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                                           xfrm_address_t *saddr,
  23                                           xfrm_address_t *daddr)
  24 {
  25         struct flowi fl = {
  26                 .nl_u = {
  27                         .ip4_u = {
  28                                 .tos = tos,
  29                                 .daddr = daddr->a4,
  30                         },
  31                 },
  32         };
  33         struct dst_entry *dst;
  34         struct rtable *rt;
  35         int err;
  36
  37         if (saddr)
  38                 fl.fl4_src = saddr->a4;
  39
  40         err = __ip_route_output_key(net, &rt, &fl);
  41         dst = &rt->dst;
  42         if (err)
  43                 dst = ERR_PTR(err);
  44         return dst;
  45 }
  46
  47 static int xfrm4_get_saddr(struct net *net,
  48                            xfrm_address_t *saddr, xfrm_address_t *daddr)
  49 {
  50         struct dst_entry *dst;
  51         struct rtable *rt;
  52
  53         dst = xfrm4_dst_lookup(net, 0, NULL, daddr);
  54         if (IS_ERR(dst))
  55                 return -EHOSTUNREACH;
  56
  57         rt = (struct rtable *)dst;
  58         saddr->a4 = rt->rt_src;
  59         dst_release(dst);
  60         return 0;
  61 }
  62
  63 static int xfrm4_get_tos(struct flowi *fl)
  64 {
  65         return IPTOS_RT_MASK & fl->fl4_tos; /* Strip ECN bits */
  66 }
  67
  68 static int xfrm4_init_path(struct xfrm_dst *path, struct dst_entry *dst,
  69                            int nfheader_len)
  70 {
  71         return 0;
  72 }
  73
  74 static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
  75                           struct flowi *fl)
  76 {
  77         struct rtable *rt = (struct rtable *)xdst->route;
  78
  79         xdst->u.rt.fl = *fl;
  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
 107         memset(fl, 0, sizeof(struct flowi));
 108         fl->mark = skb->mark;
 109
 110         if (!(iph->frag_off & htons(IP_MF | IP_OFFSET))) {
 111                 switch (iph->protocol) {
 112                 case IPPROTO_UDP:
 113                 case IPPROTO_UDPLITE:
 114                 case IPPROTO_TCP:
 115                 case IPPROTO_SCTP:
 116                 case IPPROTO_DCCP:
 117                         if (xprth + 4 < skb->data ||
 118                             pskb_may_pull(skb, xprth + 4 - skb->data)) {
 119                                 __be16 *ports = (__be16 *)xprth;
 120
 121                                 fl->fl_ip_sport = ports[!!reverse];
 122                                 fl->fl_ip_dport = ports[!reverse];
 123                         }
 124                         break;
 125
 126                 case IPPROTO_ICMP:
 127                         if (pskb_may_pull(skb, xprth + 2 - skb->data)) {
 128                                 u8 *icmp = xprth;
 129
 130                                 fl->fl_icmp_type = icmp[0];
 131                                 fl->fl_icmp_code = icmp[1];
 132                         }
 133                         break;
 134
 135                 case IPPROTO_ESP:
 136                         if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
 137                                 __be32 *ehdr = (__be32 *)xprth;
 138
 139                                 fl->fl_ipsec_spi = ehdr[0];
 140                         }
 141                         break;
 142
 143                 case IPPROTO_AH:
 144                         if (pskb_may_pull(skb, xprth + 8 - skb->data)) {
 145                                 __be32 *ah_hdr = (__be32*)xprth;
 146
 147                                 fl->fl_ipsec_spi = ah_hdr[1];
 148                         }
 149                         break;
 150
 151                 case IPPROTO_COMP:
 152                         if (pskb_may_pull(skb, xprth + 4 - skb->data)) {
 153                                 __be16 *ipcomp_hdr = (__be16 *)xprth;
 154
 155                                 fl->fl_ipsec_spi = htonl(ntohs(ipcomp_hdr[1]));
 156                         }
 157                         break;
 158
 159                 case IPPROTO_GRE:
 160                         if (pskb_may_pull(skb, xprth + 12 - skb->data)) {
 161                                 __be16 *greflags = (__be16 *)xprth;
 162                                 __be32 *gre_hdr = (__be32 *)xprth;
 163
 164                                 if (greflags[0] & GRE_KEY) {
 165                                         if (greflags[0] & GRE_CSUM)
 166                                                 gre_hdr++;
