md = geneve->cfg.collect_md;
1294 *tun_on_same_port = true;
1296 if (info->key.tun_id == geneve->cfg.info.key.tun_id &&
1297 info->key.tp_dst == geneve->cfg.info.key.tp_dst &&
1298 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u)))
1304 static bool is_tnl_info_zero(const struct ip_tunnel_info *info)
1306 return !(info->key.tun_id || info->key.tun_flags || info->key.tos ||
1307 info->key.ttl || info->key.label || info->key.tp_src ||
1308 memchr_inv(&info->key.u, 0, sizeof(info->key.u)));
1311 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a,
1312 struct ip_tunnel_info *b)
1314 if (ip_tunnel_info_af(a) == AF_INET)
1315 return a->key.u.ipv4.dst == b->key.u.ipv4.dst;
1317 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst);
1320 static int geneve_configure(struct net *net, struct net_device *dev,
1321 struct netlink_ext_ack *extack,
1322 const struct geneve_config *cfg)
1324 struct geneve_net *gn = net_generic(net, geneve_net_id);
1325 struct geneve_dev *t, *geneve = netdev_priv(dev);
1326 const struct ip_tunnel_info *info = &cfg->info;
1327 bool tun_collect_md, tun_on_same_port;
1330 if (cfg->collect_md && !is_tnl_info_zero(info)) {
1331 NL_SET_ERR_MSG(extack,
1332 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified");
1339 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md);
1343 /* make enough headroom for basic scenario */
1344 encap_len = GENEVE_BASE_HLEN + ETH_HLEN;
1345 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) {
1346 encap_len += sizeof(struct iphdr);
1347 dev->max_mtu -= sizeof(struct iphdr);
1349 encap_len += sizeof(struct ipv6hdr);
1350 dev->max_mtu -= sizeof(struct ipv6hdr);
1352 dev->needed_headroom = encap_len + ETH_HLEN;
1354 if (cfg->collect_md) {
1355 if (tun_on_same_port) {
1356 NL_SET_ERR_MSG(extack,
1357 "There can be only one externally controlled device on a destination port");
1361 if (tun_collect_md) {
1362 NL_SET_ERR_MSG(extack,
1363 "There already exists an externally controlled device on this destination port");
1368 dst_cache_reset(&geneve->cfg.info.dst_cache);
1369 memcpy(&geneve->cfg, cfg, sizeof(*cfg));
1371 err = register_netdevice(dev);
1375 list_add(&geneve->next, &gn->geneve_list);
1379 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port)
1381 memset(info, 0, sizeof(*info));
1382 info->key.tp_dst = htons(dst_port);
1385 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
1386 struct netlink_ext_ack *extack,
1387 struct geneve_config *cfg, bool changelink)
1389 struct ip_tunnel_info *info = &cfg->info;
1392 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) {
1393 NL_SET_ERR_MSG(extack,
1394 "Cannot specify both IPv4 and IPv6 Remote addresses");
1398 if (data[IFLA_GENEVE_REMOTE]) {
1399 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) {
1400 attrtype = IFLA_GENEVE_REMOTE;
1404 info->key.u.ipv4.dst =
1405 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]);
1407 if (ipv4_is_multicast(info->key.u.ipv4.dst)) {
1408 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE],
1409 "Remote IPv4 address cannot be Multicast");
1414 if (data[IFLA_GENEVE_REMOTE6]) {
1415 #if IS_ENABLED(CONFIG_IPV6)
1416 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) {
1417 attrtype = IFLA_GENEVE_REMOTE6;
1421 info->mode = IP_TUNNEL_INFO_IPV6;
1422 info->key.u.ipv6.dst =
1423 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]);
1425 if (ipv6_addr_type(&info->key.u.ipv6.dst) &
1426 IPV6_ADDR_LINKLOCAL) {
1427 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1428 "Remote IPv6 address cannot be link-local");
1431 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) {
1432 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1433 "Remote IPv6 address cannot be Multicast");
1436 info->key.tun_flags |= TUNNEL_CSUM;
1437 cfg->use_udp6_rx_checksums = true;
1439 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6],
1440 "IPv6 support not enabled in the kernel");
1441 return -EPFNOSUPPORT;
1445 if (data[IFLA_GENEVE_ID]) {
1450 vni = nla_get_u32(data[IFLA_GENEVE_ID]);
1451 tvni[0] = (vni & 0x00ff0000) >> 16;
1452 tvni[1] = (vni & 0x0000ff00) >> 8;
1453 tvni[2] = vni & 0x000000ff;
1455 tunid = vni_to_tunnel_id(tvni);
1456 if (changelink && (tunid != info->key.tun_id)) {
