xref: /openbmc/linux/net/ipv4/xfrm4_policy.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
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, int tos,
22 					  xfrm_address_t *saddr,
23 					  xfrm_address_t *daddr)
24 {
25 	struct flowi fl = {
26 		.fl4_dst = daddr->a4,
27 		.fl4_tos = tos,
28 	};
29 	struct dst_entry *dst;
30 	struct rtable *rt;
31 	int err;
32 
33 	if (saddr)
34 		fl.fl4_src = saddr->a4;
35 
36 	err = __ip_route_output_key(net, &rt, &fl);
37 	dst = &rt->dst;
38 	if (err)
39 		dst = ERR_PTR(err);
40 	return dst;
41 }
42 
43 static int xfrm4_get_saddr(struct net *net,
44 			   xfrm_address_t *saddr, xfrm_address_t *daddr)
45 {
46 	struct dst_entry *dst;
47 	struct rtable *rt;
48 
49 	dst = xfrm4_dst_lookup(net, 0, NULL, daddr);
50 	if (IS_ERR(dst))
51 		return -EHOSTUNREACH;
52 
53 	rt = (struct rtable *)dst;
54 	saddr->a4 = rt->rt_src;
55 	dst_release(dst);
56 	return 0;
57 }
58 
59 static int xfrm4_get_tos(struct flowi *fl)
60 {
61 	return IPTOS_RT_MASK & fl->fl4_tos; /* Strip ECN bits */
62 }
63 
64 static int xfrm4_init_path(struct xfrm_dst *path, struct dst_entry *dst,
65 			   int nfheader_len)
66 {
67 	return 0;
68 }
69 
70 static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
71 			  struct flowi *fl)
72 {
73 	struct rtable *rt = (struct rtable *)xdst->route;
74 
75 	xdst->u.rt.fl = *fl;
76 
77 	xdst->u.dst.dev = dev;
78 	dev_hold(dev);
79 
80 	xdst->u.rt.peer = rt->peer;
81 	if (rt->peer)
82 		atomic_inc(&rt->peer->refcnt);
83 
84 	/* Sheit... I remember I did this right. Apparently,
85 	 * it was magically lost, so this code needs audit */
86 	xdst->u.rt.rt_flags = rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST |
87 					      RTCF_LOCAL);
88 	xdst->u.rt.rt_type = rt->rt_type;
89 	xdst->u.rt.rt_src = rt->rt_src;
90 	xdst->u.rt.rt_dst = rt->rt_dst;
91 	xdst->u.rt.rt_gateway = rt->rt_gateway;
92 	xdst->u.rt.rt_spec_dst = rt->rt_spec_dst;
93 
94 	return 0;
95 }
96 
97 static void
98 _decode_session4(struct sk_buff *skb, struct flowi *fl, int reverse)
99 {
100 	struct iphdr *iph = ip_hdr(skb);
101 	u8 *xprth = skb_network_header(skb) + iph->ihl * 4;
102 
103 	memset(fl, 0, sizeof(struct flowi));
104 	fl->mark = skb->mark;
105 
106 	if (!(iph->frag_off & htons(IP_MF | IP_OFFSET))) {
107 		switch (iph->protocol) {
108 		case IPPROTO_UDP:
109 		case IPPROTO_UDPLITE:
110 		case IPPROTO_TCP:
111 		case IPPROTO_SCTP:
112 		case IPPROTO_DCCP:
113 			if (xprth + 4 < skb->data ||
114 			    pskb_may_pull(skb, xprth + 4 - skb->data)) {
115 				__be16 *ports = (__be16 *)xprth;
116 
117 				fl->fl_ip_sport = ports[!!reverse];
118 				fl->fl_ip_dport = ports[!reverse];
119 			}
120 			break;
121 
122 		case IPPROTO_ICMP:
123 			if (pskb_may_pull(skb, xprth + 2 - skb->data)) {
124 				u8 *icmp = xprth;
125 
126 				fl->fl_icmp_type = icmp[0];
127 				fl->fl_icmp_code = icmp[1];
128 			}
129 			break;
130 
131 		case IPPROTO_ESP:
132 			if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
133 				__be32 *ehdr = (__be32 *)xprth;
134 
135 				fl->fl_ipsec_spi = ehdr[0];
136 			}
137 			break;
138 
139 		case IPPROTO_AH:
140 			if (pskb_may_pull(skb, xprth + 8 - skb->data)) {
141 				__be32 *ah_hdr = (__be32*)xprth;
142 
143 				fl->fl_ipsec_spi = ah_hdr[1];
144 			}
145 			break;
146 
147 		case IPPROTO_COMP:
148 			if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
149 				__be16 *ipcomp_hdr = (__be16 *)xprth;
150 
151 				fl->fl_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
152 			}
153 			break;
154 
155 		case IPPROTO_GRE:
156 			if (pskb_may_pull(skb, xprth + 12 - skb->data)) {
157 				__be16 *greflags = (__be16 *)xprth;
158 				__be32 *gre_hdr = (__be32 *)xprth;
159 
