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 <net/dst.h> 15 #include <net/xfrm.h> 16 #include <net/ip.h> 17 18 static struct xfrm_policy_afinfo xfrm4_policy_afinfo; 19 20 static struct dst_entry *xfrm4_dst_lookup(struct net *net, int tos, 21 xfrm_address_t *saddr, 22 xfrm_address_t *daddr) 23 { 24 struct flowi fl = { 25 .nl_u = { 26 .ip4_u = { 27 .tos = tos, 28 .daddr = daddr->a4, 29 }, 30 }, 31 }; 32 struct dst_entry *dst; 33 struct rtable *rt; 34 int err; 35 36 if (saddr) 37 fl.fl4_src = saddr->a4; 38 39 err = __ip_route_output_key(net, &rt, &fl); 40 dst = &rt->u.dst; 41 if (err) 42 dst = ERR_PTR(err); 43 return dst; 44 } 45 46 static int xfrm4_get_saddr(struct net *net, 47 xfrm_address_t *saddr, xfrm_address_t *daddr) 48 { 49 struct dst_entry *dst; 50 struct rtable *rt; 51 52 dst = xfrm4_dst_lookup(net, 0, NULL, daddr); 53 if (IS_ERR(dst)) 54 return -EHOSTUNREACH; 55 56 rt = (struct rtable *)dst; 57 saddr->a4 = rt->rt_src; 58 dst_release(dst); 59 return 0; 60 } 61 62 static struct dst_entry * 63 __xfrm4_find_bundle(struct flowi *fl, struct xfrm_policy *policy) 64 { 65 struct dst_entry *dst; 66 67 read_lock_bh(&policy->lock); 68 for (dst = policy->bundles; dst; dst = dst->next) { 69 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 70 if (xdst->u.rt.fl.oif == fl->oif && /*XXX*/ 71 xdst->u.rt.fl.fl4_dst == fl->fl4_dst && 72 xdst->u.rt.fl.fl4_src == fl->fl4_src && 73 xdst->u.rt.fl.fl4_tos == fl->fl4_tos && 74 xfrm_bundle_ok(policy, xdst, fl, AF_INET, 0)) { 75 dst_clone(dst); 76 break; 77 } 78 } 79 read_unlock_bh(&policy->lock); 80 return dst; 81 } 82 83 static int xfrm4_get_tos(struct flowi *fl) 84 { 85 return fl->fl4_tos; 86 } 87 88 static int xfrm4_init_path(struct xfrm_dst *path, struct dst_entry *dst, 89 int nfheader_len) 90 { 91 return 0; 92 } 93 94 static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev) 95 { 96 struct rtable *rt = (struct rtable *)xdst->route; 97 98 xdst->u.rt.fl = rt->fl; 99 100 xdst->u.dst.dev = dev; 101 dev_hold(dev); 102 103 xdst->u.rt.idev = in_dev_get(dev); 104 if (!xdst->u.rt.idev) 105 return -ENODEV; 106 107 xdst->u.rt.peer = rt->peer; 108 if (rt->peer) 109 atomic_inc(&rt->peer->refcnt); 110 111 /* Sheit... I remember I did this right. Apparently, 112 * it was magically lost, so this code needs audit */ 113 xdst->u.rt.rt_flags = rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | 114 RTCF_LOCAL); 115 xdst->u.rt.rt_type = rt->rt_type; 116 xdst->u.rt.rt_src = rt->rt_src; 117 xdst->u.rt.rt_dst = rt->rt_dst; 118 xdst->u.rt.rt_gateway = rt->rt_gateway; 119 xdst->u.rt.rt_spec_dst = rt->rt_spec_dst; 120 121 return 0; 122 } 123 124 static void 125 _decode_session4(struct sk_buff *skb, struct flowi *fl, int reverse) 126 { 127 struct iphdr *iph = ip_hdr(skb); 128 u8 *xprth = skb_network_header(skb) + iph->ihl * 4; 129 130 memset(fl, 0, sizeof(struct flowi)); 131 if (!(iph->frag_off & htons(IP_MF | IP_OFFSET))) { 132 switch (iph->protocol) { 133 case IPPROTO_UDP: 134 case IPPROTO_UDPLITE: 135 case IPPROTO_TCP: 136 case IPPROTO_SCTP: 137 case IPPROTO_DCCP: 138 if (xprth + 4 < skb->data || 139 pskb_may_pull(skb, xprth + 4 - skb->data)) { 140 __be16 *ports = (__be16 *)xprth; 141 142 fl->fl_ip_sport = ports[!!reverse]; 143 fl->fl_ip_dport = ports[!reverse]; 144 } 145 break; 146 147 case IPPROTO_ICMP: 148 if (pskb_may_pull(skb, xprth + 2 - skb->data)) { 149 u8 *icmp = xprth; 150 151 fl->fl_icmp_type = icmp[0]; 152 fl->fl_icmp_code = icmp[1]; 153 } 154 break; 155 156 case IPPROTO_ESP: 157 if (pskb_may_pull(skb, xprth + 4 - skb->data)) { 158 __be32 *ehdr = (__be32 *)xprth; 159 160 fl->fl_ipsec_spi = ehdr[0]; 161 } 162 break; 163 164 case IPPROTO_AH: 165 if (pskb_may_pull(skb, xprth + 8 - skb->data)) { 166 __be32 *ah_hdr = (__be32*)xprth; 167 168 fl->fl_ipsec_spi = ah_hdr[1]; 169 } 170 break; 171 172 case IPPROTO_COMP: 173 if (pskb_may_pull(skb, xprth + 4 - skb->data)) { 174 __be16 *ipcomp_hdr = (__be16 *)xprth; 175 176 fl->fl_ipsec_spi = htonl(ntohs(ipcomp_hdr[1])); 177 } 178 break; 179 default: 180 fl->fl_ipsec_spi = 0; 181 break; 182 } 183 } 184 fl->proto = iph->protocol; 185 fl->fl4_dst = reverse ? iph->saddr : iph->daddr; 186 fl->fl4_src = reverse ? iph->daddr : iph->saddr; 187 fl->fl4_tos = iph->tos; 188 } 189 190 static inline int xfrm4_garbage_collect(struct dst_ops *ops) 191 { 192 struct net *net = container_of(ops, struct net, xfrm.xfrm4_dst_ops); 193 194 xfrm4_policy_afinfo.garbage_collect(net); 195 return (atomic_read(&ops->entries) > ops->gc_thresh * 2); 196 } 197 198 static void xfrm4_update_pmtu(struct dst_entry *dst, u32 mtu) 199 { 200 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 201 struct dst_entry *path = xdst->route; 202 203 path->ops->update_pmtu(path, mtu); 204 } 205 206 static void xfrm4_dst_destroy(struct dst_entry *dst) 207 { 208 struct xfrm_dst *xdst = (struct xfrm_dst *)dst; 209 210 if (likely(xdst->u.rt.idev)) 211 in_dev_put(xdst->u.rt.idev); 212 if (likely(xdst->u.rt.peer)) 213 inet_putpeer(xdst->u.rt.peer); 214 xfrm_dst_destroy(xdst); 215 } 216 217 static void xfrm4_dst_ifdown(struct dst_entry *dst, struct net_device *dev, 218 int unregister) 219 { 220 struct xfrm_dst *xdst; 221 222 if (!unregister) 223 return; 224 225 xdst = (struct xfrm_dst *)dst; 226 if (xdst->u.rt.idev->dev == dev) { 227 struct in_device *loopback_idev = 228 in_dev_get(dev_net(dev)->loopback_dev); 229 BUG_ON(!loopback_idev); 230 231 do { 232 in_dev_put(xdst->u.rt.idev); 233 xdst->u.rt.idev = loopback_idev; 234 in_dev_hold(loopback_idev); 235 xdst = (struct xfrm_dst *)xdst->u.dst.child; 236 } while (xdst->u.dst.xfrm); 237 238 __in_dev_put(loopback_idev); 239 } 240 241 xfrm_dst_ifdown(dst, dev); 242 } 243 244 static struct dst_ops xfrm4_dst_ops = { 245 .family = AF_INET, 246 .protocol = cpu_to_be16(ETH_P_IP), 247 .gc = xfrm4_garbage_collect, 248 .update_pmtu = xfrm4_update_pmtu, 249 .destroy = xfrm4_dst_destroy, 250 .ifdown = xfrm4_dst_ifdown, 251 .local_out = __ip_local_out, 252 .gc_thresh = 1024, 253 .entries = ATOMIC_INIT(0), 254 }; 255 256 static struct xfrm_policy_afinfo xfrm4_policy_afinfo = { 257 .family = AF_INET, 258 .dst_ops = &xfrm4_dst_ops, 259 .dst_lookup = xfrm4_dst_lookup, 260 .get_saddr = xfrm4_get_saddr, 261 .find_bundle = __xfrm4_find_bundle, 262 .decode_session = _decode_session4, 263 .get_tos = xfrm4_get_tos, 264 .init_path = xfrm4_init_path, 265 .fill_dst = xfrm4_fill_dst, 266 }; 267 268 #ifdef CONFIG_SYSCTL 269 static struct ctl_table xfrm4_policy_table[] = { 270 { 271 .procname = "xfrm4_gc_thresh", 272 .data = &init_net.xfrm.xfrm4_dst_ops.gc_thresh, 273 .maxlen = sizeof(int), 274 .mode = 0644, 275 .proc_handler = proc_dointvec, 276 }, 277 { } 278 }; 279 280 static struct ctl_table_header *sysctl_hdr; 281 #endif 282 283 static void __init xfrm4_policy_init(void) 284 { 285 xfrm_policy_register_afinfo(&xfrm4_policy_afinfo); 286 } 287 288 static void __exit xfrm4_policy_fini(void) 289 { 290 #ifdef CONFIG_SYSCTL 291 if (sysctl_hdr) 292 unregister_net_sysctl_table(sysctl_hdr); 293 #endif 294 xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo); 295 } 296 297 void __init xfrm4_init(int rt_max_size) 298 { 299 /* 300 * Select a default value for the gc_thresh based on the main route 301 * table hash size. It seems to me the worst case scenario is when 302 * we have ipsec operating in transport mode, in which we create a 303 * dst_entry per socket. The xfrm gc algorithm starts trying to remove 304 * entries at gc_thresh, and prevents new allocations as 2*gc_thresh 305 * so lets set an initial xfrm gc_thresh value at the rt_max_size/2. 306 * That will let us store an ipsec connection per route table entry, 307 * and start cleaning when were 1/2 full 308 */ 309 xfrm4_dst_ops.gc_thresh = rt_max_size/2; 310 311 xfrm4_state_init(); 312 xfrm4_policy_init(); 313 #ifdef CONFIG_SYSCTL 314 sysctl_hdr = register_net_sysctl_table(&init_net, net_ipv4_ctl_path, 315 xfrm4_policy_table); 316 #endif 317 } 318 319