]>
Commit | Line | Data |
---|---|---|
e905a9ed | 1 | /* |
1da177e4 LT |
2 | * xfrm4_policy.c |
3 | * | |
4 | * Changes: | |
5 | * Kazunori MIYAZAWA @USAGI | |
6 | * YOSHIFUJI Hideaki @USAGI | |
7 | * Split up af-specific portion | |
e905a9ed | 8 | * |
1da177e4 LT |
9 | */ |
10 | ||
66cdb3ca HX |
11 | #include <linux/err.h> |
12 | #include <linux/kernel.h> | |
aabc9761 | 13 | #include <linux/inetdevice.h> |
cc9ff19d | 14 | #include <linux/if_tunnel.h> |
45ff5a3f | 15 | #include <net/dst.h> |
1da177e4 LT |
16 | #include <net/xfrm.h> |
17 | #include <net/ip.h> | |
18 | ||
1da177e4 LT |
19 | static struct xfrm_policy_afinfo xfrm4_policy_afinfo; |
20 | ||
c5b3cf46 AD |
21 | static struct dst_entry *xfrm4_dst_lookup(struct net *net, int tos, |
22 | xfrm_address_t *saddr, | |
66cdb3ca | 23 | xfrm_address_t *daddr) |
1da177e4 | 24 | { |
66cdb3ca | 25 | struct flowi fl = { |
5811662b CG |
26 | .fl4_dst = daddr->a4, |
27 | .fl4_tos = tos, | |
a1e59abf | 28 | }; |
66cdb3ca HX |
29 | struct dst_entry *dst; |
30 | struct rtable *rt; | |
31 | int err; | |
a1e59abf | 32 | |
66cdb3ca HX |
33 | if (saddr) |
34 | fl.fl4_src = saddr->a4; | |
35 | ||
c5b3cf46 | 36 | err = __ip_route_output_key(net, &rt, &fl); |
d8d1f30b | 37 | dst = &rt->dst; |
66cdb3ca HX |
38 | if (err) |
39 | dst = ERR_PTR(err); | |
40 | return dst; | |
41 | } | |
42 | ||
fbda33b2 AD |
43 | static int xfrm4_get_saddr(struct net *net, |
44 | xfrm_address_t *saddr, xfrm_address_t *daddr) | |
66cdb3ca HX |
45 | { |
46 | struct dst_entry *dst; | |
47 | struct rtable *rt; | |
48 | ||
fbda33b2 | 49 | dst = xfrm4_dst_lookup(net, 0, NULL, daddr); |
66cdb3ca HX |
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; | |
a1e59abf PM |
57 | } |
58 | ||
05d84025 | 59 | static int xfrm4_get_tos(const struct flowi *fl) |
1da177e4 | 60 | { |
94e22389 | 61 | return IPTOS_RT_MASK & fl->fl4_tos; /* Strip ECN bits */ |
25ee3286 | 62 | } |
1da177e4 | 63 | |
a1b05140 MN |
64 | static int xfrm4_init_path(struct xfrm_dst *path, struct dst_entry *dst, |
65 | int nfheader_len) | |
66 | { | |
67 | return 0; | |
68 | } | |
69 | ||
87c1e12b HX |
70 | static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev, |
71 | struct flowi *fl) | |
25ee3286 HX |
72 | { |
73 | struct rtable *rt = (struct rtable *)xdst->route; | |
1da177e4 | 74 | |
87c1e12b | 75 | xdst->u.rt.fl = *fl; |
1da177e4 | 76 | |
25ee3286 HX |
77 | xdst->u.dst.dev = dev; |
78 | dev_hold(dev); | |
43372262 | 79 | |
25ee3286 HX |
80 | xdst->u.rt.peer = rt->peer; |
81 | if (rt->peer) | |
82 | atomic_inc(&rt->peer->refcnt); | |
66cdb3ca | 83 | |
25ee3286 HX |
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; | |
1da177e4 | 93 | |
1da177e4 | 94 | return 0; |
1da177e4 LT |
95 | } |
96 | ||
97 | static void | |
d5422efe | 98 | _decode_session4(struct sk_buff *skb, struct flowi *fl, int reverse) |
1da177e4 | 99 | { |
eddc9ec5 | 100 | struct iphdr *iph = ip_hdr(skb); |