1457 attrtype = IFLA_GENEVE_ID;
1460 info->key.tun_id = tunid;
1463 if (data[IFLA_GENEVE_TTL_INHERIT]) {
1464 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT]))
1465 cfg->ttl_inherit = true;
1467 cfg->ttl_inherit = false;
1468 } else if (data[IFLA_GENEVE_TTL]) {
1469 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]);
1470 cfg->ttl_inherit = false;
1473 if (data[IFLA_GENEVE_TOS])
1474 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]);
1476 if (data[IFLA_GENEVE_DF])
1477 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]);
1479 if (data[IFLA_GENEVE_LABEL]) {
1480 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) &
1481 IPV6_FLOWLABEL_MASK;
1482 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) {
1483 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL],
1484 "Label attribute only applies for IPv6 Geneve devices");
1489 if (data[IFLA_GENEVE_PORT]) {
1491 attrtype = IFLA_GENEVE_PORT;
1494 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]);
1497 if (data[IFLA_GENEVE_COLLECT_METADATA]) {
1499 attrtype = IFLA_GENEVE_COLLECT_METADATA;
1502 cfg->collect_md = true;
1505 if (data[IFLA_GENEVE_UDP_CSUM]) {
1507 attrtype = IFLA_GENEVE_UDP_CSUM;
1510 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM]))
1511 info->key.tun_flags |= TUNNEL_CSUM;
1514 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) {
1515 #if IS_ENABLED(CONFIG_IPV6)
1517 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX;
1520 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]))
1521 info->key.tun_flags &= ~TUNNEL_CSUM;
1523 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX],
1524 "IPv6 support not enabled in the kernel");
1525 return -EPFNOSUPPORT;
1529 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) {
1530 #if IS_ENABLED(CONFIG_IPV6)
1532 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX;
1535 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]))
1536 cfg->use_udp6_rx_checksums = false;
1538 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX],
1539 "IPv6 support not enabled in the kernel");
1540 return -EPFNOSUPPORT;
1546 NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
1547 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
1551 static void geneve_link_config(struct net_device *dev,
1552 struct ip_tunnel_info *info, struct nlattr *tb[])
1554 struct geneve_dev *geneve = netdev_priv(dev);
1558 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU]));
1562 switch (ip_tunnel_info_af(info)) {
1564 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst };
1565 struct rtable *rt = ip_route_output_key(geneve->net, &fl4);
1567 if (!IS_ERR(rt) && rt->dst.dev) {
1568 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN;
1573 #if IS_ENABLED(CONFIG_IPV6)
1575 struct rt6_info *rt;
1577 if (!__in6_dev_get(dev))
1580 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0,
1583 if (rt && rt->dst.dev)
1584 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN;
1594 geneve_change_mtu(dev, ldev_mtu - info->options_len);
1597 static int geneve_newlink(struct net *net, struct net_device *dev,
1598 struct nlattr *tb[], struct nlattr *data[],
1599 struct netlink_ext_ack *extack)
1601 struct geneve_config cfg = {
1602 .df = GENEVE_DF_UNSET,
1603 .use_udp6_rx_checksums = false,
1604 .ttl_inherit = false,
1605 .collect_md = false,
1609 init_tnl_info(&cfg.info, GENEVE_UDP_PORT);
1610 err = geneve_nl2info(tb, data, extack, &cfg, false);
1614 err = geneve_configure(net, dev, extack, &cfg);
1618 geneve_link_config(dev, &cfg.info, tb);
1623 /* Quiesces the geneve device data path for both TX and RX.
1625 * On transmit geneve checks for non-NULL geneve_sock before it proceeds.
1626 * So, if we set that socket to NULL under RCU and wait for synchronize_net()
1627 * to complete for the existing set of in-flight packets to be transmitted,
1628 * then we would have quiesced the transmit data path. All the future packets
1629 * will get dropped until we unquiesce the data path.
1631 * On receive geneve dereference the geneve_sock stashed in the socket. So,
1632 * if we set that to NULL under RCU and wait for synchronize_net() to
1633 * complete, then we would have quiesced the receive data path.
