1e3e4712eSRoopa Prabhu /* 2e3e4712eSRoopa Prabhu * mpls tunnels An implementation mpls tunnels using the light weight tunnel 3e3e4712eSRoopa Prabhu * infrastructure 4e3e4712eSRoopa Prabhu * 5e3e4712eSRoopa Prabhu * Authors: Roopa Prabhu, <roopa@cumulusnetworks.com> 6e3e4712eSRoopa Prabhu * 7e3e4712eSRoopa Prabhu * This program is free software; you can redistribute it and/or 8e3e4712eSRoopa Prabhu * modify it under the terms of the GNU General Public License 9e3e4712eSRoopa Prabhu * as published by the Free Software Foundation; either version 10e3e4712eSRoopa Prabhu * 2 of the License, or (at your option) any later version. 11e3e4712eSRoopa Prabhu * 12e3e4712eSRoopa Prabhu */ 13e3e4712eSRoopa Prabhu #include <linux/types.h> 14e3e4712eSRoopa Prabhu #include <linux/skbuff.h> 15e3e4712eSRoopa Prabhu #include <linux/net.h> 16e3e4712eSRoopa Prabhu #include <linux/module.h> 17e3e4712eSRoopa Prabhu #include <linux/mpls.h> 18e3e4712eSRoopa Prabhu #include <linux/vmalloc.h> 19e3e4712eSRoopa Prabhu #include <net/ip.h> 20e3e4712eSRoopa Prabhu #include <net/dst.h> 21e3e4712eSRoopa Prabhu #include <net/lwtunnel.h> 22e3e4712eSRoopa Prabhu #include <net/netevent.h> 23e3e4712eSRoopa Prabhu #include <net/netns/generic.h> 24e3e4712eSRoopa Prabhu #include <net/ip6_fib.h> 25e3e4712eSRoopa Prabhu #include <net/route.h> 26e3e4712eSRoopa Prabhu #include <net/mpls_iptunnel.h> 27e3e4712eSRoopa Prabhu #include <linux/mpls_iptunnel.h> 28e3e4712eSRoopa Prabhu #include "internal.h" 29e3e4712eSRoopa Prabhu 30e3e4712eSRoopa Prabhu static const struct nla_policy mpls_iptunnel_policy[MPLS_IPTUNNEL_MAX + 1] = { 31e3e4712eSRoopa Prabhu [MPLS_IPTUNNEL_DST] = { .type = NLA_U32 }, 32e3e4712eSRoopa Prabhu }; 33e3e4712eSRoopa Prabhu 34e3e4712eSRoopa Prabhu static unsigned int mpls_encap_size(struct mpls_iptunnel_encap *en) 35e3e4712eSRoopa Prabhu { 36e3e4712eSRoopa Prabhu /* The size of the layer 2.5 labels to be added for this route */ 37e3e4712eSRoopa Prabhu return en->labels * sizeof(struct mpls_shim_hdr); 38e3e4712eSRoopa Prabhu } 39e3e4712eSRoopa Prabhu 4014972cbdSRoopa Prabhu static int mpls_xmit(struct sk_buff *skb) 41e3e4712eSRoopa Prabhu { 42e3e4712eSRoopa Prabhu struct mpls_iptunnel_encap *tun_encap_info; 43e3e4712eSRoopa Prabhu struct mpls_shim_hdr *hdr; 44e3e4712eSRoopa Prabhu struct net_device *out_dev; 45e3e4712eSRoopa Prabhu unsigned int hh_len; 46e3e4712eSRoopa Prabhu unsigned int new_header_size; 47e3e4712eSRoopa Prabhu unsigned int mtu; 48e3e4712eSRoopa Prabhu struct dst_entry *dst = skb_dst(skb); 49e3e4712eSRoopa Prabhu struct rtable *rt = NULL; 50e3e4712eSRoopa Prabhu struct rt6_info *rt6 = NULL; 5127d69105SRobert Shearman struct mpls_dev *out_mdev; 52e3e4712eSRoopa Prabhu int err = 0; 53e3e4712eSRoopa Prabhu bool bos; 54e3e4712eSRoopa Prabhu int i; 55e3e4712eSRoopa Prabhu unsigned int ttl; 56e3e4712eSRoopa Prabhu 5727d69105SRobert Shearman /* Find the