2a8d5cfc340fd4c7d206418ec3c4f893854df856
[~shefty/rdma-dev.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_route_iif = fl4->flowi4_iif;
83         xdst->u.rt.rt_iif = fl4->flowi4_iif;
84         xdst->u.rt.rt_oif = fl4->flowi4_oif;
85         xdst->u.rt.rt_mark = fl4->flowi4_mark;
86
87         xdst->u.dst.dev = dev;
88         dev_hold(dev);
89
90         /* Sheit... I remember I did this right. Apparently,
91          * it was magically lost, so this code needs audit */
92         xdst->u.rt.rt_flags = rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST |
93                                               RTCF_LOCAL);
94         xdst->u.rt.rt_type = rt->rt_type;
95         xdst->u.rt.rt_src = rt->rt_src;
96         xdst->u.rt.rt_dst = rt->rt_dst;
97         xdst->u.rt.rt_gateway = rt->rt_gateway;
98         xdst->u.rt.rt_pmtu = rt->rt_pmtu;
99
100         return 0;
101 }
102
103 static void
104 _decode_session4(struct sk_buff *skb, struct flowi *fl, int reverse)
105 {
106         const struct iphdr *iph = ip_hdr(skb);
107         u8 *xprth = skb_network_header(skb) + iph->ihl * 4;
108         struct flowi4 *fl4 = &fl->u.ip4;
109
110         memset(fl4, 0, sizeof(struct flowi4));
111         fl4->flowi4_mark = skb->mark;
112
113         if (!ip_is_fragment(iph)) {
114                 switch (iph->protocol) {
115                 case IPPROTO_UDP:
116                 case IPPROTO_UDPLITE:
117                 case IPPROTO_TCP:
118                 case IPPROTO_SCTP:
119                 case IPPROTO_DCCP:
120                         if (xprth + 4 < skb->data ||
121                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
122                                 __be16 *ports = (__be16 *)xprth;
123
124                                 fl4->fl4_sport = ports[!!reverse];
125                                 fl4->fl4_dport = ports[!reverse];
126                         }
127                         break;
128
129                 case IPPROTO_ICMP:
130                         if (pskb_may_pull(skb, xprth + 2 - skb->data)) {
131                                 u8 *icmp = xprth;
132
133                                 fl4->fl4_icmp_type = icmp[0];
134                                 fl4->fl4_icmp_code = icmp[1];
135                         }
136                         break;
137
138                 case IPPROTO_ESP:
139                         if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
140                                 __be32 *ehdr = (__be32 *)xprth;
141
142                                 fl4->fl4_ipsec_spi = ehdr[0];
143                         }
144                         break;
145
146                 case IPPROTO_AH:
147                         if (pskb_may_pull(skb, xprth + 8 - skb->data)) {
148                                 __be32 *ah_hdr = (__be32 *)xprth;
149
150                                 fl4->fl4_ipsec_spi = ah_hdr[1];
151                         }
152                         break;
153
154                 case IPPROTO_COMP:
155                         if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
156                                 __be16 *ipcomp_hdr = (__be16 *)xprth;
157
158                                 fl4->fl4_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
159                         }
160                         break;
161
162                 case IPPROTO_GRE:
163                         if (pskb_may_pull(skb, xprth + 12 - skb->data)) {
164                                 __be16 *greflags = (__be16 *)xprth;
165                                 __be32 *gre_hdr = (__be32 *)xprth;
166
167                                 if (greflags[0] & GRE_KEY) {
168                                         if (greflags[0] & GRE_CSUM)
169                                                 gre_hdr++;
170                                         fl4->fl4_gre_key = gre_hdr[1];
171                                 }
172                         }
173                         break;
174
175                 default:
176                         fl4->fl4_ipsec_spi = 0;
177                         break;
178                 }
179         }
