net/ipv4/xfrm4_policy.c

   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/compiler.h>
  12 #include <linux/inetdevice.h>
  13 #include <net/xfrm.h>
  14 #include <net/ip.h>
  15
  16 static struct dst_ops xfrm4_dst_ops;
  17 static struct xfrm_policy_afinfo xfrm4_policy_afinfo;
  18
  19 static int xfrm4_dst_lookup(struct xfrm_dst **dst, struct flowi *fl)
  20 {
  21         return __ip_route_output_key((struct rtable**)dst, fl);
  22 }
  23
  24 static int xfrm4_get_saddr(xfrm_address_t *saddr, xfrm_address_t *daddr)
  25 {
  26         struct rtable *rt;
  27         struct flowi fl_tunnel = {
  28                 .nl_u = {
  29                         .ip4_u = {
  30                                 .daddr = daddr->a4,
  31                         },
  32                 },
  33         };
  34
  35         if (!xfrm4_dst_lookup((struct xfrm_dst **)&rt, &fl_tunnel)) {
  36                 saddr->a4 = rt->rt_src;
  37                 dst_release(&rt->u.dst);
  38                 return 0;
  39         }
  40         return -EHOSTUNREACH;
  41 }
  42
  43 static struct dst_entry *
  44 __xfrm4_find_bundle(struct flowi *fl, struct xfrm_policy *policy)
  45 {
  46         struct dst_entry *dst;
  47
  48         read_lock_bh(&policy->lock);
  49         for (dst = policy->bundles; dst; dst = dst->next) {
  50                 struct xfrm_dst *xdst = (struct xfrm_dst*)dst;
  51                 if (xdst->u.rt.fl.oif == fl->oif &&     /*XXX*/
  52                     xdst->u.rt.fl.fl4_dst == fl->fl4_dst &&
  53                     xdst->u.rt.fl.fl4_src == fl->fl4_src &&
  54                     xdst->u.rt.fl.fl4_tos == fl->fl4_tos &&
  55                     xfrm_bundle_ok(policy, xdst, fl, AF_INET, 0)) {
  56                         dst_clone(dst);
  57                         break;
  58                 }
  59         }
  60         read_unlock_bh(&policy->lock);
  61         return dst;
  62 }
  63
  64 /* Allocate chain of dst_entry's, attach known xfrm's, calculate
  65  * all the metrics... Shortly, bundle a bundle.
  66  */
  67
  68 static int
  69 __xfrm4_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
  70                       struct flowi *fl, struct dst_entry **dst_p)
  71 {
  72         struct dst_entry *dst, *dst_prev;
  73         struct rtable *rt0 = (struct rtable*)(*dst_p);
  74         struct rtable *rt = rt0;
  75         struct flowi fl_tunnel = {
  76                 .nl_u = {
  77                         .ip4_u = {
  78                                 .saddr = fl->fl4_src,
  79                                 .daddr = fl->fl4_dst,
  80                                 .tos = fl->fl4_tos
  81                         }
  82                 }
  83         };
  84         int i;
  85         int err;
  86         int header_len = 0;
  87         int trailer_len = 0;
  88
  89         dst = dst_prev = NULL;
  90         dst_hold(&rt->u.dst);
  91
  92         for (i = 0; i < nx; i++) {
  93                 struct dst_entry *dst1 = dst_alloc(&xfrm4_dst_ops);
  94                 struct xfrm_dst *xdst;
  95
  96                 if (unlikely(dst1 == NULL)) {
  97                         err = -ENOBUFS;
  98                         dst_release(&rt->u.dst);
  99                         goto error;
 100                 }
 101
 102                 if (!dst)
 103                         dst = dst1;
 104                 else {
 105                         dst_prev->child = dst1;
 106                         dst1->flags |= DST_NOHASH;
 107                         dst_clone(dst1);
 108                 }
 109
 110                 xdst = (struct xfrm_dst *)dst1;
 111                 xdst->route = &rt->u.dst;
 112                 xdst->genid = xfrm[i]->genid;
 113
 114                 dst1->next = dst_prev;
 115                 dst_prev = dst1;
 116
 117                 header_len += xfrm[i]->props.header_len;
 118                 trailer_len += xfrm[i]->props.trailer_len;
 119
 120                 if (xfrm[i]->props.mode != XFRM_MODE_TRANSPORT) {
 121                         unsigned short encap_family = xfrm[i]->props.family;
 122                         switch (encap_family) {
 123                         case AF_INET:
 124                                 fl_tunnel.fl4_dst = xfrm[i]->id.daddr.a4;
 125                                 fl_tunnel.fl4_src = xfrm[i]->props.saddr.a4;
 126                                 break;
 127 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
