net/sched/act_nat.c

   1 /*
   2  * Stateless NAT actions
   3  *
   4  * Copyright (c) 2007 Herbert Xu <herbert@gondor.apana.org.au>
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the Free
   8  * Software Foundation; either version 2 of the License, or (at your option)
   9  * any later version.
  10  */
  11
  12 #include <linux/errno.h>
  13 #include <linux/init.h>
  14 #include <linux/kernel.h>
  15 #include <linux/module.h>
  16 #include <linux/netfilter.h>
  17 #include <linux/rtnetlink.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/slab.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/string.h>
  22 #include <linux/tc_act/tc_nat.h>
  23 #include <net/act_api.h>
  24 #include <net/icmp.h>
  25 #include <net/ip.h>
  26 #include <net/netlink.h>
  27 #include <net/tc_act/tc_nat.h>
  28 #include <net/tcp.h>
  29 #include <net/udp.h>
  30
  31
  32 #define NAT_TAB_MASK    15
  33
  34 static const struct nla_policy nat_policy[TCA_NAT_MAX + 1] = {
  35         [TCA_NAT_PARMS] = { .len = sizeof(struct tc_nat) },
  36 };
  37
  38 static int tcf_nat_init(struct net *net, struct nlattr *nla, struct nlattr *est,
  39                         struct tc_action *a, int ovr, int bind)
  40 {
  41         struct nlattr *tb[TCA_NAT_MAX + 1];
  42         struct tc_nat *parm;
  43         int ret = 0, err;
  44         struct tcf_nat *p;
  45
  46         if (nla == NULL)
  47                 return -EINVAL;
  48
  49         err = nla_parse_nested(tb, TCA_NAT_MAX, nla, nat_policy);
  50         if (err < 0)
  51                 return err;
  52
  53         if (tb[TCA_NAT_PARMS] == NULL)
  54                 return -EINVAL;
  55         parm = nla_data(tb[TCA_NAT_PARMS]);
  56
  57         if (!tcf_hash_check(parm->index, a, bind)) {
  58                 ret = tcf_hash_create(parm->index, est, a, sizeof(*p),
  59                                       bind, false);
  60                 if (ret)
  61                         return ret;
  62                 ret = ACT_P_CREATED;
  63         } else {
  64                 if (bind)
  65                         return 0;
  66                 tcf_hash_release(a, bind);
  67                 if (!ovr)
  68                         return -EEXIST;
  69         }
  70         p = to_tcf_nat(a);
  71
  72         spin_lock_bh(&p->tcf_lock);
  73         p->old_addr = parm->old_addr;
  74         p->new_addr = parm->new_addr;
  75         p->mask = parm->mask;
  76         p->flags = parm->flags;
  77
  78         p->tcf_action = parm->action;
  79         spin_unlock_bh(&p->tcf_lock);
  80
  81         if (ret == ACT_P_CREATED)
  82                 tcf_hash_insert(a);
  83
  84         return ret;
  85 }
  86
  87 static int tcf_nat(struct sk_buff *skb, const struct tc_action *a,
  88                    struct tcf_result *res)
  89 {
  90         struct tcf_nat *p = a->priv;
  91         struct iphdr *iph;
  92         __be32 old_addr;
  93         __be32 new_addr;
  94         __be32 mask;
  95         __be32 addr;
  96         int egress;
  97         int action;
  98         int ihl;
  99         int noff;
 100
 101         spin_lock(&p->tcf_lock);
 102
 103         p->tcf_tm.lastuse = jiffies;
 104         old_addr = p->old_addr;
 105         new_addr = p->new_addr;
 106         mask = p->mask;
 107         egress = p->flags & TCA_NAT_FLAG_EGRESS;
 108         action = p->tcf_action;
 109
 110         bstats_update(&p->tcf_bstats, skb);
 111
 112         spin_unlock(&p->tcf_lock);
 113
 114         if (unlikely(action == TC_ACT_SHOT))
 115                 goto drop;
 116
 117         noff = skb_network_offset(skb);
 118         if (!pskb_may_pull(skb, sizeof(*iph) + noff))
 119                 goto drop;
 120
 121         iph = ip_hdr(skb);
 122
 123         if (egress)
 124                 addr = iph->saddr;
 125         else
 126                 addr = iph->daddr;
 127
 128         if (!((old_addr ^ addr) & mask)) {
 129                 if (skb_try_make_writable(skb, sizeof(*iph) + noff))
 130                         goto drop;
 131
 132                 new_addr &= mask;
 133                 new_addr |= addr & ~mask;
 134
 135                 /* Rewrite IP header */
 136                 iph = ip_hdr(skb);
 137                 if (egress)
 138                         iph->saddr = new_addr;
 139                 else
 140                         iph->daddr = new_addr;
 141
 142                 csum_replace4(&iph->check, addr, new_addr);
 143         } else if ((iph->frag_off & htons(IP_OFFSET)) ||
 144                    iph->protocol != IPPROTO_ICMP) {
 145                 goto out;
 146         }
 147
 148         ihl = iph->ihl * 4;
 149
 150         /* It would be nice to share code with stateful NAT. */
 151         switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
 152         case IPPROTO_TCP:
 153         {
 154                 struct tcphdr *tcph;
 155
 156                 if (!pskb_may_pull(skb, ihl + sizeof(*tcph) + noff) ||
 157                     skb_try_make_writable(skb, ihl + sizeof(*tcph) + noff))
 158                         goto drop;
 159
 160                 tcph = (void *)(skb_network_header(skb) + ihl);