160 				if (greflags[0] & GRE_KEY) {
161 					if (greflags[0] & GRE_CSUM)
162 						gre_hdr++;
163 					fl->fl_gre_key = gre_hdr[1];
164 				}
165 			}
166 			break;
167 
168 		default:
169 			fl->fl_ipsec_spi = 0;
170 			break;
171 		}
172 	}
173 	fl->proto = iph->protocol;
174 	fl->fl4_dst = reverse ? iph->saddr : iph->daddr;
175 	fl->fl4_src = reverse ? iph->daddr : iph->saddr;
176 	fl->fl4_tos = iph->tos;
177 }
178 
179 static inline int xfrm4_garbage_collect(struct dst_ops *ops)
180 {
181 	struct net *net = container_of(ops, struct net, xfrm.xfrm4_dst_ops);
182 
183 	xfrm4_policy_afinfo.garbage_collect(net);
184 	return (dst_entries_get_slow(ops) > ops->gc_thresh * 2);
185 }
186 
187 static void xfrm4_update_pmtu(struct dst_entry *dst, u32 mtu)
188 {
189 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
190 	struct dst_entry *path = xdst->route;
191 
192 	path->ops->update_pmtu(path, mtu);
193 }
194 
195 static void xfrm4_dst_destroy(struct dst_entry *dst)
196 {
197 	struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
198 
199 	if (likely(xdst->u.rt.peer))
200 		inet_putpeer(xdst->u.rt.peer);
201 	xfrm_dst_destroy(xdst);
202 }
203 
204 static void xfrm4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
205 			     int unregister)
206 {
207 	if (!unregister)
208 		return;
209 
210 	xfrm_dst_ifdown(dst, dev);
211 }
212 
213 static struct dst_ops xfrm4_dst_ops = {
214 	.family =		AF_INET,
215 	.protocol =		cpu_to_be16(ETH_P_IP),
216 	.gc =			xfrm4_garbage_collect,
217 	.update_pmtu =		xfrm4_update_pmtu,
218 	.destroy =		xfrm4_dst_destroy,
219 	.ifdown =		xfrm4_dst_ifdown,
220 	.local_out =		__ip_local_out,
221 	.gc_thresh =		1024,
222 };
223 
224 static struct xfrm_policy_afinfo xfrm4_policy_afinfo = {
225 	.family = 		AF_INET,
226 	.dst_ops =		&xfrm4_dst_ops,
227 	.dst_lookup =		xfrm4_dst_lookup,
228 	.get_saddr =		xfrm4_get_saddr,
229 	.decode_session =	_decode_session4,
230 	.get_tos =		xfrm4_get_tos,
231 	.init_path =		xfrm4_init_path,
232 	.fill_dst =		xfrm4_fill_dst,
233 };
234 
235 #ifdef CONFIG_SYSCTL
236 static struct ctl_table xfrm4_policy_table[] = {
237 	{
238 		.procname       = "xfrm4_gc_thresh",
239 		.data           = &init_net.xfrm.xfrm4_dst_ops.gc_thresh,
240 		.maxlen         = sizeof(int),
241 		.mode           = 0644,
242 		.proc_handler   = proc_dointvec,
243 	},
244 	{ }
245 };
246 
247 static struct ctl_table_header *sysctl_hdr;
248 #endif
249 
250 static void __init xfrm4_policy_init(void)
251 {
252 	xfrm_policy_register_afinfo(&xfrm4_policy_afinfo);
253 }
254 
255 static void __exit xfrm4_policy_fini(void)
256 {
257 #ifdef CONFIG_SYSCTL
258 	if (sysctl_hdr)
259 		unregister_net_sysctl_table(sysctl_hdr);
260 #endif
261 	xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo);
262 }
263 
264 void __init xfrm4_init(int rt_max_size)
265 {
266 	/*
267 	 * Select a default value for the gc_thresh based on the main route
268 	 * table hash size.  It seems to me the worst case scenario is when
269 	 * we have ipsec operating in transport mode, in which we create a
270 	 * dst_entry per socket.  The xfrm gc algorithm starts trying to remove
271 	 * entries at gc_thresh, and prevents new allocations as 2*gc_thresh
272 	 * so lets set an initial xfrm gc_thresh value at the rt_max_size/2.
273 	 * That will let us store an ipsec connection per route table entry,
274 	 * and start cleaning when were 1/2 full
275 	 */
276 	xfrm4_dst_ops.gc_thresh = rt_max_size/2;
277 	dst_entries_init(&xfrm4_dst_ops);
278 
279 	xfrm4_state_init();
280 	xfrm4_policy_init();
281 #ifdef CONFIG_SYSCTL
282 	sysctl_hdr = register_net_sysctl_table(&init_net, net_ipv4_ctl_path,
283 						xfrm4_policy_table);
284 #endif
285 }
286 
287