| 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, struct flowi4 *fl4, |
| 22 | int tos, |
| 23 | const xfrm_address_t *saddr, |
| 24 | const xfrm_address_t *daddr) |
| 25 | { |
| 26 | struct rtable *rt; |
| 27 | |
| 28 | memset(fl4, 0, sizeof(*fl4)); |
| 29 | fl4->daddr = daddr->a4; |
| 30 | fl4->flowi4_tos = tos; |
| 31 | if (saddr) |
| 32 | fl4->saddr = saddr->a4; |
| 33 | |
| 34 | rt = __ip_route_output_key(net, fl4); |
| 35 | if (!IS_ERR(rt)) |
| 36 | return &rt->dst; |
| 37 | |
| 38 | return ERR_CAST(rt); |
| 39 | } |
| 40 | |
| 41 | static struct dst_entry *xfrm4_dst_lookup(struct net *net, int tos, |
| 42 | const xfrm_address_t *saddr, |
| 43 | const xfrm_address_t *daddr) |
| 44 | { |
| 45 | struct flowi4 fl4; |
| 46 | |
| 47 | return __xfrm4_dst_lookup(net, &fl4, tos, saddr, daddr); |
| 48 | } |
| 49 | |
| 50 | static int xfrm4_get_saddr(struct net *net, |
| 51 | xfrm_address_t *saddr, xfrm_address_t *daddr) |
| 52 | { |
| 53 | struct dst_entry *dst; |
| 54 | struct flowi4 fl4; |
| 55 | |
| 56 | dst = __xfrm4_dst_lookup(net, &fl4, 0, NULL, daddr); |
| 57 | if (IS_ERR(dst)) |
| 58 | return -EHOSTUNREACH; |
| 59 | |
| 60 | saddr->a4 = fl4.saddr; |
| 61 | dst_release(dst); |
| 62 | return 0; |
| 63 | } |
| 64 | |
| 65 | static int xfrm4_get_tos(const struct flowi *fl) |
| 66 | { |
| 67 | return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos; /* Strip ECN bits */ |
| 68 | } |
| 69 | |
| 70 | static int xfrm4_init_path(struct xfrm_dst *path, struct dst_entry *dst, |
| 71 | int nfheader_len) |
| 72 | { |
| 73 | return 0; |
| 74 | } |
| 75 | |
| 76 | static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev, |
| 77 | const struct flowi *fl) |
| 78 | { |
| 79 | struct rtable *rt = (struct rtable *)xdst->route; |
| 80 | const struct flowi4 *fl4 = &fl->u.ip4; |
| 81 | |
| 82 | xdst->u.rt.rt_iif = fl4->flowi4_iif; |
| 83 | |
| 84 | xdst->u.dst.dev = dev; |
| 85 | dev_hold(dev); |
| 86 | |
| 87 | /* Sheit... I remember I did this right. Apparently, |
| 88 | * it was magically lost, so this code needs audit */ |
| 89 | xdst->u.rt.rt_is_input = rt->rt_is_input; |
| 90 | xdst->u.rt.rt_flags = rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | |
| 91 | RTCF_LOCAL); |
| 92 | xdst->u.rt.rt_type = rt->rt_type; |
| 93 | xdst->u.rt.rt_gateway = rt->rt_gateway; |
| 94 | xdst->u.rt.rt_uses_gateway = rt->rt_uses_gateway; |
| 95 | xdst->u.rt.rt_pmtu = rt->rt_pmtu; |
| 96 | INIT_LIST_HEAD(&xdst->u.rt.rt_uncached); |
| 97 | |
| 98 | return 0; |
| 99 | } |
| 100 | |
| 101 | static void |
| 102 | _decode_session4(struct sk_buff *skb, struct flowi *fl, int reverse) |
| 103 | { |
| 104 | const struct iphdr *iph = ip_hdr(skb); |
| 105 | u8 *xprth = skb_network_header(skb) + iph->ihl * 4; |
| 106 | struct flowi4 *fl4 = &fl->u.ip4; |
