net/ipv6/xfrm6_policy.c

   1 /*
   2  * xfrm6_policy.c: based on xfrm4_policy.c
   3  *
   4  * Authors:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      YOSHIFUJI Hideaki
  10  *              Split up af-specific portion
  11  *
  12  */
  13
  14 #include <linux/err.h>
  15 #include <linux/kernel.h>
  16 #include <linux/netdevice.h>
  17 #include <net/addrconf.h>
  18 #include <net/dst.h>
  19 #include <net/xfrm.h>
  20 #include <net/ip.h>
  21 #include <net/ipv6.h>
  22 #include <net/ip6_route.h>
  23 #if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
  24 #include <net/mip6.h>
  25 #endif
  26
  27 static struct xfrm_policy_afinfo xfrm6_policy_afinfo;
  28
  29 static struct dst_entry *xfrm6_dst_lookup(struct net *net, int tos,
  30                                           const xfrm_address_t *saddr,
  31                                           const xfrm_address_t *daddr)
  32 {
  33         struct flowi6 fl6;
  34         struct dst_entry *dst;
  35         int err;
  36
  37         memset(&fl6, 0, sizeof(fl6));
  38         memcpy(&fl6.daddr, daddr, sizeof(fl6.daddr));
  39         if (saddr)
  40                 memcpy(&fl6.saddr, saddr, sizeof(fl6.saddr));
  41
  42         dst = ip6_route_output(net, NULL, &fl6);
  43
  44         err = dst->error;
  45         if (dst->error) {
  46                 dst_release(dst);
  47                 dst = ERR_PTR(err);
  48         }
  49
  50         return dst;
  51 }
  52
  53 static int xfrm6_get_saddr(struct net *net,
  54                            xfrm_address_t *saddr, xfrm_address_t *daddr)
  55 {
  56         struct dst_entry *dst;
  57         struct net_device *dev;
  58
  59         dst = xfrm6_dst_lookup(net, 0, NULL, daddr);
  60         if (IS_ERR(dst))
  61                 return -EHOSTUNREACH;
  62
  63         dev = ip6_dst_idev(dst)->dev;
  64         ipv6_dev_get_saddr(dev_net(dev), dev,
  65                            (struct in6_addr *)&daddr->a6, 0,
  66                            (struct in6_addr *)&saddr->a6);
  67         dst_release(dst);
  68         return 0;
  69 }
  70
  71 static int xfrm6_get_tos(const struct flowi *fl)
  72 {
  73         return 0;
  74 }
  75
  76 static int xfrm6_init_path(struct xfrm_dst *path, struct dst_entry *dst,
  77                            int nfheader_len)
  78 {
  79         if (dst->ops->family == AF_INET6) {
  80                 struct rt6_info *rt = (struct rt6_info*)dst;
  81                 if (rt->rt6i_node)
  82                         path->path_cookie = rt->rt6i_node->fn_sernum;
  83         }
  84
  85         path->u.rt6.rt6i_nfheader_len = nfheader_len;
  86
  87         return 0;
  88 }
  89
  90 static int xfrm6_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
  91                           const struct flowi *fl)
  92 {
  93         struct rt6_info *rt = (struct rt6_info*)xdst->route;
  94
  95         xdst->u.dst.dev = dev;
  96         dev_hold(dev);
  97
  98         xdst->u.rt6.rt6i_idev = in6_dev_get(dev);
  99         if (!xdst->u.rt6.rt6i_idev)
 100                 return -ENODEV;
 101
 102         rt6_transfer_peer(&xdst->u.rt6, rt);
 103
 104         /* Sheit... I remember I did this right. Apparently,
 105          * it was magically lost, so this code needs audit */
 106         xdst->u.rt6.n = neigh_clone(rt->n);
 107         xdst->u.rt6.rt6i_flags = rt->rt6i_flags & (RTF_ANYCAST |
 108                                                    RTF_LOCAL);
 109         xdst->u.rt6.rt6i_metric = rt->rt6i_metric;
 110         xdst->u.rt6.rt6i_node = rt->rt6i_node;
 111         if (rt->rt6i_node)
 112                 xdst->route_cookie = rt->rt6i_node->fn_sernum;
 113         xdst->u.rt6.rt6i_gateway = rt->rt6i_gateway;
 114         xdst->u.rt6.rt6i_dst = rt->rt6i_dst;
 115         xdst->u.rt6.rt6i_src = rt->rt6i_src;
 116
 117         return 0;
 118 }
 119
 120 static inline void
 121 _decode_session6(struct sk_buff *skb, struct flowi *fl, int reverse)
 122 {
 123         struct flowi6 *fl6 = &fl->u.ip6;
 124         int onlyproto = 0;
 125         u16 offset = skb_network_header_len(skb);
 126         const struct ipv6hdr *hdr = ipv6_hdr(skb);
 127         struct ipv6_opt_hdr *exthdr;
 128         const unsigned char *nh = skb_network_header(skb);
 129         u8 nexthdr = nh[IP6CB(skb)->nhoff];
 130
 131         memset(fl6, 0, sizeof(struct flowi6));
 132         fl6->flowi6_mark = skb->mark;
 133
 134         fl6->daddr = reverse ? hdr->saddr : hdr->daddr;
 135         fl6->saddr = reverse ? hdr->daddr : hdr->saddr;
