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