| 107 | |
| 108 | memset(fl4, 0, sizeof(struct flowi4)); |
| 109 | fl4->flowi4_mark = skb->mark; |
| 110 | |
| 111 | if (!ip_is_fragment(iph)) { |
| 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 | fl4->fl4_sport = ports[!!reverse]; |
| 123 | fl4->fl4_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 | fl4->fl4_icmp_type = icmp[0]; |
| 132 | fl4->fl4_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 | fl4->fl4_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 | fl4->fl4_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 | fl4->fl4_ipsec_spi = htonl(ntohs(ipcomp_hdr[1])); |
| 157 | } |
| 158 | break; |
| 159 | |
| 160 | case IPPROTO_GRE: |
| 161 | if (pskb_may_pull(skb, xprth + 12 - skb->data)) { |
| 162 | __be16 *greflags = (__be16 *)xprth; |
| 163 | __be32 *gre_hdr = (__be32 *)xprth; |
| 164 | |
| 165 | if (greflags[0] & GRE_KEY) { |
| 166 | if (greflags[0] & GRE_CSUM) |
| 167 | gre_hdr++; |
| 168 | fl4->fl4_gre_key = gre_hdr[1]; |
| 169 | } |
| 170 | } |
| 171 | break; |
| 172 | |
| 173 | default: |
| 174 | fl4->fl4_ipsec_spi = 0; |
| 175 | break; |
| 176 | } |
| 177 | } |
| 178 | fl4->flowi4_proto = iph->protocol; |
| 179 | fl4->daddr = reverse ? iph->saddr : iph->daddr; |
| 180 | fl4->saddr = reverse ? iph->daddr : iph->saddr; |
| 181 | fl4->flowi4_tos = iph->tos; |
| 182 | } |
| 183 | |
| 184 | static inline int xfrm4_garbage_collect(struct dst_ops *ops) |
| 185 | { |
| 186 | struct net *net = container_of(ops, struct net, xfrm.xfrm4_dst_ops); |
| 187 | |
| 188 | xfrm4_policy_afinfo.garbage_collect(net); |
| 189 | return (dst_entries_get_slow(ops) > ops->gc_thresh * 2); |
| 190 | } |
| 191 | |
| 192 | static void xfrm4_update_pmtu(struct dst_entry *dst, struct sock *sk, |
| 193 | struct sk_buff *skb, u32 mtu) |
| 194 | { |
| 195 | struct xfrm_dst *xdst = (struct xfrm_dst *)dst; |
| 196 | struct dst_entry *path = xdst->route; |
| 197 | |
| 198 | path->ops->update_pmtu(path, sk, skb, mtu); |
| 199 | } |
| 200 | |
| 201 | static void xfrm4_redirect(struct dst_entry *dst, struct sock *sk, |
| 202 | struct sk_buff *skb) |
| 203 | { |
| 204 | struct xfrm_dst *xdst = (struct xfrm_dst *)dst; |
| 205 | struct dst_entry *path = xdst->route; |
| 206 | |
| 207 | path->ops->redirect(path, sk, skb); |
| 208 | } |
| 209 | |
| 210 | static void xfrm4_dst_destroy(struct dst_entry *dst) |
| 211 | { |
| 212 | struct xfrm_dst *xdst = (struct xfrm_dst *)dst; |
| 213 | |
| 214 | dst_destroy_metrics_generic(dst); |
| 215 | |
| 216 | xfrm_dst_destroy(xdst); |
| 217 | } |
| 218 | |
| 219 | static void xfrm4_dst_ifdown(struct dst_entry *dst, struct net_device *dev, |
| 220 | int unregister) |
| 221 | { |
| 222 | if (!unregister) |
| 223 | return; |
| 224 | |
| 225 | xfrm_dst_ifdown(dst, dev); |
| 226 | } |
| 227 | |