output device */ 5827d69105SRobert Shearman out_dev = dst->dev; 5927d69105SRobert Shearman 60e3e4712eSRoopa Prabhu /* Obtain the ttl */ 61fe82b330SRobert Shearman if (dst->ops->family == AF_INET) { 62e3e4712eSRoopa Prabhu ttl = ip_hdr(skb)->ttl; 63e3e4712eSRoopa Prabhu rt = (struct rtable *)dst; 64fe82b330SRobert Shearman } else if (dst->ops->family == AF_INET6) { 65e3e4712eSRoopa Prabhu ttl = ipv6_hdr(skb)->hop_limit; 66e3e4712eSRoopa Prabhu rt6 = (struct rt6_info *)dst; 67e3e4712eSRoopa Prabhu } else { 68e3e4712eSRoopa Prabhu goto drop; 69e3e4712eSRoopa Prabhu } 70e3e4712eSRoopa Prabhu 71e3e4712eSRoopa Prabhu skb_orphan(skb); 72e3e4712eSRoopa Prabhu 73e3e4712eSRoopa Prabhu if (!mpls_output_possible(out_dev) || 7461adedf3SJiri Benc !dst->lwtstate || skb_warn_if_lro(skb)) 75e3e4712eSRoopa Prabhu goto drop; 76e3e4712eSRoopa Prabhu 77e3e4712eSRoopa Prabhu skb_forward_csum(skb); 78e3e4712eSRoopa Prabhu 7961adedf3SJiri Benc tun_encap_info = mpls_lwtunnel_encap(dst->lwtstate); 80e3e4712eSRoopa Prabhu 81e3e4712eSRoopa Prabhu /* Verify the destination can hold the packet */ 82e3e4712eSRoopa Prabhu new_header_size = mpls_encap_size(tun_encap_info); 83e3e4712eSRoopa Prabhu mtu = mpls_dev_mtu(out_dev); 84e3e4712eSRoopa Prabhu if (mpls_pkt_too_big(skb, mtu - new_header_size)) 85e3e4712eSRoopa Prabhu goto drop; 86e3e4712eSRoopa Prabhu 87e3e4712eSRoopa Prabhu hh_len = LL_RESERVED_SPACE(out_dev); 88e3e4712eSRoopa Prabhu if (!out_dev->header_ops) 89e3e4712eSRoopa Prabhu hh_len = 0; 90e3e4712eSRoopa Prabhu 91e3e4712eSRoopa Prabhu /* Ensure there is enough space for the headers in the skb */ 92e3e4712eSRoopa Prabhu if (skb_cow(skb, hh_len + new_header_size)) 93e3e4712eSRoopa Prabhu goto drop; 94e3e4712eSRoopa Prabhu 9548d2ab60SDavid Ahern skb_set_inner_protocol(skb, skb->protocol); 9648d2ab60SDavid Ahern skb_reset_inner_network_header(skb); 9748d2ab60SDavid Ahern 98e3e4712eSRoopa Prabhu skb_push(skb, new_header_size); 9948d2ab60SDavid Ahern 100e3e4712eSRoopa Prabhu skb_reset_network_header(skb); 101e3e4712eSRoopa Prabhu 102e3e4712eSRoopa Prabhu skb->dev = out_dev; 103e3e4712eSRoopa Prabhu skb->protocol = htons(ETH_P_MPLS_UC); 104e3e4712eSRoopa Prabhu 105e3e4712eSRoopa Prabhu /* Push the new labels */ 106e3e4712eSRoopa Prabhu hdr = mpls_hdr(skb); 107e3e4712eSRoopa Prabhu bos = true; 108e3e4712eSRoopa Prabhu for (i = tun_encap_info->labels - 1; i >= 0; i--) { 109e3e4712eSRoopa Prabhu hdr[i] = mpls_entry_encode(tun_encap_info->label[i], 110e3e4712eSRoopa Prabhu ttl, 0, bos); 111e3e4712eSRoopa Prabhu bos = false; 112e3e4712eSRoopa Prabhu } 113e3e4712eSRoopa Prabhu 11427d69105SRobert Shearman mpls_stats_inc_outucastpkts(out_dev, skb); 11527d69105SRobert Shearman 116e3e4712eSRoopa Prabhu if (rt) 117e3e4712eSRoopa Prabhu err = neigh_xmit(NEIGH_ARP_TABLE, out_dev, &rt->rt_gateway, 118e3e4712eSRoopa Prabhu skb); 119e3e4712eSRoopa Prabhu else if (rt6) 