 136
 137         while (nh + offset + 1 < skb->data ||
 138                pskb_may_pull(skb, nh + offset + 1 - skb->data)) {
 139                 nh = skb_network_header(skb);
 140                 exthdr = (struct ipv6_opt_hdr *)(nh + offset);
 141
 142                 switch (nexthdr) {
 143                 case NEXTHDR_FRAGMENT:
 144                         onlyproto = 1;
 145                 case NEXTHDR_ROUTING:
 146                 case NEXTHDR_HOP:
 147                 case NEXTHDR_DEST:
 148                         offset += ipv6_optlen(exthdr);
 149                         nexthdr = exthdr->nexthdr;
 150                         exthdr = (struct ipv6_opt_hdr *)(nh + offset);
 151                         break;
 152
 153                 case IPPROTO_UDP:
 154                 case IPPROTO_UDPLITE:
 155                 case IPPROTO_TCP:
 156                 case IPPROTO_SCTP:
 157                 case IPPROTO_DCCP:
 158                         if (!onlyproto && (nh + offset + 4 < skb->data ||
 159                              pskb_may_pull(skb, nh + offset + 4 - skb->data))) {
 160                                 __be16 *ports = (__be16 *)exthdr;
 161
 162                                 fl6->fl6_sport = ports[!!reverse];
 163                                 fl6->fl6_dport = ports[!reverse];
 164                         }
 165                         fl6->flowi6_proto = nexthdr;
 166                         return;
 167
 168                 case IPPROTO_ICMPV6:
 169                         if (!onlyproto && pskb_may_pull(skb, nh + offset + 2 - skb->data)) {
 170                                 u8 *icmp = (u8 *)exthdr;
 171
 172                                 fl6->fl6_icmp_type = icmp[0];
 173                                 fl6->fl6_icmp_code = icmp[1];
 174                         }
 175                         fl6->flowi6_proto = nexthdr;
 176                         return;
 177
 178 #if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
 179                 case IPPROTO_MH:
 180                         if (!onlyproto && pskb_may_pull(skb, nh + offset + 3 - skb->data)) {
 181                                 struct ip6_mh *mh;
 182                                 mh = (struct ip6_mh *)exthdr;
 183
 184                                 fl6->fl6_mh_type = mh->ip6mh_type;
 185                         }
 186                         fl6->flowi6_proto = nexthdr;
 187                         return;
 188 #endif
 189
 190                 /* XXX Why are there these headers? */
 191                 case IPPROTO_AH:
 192                 case IPPROTO_ESP:
 193                 case IPPROTO_COMP:
 194                 default:
 195                         fl6->fl6_ipsec_spi = 0;
 196                         fl6->flowi6_proto = nexthdr;
 197                         return;
 198                 }
 199         }
 200 }
 201
 202 static inline int xfrm6_garbage_collect(struct dst_ops *ops)
 203 {
 204         struct net *net = container_of(ops, struct net, xfrm.xfrm6_dst_ops);
 205
 206         xfrm6_policy_afinfo.garbage_collect(net);
 207         return dst_entries_get_fast(ops) > ops->gc_thresh * 2;
 208 }
 209
 210 static void xfrm6_update_pmtu(struct dst_entry *dst, struct sock *sk,
 211                               struct sk_buff *skb, u32 mtu)
 212 {
 213         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 214         struct dst_entry *path = xdst->route;
 215
 216         path->ops->update_pmtu(path, sk, skb, mtu);
 217 }
 218
 219 static void xfrm6_redirect(struct dst_entry *dst, struct sock *sk,
 220                            struct sk_buff *skb)
 221 {
 222         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 223         struct dst_entry *path = xdst->route;
 224
 225         path->ops->redirect(path, sk, skb);
 226 }
 227
 228 static void xfrm6_dst_destroy(struct dst_entry *dst)
 229 {
 230         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 231
 232         if (likely(xdst->u.rt6.rt6i_idev))
 233                 in6_dev_put(xdst->u.rt6.rt6i_idev);
 234         dst_destroy_metrics_generic(dst);
 235         if (rt6_has_peer(&xdst->u.rt6)) {
 236                 struct inet_peer *peer = rt6_peer_ptr(&xdst->u.rt6);
 237                 inet_putpeer(peer);
 238         }
 239         xfrm_dst_destroy(xdst);
 240 }
 241
 242 static void xfrm6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 243                              int unregister)
 244 {
 245         struct xfrm_dst *xdst;
 246
 247         if (!unregister)
 248                 return;