d56f90a7 | 101 | u8 *xprth = skb_network_header(skb) + iph->ihl * 4; |
1da177e4 LT |
102 | |
103 | memset(fl, 0, sizeof(struct flowi)); | |
44b451f1 PK |
104 | fl->mark = skb->mark; |
105 | ||
1da177e4 LT |
106 | if (!(iph->frag_off & htons(IP_MF | IP_OFFSET))) { |
107 | switch (iph->protocol) { | |
108 | case IPPROTO_UDP: | |
ba4e58ec | 109 | case IPPROTO_UDPLITE: |
1da177e4 LT |
110 | case IPPROTO_TCP: |
111 | case IPPROTO_SCTP: | |
9e999993 | 112 | case IPPROTO_DCCP: |
c615c9f3 WY |
113 | if (xprth + 4 < skb->data || |
114 | pskb_may_pull(skb, xprth + 4 - skb->data)) { | |
8c689a6e | 115 | __be16 *ports = (__be16 *)xprth; |
1da177e4 | 116 | |
d5422efe HX |
117 | fl->fl_ip_sport = ports[!!reverse]; |
118 | fl->fl_ip_dport = ports[!reverse]; | |
1da177e4 LT |
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)) { | |
4324a174 | 133 | __be32 *ehdr = (__be32 *)xprth; |
1da177e4 LT |
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)) { | |
4324a174 | 141 | __be32 *ah_hdr = (__be32*)xprth; |
1da177e4 LT |
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)) { | |
4324a174 | 149 | __be16 *ipcomp_hdr = (__be16 *)xprth; |
1da177e4 | 150 | |
4195f814 | 151 | fl->fl_ipsec_spi = htonl(ntohs(ipcomp_hdr[1])); |
1da177e4 LT |
152 | } |
153 | break; | |
cc9ff19d TT |
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 | ||
1da177e4 LT |
168 | default: |
169 | fl->fl_ipsec_spi = 0; | |
170 | break; | |
3ff50b79 | 171 | } |
1da177e4 LT |
172 | } |
173 | fl->proto = iph->protocol; | |
d5422efe HX |
174 | fl->fl4_dst = reverse ? iph->saddr : iph->daddr; |
175 | fl->fl4_src = reverse ? iph->daddr : iph->saddr; | |
4da3089f | 176 | fl->fl4_tos = iph->tos; |
1da177e4 LT |
177 | } |
178 | ||
569d3645 | 179 | static inline int xfrm4_garbage_collect(struct dst_ops *ops) |
1da177e4 | 180 | { |
d7c7544c AD |
181 | struct net *net = container_of(ops, struct net, xfrm.xfrm4_dst_ops); |
182 | ||
183 | xfrm4_policy_afinfo.garbage_collect(net); | |
fc66f95c | 184 | return (dst_entries_get_slow(ops) > ops->gc_thresh * 2); |
1da177e4 LT |
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 | ||
aabc9761 HX |
195 | static void xfrm4_dst_destroy(struct dst_entry *dst) |
196 | { | |
197 | struct xfrm_dst *xdst = (struct xfrm_dst *)dst; | |
198 | ||
62fa8a84 DM |
199 | dst_destroy_metrics_generic(dst); |
200 | ||
ed3e37dd | 201 | if (likely(xdst->u.rt.peer)) |
26db1677 | 202 | inet_putpeer(xdst->u.rt.peer); |
62fa8a84 | 203 | |
aabc9761 HX |
204 | xfrm_dst_destroy(xdst); |
205 | } | |
206 | ||
207 | static void xfrm4_dst_ifdown(struct dst_entry *dst, struct net_device *dev, | |
208 | int unregister) | |
209 | { | |
aabc9761 HX |
210 | if (!unregister) |
211 | return; | |
212 | ||
aabc9761 HX |
213 | xfrm_dst_ifdown(dst, dev); |
214 | } | |
215 | ||
1da177e4 LT |
216 | static struct dst_ops xfrm4_dst_ops = { |
217 | .family = AF_INET, | |
09640e63 | 218 | .protocol = cpu_to_be16(ETH_P_IP), |
