10e6fbc5bSPravin B Shelar /* 20e6fbc5bSPravin B Shelar * Copyright (c) 2013 Nicira, Inc. 30e6fbc5bSPravin B Shelar * 40e6fbc5bSPravin B Shelar * This program is free software; you can redistribute it and/or 50e6fbc5bSPravin B Shelar * modify it under the terms of version 2 of the GNU General Public 60e6fbc5bSPravin B Shelar * License as published by the Free Software Foundation. 70e6fbc5bSPravin B Shelar * 80e6fbc5bSPravin B Shelar * This program is distributed in the hope that it will be useful, but 90e6fbc5bSPravin B Shelar * WITHOUT ANY WARRANTY; without even the implied warranty of 100e6fbc5bSPravin B Shelar * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 110e6fbc5bSPravin B Shelar * General Public License for more details. 120e6fbc5bSPravin B Shelar * 130e6fbc5bSPravin B Shelar * You should have received a copy of the GNU General Public License 140e6fbc5bSPravin B Shelar * along with this program; if not, write to the Free Software 150e6fbc5bSPravin B Shelar * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 160e6fbc5bSPravin B Shelar * 02110-1301, USA 170e6fbc5bSPravin B Shelar */ 180e6fbc5bSPravin B Shelar 190e6fbc5bSPravin B Shelar #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 200e6fbc5bSPravin B Shelar 210e6fbc5bSPravin B Shelar #include <linux/types.h> 220e6fbc5bSPravin B Shelar #include <linux/kernel.h> 230e6fbc5bSPravin B Shelar #include <linux/skbuff.h> 240e6fbc5bSPravin B Shelar #include <linux/netdevice.h> 250e6fbc5bSPravin B Shelar #include <linux/in.h> 260e6fbc5bSPravin B Shelar #include <linux/if_arp.h> 270e6fbc5bSPravin B Shelar #include <linux/init.h> 280e6fbc5bSPravin B Shelar #include <linux/in6.h> 290e6fbc5bSPravin B Shelar #include <linux/inetdevice.h> 300e6fbc5bSPravin B Shelar #include <linux/netfilter_ipv4.h> 310e6fbc5bSPravin B Shelar #include <linux/etherdevice.h> 320e6fbc5bSPravin B Shelar #include <linux/if_ether.h> 330e6fbc5bSPravin B Shelar #include <linux/if_vlan.h> 34e7030878SThomas Graf #include <linux/static_key.h> 350e6fbc5bSPravin B Shelar 360e6fbc5bSPravin B Shelar #include <net/ip.h> 370e6fbc5bSPravin B Shelar #include <net/icmp.h> 380e6fbc5bSPravin B Shelar #include <net/protocol.h> 390e6fbc5bSPravin B Shelar #include <net/ip_tunnels.h> 40058214a4STom Herbert #include <net/ip6_tunnel.h> 410e6fbc5bSPravin B Shelar #include <net/arp.h> 420e6fbc5bSPravin B Shelar #include <net/checksum.h> 430e6fbc5bSPravin B Shelar #include <net/dsfield.h> 440e6fbc5bSPravin B Shelar #include <net/inet_ecn.h> 450e6fbc5bSPravin B Shelar #include <net/xfrm.h> 460e6fbc5bSPravin B Shelar #include <net/net_namespace.h> 470e6fbc5bSPravin B Shelar #include <net/netns/generic.h> 480e6fbc5bSPravin B Shelar #include <net/rtnetlink.h> 4963d008a4SJiri Benc #include <net/dst_metadata.h> 500e6fbc5bSPravin B Shelar 5155c2bc14STom Herbert const struct ip_tunnel_encap_ops __rcu * 5255c2bc14STom Herbert iptun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly; 5355c2bc14STom Herbert EXPORT_SYMBOL(iptun_encaps); 5455c2bc14STom Herbert 55058214a4STom Herbert const struct ip6_tnl_encap_ops __rcu * 56058214a4STom Herbert ip6tun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly; 57058214a4STom Herbert EXPORT_SYMBOL(ip6tun_encaps); 58058214a4STom Herbert 59039f5062SPravin B Shelar