 167                                         fl->fl_gre_key = gre_hdr[1];
 168                                 }
 169                         }
 170                         break;
 171
 172                 default:
 173                         fl->fl_ipsec_spi = 0;
 174                         break;
 175                 }
 176         }
 177         fl->proto = iph->protocol;
 178         fl->fl4_dst = reverse ? iph->saddr : iph->daddr;
 179         fl->fl4_src = reverse ? iph->daddr : iph->saddr;
 180         fl->fl4_tos = iph->tos;
 181 }
 182
 183 static inline int xfrm4_garbage_collect(struct dst_ops *ops)
 184 {
 185         struct net *net = container_of(ops, struct net, xfrm.xfrm4_dst_ops);
 186
 187         xfrm4_policy_afinfo.garbage_collect(net);
 188         return (dst_entries_get_slow(ops) > ops->gc_thresh * 2);
 189 }
 190
 191 static void xfrm4_update_pmtu(struct dst_entry *dst, u32 mtu)
 192 {
 193         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 194         struct dst_entry *path = xdst->route;
 195
 196         path->ops->update_pmtu(path, mtu);
 197 }
 198
 199 static void xfrm4_dst_destroy(struct dst_entry *dst)
 200 {
 201         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 202
 203         if (likely(xdst->u.rt.peer))
 204                 inet_putpeer(xdst->u.rt.peer);
 205         xfrm_dst_destroy(xdst);
 206 }
 207
 208 static void xfrm4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 209                              int unregister)
 210 {
 211         if (!unregister)
 212                 return;
 213
 214         xfrm_dst_ifdown(dst, dev);
 215 }
 216
 217 static struct dst_ops xfrm4_dst_ops = {
 218         .family =               AF_INET,
 219         .protocol =             cpu_to_be16(ETH_P_IP),
 220         .gc =                   xfrm4_garbage_collect,
 221         .update_pmtu =          xfrm4_update_pmtu,
 222         .destroy =              xfrm4_dst_destroy,
 223         .ifdown =               xfrm4_dst_ifdown,
 224         .local_out =            __ip_local_out,
 225         .gc_thresh =            1024,
 226 };
 227
 228 static struct xfrm_policy_afinfo xfrm4_policy_afinfo = {
 229         .family =               AF_INET,
 230         .dst_ops =              &xfrm4_dst_ops,
 231         .dst_lookup =           xfrm4_dst_lookup,
 232         .get_saddr =            xfrm4_get_saddr,
 233         .decode_session =       _decode_session4,
 234         .get_tos =              xfrm4_get_tos,
 235         .init_path =            xfrm4_init_path,
 236         .fill_dst =             xfrm4_fill_dst,
 237 };
 238
 239 #ifdef CONFIG_SYSCTL
 240 static struct ctl_table xfrm4_policy_table[] = {
 241         {
 242                 .procname       = "xfrm4_gc_thresh",
 243                 .data           = &init_net.xfrm.xfrm4_dst_ops.gc_thresh,
 244                 .maxlen         = sizeof(int),
 245                 .mode           = 0644,
 246                 .proc_handler   = proc_dointvec,
 247         },
 248         { }
 249 };
 250
 251 static struct ctl_table_header *sysctl_hdr;
 252 #endif
 253
 254 static void __init xfrm4_policy_init(void)
 255 {
 256         xfrm_policy_register_afinfo(&xfrm4_policy_afinfo);
 257 }
 258
 259 static void __exit xfrm4_policy_fini(void)
 260 {
 261 #ifdef CONFIG_SYSCTL
 262         if (sysctl_hdr)
 263                 unregister_net_sysctl_table(sysctl_hdr);
 264 #endif
 265         xfrm_policy_unregister_afinfo(&xfrm4_policy_afinfo);
 266 }
 267
 268 void __init xfrm4_init(int rt_max_size)
 269 {
 270         /*
 271          * Select a default value for the gc_thresh based on the main route
 272          * table hash size.  It seems to me the worst case scenario is when
 273          * we have ipsec operating in transport mode, in which we create a
 274          * dst_entry per socket.  The xfrm gc algorithm starts trying to remove
 275          * entries at gc_thresh, and prevents new allocations as 2*gc_thresh
 276          * so lets set an initial xfrm gc_thresh value at the rt_max_size/2.
 277          * That will let us store an ipsec connection per route table entry,
 278          * and start cleaning when were 1/2 full
 279          */
 280         xfrm4_dst_ops.gc_thresh = rt_max_size/2;
 281         dst_entries_init(&xfrm4_dst_ops);
 282
 283         xfrm4_state_init();
 284         xfrm4_policy_init();
 285 #ifdef CONFIG_SYSCTL
 286         sysctl_hdr = register_net_sysctl_table(&init_net, net_ipv4_ctl_path,
 287                                                 xfrm4_policy_table);
 288 #endif
 289 }
 290