120e3e4712eSRoopa Prabhu err = neigh_xmit(NEIGH_ND_TABLE, out_dev, &rt6->rt6i_gateway, 121e3e4712eSRoopa Prabhu skb); 122e3e4712eSRoopa Prabhu if (err) 123e3e4712eSRoopa Prabhu net_dbg_ratelimited("%s: packet transmission failed: %d\n", 124e3e4712eSRoopa Prabhu __func__, err); 125e3e4712eSRoopa Prabhu 12614972cbdSRoopa Prabhu return LWTUNNEL_XMIT_DONE; 127e3e4712eSRoopa Prabhu 128e3e4712eSRoopa Prabhu drop: 12927d69105SRobert Shearman out_mdev = out_dev ? mpls_dev_get(out_dev) : NULL; 13027d69105SRobert Shearman if (out_mdev) 13127d69105SRobert Shearman MPLS_INC_STATS(out_mdev, tx_errors); 132e3e4712eSRoopa Prabhu kfree_skb(skb); 133e3e4712eSRoopa Prabhu return -EINVAL; 134e3e4712eSRoopa Prabhu } 135e3e4712eSRoopa Prabhu 136e3e4712eSRoopa Prabhu static int mpls_build_state(struct net_device *dev, struct nlattr *nla, 137127eb7cdSTom Herbert unsigned int family, const void *cfg, 138e3e4712eSRoopa Prabhu struct lwtunnel_state **ts) 139e3e4712eSRoopa Prabhu { 140e3e4712eSRoopa Prabhu struct mpls_iptunnel_encap *tun_encap_info; 141e3e4712eSRoopa Prabhu struct nlattr *tb[MPLS_IPTUNNEL_MAX + 1]; 142e3e4712eSRoopa Prabhu struct lwtunnel_state *newts; 143e3e4712eSRoopa Prabhu int ret; 144e3e4712eSRoopa Prabhu 145e3e4712eSRoopa Prabhu ret = nla_parse_nested(tb, MPLS_IPTUNNEL_MAX, nla, 146e3e4712eSRoopa Prabhu mpls_iptunnel_policy); 147e3e4712eSRoopa Prabhu if (ret < 0) 148e3e4712eSRoopa Prabhu return ret; 149e3e4712eSRoopa Prabhu 150e3e4712eSRoopa Prabhu if (!tb[MPLS_IPTUNNEL_DST]) 151e3e4712eSRoopa Prabhu return -EINVAL; 152e3e4712eSRoopa Prabhu 153e3e4712eSRoopa Prabhu 154f76a9db3SThomas Graf newts = lwtunnel_state_alloc(sizeof(*tun_encap_info)); 155e3e4712eSRoopa Prabhu if (!newts) 156e3e4712eSRoopa Prabhu return -ENOMEM; 157e3e4712eSRoopa Prabhu 158e3e4712eSRoopa Prabhu tun_encap_info = mpls_lwtunnel_encap(newts); 159e3e4712eSRoopa Prabhu ret = nla_get_labels(tb[MPLS_IPTUNNEL_DST], MAX_NEW_LABELS, 160e3e4712eSRoopa Prabhu &tun_encap_info->labels, tun_encap_info->label); 161e3e4712eSRoopa Prabhu if (ret) 162e3e4712eSRoopa Prabhu goto errout; 163e3e4712eSRoopa Prabhu newts->type = LWTUNNEL_ENCAP_MPLS; 16414972cbdSRoopa Prabhu newts->flags |= LWTUNNEL_STATE_XMIT_REDIRECT; 16514972cbdSRoopa Prabhu newts->headroom = mpls_encap_size(tun_encap_info); 166e3e4712eSRoopa Prabhu 167e3e4712eSRoopa Prabhu *ts = newts; 168e3e4712eSRoopa Prabhu 169e3e4712eSRoopa Prabhu return 0; 170e3e4712eSRoopa Prabhu 171e3e4712eSRoopa Prabhu errout: 172e3e4712eSRoopa Prabhu kfree(newts); 173e3e4712eSRoopa Prabhu *ts = NULL; 174e3e4712eSRoopa Prabhu 175e3e4712eSRoopa Prabhu return ret; 176e3e4712eSRoopa Prabhu } 177e3e4712eSRoopa Prabhu 178e3e4712eSRoopa Prabhu static int mpls_fill_encap_info(struct sk_buff *skb, 179e3e4712eSRoopa Prabhu struct lwtunnel_state *lwtstate) 180e3e4712eSRoopa Prabhu { 181e3e4712eSRoopa Prabhu struct mpls_iptunnel_encap *tun_encap_info; 182e3e4712eSRoopa