 249
 250         xdst = (struct xfrm_dst *)dst;
 251         if (xdst->u.rt6.rt6i_idev->dev == dev) {
 252                 struct inet6_dev *loopback_idev =
 253                         in6_dev_get(dev_net(dev)->loopback_dev);
 254                 BUG_ON(!loopback_idev);
 255
 256                 do {
 257                         in6_dev_put(xdst->u.rt6.rt6i_idev);
 258                         xdst->u.rt6.rt6i_idev = loopback_idev;
 259                         in6_dev_hold(loopback_idev);
 260                         xdst = (struct xfrm_dst *)xdst->u.dst.child;
 261                 } while (xdst->u.dst.xfrm);
 262
 263                 __in6_dev_put(loopback_idev);
 264         }
 265
 266         xfrm_dst_ifdown(dst, dev);
 267 }
 268
 269 static struct dst_ops xfrm6_dst_ops = {
 270         .family =               AF_INET6,
 271         .protocol =             cpu_to_be16(ETH_P_IPV6),
 272         .gc =                   xfrm6_garbage_collect,
 273         .update_pmtu =          xfrm6_update_pmtu,
 274         .redirect =             xfrm6_redirect,
 275         .cow_metrics =          dst_cow_metrics_generic,
 276         .destroy =              xfrm6_dst_destroy,
 277         .ifdown =               xfrm6_dst_ifdown,
 278         .local_out =            __ip6_local_out,
 279         .gc_thresh =            1024,
 280 };
 281
 282 static struct xfrm_policy_afinfo xfrm6_policy_afinfo = {
 283         .family =               AF_INET6,
 284         .dst_ops =              &xfrm6_dst_ops,
 285         .dst_lookup =           xfrm6_dst_lookup,
 286         .get_saddr =            xfrm6_get_saddr,
 287         .decode_session =       _decode_session6,
 288         .get_tos =              xfrm6_get_tos,
 289         .init_path =            xfrm6_init_path,
 290         .fill_dst =             xfrm6_fill_dst,
 291         .blackhole_route =      ip6_blackhole_route,
 292 };
 293
 294 static int __init xfrm6_policy_init(void)
 295 {
 296         return xfrm_policy_register_afinfo(&xfrm6_policy_afinfo);
 297 }
 298
 299 static void xfrm6_policy_fini(void)
 300 {
 301         xfrm_policy_unregister_afinfo(&xfrm6_policy_afinfo);
 302 }
 303
 304 #ifdef CONFIG_SYSCTL
 305 static struct ctl_table xfrm6_policy_table[] = {
 306         {
 307                 .procname       = "xfrm6_gc_thresh",
 308                 .data           = &init_net.xfrm.xfrm6_dst_ops.gc_thresh,
 309                 .maxlen         = sizeof(int),
 310                 .mode           = 0644,
 311                 .proc_handler   = proc_dointvec,
 312         },
 313         { }
 314 };
 315
 316 static struct ctl_table_header *sysctl_hdr;
 317 #endif
 318
 319 int __init xfrm6_init(void)
 320 {
 321         int ret;
 322         unsigned int gc_thresh;
 323
 324         /*
 325          * We need a good default value for the xfrm6 gc threshold.
 326          * In ipv4 we set it to the route hash table size * 8, which
 327          * is half the size of the maximaum route cache for ipv4.  It
 328          * would be good to do the same thing for v6, except the table is
 329          * constructed differently here.  Here each table for a net namespace
 330          * can have FIB_TABLE_HASHSZ entries, so lets go with the same
 331          * computation that we used for ipv4 here.  Also, lets keep the initial
 332          * gc_thresh to a minimum of 1024, since, the ipv6 route cache defaults
 333          * to that as a minimum as well
 334          */
 335         gc_thresh = FIB6_TABLE_HASHSZ * 8;
 336         xfrm6_dst_ops.gc_thresh = (gc_thresh < 1024) ? 1024 : gc_thresh;
 337         dst_entries_init(&xfrm6_dst_ops);
 338
 339         ret = xfrm6_policy_init();
 340         if (ret) {
 341                 dst_entries_destroy(&xfrm6_dst_ops);
 342                 goto out;
 343         }
 344         ret = xfrm6_state_init();
 345         if (ret)
 346                 goto out_policy;
 347
 348 #ifdef CONFIG_SYSCTL
 349         sysctl_hdr = register_net_sysctl(&init_net, "net/ipv6",
 350                                          xfrm6_policy_table);
 351 #endif
 352 out:
 353         return ret;
 354 out_policy:
 355         xfrm6_policy_fini();
 356         goto out;
 357 }
 358
 359 void xfrm6_fini(void)
 360 {
 361 #ifdef CONFIG_SYSCTL
 362         if (sysctl_hdr)
 363                 unregister_net_sysctl_table(sysctl_hdr);
 364 #endif
 365         //xfrm6_input_fini();
 366         xfrm6_policy_fini();
 367         xfrm6_state_fini();
 368         dst_entries_destroy(&xfrm6_dst_ops);
 369 }