void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb, 600e6fbc5bSPravin B Shelar __be32 src, __be32 dst, __u8 proto, 61963a88b3SNicolas Dichtel __u8 tos, __u8 ttl, __be16 df, bool xnet) 620e6fbc5bSPravin B Shelar { 63bc22a0e2SNicolas Dichtel int pkt_len = skb->len - skb_inner_network_offset(skb); 64f859b0f6SEric W. Biederman struct net *net = dev_net(rt->dst.dev); 65039f5062SPravin B Shelar struct net_device *dev = skb->dev; 66b8247f09SShmulik Ladkani int skb_iif = skb->skb_iif; 670e6fbc5bSPravin B Shelar struct iphdr *iph; 680e6fbc5bSPravin B Shelar int err; 690e6fbc5bSPravin B Shelar 70963a88b3SNicolas Dichtel skb_scrub_packet(skb, xnet); 71963a88b3SNicolas Dichtel 727539fadcSTom Herbert skb_clear_hash(skb); 730e6fbc5bSPravin B Shelar skb_dst_set(skb, &rt->dst); 740e6fbc5bSPravin B Shelar memset(IPCB(skb), 0, sizeof(*IPCB(skb))); 750e6fbc5bSPravin B Shelar 76c0451fe1SShmulik Ladkani if (skb_iif && !(df & htons(IP_DF))) { 77c0451fe1SShmulik Ladkani /* Arrived from an ingress interface, got encapsulated, with 78c0451fe1SShmulik Ladkani * fragmentation of encapulating frames allowed. 79c0451fe1SShmulik Ladkani * If skb is gso, the resulting encapsulated network segments 80c0451fe1SShmulik Ladkani * may exceed dst mtu. 81b8247f09SShmulik Ladkani * Allow IP Fragmentation of segments. 82b8247f09SShmulik Ladkani */ 83b8247f09SShmulik Ladkani IPCB(skb)->flags |= IPSKB_FRAG_SEGS; 84b8247f09SShmulik Ladkani } 85b8247f09SShmulik Ladkani 860e6fbc5bSPravin B Shelar /* Push down and install the IP header. */ 8778a3694dSSteffen Klassert skb_push(skb, sizeof(struct iphdr)); 880e6fbc5bSPravin B Shelar skb_reset_network_header(skb); 890e6fbc5bSPravin B Shelar 900e6fbc5bSPravin B Shelar iph = ip_hdr(skb); 910e6fbc5bSPravin B Shelar 920e6fbc5bSPravin B Shelar iph->version = 4; 930e6fbc5bSPravin B Shelar iph->ihl = sizeof(struct iphdr) >> 2; 940e6fbc5bSPravin B Shelar iph->frag_off = df; 950e6fbc5bSPravin B Shelar iph->protocol = proto; 960e6fbc5bSPravin B Shelar iph->tos = tos; 970e6fbc5bSPravin B Shelar iph->daddr = dst; 980e6fbc5bSPravin B Shelar iph->saddr = src; 990e6fbc5bSPravin B Shelar iph->ttl = ttl; 100f859b0f6SEric W. Biederman __ip_select_ident(net, iph, skb_shinfo(skb)->gso_segs ?: 1); 1010e6fbc5bSPravin B Shelar 10233224b16SEric W. Biederman err = ip_local_out(net, sk, skb); 1030e6fbc5bSPravin B Shelar if (unlikely(net_xmit_eval(err))) 1040e6fbc5bSPravin B Shelar pkt_len = 0; 105039f5062SPravin B Shelar iptunnel_xmit_stats(dev, pkt_len); 1060e6fbc5bSPravin B Shelar } 1070e6fbc5bSPravin B Shelar EXPORT_SYMBOL_GPL(iptunnel_xmit); 1083d7b46cdSPravin B Shelar 109a6d5bbf3SJiri Benc int __iptunnel_pull_header(struct sk_buff *skb, int hdr_len, 110a6d5bbf3SJiri Benc __be16 inner_proto, bool raw_proto, bool xnet) 1113d7b46cdSPravin B Shelar { 1123d7b46cdSPravin B Shelar if (unlikely(!pskb_may_pull(skb, hdr_len))) 1133d7b46cdSPravin B Shelar return -ENOMEM; 1143d7b46cdSPravin B Shelar 1153d7b46cdSPravin B Shelar skb_pull_rcsum(skb, hdr_len); 1163d7b46cdSPravin B Shelar 117a6d5bbf3SJiri Benc if (!raw_proto && inner_proto == htons(ETH_P_TEB)) { 1181245dfc8SLi RongQing struct ethhdr *eh; 1193d7b46cdSPravin B Shelar 1203d7b46cdSPravin B Shelar if (unlikely(!pskb_may_pull(skb, ETH_HLEN))) 1213d7b46cdSPravin B Shelar return -ENOMEM; 