 128                         case AF_INET6:
 129                                 ipv6_addr_copy(&fl_tunnel.fl6_dst, (struct in6_addr*)&xfrm[i]->id.daddr.a6);
 130                                 ipv6_addr_copy(&fl_tunnel.fl6_src, (struct in6_addr*)&xfrm[i]->props.saddr.a6);
 131                                 break;
 132 #endif
 133                         default:
 134                                 BUG_ON(1);
 135                         }
 136                         err = xfrm_dst_lookup((struct xfrm_dst **)&rt,
 137                                               &fl_tunnel, encap_family);
 138                         if (err)
 139                                 goto error;
 140                 } else
 141                         dst_hold(&rt->u.dst);
 142         }
 143
 144         dst_prev->child = &rt->u.dst;
 145         dst->path = &rt->u.dst;
 146
 147         *dst_p = dst;
 148         dst = dst_prev;
 149
 150         dst_prev = *dst_p;
 151         i = 0;
 152         for (; dst_prev != &rt->u.dst; dst_prev = dst_prev->child) {
 153                 struct xfrm_dst *x = (struct xfrm_dst*)dst_prev;
 154                 x->u.rt.fl = *fl;
 155
 156                 dst_prev->xfrm = xfrm[i++];
 157                 dst_prev->dev = rt->u.dst.dev;
 158                 if (rt->u.dst.dev)
 159                         dev_hold(rt->u.dst.dev);
 160                 dst_prev->obsolete      = -1;
 161                 dst_prev->flags        |= DST_HOST;
 162                 dst_prev->lastuse       = jiffies;
 163                 dst_prev->header_len    = header_len;
 164                 dst_prev->nfheader_len  = 0;
 165                 dst_prev->trailer_len   = trailer_len;
 166                 memcpy(&dst_prev->metrics, &x->route->metrics, sizeof(dst_prev->metrics));
 167
 168                 /* Copy neighbout for reachability confirmation */
 169                 dst_prev->neighbour     = neigh_clone(rt->u.dst.neighbour);
 170                 dst_prev->input         = rt->u.dst.input;
 171                 dst_prev->output = dst_prev->xfrm->outer_mode->afinfo->output;
 172                 if (rt0->peer)
 173                         atomic_inc(&rt0->peer->refcnt);
 174                 x->u.rt.peer = rt0->peer;
 175                 /* Sheit... I remember I did this right. Apparently,
 176                  * it was magically lost, so this code needs audit */
 177                 x->u.rt.rt_flags = rt0->rt_flags&(RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL);
 178                 x->u.rt.rt_type = rt0->rt_type;
 179                 x->u.rt.rt_src = rt0->rt_src;
 180                 x->u.rt.rt_dst = rt0->rt_dst;
 181                 x->u.rt.rt_gateway = rt0->rt_gateway;
 182                 x->u.rt.rt_spec_dst = rt0->rt_spec_dst;
 183                 x->u.rt.idev = rt0->idev;
 184                 in_dev_hold(rt0->idev);
 185                 header_len -= x->u.dst.xfrm->props.header_len;
 186                 trailer_len -= x->u.dst.xfrm->props.trailer_len;
 187         }
 188
 189         xfrm_init_pmtu(dst);
 190         return 0;
 191
 192 error:
 193         if (dst)
 194                 dst_free(dst);
 195         return err;
 196 }
 197
 198 static void
 199 _decode_session4(struct sk_buff *skb, struct flowi *fl)
 200 {
 201         struct iphdr *iph = ip_hdr(skb);
 202         u8 *xprth = skb_network_header(skb) + iph->ihl * 4;
 203
 204         memset(fl, 0, sizeof(struct flowi));
 205         if (!(iph->frag_off & htons(IP_MF | IP_OFFSET))) {
 206                 switch (iph->protocol) {
 207                 case IPPROTO_UDP:
 208                 case IPPROTO_UDPLITE:
 209                 case IPPROTO_TCP:
 210                 case IPPROTO_SCTP:
 211                 case IPPROTO_DCCP:
 212                         if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
 213                                 __be16 *ports = (__be16 *)xprth;
 214
 215                                 fl->fl_ip_sport = ports[0];
 216                                 fl->fl_ip_dport = ports[1];
 217                         }
 218                         break;
 219
 220                 case IPPROTO_ICMP:
 221                         if (pskb_may_pull(skb, xprth + 2 - skb->data)) {
 222                                 u8 *icmp = xprth;
 223
 224                                 fl->fl_icmp_type = icmp[0];
 225                                 fl->fl_icmp_code = icmp[1];
 226                         }
 227                         break;