 161                 inet_proto_csum_replace4(&tcph->check, skb, addr, new_addr,
 162                                          true);
 163                 break;
 164         }
 165         case IPPROTO_UDP:
 166         {
 167                 struct udphdr *udph;
 168
 169                 if (!pskb_may_pull(skb, ihl + sizeof(*udph) + noff) ||
 170                     skb_try_make_writable(skb, ihl + sizeof(*udph) + noff))
 171                         goto drop;
 172
 173                 udph = (void *)(skb_network_header(skb) + ihl);
 174                 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 175                         inet_proto_csum_replace4(&udph->check, skb, addr,
 176                                                  new_addr, true);
 177                         if (!udph->check)
 178                                 udph->check = CSUM_MANGLED_0;
 179                 }
 180                 break;
 181         }
 182         case IPPROTO_ICMP:
 183         {
 184                 struct icmphdr *icmph;
 185
 186                 if (!pskb_may_pull(skb, ihl + sizeof(*icmph) + noff))
 187                         goto drop;
 188
 189                 icmph = (void *)(skb_network_header(skb) + ihl);
 190
 191                 if ((icmph->type != ICMP_DEST_UNREACH) &&
 192                     (icmph->type != ICMP_TIME_EXCEEDED) &&
 193                     (icmph->type != ICMP_PARAMETERPROB))
 194                         break;
 195
 196                 if (!pskb_may_pull(skb, ihl + sizeof(*icmph) + sizeof(*iph) +
 197                                         noff))
 198                         goto drop;
 199
 200                 icmph = (void *)(skb_network_header(skb) + ihl);
 201                 iph = (void *)(icmph + 1);
 202                 if (egress)
 203                         addr = iph->daddr;
 204                 else
 205                         addr = iph->saddr;
 206
 207                 if ((old_addr ^ addr) & mask)
 208                         break;
 209
 210                 if (skb_try_make_writable(skb, ihl + sizeof(*icmph) +
 211                                           sizeof(*iph) + noff))
 212                         goto drop;
 213
 214                 icmph = (void *)(skb_network_header(skb) + ihl);
 215                 iph = (void *)(icmph + 1);
 216
 217                 new_addr &= mask;
 218                 new_addr |= addr & ~mask;
 219
 220                 /* XXX Fix up the inner checksums. */
 221                 if (egress)
 222                         iph->daddr = new_addr;
 223                 else
 224                         iph->saddr = new_addr;
 225
 226                 inet_proto_csum_replace4(&icmph->checksum, skb, addr, new_addr,
 227                                          false);
 228                 break;
 229         }
 230         default:
 231                 break;
 232         }
 233
 234 out:
 235         return action;
 236
 237 drop:
 238         spin_lock(&p->tcf_lock);
 239         p->tcf_qstats.drops++;
 240         spin_unlock(&p->tcf_lock);
 241         return TC_ACT_SHOT;
 242 }
 243
 244 static int tcf_nat_dump(struct sk_buff *skb, struct tc_action *a,
 245                         int bind, int ref)
 246 {
 247         unsigned char *b = skb_tail_pointer(skb);
 248         struct tcf_nat *p = a->priv;
 249         struct tc_nat opt = {
 250                 .old_addr = p->old_addr,
 251                 .new_addr = p->new_addr,
 252                 .mask     = p->mask,
 253                 .flags    = p->flags,
 254
 255                 .index    = p->tcf_index,
 256                 .action   = p->tcf_action,
 257                 .refcnt   = p->tcf_refcnt - ref,
 258                 .bindcnt  = p->tcf_bindcnt - bind,
 259         };
 260         struct tcf_t t;
 261
 262         if (nla_put(skb, TCA_NAT_PARMS, sizeof(opt), &opt))
 263                 goto nla_put_failure;
 264         t.install = jiffies_to_clock_t(jiffies - p->tcf_tm.install);
 265         t.lastuse = jiffies_to_clock_t(jiffies - p->tcf_tm.lastuse);
 266         t.expires = jiffies_to_clock_t(p->tcf_tm.expires);
 267         if (nla_put(skb, TCA_NAT_TM, sizeof(t), &t))
 268                 goto nla_put_failure;
 269
 270         return skb->len;
 271
 272 nla_put_failure:
 273         nlmsg_trim(skb, b);
 274         return -1;
 275 }
 276
 277 static struct tc_action_ops act_nat_ops = {
 278         .kind           =       "nat",
 279         .type           =       TCA_ACT_NAT,
 280         .owner          =       THIS_MODULE,
 281         .act            =       tcf_nat,
 282         .dump           =       tcf_nat_dump,
 283         .init           =       tcf_nat_init,
 284 };
 285
 286 MODULE_DESCRIPTION("Stateless NAT actions");
 287 MODULE_LICENSE("GPL");
 288
 289 static int __init nat_init_module(void)
 290 {
 291         return tcf_register_action(&act_nat_ops, NAT_TAB_MASK);
 292 }
 293
 294 static void __exit nat_cleanup_module(void)
 295 {
 296         tcf_unregister_action(&act_nat_ops);
 297 }
 298
 299 module_init(nat_init_module);
 300 module_exit(nat_cleanup_module);