1223d7b46cdSPravin B Shelar 1231245dfc8SLi RongQing eh = (struct ethhdr *)skb->data; 124d181ddcaSAlexander Duyck if (likely(eth_proto_is_802_3(eh->h_proto))) 1253d7b46cdSPravin B Shelar skb->protocol = eh->h_proto; 1263d7b46cdSPravin B Shelar else 1273d7b46cdSPravin B Shelar skb->protocol = htons(ETH_P_802_2); 1283d7b46cdSPravin B Shelar 1293d7b46cdSPravin B Shelar } else { 1303d7b46cdSPravin B Shelar skb->protocol = inner_proto; 1313d7b46cdSPravin B Shelar } 1323d7b46cdSPravin B Shelar 1337539fadcSTom Herbert skb_clear_hash_if_not_l4(skb); 1343d7b46cdSPravin B Shelar skb->vlan_tci = 0; 1353d7b46cdSPravin B Shelar skb_set_queue_mapping(skb, 0); 1367f290c94SJiri Benc skb_scrub_packet(skb, xnet); 137a09a4c8dSJesse Gross 138a09a4c8dSJesse Gross return iptunnel_pull_offloads(skb); 1393d7b46cdSPravin B Shelar } 140a6d5bbf3SJiri Benc EXPORT_SYMBOL_GPL(__iptunnel_pull_header); 1412d26f0a3SEric Dumazet 14263d008a4SJiri Benc struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md, 14363d008a4SJiri Benc gfp_t flags) 14463d008a4SJiri Benc { 14563d008a4SJiri Benc struct metadata_dst *res; 14663d008a4SJiri Benc struct ip_tunnel_info *dst, *src; 14763d008a4SJiri Benc 14863d008a4SJiri Benc if (!md || md->u.tun_info.mode & IP_TUNNEL_INFO_TX) 14963d008a4SJiri Benc return NULL; 15063d008a4SJiri Benc 15163d008a4SJiri Benc res = metadata_dst_alloc(0, flags); 15263d008a4SJiri Benc if (!res) 15363d008a4SJiri Benc return NULL; 15463d008a4SJiri Benc 15563d008a4SJiri Benc dst = &res->u.tun_info; 15663d008a4SJiri Benc src = &md->u.tun_info; 15763d008a4SJiri Benc dst->key.tun_id = src->key.tun_id; 15863d008a4SJiri Benc if (src->mode & IP_TUNNEL_INFO_IPV6) 15963d008a4SJiri Benc memcpy(&dst->key.u.ipv6.dst, &src->key.u.ipv6.src, 16063d008a4SJiri Benc sizeof(struct in6_addr)); 16163d008a4SJiri Benc else 16263d008a4SJiri Benc dst->key.u.ipv4.dst = src->key.u.ipv4.src; 16363d008a4SJiri Benc dst->mode = src->mode | IP_TUNNEL_INFO_TX; 16463d008a4SJiri Benc 16563d008a4SJiri Benc return res; 16663d008a4SJiri Benc } 16763d008a4SJiri Benc EXPORT_SYMBOL_GPL(iptunnel_metadata_reply); 16863d008a4SJiri Benc 169aed069dfSAlexander Duyck int iptunnel_handle_offloads(struct sk_buff *skb, 1702d26f0a3SEric Dumazet int gso_type_mask) 1712d26f0a3SEric Dumazet { 1722d26f0a3SEric Dumazet int err; 1732d26f0a3SEric Dumazet 1742d26f0a3SEric Dumazet if (likely(!skb->encapsulation)) { 1752d26f0a3SEric Dumazet skb_reset_inner_headers(skb); 1762d26f0a3SEric Dumazet skb->encapsulation = 1; 1772d26f0a3SEric Dumazet } 1782d26f0a3SEric Dumazet 1792d26f0a3SEric Dumazet if (skb_is_gso(skb)) { 1809580bf2eSEric Dumazet err = skb_header_unclone(skb, GFP_ATOMIC); 1812d26f0a3SEric Dumazet if (unlikely(err)) 182aed069dfSAlexander Duyck return err; 1832d26f0a3SEric Dumazet skb_shinfo(skb)->gso_type |= gso_type_mask; 184aed069dfSAlexander Duyck return 0; 1852d26f0a3SEric Dumazet } 1862d26f0a3SEric Dumazet 1876fa79666SEdward Cree if (skb->ip_summed != CHECKSUM_PARTIAL) { 1882d26f0a3SEric Dumazet skb->ip_summed = CHECKSUM_NONE; 1896fa79666SEdward Cree /* We clear encapsulation here to prevent badly-written 1906fa79666SEdward Cree * drivers potentially deciding to offload an inner checksum 1916fa79666SEdward Cree * if we set CHECKSUM_PARTIAL on the outer header. 