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