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>
18 static struct xfrm_policy_afinfo xfrm4_policy_afinfo;
20 static struct dst_entry *xfrm4_dst_lookup(struct net *net, int tos,
21 xfrm_address_t *saddr,
22 xfrm_address_t *daddr)
32 struct dst_entry *dst;
37 fl.fl4_src = saddr->a4;
39 err = __ip_route_output_key(net, &rt, &fl);
46 static int xfrm4_get_saddr(struct net *net,
47 xfrm_address_t *saddr, xfrm_address_t *daddr)
49 struct dst_entry *dst;
52 dst = xfrm4_dst_lookup(net, 0, NULL, daddr);
56 rt = (struct rtable *)dst;
57 saddr->a4 = rt->rt_src;
62 static int xfrm4_get_tos(struct flowi *fl)
64 return IPTOS_RT_MASK & fl->fl4_tos; /* Strip ECN bits */
67 static int xfrm4_init_path(struct xfrm_dst *path, struct dst_entry *dst,
73 static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
76 struct rtable *rt = (struct rtable *)xdst->route;
80 xdst->u.dst.dev = dev;
83 xdst->u.rt.peer = rt->peer;
85 atomic_inc(&rt->peer->refcnt);
87 /* Sheit... I remember I did this right. Apparently,
88 * it was magically lost, so this code needs audit */
89 xdst->u.rt.rt_flags = rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST |
91 xdst->u.rt.rt_type = rt->rt_type;
92 xdst->u.rt.rt_src = rt->rt_src;
93 xdst->u.rt.rt_dst = rt->rt_dst;
94 xdst->u.rt.rt_gateway = rt->rt_gateway;
95 xdst->u.rt.rt_spec_dst = rt->rt_spec_dst;
101 _decode_session4(struct sk_buff *skb, struct flowi *fl, int reverse)
103 struct iphdr *iph = ip_hdr(skb);
104 u8 *xprth = skb_network_header(skb) + iph->ihl * 4;
106 memset(fl, 0, sizeof(struct flowi));
107 fl->mark = skb->mark;
109 if (!(iph->frag_off & htons(IP_MF | IP_OFFSET))) {
110 switch (iph->protocol) {
112 case IPPROTO_UDPLITE:
116 if (xprth + 4 < skb->data ||
117 pskb_may_pull(skb, xprth + 4 - skb->data)) {
118 __be16 *ports = (__be16 *)xprth;
120 fl->fl_ip_sport = ports[!!reverse];
121 fl->fl_ip_dport = ports[!reverse];
126 if (pskb_may_pull(skb, xprth + 2 - skb->data)) {
129 fl->fl_icmp_type = icmp[0];
130 fl->fl_icmp_code = icmp[1];
135 if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
136 __be32 *ehdr = (__be32 *)xprth;
138 fl->fl_ipsec_spi = ehdr[0];
143 if (pskb_may_pull(skb, xprth + 8 - skb->data)) {
144 __be32 *ah_hdr = (__be32*)xprth;
146 fl->fl_ipsec_spi = ah_hdr[1];
151 if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
152 __be16 *ipcomp_hdr = (__be16 *)xprth;
154 fl->fl_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
158 fl->fl_ipsec_spi = 0;
162 fl->proto = iph->protocol;
163 fl->fl4_dst = reverse ? iph->saddr : iph->daddr;
164 fl->fl4_src = reverse ? iph->daddr : iph->saddr;
165 fl->fl4_tos = iph->tos;
168 static inline int xfrm4_garbage_collect(struct dst_ops *ops)
170 struct net *net = container_of(ops, struct net, xfrm.xfrm4_dst_ops);
172 xfrm4_policy_afinfo.garbage_collect(net);
173 return (dst_entries_get_slow(ops) > ops->gc_thresh * 2);
176 static void xfrm4_update_pmtu(struct dst_entry *dst, u32 mtu)
178 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
179 struct dst_entry *path = xdst->route;
181 path->ops->update_pmtu(path, mtu);
184 static void xfrm4_dst_destroy(struct dst_entry *dst)
186 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
188 if (likely(xdst->u.rt.peer))
189 inet_putpeer(xdst->u.rt.peer);
190 xfrm_dst_destroy(xdst);
193 static void xfrm4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
199 xfrm_dst_ifdown(dst, dev);
202 static struct dst_ops xfrm4_dst_ops = {
204 .protocol = cpu_to_be16(ETH_P_IP),
205 .gc = xfrm4_garbage_collect,
206 .update_pmtu = xfrm4_update_pmtu,
207 .destroy = xfrm4_dst_destroy,
208 .ifdown = xfrm4_dst_ifdown,
209 .local_out = __ip_local_out,
213 static struct xfrm_policy_afinfo xfrm4_policy_afinfo = {
215 .dst_ops = &xfrm4_dst_ops,
216 .dst_lookup = xfrm4_dst_lookup,
217 .get_saddr = xfrm4_get_saddr,
218 .decode_session = _decode_session4,
219 .get_tos = xfrm4_get_tos,
220 .init_path = xfrm4_init_path,
221 .fill_dst = xfrm4_fill_dst,
225 static struct ctl_table xfrm4_policy_table[] = {
227 .procname = "xfrm4_gc_thresh",
228 .data = &init_net.xfrm.xfrm4_dst_ops.gc_thresh,
229 .maxlen = sizeof(int),
231 .proc_handler = proc_dointvec,
236 static struct ctl_table_header *sysctl_hdr;
239 static void __init xfrm4_policy_init(void)
241 xfrm_policy_register_afinfo(&xfrm4_policy_afinfo);
244 static void __exit xfrm4_policy_fini(void)
248 unregister_net_sysctl_table(sysctl_hdr);
250 xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo);
253 void __init xfrm4_init(int rt_max_size)
256 * Select a default value for the gc_thresh based on the main route
257 * table hash size. It seems to me the worst case scenario is when
258 * we have ipsec operating in transport mode, in which we create a
259 * dst_entry per socket. The xfrm gc algorithm starts trying to remove
260 * entries at gc_thresh, and prevents new allocations as 2*gc_thresh
261 * so lets set an initial xfrm gc_thresh value at the rt_max_size/2.
262 * That will let us store an ipsec connection per route table entry,
263 * and start cleaning when were 1/2 full
265 xfrm4_dst_ops.gc_thresh = rt_max_size/2;
266 dst_entries_init(&xfrm4_dst_ops);
271 sysctl_hdr = register_net_sysctl_table(&init_net, net_ipv4_ctl_path,