1926fa79666SEdward Cree * This should go away when the drivers are all fixed. 1936fa79666SEdward Cree */ 194179bc67fSEdward Cree skb->encapsulation = 0; 195179bc67fSEdward Cree } 1962d26f0a3SEric Dumazet 197aed069dfSAlexander Duyck return 0; 1982d26f0a3SEric Dumazet } 1992d26f0a3SEric Dumazet EXPORT_SYMBOL_GPL(iptunnel_handle_offloads); 200ebe44f35SDavid S. Miller 201ebe44f35SDavid S. Miller /* Often modified stats are per cpu, other are shared (netdev->stats) */ 202ebe44f35SDavid S. Miller struct rtnl_link_stats64 *ip_tunnel_get_stats64(struct net_device *dev, 203ebe44f35SDavid S. Miller struct rtnl_link_stats64 *tot) 204ebe44f35SDavid S. Miller { 205ebe44f35SDavid S. Miller int i; 206ebe44f35SDavid S. Miller 207c24a5964SAlexander Duyck netdev_stats_to_stats64(tot, &dev->stats); 208c24a5964SAlexander Duyck 209ebe44f35SDavid S. Miller for_each_possible_cpu(i) { 210ebe44f35SDavid S. Miller const struct pcpu_sw_netstats *tstats = 211ebe44f35SDavid S. Miller per_cpu_ptr(dev->tstats, i); 212ebe44f35SDavid S. Miller u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 213ebe44f35SDavid S. Miller unsigned int start; 214ebe44f35SDavid S. Miller 215ebe44f35SDavid S. Miller do { 21657a7744eSEric W. Biederman start = u64_stats_fetch_begin_irq(&tstats->syncp); 217ebe44f35SDavid S. Miller rx_packets = tstats->rx_packets; 218ebe44f35SDavid S. Miller tx_packets = tstats->tx_packets; 219ebe44f35SDavid S. Miller rx_bytes = tstats->rx_bytes; 220ebe44f35SDavid S. Miller tx_bytes = tstats->tx_bytes; 22157a7744eSEric W. Biederman } while (u64_stats_fetch_retry_irq(&tstats->syncp, start)); 222ebe44f35SDavid S. Miller 223ebe44f35SDavid S. Miller tot->rx_packets += rx_packets; 224ebe44f35SDavid S. Miller tot->tx_packets += tx_packets; 225ebe44f35SDavid S. Miller tot->rx_bytes += rx_bytes; 226ebe44f35SDavid S. Miller tot->tx_bytes += tx_bytes; 227ebe44f35SDavid S. Miller } 228ebe44f35SDavid S. Miller 229ebe44f35SDavid S. Miller return tot; 230ebe44f35SDavid S. Miller } 231ebe44f35SDavid S. Miller EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64); 2323093fbe7SThomas Graf 233a1c234f9SJiri Benc static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = { 234a1c234f9SJiri Benc [LWTUNNEL_IP_ID] = { .type = NLA_U64 }, 235a1c234f9SJiri Benc [LWTUNNEL_IP_DST] = { .type = NLA_U32 }, 236a1c234f9SJiri Benc [LWTUNNEL_IP_SRC] = { .type = NLA_U32 }, 237a1c234f9SJiri Benc [LWTUNNEL_IP_TTL] = { .type = NLA_U8 }, 238a1c234f9SJiri Benc [LWTUNNEL_IP_TOS] = { .type = NLA_U8 }, 239a1c234f9SJiri Benc [LWTUNNEL_IP_FLAGS] = { .type = NLA_U16 }, 2403093fbe7SThomas Graf }; 2413093fbe7SThomas Graf 2423093fbe7SThomas Graf static int ip_tun_build_state(struct net_device *dev, struct nlattr *attr, 243127eb7cdSTom Herbert unsigned int family, const void *cfg, 2443093fbe7SThomas Graf struct lwtunnel_state **ts) 2453093fbe7SThomas Graf { 2463093fbe7SThomas Graf struct ip_tunnel_info *tun_info; 2473093fbe7SThomas Graf struct lwtunnel_state *new_state; 248a1c234f9SJiri Benc struct nlattr *tb[LWTUNNEL_IP_MAX + 1]; 2493093fbe7SThomas Graf int err; 2503093fbe7SThomas Graf 251a1c234f9SJiri Benc err = nla_parse_nested(tb, LWTUNNEL_IP_MAX, attr, ip_tun_policy); 2523093fbe7SThomas Graf if (err < 0) 2533093fbe7SThomas Graf return err; 2543093fbe7SThomas Graf 2553093fbe7SThomas Graf