ipv4: fix ipsec forward performance regression
[linux-2.6.git] / net / ipv4 / xfrm4_policy.c
1 /*
2  * xfrm4_policy.c
3  *
4  * Changes:
5  *      Kazunori MIYAZAWA @USAGI
6  *      YOSHIFUJI Hideaki @USAGI
7  *              Split up af-specific portion
8  *
9  */
10
11 #include <linux/err.h>
12 #include <linux/kernel.h>
13 #include <linux/inetdevice.h>
14 #include <linux/if_tunnel.h>
15 #include <net/dst.h>
16 #include <net/xfrm.h>
17 #include <net/ip.h>
18
19 static struct xfrm_policy_afinfo xfrm4_policy_afinfo;
20
21 static struct dst_entry *__xfrm4_dst_lookup(struct net *net, struct flowi4 *fl4,
22                                             int tos,
23                                             const xfrm_address_t *saddr,
24                                             const xfrm_address_t *daddr)
25 {
26         struct rtable *rt;
27
28         memset(fl4, 0, sizeof(*fl4));
29         fl4->daddr = daddr->a4;
30         fl4->flowi4_tos = tos;
31         if (saddr)
32                 fl4->saddr = saddr->a4;
33
34         rt = __ip_route_output_key(net, fl4);
35         if (!IS_ERR(rt))
36                 return &rt->dst;
37
38         return ERR_CAST(rt);
39 }
40
41 static struct dst_entry *xfrm4_dst_lookup(struct net *net, int tos,
42                                           const xfrm_address_t *saddr,
43                                           const xfrm_address_t *daddr)
44 {
45         struct flowi4 fl4;
46
47         return __xfrm4_dst_lookup(net, &fl4, tos, saddr, daddr);
48 }
49
50 static int xfrm4_get_saddr(struct net *net,
51                            xfrm_address_t *saddr, xfrm_address_t *daddr)
52 {
53         struct dst_entry *dst;
54         struct flowi4 fl4;
55
56         dst = __xfrm4_dst_lookup(net, &fl4, 0, NULL, daddr);
57         if (IS_ERR(dst))
58                 return -EHOSTUNREACH;
59
60         saddr->a4 = fl4.saddr;
61         dst_release(dst);
62         return 0;
63 }
64
65 static int xfrm4_get_tos(const struct flowi *fl)
66 {
67         return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos; /* Strip ECN bits */
68 }
69
70 static int xfrm4_init_path(struct xfrm_dst *path, struct dst_entry *dst,
71                            int nfheader_len)
72 {
73         return 0;
74 }
75
76 static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
77                           const struct flowi *fl)
78 {
79         struct rtable *rt = (struct rtable *)xdst->route;
80         const struct flowi4 *fl4 = &fl->u.ip4;
81
82         xdst->u.rt.rt_key_dst = fl4->daddr;
83         xdst->u.rt.rt_key_src = fl4->saddr;
84         xdst->u.rt.rt_key_tos = fl4->flowi4_tos;
85         xdst->u.rt.rt_route_iif = fl4->flowi4_iif;
86         xdst->u.rt.rt_iif = fl4->flowi4_iif;
87         xdst->u.rt.rt_oif = fl4->flowi4_oif;
88         xdst->u.rt.rt_mark = fl4->flowi4_mark;
89
90         xdst->u.dst.dev = dev;
91         dev_hold(dev);
92
93         xdst->u.rt.peer = rt->peer;
94         if (rt->peer)
95                 atomic_inc(&rt->peer->refcnt);
96
97         /* Sheit... I remember I did this right. Apparently,
98          * it was magically lost, so this code needs audit */
99         xdst->u.rt.rt_flags = rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST |
100                                               RTCF_LOCAL);
101         xdst->u.rt.rt_type = rt->rt_type;
102         xdst->u.rt.rt_src = rt->rt_src;
103         xdst->u.rt.rt_dst = rt->rt_dst;
104         xdst->u.rt.rt_gateway = rt->rt_gateway;
105         xdst->u.rt.rt_spec_dst = rt->rt_spec_dst;
106
107         return 0;
108 }
109
110 static void
111 _decode_session4(struct sk_buff *skb, struct flowi *fl, int reverse)
112 {
113         const struct iphdr *iph = ip_hdr(skb);
114         u8 *xprth = skb_network_header(skb) + iph->ihl * 4;
115         struct flowi4 *fl4 = &fl->u.ip4;
116
117         memset(fl4, 0, sizeof(struct flowi4));
118         fl4->flowi4_mark = skb->mark;
119
120         if (!ip_is_fragment(iph)) {
121                 switch (iph->protocol) {
122                 case IPPROTO_UDP:
123                 case IPPROTO_UDPLITE:
124                 case IPPROTO_TCP:
125                 case IPPROTO_SCTP:
126                 case IPPROTO_DCCP:
127                         if (xprth + 4 < skb->data ||
128                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
129                                 __be16 *ports = (__be16 *)xprth;
130
131                                 fl4->fl4_sport = ports[!!reverse];
132                                 fl4->fl4_dport = ports[!reverse];
133                         }
134                         break;
135
136                 case IPPROTO_ICMP:
137                         if (pskb_may_pull(skb, xprth + 2 - skb->data)) {
138                                 u8 *icmp = xprth;
139
140                                 fl4->fl4_icmp_type = icmp[0];
141                                 fl4->fl4_icmp_code = icmp[1];
142                         }
143                         break;
144
145                 case IPPROTO_ESP:
146                         if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
147                                 __be32 *ehdr = (__be32 *)xprth;
148
149                                 fl4->fl4_ipsec_spi = ehdr[0];
150                         }
151                         break;
152
153                 case IPPROTO_AH:
154                         if (pskb_may_pull(skb, xprth + 8 - skb->data)) {
155                                 __be32 *ah_hdr = (__be32*)xprth;
156
157                                 fl4->fl4_ipsec_spi = ah_hdr[1];
