5 * Kazunori MIYAZAWA @USAGI
6 * YOSHIFUJI Hideaki @USAGI
7 * Split up af-specific portion
11 #include <linux/err.h>
12 #include <linux/kernel.h>
13 #include <linux/inetdevice.h>
14 #include <linux/if_tunnel.h>
19 static struct xfrm_policy_afinfo xfrm4_policy_afinfo;
21 static struct dst_entry *xfrm4_dst_lookup(struct net *net, int tos,
22 const xfrm_address_t *saddr,
23 const xfrm_address_t *daddr)
32 fl.fl4_src = saddr->a4;
34 rt = __ip_route_output_key(net, &fl);
41 static int xfrm4_get_saddr(struct net *net,
42 xfrm_address_t *saddr, xfrm_address_t *daddr)
44 struct dst_entry *dst;
47 dst = xfrm4_dst_lookup(net, 0, NULL, daddr);
51 rt = (struct rtable *)dst;
52 saddr->a4 = rt->rt_src;
57 static int xfrm4_get_tos(const struct flowi *fl)
59 return IPTOS_RT_MASK & fl->fl4_tos; /* Strip ECN bits */
62 static int xfrm4_init_path(struct xfrm_dst *path, struct dst_entry *dst,
68 static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
69 const struct flowi *fl)
71 struct rtable *rt = (struct rtable *)xdst->route;
75 xdst->u.dst.dev = dev;
78 xdst->u.rt.peer = rt->peer;
80 atomic_inc(&rt->peer->refcnt);
82 /* Sheit... I remember I did this right. Apparently,
83 * it was magically lost, so this code needs audit */
84 xdst->u.rt.rt_flags = rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST |
86 xdst->u.rt.rt_type = rt->rt_type;
87 xdst->u.rt.rt_src = rt->rt_src;
88 xdst->u.rt.rt_dst = rt->rt_dst;
89 xdst->u.rt.rt_gateway = rt->rt_gateway;
90 xdst->u.rt.rt_spec_dst = rt->rt_spec_dst;
96 _decode_session4(struct sk_buff *skb, struct flowi *fl, int reverse)
98 struct iphdr *iph = ip_hdr(skb);
99 u8 *xprth = skb_network_header(skb) + iph->ihl * 4;
101 memset(fl, 0, sizeof(struct flowi));
102 fl->mark = skb->mark;
104 if (!(iph->frag_off & htons(IP_MF | IP_OFFSET))) {
105 switch (iph->protocol) {
107 case IPPROTO_UDPLITE:
111 if (xprth + 4 < skb->data ||
112 pskb_may_pull(skb, xprth + 4 - skb->data)) {
113 __be16 *ports = (__be16 *)xprth;
115 fl->fl_ip_sport = ports[!!reverse];
116 fl->fl_ip_dport = ports[!reverse];
121 if (pskb_may_pull(skb, xprth + 2 - skb->data)) {
124 fl->fl_icmp_type = icmp[0];
125 fl->fl_icmp_code = icmp[1];
130 if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
131 __be32 *ehdr = (__be32 *)xprth;
133 fl->fl_ipsec_spi = ehdr[0];
138 if (pskb_may_pull(skb, xprth + 8 - skb->data)) {
139 __be32 *ah_hdr = (__be32*)xprth;
141 fl->fl_ipsec_spi = ah_hdr[1];
146 if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
147 __be16 *ipcomp_hdr = (__be16 *)xprth;
149 fl->fl_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
154 if (pskb_may_pull(skb, xprth + 12 - skb->data)) {
155 __be16 *greflags = (__be16 *)xprth;
156 __be32 *gre_hdr = (__be32 *)xprth;
158 if (greflags[0] & GRE_KEY) {
159 if (greflags[0] & GRE_CSUM)
161 fl->fl_gre_key = gre_hdr[1];
167 fl->fl_ipsec_spi = 0;
171 fl->proto = iph->protocol;
172 fl->fl4_dst = reverse ? iph->saddr : iph->daddr;
173 fl->fl4_src = reverse ? iph->daddr : iph->saddr;
174 fl->fl4_tos = iph->tos;
177 static inline int xfrm4_garbage_collect(struct dst_ops *ops)
179 struct net *net = container_of(ops, struct net, xfrm.xfrm4_dst_ops);
181 xfrm4_policy_afinfo.garbage_collect(net);
182 return (dst_entries_get_slow(ops) > ops->gc_thresh * 2);
185 static void xfrm4_update_pmtu(struct dst_entry *dst, u32 mtu)
187 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
188 struct dst_entry *path = xdst->route;
190 path->ops->update_pmtu(path, mtu);
193 static void xfrm4_dst_destroy(struct dst_entry *dst)
195 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
197 dst_destroy_metrics_generic(dst);
199 if (likely(xdst->u.rt.peer))
200 inet_putpeer(xdst->u.rt.peer);
202 xfrm_dst_destroy(xdst);
205 static void xfrm4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
211 xfrm_dst_ifdown(dst, dev);
214 static struct dst_ops xfrm4_dst_ops = {
216 .protocol = cpu_to_be16(ETH_P_IP),
217 .gc = xfrm4_garbage_collect,
218 .update_pmtu = xfrm4_update_pmtu,
219 .cow_metrics = dst_cow_metrics_generic,
220 .destroy = xfrm4_dst_destroy,
221 .ifdown = xfrm4_dst_ifdown,
222 .local_out = __ip_local_out,
226 static struct xfrm_policy_afinfo xfrm4_policy_afinfo = {
228 .dst_ops = &xfrm4_dst_ops,
229 .dst_lookup = xfrm4_dst_lookup,
230 .get_saddr = xfrm4_get_saddr,
231 .decode_session = _decode_session4,
232 .get_tos = xfrm4_get_tos,
233 .init_path = xfrm4_init_path,
234 .fill_dst = xfrm4_fill_dst,
235 .blackhole_route = ipv4_blackhole_route,
239 static struct ctl_table xfrm4_policy_table[] = {
241 .procname = "xfrm4_gc_thresh",
242 .data = &init_net.xfrm.xfrm4_dst_ops.gc_thresh,
243 .maxlen = sizeof(int),
245 .proc_handler = proc_dointvec,
250 static struct ctl_table_header *sysctl_hdr;
253 static void __init xfrm4_policy_init(void)
255 xfrm_policy_register_afinfo(&xfrm4_policy_afinfo);
258 static void __exit xfrm4_policy_fini(void)
262 unregister_net_sysctl_table(sysctl_hdr);
264 xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo);
267 void __init xfrm4_init(int rt_max_size)
270 * Select a default value for the gc_thresh based on the main route
271 * table hash size. It seems to me the worst case scenario is when
272 * we have ipsec operating in transport mode, in which we create a
273 * dst_entry per socket. The xfrm gc algorithm starts trying to remove
274 * entries at gc_thresh, and prevents new allocations as 2*gc_thresh
275 * so lets set an initial xfrm gc_thresh value at the rt_max_size/2.
276 * That will let us store an ipsec connection per route table entry,
277 * and start cleaning when were 1/2 full
279 xfrm4_dst_ops.gc_thresh = rt_max_size/2;
280 dst_entries_init(&xfrm4_dst_ops);
285 sysctl_hdr = register_net_sysctl_table(&init_net, net_ipv4_ctl_path,