new_state = lwtunnel_state_alloc(sizeof(*tun_info)); 2563093fbe7SThomas Graf if (!new_state) 2573093fbe7SThomas Graf return -ENOMEM; 2583093fbe7SThomas Graf 2593093fbe7SThomas Graf new_state->type = LWTUNNEL_ENCAP_IP; 2603093fbe7SThomas Graf 2613093fbe7SThomas Graf tun_info = lwt_tun_info(new_state); 2623093fbe7SThomas Graf 263a1c234f9SJiri Benc if (tb[LWTUNNEL_IP_ID]) 26430d3d83aSLance Richardson tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]); 2653093fbe7SThomas Graf 266a1c234f9SJiri Benc if (tb[LWTUNNEL_IP_DST]) 2677822ce73SHaishuang Yan tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]); 2683093fbe7SThomas Graf 269a1c234f9SJiri Benc if (tb[LWTUNNEL_IP_SRC]) 2707822ce73SHaishuang Yan tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]); 2713093fbe7SThomas Graf 272a1c234f9SJiri Benc if (tb[LWTUNNEL_IP_TTL]) 2737c383fb2SJiri Benc tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]); 2743093fbe7SThomas Graf 275a1c234f9SJiri Benc if (tb[LWTUNNEL_IP_TOS]) 2767c383fb2SJiri Benc tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]); 2773093fbe7SThomas Graf 278a1c234f9SJiri Benc if (tb[LWTUNNEL_IP_FLAGS]) 27930d3d83aSLance Richardson tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP_FLAGS]); 2803093fbe7SThomas Graf 2813093fbe7SThomas Graf tun_info->mode = IP_TUNNEL_INFO_TX; 2823093fbe7SThomas Graf tun_info->options_len = 0; 2833093fbe7SThomas Graf 2843093fbe7SThomas Graf *ts = new_state; 2853093fbe7SThomas Graf 2863093fbe7SThomas Graf return 0; 2873093fbe7SThomas Graf } 2883093fbe7SThomas Graf 2893093fbe7SThomas Graf static int ip_tun_fill_encap_info(struct sk_buff *skb, 2903093fbe7SThomas Graf struct lwtunnel_state *lwtstate) 2913093fbe7SThomas Graf { 2923093fbe7SThomas Graf struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate); 2933093fbe7SThomas Graf 294b46f6dedSNicolas Dichtel if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id, 295b46f6dedSNicolas Dichtel LWTUNNEL_IP_PAD) || 2967822ce73SHaishuang Yan nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) || 2977822ce73SHaishuang Yan nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) || 2987c383fb2SJiri Benc nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) || 2997c383fb2SJiri Benc nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) || 30030d3d83aSLance Richardson nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags)) 3013093fbe7SThomas Graf return -ENOMEM; 3023093fbe7SThomas Graf 3033093fbe7SThomas Graf return 0; 3043093fbe7SThomas Graf } 3053093fbe7SThomas Graf 3063093fbe7SThomas Graf static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate) 3073093fbe7SThomas Graf { 308b46f6dedSNicolas Dichtel return nla_total_size_64bit(8) /* LWTUNNEL_IP_ID */ 309a1c234f9SJiri Benc + nla_total_size(4) /* LWTUNNEL_IP_DST */ 310a1c234f9SJiri Benc + nla_total_size(4) /* LWTUNNEL_IP_SRC */ 311a1c234f9SJiri Benc + nla_total_size(1) /* LWTUNNEL_IP_TOS */ 312a1c234f9SJiri Benc + nla_total_size(1) /* LWTUNNEL_IP_TTL */ 313a1c234f9SJiri Benc + nla_total_size(2); /* LWTUNNEL_IP_FLAGS */ 3143093fbe7SThomas Graf } 3153093fbe7SThomas Graf 3162d798499SJiri Benc static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b) 3172d798499SJiri Benc { 3182d798499SJiri Benc return memcmp(lwt_tun_info(a), lwt_tun_info(b), 3192d798499SJiri