 228
 229                 case IPPROTO_ESP:
 230                         if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
 231                                 __be32 *ehdr = (__be32 *)xprth;
 232
 233                                 fl->fl_ipsec_spi = ehdr[0];
 234                         }
 235                         break;
 236
 237                 case IPPROTO_AH:
 238                         if (pskb_may_pull(skb, xprth + 8 - skb->data)) {
 239                                 __be32 *ah_hdr = (__be32*)xprth;
 240
 241                                 fl->fl_ipsec_spi = ah_hdr[1];
 242                         }
 243                         break;
 244
 245                 case IPPROTO_COMP:
 246                         if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
 247                                 __be16 *ipcomp_hdr = (__be16 *)xprth;
 248
 249                                 fl->fl_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
 250                         }
 251                         break;
 252                 default:
 253                         fl->fl_ipsec_spi = 0;
 254                         break;
 255                 }
 256         }
 257         fl->proto = iph->protocol;
 258         fl->fl4_dst = iph->daddr;
 259         fl->fl4_src = iph->saddr;
 260         fl->fl4_tos = iph->tos;
 261 }
 262
 263 static inline int xfrm4_garbage_collect(void)
 264 {
 265         xfrm4_policy_afinfo.garbage_collect();
 266         return (atomic_read(&xfrm4_dst_ops.entries) > xfrm4_dst_ops.gc_thresh*2);
 267 }
 268
 269 static void xfrm4_update_pmtu(struct dst_entry *dst, u32 mtu)
 270 {
 271         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 272         struct dst_entry *path = xdst->route;
 273
 274         path->ops->update_pmtu(path, mtu);
 275 }
 276
 277 static void xfrm4_dst_destroy(struct dst_entry *dst)
 278 {
 279         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 280
 281         if (likely(xdst->u.rt.idev))
 282                 in_dev_put(xdst->u.rt.idev);
 283         if (likely(xdst->u.rt.peer))
 284                 inet_putpeer(xdst->u.rt.peer);
 285         xfrm_dst_destroy(xdst);
 286 }
 287
 288 static void xfrm4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 289                              int unregister)
 290 {
 291         struct xfrm_dst *xdst;
 292
 293         if (!unregister)
 294                 return;
 295
 296         xdst = (struct xfrm_dst *)dst;
 297         if (xdst->u.rt.idev->dev == dev) {
 298                 struct in_device *loopback_idev = in_dev_get(init_net.loopback_dev);
 299                 BUG_ON(!loopback_idev);
 300
 301                 do {
 302                         in_dev_put(xdst->u.rt.idev);
 303                         xdst->u.rt.idev = loopback_idev;
 304                         in_dev_hold(loopback_idev);
 305                         xdst = (struct xfrm_dst *)xdst->u.dst.child;
 306                 } while (xdst->u.dst.xfrm);
 307
 308                 __in_dev_put(loopback_idev);
 309         }
 310
 311         xfrm_dst_ifdown(dst, dev);
 312 }
 313
 314 static struct dst_ops xfrm4_dst_ops = {
 315         .family =               AF_INET,
 316         .protocol =             __constant_htons(ETH_P_IP),
 317         .gc =                   xfrm4_garbage_collect,
 318         .update_pmtu =          xfrm4_update_pmtu,
 319         .destroy =              xfrm4_dst_destroy,
 320         .ifdown =               xfrm4_dst_ifdown,
 321         .gc_thresh =            1024,
 322         .entry_size =           sizeof(struct xfrm_dst),
 323 };
 324
 325 static struct xfrm_policy_afinfo xfrm4_policy_afinfo = {
 326         .family =               AF_INET,
 327         .dst_ops =              &xfrm4_dst_ops,
 328         .dst_lookup =           xfrm4_dst_lookup,
 329         .get_saddr =            xfrm4_get_saddr,
 330         .find_bundle =          __xfrm4_find_bundle,
 331         .bundle_create =        __xfrm4_bundle_create,
 332         .decode_session =       _decode_session4,
 333 };
 334
 335 static void __init xfrm4_policy_init(void)
 336 {
 337         xfrm_policy_register_afinfo(&xfrm4_policy_afinfo);
 338 }
 339
 340 static void __exit xfrm4_policy_fini(void)
 341 {
 342         xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo);
 343 }
 344
 345 void __init xfrm4_init(void)
 346 {
 347         xfrm4_state_init();
 348         xfrm4_policy_init();
 349 }
 350