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/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, int tos,
  22                                           const xfrm_address_t *saddr,
  23                                           const xfrm_address_t *daddr)
  24 {
  25         struct flowi4 fl4 = {
  26                 .daddr = daddr->a4,
  27                 .flowi4_tos = tos,
  28         };
  29         struct rtable *rt;
  30
  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 int xfrm4_get_saddr(struct net *net,
  42                            xfrm_address_t *saddr, xfrm_address_t *daddr)
  43 {
  44         struct dst_entry *dst;
  45         struct rtable *rt;
  46
  47         dst = xfrm4_dst_lookup(net, 0, NULL, daddr);
  48         if (IS_ERR(dst))
  49                 return -EHOSTUNREACH;
  50
  51         rt = (struct rtable *)dst;
  52         saddr->a4 = rt->rt_src;
  53         dst_release(dst);
  54         return 0;
  55 }
  56
  57 static int xfrm4_get_tos(const struct flowi *fl)
  58 {
  59         return IPTOS_RT_MASK & fl->u.ip4.flowi4_tos; /* Strip ECN bits */
  60 }
  61
  62 static int xfrm4_init_path(struct xfrm_dst *path, struct dst_entry *dst,
  63                            int nfheader_len)
  64 {
  65         return 0;
  66 }
  67
  68 static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
  69                           const struct flowi *fl)
  70 {
  71         struct rtable *rt = (struct rtable *)xdst->route;
  72         const struct flowi4 *fl4 = &fl->u.ip4;
  73
  74         rt->rt_key_dst = fl4->daddr;
  75         rt->rt_key_src = fl4->saddr;
  76         rt->rt_tos = fl4->flowi4_tos;
  77         rt->rt_iif = fl4->flowi4_iif;
  78         rt->rt_oif = fl4->flowi4_oif;
  79         rt->rt_mark = fl4->flowi4_mark;
  80
  81         xdst->u.dst.dev = dev;
  82         dev_hold(dev);
  83
  84         xdst->u.rt.peer = rt->peer;
  85         if (rt->peer)
  86                 atomic_inc(&rt->peer->refcnt);
  87
  88         /* Sheit... I remember I did this right. Apparently,
  89          * it was magically lost, so this code needs audit */
  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_src = rt->rt_src;
  94         xdst->u.rt.rt_dst = rt->rt_dst;
  95         xdst->u.rt.rt_gateway = rt->rt_gateway;
  96         xdst->u.rt.rt_spec_dst = rt->rt_spec_dst;
  97
  98         return 0;
  99 }
 100
 101 static void
 102 _decode_session4(struct sk_buff *skb, struct flowi *fl, int reverse)
 103 {
 104         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 (!(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                                 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, u32 mtu)
 193 {
 194         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 195         struct dst_entry *path = xdst->route;
 196
 197         path->ops->update_pmtu(path, mtu);
 198 }
 199
 200 static void xfrm4_dst_destroy(struct dst_entry *dst)
 201 {
 202         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 203
 204         dst_destroy_metrics_generic(dst);
 205
 206         if (likely(xdst->u.rt.peer))
 207                 inet_putpeer(xdst->u.rt.peer);
 208
 209         xfrm_dst_destroy(xdst);
 210 }
 211
 212 static void xfrm4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 213                              int unregister)
 214 {
 215         if (!unregister)
 216                 return;
 217
 218         xfrm_dst_ifdown(dst, dev);
 219 }
 220
 221 static struct dst_ops xfrm4_dst_ops = {
 222         .family =               AF_INET,
 223         .protocol =             cpu_to_be16(ETH_P_IP),
 224         .gc =                   xfrm4_garbage_collect,
 225         .update_pmtu =          xfrm4_update_pmtu,
 226         .cow_metrics =          dst_cow_metrics_generic,
 227         .destroy =              xfrm4_dst_destroy,
 228         .ifdown =               xfrm4_dst_ifdown,
 229         .local_out =            __ip_local_out,
 230         .gc_thresh =            1024,
 231 };
 232
 233 static struct xfrm_policy_afinfo xfrm4_policy_afinfo = {
 234         .family =               AF_INET,
 235         .dst_ops =              &xfrm4_dst_ops,
 236         .dst_lookup =           xfrm4_dst_lookup,
 237         .get_saddr =            xfrm4_get_saddr,
 238         .decode_session =       _decode_session4,
 239         .get_tos =              xfrm4_get_tos,
 240         .init_path =            xfrm4_init_path,
 241         .fill_dst =             xfrm4_fill_dst,
 242         .blackhole_route =      ipv4_blackhole_route,
 243 };
 244
 245 #ifdef CONFIG_SYSCTL
 246 static struct ctl_table xfrm4_policy_table[] = {
 247         {
 248                 .procname       = "xfrm4_gc_thresh",
 249                 .data           = &init_net.xfrm.xfrm4_dst_ops.gc_thresh,
 250                 .maxlen         = sizeof(int),
 251                 .mode           = 0644,
 252                 .proc_handler   = proc_dointvec,
 253         },
 254         { }
 255 };
 256
 257 static struct ctl_table_header *sysctl_hdr;
 258 #endif
 259
 260 static void __init xfrm4_policy_init(void)
 261 {
 262         xfrm_policy_register_afinfo(&xfrm4_policy_afinfo);
 263 }
 264
 265 static void __exit xfrm4_policy_fini(void)
 266 {
 267 #ifdef CONFIG_SYSCTL
 268         if (sysctl_hdr)
 269                 unregister_net_sysctl_table(sysctl_hdr);
 270 #endif
 271         xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo);
 272 }
 273
 274 void __init xfrm4_init(int rt_max_size)
 275 {
 276         /*
 277          * Select a default value for the gc_thresh based on the main route
 278          * table hash size.  It seems to me the worst case scenario is when
 279          * we have ipsec operating in transport mode, in which we create a
 280          * dst_entry per socket.  The xfrm gc algorithm starts trying to remove
 281          * entries at gc_thresh, and prevents new allocations as 2*gc_thresh
 282          * so lets set an initial xfrm gc_thresh value at the rt_max_size/2.
 283          * That will let us store an ipsec connection per route table entry,
 284          * and start cleaning when were 1/2 full
 285          */
 286         xfrm4_dst_ops.gc_thresh = rt_max_size/2;
 287         dst_entries_init(&xfrm4_dst_ops);
 288
 289         xfrm4_state_init();
 290         xfrm4_policy_init();
 291 #ifdef CONFIG_SYSCTL
 292         sysctl_hdr = register_net_sysctl_table(&init_net, net_ipv4_ctl_path,
 293                                                 xfrm4_policy_table);
 294 #endif
 295 }
 296