1635 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4,
1636 struct geneve_sock **gs6)
1638 *gs4 = rtnl_dereference(geneve->sock4);
1639 rcu_assign_pointer(geneve->sock4, NULL);
1641 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL);
1642 #if IS_ENABLED(CONFIG_IPV6)
1643 *gs6 = rtnl_dereference(geneve->sock6);
1644 rcu_assign_pointer(geneve->sock6, NULL);
1646 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL);
1653 /* Resumes the geneve device data path for both TX and RX. */
1654 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4,
1655 struct geneve_sock __maybe_unused *gs6)
1657 rcu_assign_pointer(geneve->sock4, gs4);
1659 rcu_assign_sk_user_data(gs4->sock->sk, gs4);
1660 #if IS_ENABLED(CONFIG_IPV6)
1661 rcu_assign_pointer(geneve->sock6, gs6);
1663 rcu_assign_sk_user_data(gs6->sock->sk, gs6);
1668 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
1669 struct nlattr *data[],
1670 struct netlink_ext_ack *extack)
1672 struct geneve_dev *geneve = netdev_priv(dev);
1673 struct geneve_sock *gs4, *gs6;
1674 struct geneve_config cfg;
1677 /* If the geneve device is configured for metadata (or externally
1678 * controlled, for example, OVS), then nothing can be changed.
1680 if (geneve->cfg.collect_md)
1683 /* Start with the existing info. */
1684 memcpy(&cfg, &geneve->cfg, sizeof(cfg));
1685 err = geneve_nl2info(tb, data, extack, &cfg, true);
1689 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) {
1690 dst_cache_reset(&cfg.info.dst_cache);
1691 geneve_link_config(dev, &cfg.info, tb);
1694 geneve_quiesce(geneve, &gs4, &gs6);
1695 memcpy(&geneve->cfg, &cfg, sizeof(cfg));
1696 geneve_unquiesce(geneve, gs4, gs6);
1701 static void geneve_dellink(struct net_device *dev, struct list_head *head)
1703 struct geneve_dev *geneve = netdev_priv(dev);
1705 list_del(&geneve->next);
1706 unregister_netdevice_queue(dev, head);
1709 static size_t geneve_get_size(const struct net_device *dev)
1711 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */
1712 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */
1713 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */
1714 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */
1715 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */
1716 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */
1717 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */
1718 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */
1719 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */
1720 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */
1721 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */
1722 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */
1726 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
1728 struct geneve_dev *geneve = netdev_priv(dev);
1729 struct ip_tunnel_info *info = &geneve->cfg.info;
1730 bool ttl_inherit = geneve->cfg.ttl_inherit;
1731 bool metadata = geneve->cfg.collect_md;
1735 tunnel_id_to_vni(info->key.tun_id, tmp_vni);
1736 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2];
1737 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni))
1738 goto nla_put_failure;
1740 if (!metadata && ip_tunnel_info_af(info) == AF_INET) {
1741 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE,
1742 info->key.u.ipv4.dst))
1743 goto nla_put_failure;
1744 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM,
1745 !!(info->key.tun_flags & TUNNEL_CSUM)))
1746 goto nla_put_failure;
1748 #if IS_ENABLED(CONFIG_IPV6)
1749 } else if (!metadata) {
1750 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6,
1751 &info->key.u.ipv6.dst))
1752 goto nla_put_failure;
1753 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX,
1754 !(info->key.tun_flags & TUNNEL_CSUM)))
1755 goto nla_put_failure;
1759 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) ||
1760 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) ||
1761 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label))
1762 goto nla_put_failure;
1764 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df))
1765 goto nla_put_failure;
1767 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst))
1768 goto nla_put_failure;
1770 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA))
1771 goto nla_put_failure;
1773 #if IS_ENABLED(CONFIG_IPV6)
1774 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX,
1775 !geneve->cfg.use_udp6_rx_checksums))
1776 goto nla_put_failure;
1779 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
1780 goto nla_put_failure;
1788 static struct rtnl_link_ops geneve_link_ops __read_mostly = {
1790 .maxtype = IFLA_GENEVE_MAX,
1791 .policy = geneve_policy,
1792 .priv_size = sizeof(struct geneve_dev),
1793 .setup = geneve_setup,
1794 .validate = geneve_validate,
1795 .newlink = geneve_newlink,
1796 .changelink = geneve_changelink,
1797 .dellink = geneve_dellink,
1798 .get_size = geneve_get_size,
1799 .fill_info = geneve_fill_info,
1802 struct net_device *geneve_dev_create_fb(struct net *net, const char *name,
1803 u8 name_assign_type, u16 dst_port)
1805 struct nlattr *tb[IFLA_MAX + 1];
1806 struct net_device *dev;
1807 LIST_HEAD(list_kill);
1809 struct geneve_config cfg = {
1810 .df = GENEVE_DF_UNSET,
1811 .use_udp6_rx_checksums = true,
1812 .ttl_inherit = false,
1816 memset(tb, 0, sizeof(tb));
1817 dev = rtnl_create_link(net, name, name_assign_type,
1818 &geneve_link_ops, tb, NULL);
1822 init_tnl_info(&cfg.info, dst_port);
1823 err = geneve_configure(net, dev, NULL, &cfg);
1826 return ERR_PTR(err);
1829 /* openvswitch users expect packet sizes to be unrestricted,
1830 * so set the largest MTU we can.