Benc sizeof(struct ip_tunnel_info)); 3202d798499SJiri Benc } 3212d798499SJiri Benc 3223093fbe7SThomas Graf static const struct lwtunnel_encap_ops ip_tun_lwt_ops = { 3233093fbe7SThomas Graf .build_state = ip_tun_build_state, 3243093fbe7SThomas Graf .fill_encap = ip_tun_fill_encap_info, 3253093fbe7SThomas Graf .get_encap_size = ip_tun_encap_nlsize, 3262d798499SJiri Benc .cmp_encap = ip_tun_cmp_encap, 3273093fbe7SThomas Graf }; 3283093fbe7SThomas Graf 32932a2b002SJiri Benc static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = { 33032a2b002SJiri Benc [LWTUNNEL_IP6_ID] = { .type = NLA_U64 }, 33132a2b002SJiri Benc [LWTUNNEL_IP6_DST] = { .len = sizeof(struct in6_addr) }, 33232a2b002SJiri Benc [LWTUNNEL_IP6_SRC] = { .len = sizeof(struct in6_addr) }, 33332a2b002SJiri Benc [LWTUNNEL_IP6_HOPLIMIT] = { .type = NLA_U8 }, 33432a2b002SJiri Benc [LWTUNNEL_IP6_TC] = { .type = NLA_U8 }, 33532a2b002SJiri Benc [LWTUNNEL_IP6_FLAGS] = { .type = NLA_U16 }, 33632a2b002SJiri Benc }; 33732a2b002SJiri Benc 33832a2b002SJiri Benc static int ip6_tun_build_state(struct net_device *dev, struct nlattr *attr, 339127eb7cdSTom Herbert unsigned int family, const void *cfg, 34032a2b002SJiri Benc struct lwtunnel_state **ts) 34132a2b002SJiri Benc { 34232a2b002SJiri Benc struct ip_tunnel_info *tun_info; 34332a2b002SJiri Benc struct lwtunnel_state *new_state; 34432a2b002SJiri Benc struct nlattr *tb[LWTUNNEL_IP6_MAX + 1]; 34532a2b002SJiri Benc int err; 34632a2b002SJiri Benc 34732a2b002SJiri Benc err = nla_parse_nested(tb, LWTUNNEL_IP6_MAX, attr, ip6_tun_policy); 34832a2b002SJiri Benc if (err < 0) 34932a2b002SJiri Benc return err; 35032a2b002SJiri Benc 35132a2b002SJiri Benc new_state = lwtunnel_state_alloc(sizeof(*tun_info)); 35232a2b002SJiri Benc if (!new_state) 35332a2b002SJiri Benc return -ENOMEM; 35432a2b002SJiri Benc 35532a2b002SJiri Benc new_state->type = LWTUNNEL_ENCAP_IP6; 35632a2b002SJiri Benc 35732a2b002SJiri Benc tun_info = lwt_tun_info(new_state); 35832a2b002SJiri Benc 35932a2b002SJiri Benc if (tb[LWTUNNEL_IP6_ID]) 36030d3d83aSLance Richardson tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]); 36132a2b002SJiri Benc 36232a2b002SJiri Benc if (tb[LWTUNNEL_IP6_DST]) 36332a2b002SJiri Benc tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]); 36432a2b002SJiri Benc 36532a2b002SJiri Benc if (tb[LWTUNNEL_IP6_SRC]) 36632a2b002SJiri Benc tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]); 36732a2b002SJiri Benc 36832a2b002SJiri Benc if (tb[LWTUNNEL_IP6_HOPLIMIT]) 36932a2b002SJiri Benc tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]); 37032a2b002SJiri Benc 37132a2b002SJiri Benc if (tb[LWTUNNEL_IP6_TC]) 37232a2b002SJiri Benc tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]); 37332a2b002SJiri Benc 37432a2b002SJiri Benc if (tb[LWTUNNEL_IP6_FLAGS]) 37530d3d83aSLance Richardson tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]); 37632a2b002SJiri Benc 3777f9562a1SJiri Benc tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6; 37832a2b002SJiri Benc tun_info->options_len = 0; 37932a2b002SJiri Benc 38032a2b002SJiri Benc *ts = new_state; 38132a2b002SJiri Benc 38232a2b002SJiri Benc return 0; 38332a2b002SJiri Benc } 38432a2b002SJiri Benc 38532a2b002SJiri Benc static int ip6_tun_fill_encap_info(struct sk_buff *skb, 