180         fl4->flowi4_proto = iph->protocol;
181         fl4->daddr = reverse ? iph->saddr : iph->daddr;
182         fl4->saddr = reverse ? iph->daddr : iph->saddr;
183         fl4->flowi4_tos = iph->tos;
184 }
185
186 static inline int xfrm4_garbage_collect(struct dst_ops *ops)
187 {
188         struct net *net = container_of(ops, struct net, xfrm.xfrm4_dst_ops);
189
190         xfrm4_policy_afinfo.garbage_collect(net);
191         return (dst_entries_get_slow(ops) > ops->gc_thresh * 2);
192 }
193
194 static void xfrm4_update_pmtu(struct dst_entry *dst, struct sock *sk,
195                               struct sk_buff *skb, u32 mtu)
196 {
197         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
198         struct dst_entry *path = xdst->route;
199
200         path->ops->update_pmtu(path, sk, skb, mtu);
201 }
202
203 static void xfrm4_redirect(struct dst_entry *dst, struct sock *sk,
204                            struct sk_buff *skb)
205 {
206         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
207         struct dst_entry *path = xdst->route;
208
209         path->ops->redirect(path, sk, skb);
210 }
211
212 static void xfrm4_dst_destroy(struct dst_entry *dst)
213 {
214         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
215
216         dst_destroy_metrics_generic(dst);
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         .redirect =             xfrm4_redirect,
236         .cow_metrics =          dst_cow_metrics_generic,
237         .destroy =              xfrm4_dst_destroy,
238         .ifdown =               xfrm4_dst_ifdown,
239         .local_out =            __ip_local_out,
240         .gc_thresh =            1024,
241 };
242
243 static struct xfrm_policy_afinfo xfrm4_policy_afinfo = {
244         .family =               AF_INET,
245         .dst_ops =              &xfrm4_dst_ops,
246         .dst_lookup =           xfrm4_dst_lookup,
247         .get_saddr =            xfrm4_get_saddr,
248         .decode_session =       _decode_session4,
249         .get_tos =              xfrm4_get_tos,
250         .init_path =            xfrm4_init_path,
251         .fill_dst =             xfrm4_fill_dst,
252         .blackhole_route =      ipv4_blackhole_route,
253 };
254
255 #ifdef CONFIG_SYSCTL
256 static struct ctl_table xfrm4_policy_table[] = {
257         {
258                 .procname       = "xfrm4_gc_thresh",
259                 .data           = &init_net.xfrm.xfrm4_dst_ops.gc_thresh,
260                 .maxlen         = sizeof(int),
261                 .mode           = 0644,
262                 .proc_handler   = proc_dointvec,
263         },
264         { }
265 };
266
267 static struct ctl_table_header *sysctl_hdr;
268 #endif
269
270 static void __init xfrm4_policy_init(void)
271 {
272         xfrm_policy_register_afinfo(&xfrm4_policy_afinfo);
273 }
274
275 static void __exit xfrm4_policy_fini(void)
276 {
277 #ifdef CONFIG_SYSCTL
278         if (sysctl_hdr)
279                 unregister_net_sysctl_table(sysctl_hdr);
280 #endif
281         xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo);
282 }
283
284 void __init xfrm4_init(int rt_max_size)
285 {
286         /*
287          * Select a default value for the gc_thresh based on the main route
288          * table hash size.  It seems to me the worst case scenario is when
289          * we have ipsec operating in transport mode, in which we create a
290          * dst_entry per socket.  The xfrm gc algorithm starts trying to remove
291          * entries at gc_thresh, and prevents new allocations as 2*gc_thresh
292          * so lets set an initial xfrm gc_thresh value at the rt_max_size/2.
293          * That will let us store an ipsec connection per route table entry,
294          * and start cleaning when were 1/2 full
295          */
296         xfrm4_dst_ops.gc_thresh = rt_max_size/2;
297         dst_entries_init(&xfrm4_dst_ops);
298
299         xfrm4_state_init();
300         xfrm4_policy_init();
301 #ifdef CONFIG_SYSCTL
302         sysctl_hdr = register_net_sysctl(&init_net, "net/ipv4",
303                                          xfrm4_policy_table);
304 #endif
305 }
306