1832 err = geneve_change_mtu(dev, IP_MAX_MTU);
1836 err = rtnl_configure_link(dev, NULL);
1842 geneve_dellink(dev, &list_kill);
1843 unregister_netdevice_many(&list_kill);
1844 return ERR_PTR(err);
1846 EXPORT_SYMBOL_GPL(geneve_dev_create_fb);
1848 static int geneve_netdevice_event(struct notifier_block *unused,
1849 unsigned long event, void *ptr)
1851 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1853 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
1854 event == NETDEV_UDP_TUNNEL_DROP_INFO) {
1855 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
1856 } else if (event == NETDEV_UNREGISTER) {
1857 if (!dev->udp_tunnel_nic_info)
1858 geneve_offload_rx_ports(dev, false);
1859 } else if (event == NETDEV_REGISTER) {
1860 if (!dev->udp_tunnel_nic_info)
1861 geneve_offload_rx_ports(dev, true);
1867 static struct notifier_block geneve_notifier_block __read_mostly = {
1868 .notifier_call = geneve_netdevice_event,
1871 static __net_init int geneve_init_net(struct net *net)
1873 struct geneve_net *gn = net_generic(net, geneve_net_id);
1875 INIT_LIST_HEAD(&gn->geneve_list);
1876 INIT_LIST_HEAD(&gn->sock_list);
1880 static void geneve_destroy_tunnels(struct net *net, struct list_head *head)
1882 struct geneve_net *gn = net_generic(net, geneve_net_id);
1883 struct geneve_dev *geneve, *next;
1884 struct net_device *dev, *aux;
1886 /* gather any geneve devices that were moved into this ns */
1887 for_each_netdev_safe(net, dev, aux)
1888 if (dev->rtnl_link_ops == &geneve_link_ops)
1889 unregister_netdevice_queue(dev, head);
1891 /* now gather any other geneve devices that were created in this ns */
1892 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) {
1893 /* If geneve->dev is in the same netns, it was already added
1894 * to the list by the previous loop.
1896 if (!net_eq(dev_net(geneve->dev), net))
1897 unregister_netdevice_queue(geneve->dev, head);
1901 static void __net_exit geneve_exit_batch_net(struct list_head *net_list)
1907 list_for_each_entry(net, net_list, exit_list)
1908 geneve_destroy_tunnels(net, &list);
1910 /* unregister the devices gathered above */
1911 unregister_netdevice_many(&list);
1914 list_for_each_entry(net, net_list, exit_list) {
1915 const struct geneve_net *gn = net_generic(net, geneve_net_id);
1917 WARN_ON_ONCE(!list_empty(&gn->sock_list));
1921 static struct pernet_operations geneve_net_ops = {
1922 .init = geneve_init_net,
1923 .exit_batch = geneve_exit_batch_net,
1924 .id = &geneve_net_id,
1925 .size = sizeof(struct geneve_net),
1928 static int __init geneve_init_module(void)
1932 rc = register_pernet_subsys(&geneve_net_ops);
1936 rc = register_netdevice_notifier(&geneve_notifier_block);
1940 rc = rtnl_link_register(&geneve_link_ops);
1946 unregister_netdevice_notifier(&geneve_notifier_block);
1948 unregister_pernet_subsys(&geneve_net_ops);
1952 late_initcall(geneve_init_module);
1954 static void __exit geneve_cleanup_module(void)
1956 rtnl_link_unregister(&geneve_link_ops);
1957 unregister_netdevice_notifier(&geneve_notifier_block);
1958 unregister_pernet_subsys(&geneve_net_ops);
1960 module_exit(geneve_cleanup_module);
1962 MODULE_LICENSE("GPL");
1963 MODULE_VERSION(GENEVE_NETDEV_VER);
1964 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>");
1965 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic");
1966 MODULE_ALIAS_RTNL_LINK("geneve");