Prabhu 183e3e4712eSRoopa Prabhu tun_encap_info = mpls_lwtunnel_encap(lwtstate); 184e3e4712eSRoopa Prabhu 185e3e4712eSRoopa Prabhu if (nla_put_labels(skb, MPLS_IPTUNNEL_DST, tun_encap_info->labels, 186e3e4712eSRoopa Prabhu tun_encap_info->label)) 187e3e4712eSRoopa Prabhu goto nla_put_failure; 188e3e4712eSRoopa Prabhu 189e3e4712eSRoopa Prabhu return 0; 190e3e4712eSRoopa Prabhu 191e3e4712eSRoopa Prabhu nla_put_failure: 192e3e4712eSRoopa Prabhu return -EMSGSIZE; 193e3e4712eSRoopa Prabhu } 194e3e4712eSRoopa Prabhu 195e3e4712eSRoopa Prabhu static int mpls_encap_nlsize(struct lwtunnel_state *lwtstate) 196e3e4712eSRoopa Prabhu { 197e3e4712eSRoopa Prabhu struct mpls_iptunnel_encap *tun_encap_info; 198e3e4712eSRoopa Prabhu 199e3e4712eSRoopa Prabhu tun_encap_info = mpls_lwtunnel_encap(lwtstate); 200e3e4712eSRoopa Prabhu 201e3e4712eSRoopa Prabhu return nla_total_size(tun_encap_info->labels * 4); 202e3e4712eSRoopa Prabhu } 203e3e4712eSRoopa Prabhu 204e3e4712eSRoopa Prabhu static int mpls_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b) 205e3e4712eSRoopa Prabhu { 206e3e4712eSRoopa Prabhu struct mpls_iptunnel_encap *a_hdr = mpls_lwtunnel_encap(a); 207e3e4712eSRoopa Prabhu struct mpls_iptunnel_encap *b_hdr = mpls_lwtunnel_encap(b); 208e3e4712eSRoopa Prabhu int l; 209e3e4712eSRoopa Prabhu 210e3e4712eSRoopa Prabhu if (a_hdr->labels != b_hdr->labels) 211e3e4712eSRoopa Prabhu return 1; 212e3e4712eSRoopa Prabhu 213e3e4712eSRoopa Prabhu for (l = 0; l < MAX_NEW_LABELS; l++) 214e3e4712eSRoopa Prabhu if (a_hdr->label[l] != b_hdr->label[l]) 215e3e4712eSRoopa Prabhu return 1; 216e3e4712eSRoopa Prabhu return 0; 217e3e4712eSRoopa Prabhu } 218e3e4712eSRoopa Prabhu 219e3e4712eSRoopa Prabhu static const struct lwtunnel_encap_ops mpls_iptun_ops = { 220e3e4712eSRoopa Prabhu .build_state = mpls_build_state, 22114972cbdSRoopa Prabhu .xmit = mpls_xmit, 222e3e4712eSRoopa Prabhu .fill_encap = mpls_fill_encap_info, 223e3e4712eSRoopa Prabhu .get_encap_size = mpls_encap_nlsize, 224e3e4712eSRoopa Prabhu .cmp_encap = mpls_encap_cmp, 225e3e4712eSRoopa Prabhu }; 226e3e4712eSRoopa Prabhu 227e3e4712eSRoopa Prabhu static int __init mpls_iptunnel_init(void) 228e3e4712eSRoopa Prabhu { 229e3e4712eSRoopa Prabhu return lwtunnel_encap_add_ops(&mpls_iptun_ops, LWTUNNEL_ENCAP_MPLS); 230e3e4712eSRoopa Prabhu } 231e3e4712eSRoopa Prabhu module_init(mpls_iptunnel_init); 232e3e4712eSRoopa Prabhu 233e3e4712eSRoopa Prabhu static void __exit mpls_iptunnel_exit(void) 234e3e4712eSRoopa Prabhu { 235e3e4712eSRoopa Prabhu lwtunnel_encap_del_ops(&mpls_iptun_ops, LWTUNNEL_ENCAP_MPLS); 236e3e4712eSRoopa Prabhu } 237e3e4712eSRoopa Prabhu module_exit(mpls_iptunnel_exit); 238e3e4712eSRoopa Prabhu 239b2b04edcSRobert Shearman MODULE_ALIAS_RTNL_LWT(MPLS); 240e3e4712eSRoopa Prabhu MODULE_DESCRIPTION("MultiProtocol Label Switching IP Tunnels"); 241e3e4712eSRoopa Prabhu MODULE_LICENSE("GPL v2"); 242