38632a2b002SJiri Benc struct lwtunnel_state *lwtstate) 38732a2b002SJiri Benc { 38832a2b002SJiri Benc struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate); 38932a2b002SJiri Benc 390b46f6dedSNicolas Dichtel if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id, 391b46f6dedSNicolas Dichtel LWTUNNEL_IP6_PAD) || 39232a2b002SJiri Benc nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) || 39332a2b002SJiri Benc nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) || 394995096a0SQuentin Armitage nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.tos) || 395995096a0SQuentin Armitage nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.ttl) || 39630d3d83aSLance Richardson nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags)) 39732a2b002SJiri Benc return -ENOMEM; 39832a2b002SJiri Benc 39932a2b002SJiri Benc return 0; 40032a2b002SJiri Benc } 40132a2b002SJiri Benc 40232a2b002SJiri Benc static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate) 40332a2b002SJiri Benc { 404b46f6dedSNicolas Dichtel return nla_total_size_64bit(8) /* LWTUNNEL_IP6_ID */ 40532a2b002SJiri Benc + nla_total_size(16) /* LWTUNNEL_IP6_DST */ 40632a2b002SJiri Benc + nla_total_size(16) /* LWTUNNEL_IP6_SRC */ 40732a2b002SJiri Benc + nla_total_size(1) /* LWTUNNEL_IP6_HOPLIMIT */ 40832a2b002SJiri Benc + nla_total_size(1) /* LWTUNNEL_IP6_TC */ 40932a2b002SJiri Benc + nla_total_size(2); /* LWTUNNEL_IP6_FLAGS */ 41032a2b002SJiri Benc } 41132a2b002SJiri Benc 41232a2b002SJiri Benc static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = { 41332a2b002SJiri Benc .build_state = ip6_tun_build_state, 41432a2b002SJiri Benc .fill_encap = ip6_tun_fill_encap_info, 41532a2b002SJiri Benc .get_encap_size = ip6_tun_encap_nlsize, 41632a2b002SJiri Benc .cmp_encap = ip_tun_cmp_encap, 41732a2b002SJiri Benc }; 41832a2b002SJiri Benc 419045a0fa0SThomas Graf void __init ip_tunnel_core_init(void) 4203093fbe7SThomas Graf { 421fca5fdf6SDaniel Borkmann /* If you land here, make sure whether increasing ip_tunnel_info's 422fca5fdf6SDaniel Borkmann * options_len is a reasonable choice with its usage in front ends 423fca5fdf6SDaniel Borkmann * (f.e., it's part of flow keys, etc). 424fca5fdf6SDaniel Borkmann */ 425fca5fdf6SDaniel Borkmann BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255); 426fca5fdf6SDaniel Borkmann 4273093fbe7SThomas Graf lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP); 42832a2b002SJiri Benc lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6); 4293093fbe7SThomas Graf } 430e7030878SThomas Graf 431e7030878SThomas Graf struct static_key ip_tunnel_metadata_cnt = STATIC_KEY_INIT_FALSE; 432e7030878SThomas Graf EXPORT_SYMBOL(ip_tunnel_metadata_cnt); 433e7030878SThomas Graf 434e7030878SThomas Graf void ip_tunnel_need_metadata(void) 435e7030878SThomas Graf { 436e7030878SThomas Graf static_key_slow_inc(&ip_tunnel_metadata_cnt); 437e7030878SThomas Graf } 438e7030878SThomas Graf EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata); 439e7030878SThomas Graf 440e7030878SThomas Graf void ip_tunnel_unneed_metadata(void) 441e7030878SThomas Graf { 442e7030878SThomas Graf static_key_slow_dec(&ip_tunnel_metadata_cnt); 443e7030878SThomas Graf } 444e7030878SThomas Graf EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata); 445