| 228 | static struct dst_ops xfrm4_dst_ops = { |
| 229 | .family = AF_INET, |
| 230 | .protocol = cpu_to_be16(ETH_P_IP), |
| 231 | .gc = xfrm4_garbage_collect, |
| 232 | .update_pmtu = xfrm4_update_pmtu, |
| 233 | .redirect = xfrm4_redirect, |
| 234 | .cow_metrics = dst_cow_metrics_generic, |
| 235 | .destroy = xfrm4_dst_destroy, |
| 236 | .ifdown = xfrm4_dst_ifdown, |
| 237 | .local_out = __ip_local_out, |
| 238 | .gc_thresh = 1024, |
| 239 | }; |
| 240 | |
| 241 | static struct xfrm_policy_afinfo xfrm4_policy_afinfo = { |
| 242 | .family = AF_INET, |
| 243 | .dst_ops = &xfrm4_dst_ops, |
| 244 | .dst_lookup = xfrm4_dst_lookup, |
| 245 | .get_saddr = xfrm4_get_saddr, |
| 246 | .decode_session = _decode_session4, |
| 247 | .get_tos = xfrm4_get_tos, |
| 248 | .init_path = xfrm4_init_path, |
| 249 | .fill_dst = xfrm4_fill_dst, |
| 250 | .blackhole_route = ipv4_blackhole_route, |
| 251 | }; |
| 252 | |
| 253 | #ifdef CONFIG_SYSCTL |
| 254 | static struct ctl_table xfrm4_policy_table[] = { |
| 255 | { |
| 256 | .procname = "xfrm4_gc_thresh", |
| 257 | .data = &init_net.xfrm.xfrm4_dst_ops.gc_thresh, |
| 258 | .maxlen = sizeof(int), |
| 259 | .mode = 0644, |
| 260 | .proc_handler = proc_dointvec, |
| 261 | }, |
| 262 | { } |
| 263 | }; |
| 264 | |
| 265 | static struct ctl_table_header *sysctl_hdr; |
| 266 | #endif |
| 267 | |
| 268 | static void __init xfrm4_policy_init(void) |
| 269 | { |
| 270 | xfrm_policy_register_afinfo(&xfrm4_policy_afinfo); |
| 271 | } |
| 272 | |
| 273 | static void __exit xfrm4_policy_fini(void) |
| 274 | { |
| 275 | #ifdef CONFIG_SYSCTL |
| 276 | if (sysctl_hdr) |
| 277 | unregister_net_sysctl_table(sysctl_hdr); |
| 278 | #endif |
| 279 | xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo); |
| 280 | } |
| 281 | |
| 282 | void __init xfrm4_init(int rt_max_size) |
| 283 | { |
| 284 | /* |
| 285 | * Select a default value for the gc_thresh based on the main route |
| 286 | * table hash size. It seems to me the worst case scenario is when |
| 287 | * we have ipsec operating in transport mode, in which we create a |
| 288 | * dst_entry per socket. The xfrm gc algorithm starts trying to remove |
| 289 | * entries at gc_thresh, and prevents new allocations as 2*gc_thresh |
| 290 | * so lets set an initial xfrm gc_thresh value at the rt_max_size/2. |
| 291 | * That will let us store an ipsec connection per route table entry, |
| 292 | * and start cleaning when were 1/2 full |
| 293 | */ |
| 294 | xfrm4_dst_ops.gc_thresh = rt_max_size/2; |
| 295 | dst_entries_init(&xfrm4_dst_ops); |
| 296 | |
| 297 | xfrm4_state_init(); |
| 298 | xfrm4_policy_init(); |
| 299 | #ifdef CONFIG_SYSCTL |
| 300 | sysctl_hdr = register_net_sysctl(&init_net, "net/ipv4", |
| 301 | xfrm4_policy_table); |
| 302 | #endif |
| 303 | } |
| 304 | |