1da177e4 LT |
219 | .gc = xfrm4_garbage_collect, |
220 | .update_pmtu = xfrm4_update_pmtu, | |
62fa8a84 | 221 | .cow_metrics = dst_cow_metrics_generic, |
aabc9761 HX |
222 | .destroy = xfrm4_dst_destroy, |
223 | .ifdown = xfrm4_dst_ifdown, | |
862b82c6 | 224 | .local_out = __ip_local_out, |
1da177e4 | 225 | .gc_thresh = 1024, |
1da177e4 LT |
226 | }; |
227 | ||
228 | static struct xfrm_policy_afinfo xfrm4_policy_afinfo = { | |
229 | .family = AF_INET, | |
1da177e4 LT |
230 | .dst_ops = &xfrm4_dst_ops, |
231 | .dst_lookup = xfrm4_dst_lookup, | |
a1e59abf | 232 | .get_saddr = xfrm4_get_saddr, |
1da177e4 | 233 | .decode_session = _decode_session4, |
25ee3286 | 234 | .get_tos = xfrm4_get_tos, |
a1b05140 | 235 | .init_path = xfrm4_init_path, |
25ee3286 | 236 | .fill_dst = xfrm4_fill_dst, |
1da177e4 LT |
237 | }; |
238 | ||
f816700a | 239 | #ifdef CONFIG_SYSCTL |
a44a4a00 NH |
240 | static struct ctl_table xfrm4_policy_table[] = { |
241 | { | |
a44a4a00 | 242 | .procname = "xfrm4_gc_thresh", |
d7c7544c | 243 | .data = &init_net.xfrm.xfrm4_dst_ops.gc_thresh, |
a44a4a00 NH |
244 | .maxlen = sizeof(int), |
245 | .mode = 0644, | |
246 | .proc_handler = proc_dointvec, | |
247 | }, | |
248 | { } | |
249 | }; | |
250 | ||
251 | static struct ctl_table_header *sysctl_hdr; | |
f816700a | 252 | #endif |
a44a4a00 | 253 | |
1da177e4 LT |
254 | static void __init xfrm4_policy_init(void) |
255 | { | |
256 | xfrm_policy_register_afinfo(&xfrm4_policy_afinfo); | |
257 | } | |
258 | ||
259 | static void __exit xfrm4_policy_fini(void) | |
260 | { | |
f816700a | 261 | #ifdef CONFIG_SYSCTL |
a44a4a00 NH |
262 | if (sysctl_hdr) |
263 | unregister_net_sysctl_table(sysctl_hdr); | |
f816700a | 264 | #endif |
1da177e4 LT |
265 | xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo); |
266 | } | |
267 | ||
a33bc5c1 | 268 | void __init xfrm4_init(int rt_max_size) |
1da177e4 | 269 | { |
a33bc5c1 NH |
270 | /* |
271 | * Select a default value for the gc_thresh based on the main route | |
272 | * table hash size. It seems to me the worst case scenario is when | |
273 | * we have ipsec operating in transport mode, in which we create a | |
274 | * dst_entry per socket. The xfrm gc algorithm starts trying to remove | |
275 | * entries at gc_thresh, and prevents new allocations as 2*gc_thresh | |
276 | * so lets set an initial xfrm gc_thresh value at the rt_max_size/2. | |
277 | * That will let us store an ipsec connection per route table entry, | |
278 | * and start cleaning when were 1/2 full | |
279 | */ | |
280 | xfrm4_dst_ops.gc_thresh = rt_max_size/2; | |
fc66f95c | 281 | dst_entries_init(&xfrm4_dst_ops); |
d7c7544c AD |
282 | |
283 | xfrm4_state_init(); | |
284 | xfrm4_policy_init(); | |
f816700a | 285 | #ifdef CONFIG_SYSCTL |
a44a4a00 NH |
286 | sysctl_hdr = register_net_sysctl_table(&init_net, net_ipv4_ctl_path, |
287 | xfrm4_policy_table); | |
f816700a | 288 | #endif |
1da177e4 LT |
289 | } |
290 |