158                         }
159                         break;
160
161                 case IPPROTO_COMP:
162                         if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
163                                 __be16 *ipcomp_hdr = (__be16 *)xprth;
164
165                                 fl4->fl4_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
166                         }
167                         break;
168
169                 case IPPROTO_GRE:
170                         if (pskb_may_pull(skb, xprth + 12 - skb->data)) {
171                                 __be16 *greflags = (__be16 *)xprth;
172                                 __be32 *gre_hdr = (__be32 *)xprth;
173
174                                 if (greflags[0] & GRE_KEY) {
175                                         if (greflags[0] & GRE_CSUM)
176                                                 gre_hdr++;
177                                         fl4->fl4_gre_key = gre_hdr[1];
178                                 }
179                         }
180                         break;
181
182                 default:
183                         fl4->fl4_ipsec_spi = 0;
184                         break;
185                 }
186         }
187         fl4->flowi4_proto = iph->protocol;
188         fl4->daddr = reverse ? iph->saddr : iph->daddr;
189         fl4->saddr = reverse ? iph->daddr : iph->saddr;
190         fl4->flowi4_tos = iph->tos;
191 }
192
193 static inline int xfrm4_garbage_collect(struct dst_ops *ops)
194 {
195         struct net *net = container_of(ops, struct net, xfrm.xfrm4_dst_ops);
196
197         xfrm4_policy_afinfo.garbage_collect(net);
198         return (dst_entries_get_slow(ops) > ops->gc_thresh * 2);
199 }
200
201 static void xfrm4_update_pmtu(struct dst_entry *dst, u32 mtu)
202 {
203         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
204         struct dst_entry *path = xdst->route;
205
206         path->ops->update_pmtu(path, mtu);
207 }
208
209 static void xfrm4_dst_destroy(struct dst_entry *dst)
210 {
211         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
212
213         dst_destroy_metrics_generic(dst);
214
215         if (likely(xdst->u.rt.peer))
216                 inet_putpeer(xdst->u.rt.peer);
217
218         xfrm_dst_destroy(xdst);
219 }
220
221 static void xfrm4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
222                              int unregister)
223 {
224         if (!unregister)
225                 return;
226
227         xfrm_dst_ifdown(dst, dev);
228 }
229
230 static struct dst_ops xfrm4_dst_ops = {
231         .family =               AF_INET,
232         .protocol =             cpu_to_be16(ETH_P_IP),
233         .gc =                   xfrm4_garbage_collect,
234         .update_pmtu =          xfrm4_update_pmtu,
235         .cow_metrics =          dst_cow_metrics_generic,
236         .destroy =              xfrm4_dst_destroy,
237         .ifdown =               xfrm4_dst_ifdown,
238         .local_out =            __ip_local_out,
239         .gc_thresh =            1024,
240 };
241
242 static struct xfrm_policy_afinfo xfrm4_policy_afinfo = {
243         .family =               AF_INET,
244         .dst_ops =              &xfrm4_dst_ops,
245         .dst_lookup =           xfrm4_dst_lookup,
246         .get_saddr =            xfrm4_get_saddr,
247         .decode_session =       _decode_session4,
248         .get_tos =              xfrm4_get_tos,
249         .init_path =            xfrm4_init_path,
250         .fill_dst =             xfrm4_fill_dst,
251         .blackhole_route =      ipv4_blackhole_route,
252 };
253
254 #ifdef CONFIG_SYSCTL
255 static struct ctl_table xfrm4_policy_table[] = {
256         {
257                 .procname       = "xfrm4_gc_thresh",
258                 .data           = &init_net.xfrm.xfrm4_dst_ops.gc_thresh,
259                 .maxlen         = sizeof(int),
260                 .mode           = 0644,
261                 .proc_handler   = proc_dointvec,
262         },
263         { }
264 };
265
266 static struct ctl_table_header *sysctl_hdr;
267 #endif
268
269 static void __init xfrm4_policy_init(void)
270 {
271         xfrm_policy_register_afinfo(&xfrm4_policy_afinfo);
272 }
273
274 static void __exit xfrm4_policy_fini(void)
275 {
276 #ifdef CONFIG_SYSCTL
277         if (sysctl_hdr)
278                 unregister_net_sysctl_table(sysctl_hdr);
279 #endif
280         xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo);
281 }
282
283 void __init xfrm4_init(int rt_max_size)
284 {
285         /*
286          * Select a default value for the gc_thresh based on the main route
287          * table hash size.  It seems to me the worst case scenario is when
288          * we have ipsec operating in transport mode, in which we create a
289          * dst_entry per socket.  The xfrm gc algorithm starts trying to remove
290          * entries at gc_thresh, and prevents new allocations as 2*gc_thresh
291          * so lets set an initial xfrm gc_thresh value at the rt_max_size/2.
292          * That will let us store an ipsec connection per route table entry,
293          * and start cleaning when were 1/2 full
294          */
295         xfrm4_dst_ops.gc_thresh = rt_max_size/2;
296         dst_entries_init(&xfrm4_dst_ops);
297
298         xfrm4_state_init();
299         xfrm4_policy_init();
300 #ifdef CONFIG_SYSCTL
301         sysctl_hdr = register_net_sysctl_table(&init_net, net_ipv4_ctl_path,
302                                                 xfrm4_policy_table);
303 #endif
304 }
305