net/sched/act_nat.c

   1 /*
   2  * Stateless NAT actions
   3  *
   4  * Copyright (c) 2007 Herbert Xu <[email protected]>
   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 struct tcf_hashinfo nat_hash_info;
  35
  36 static const struct nla_policy nat_policy[TCA_NAT_MAX + 1] = {
  37         [TCA_NAT_PARMS] = { .len = sizeof(struct tc_nat) },
  38 };
  39
  40 static int tcf_nat_init(struct net *net, struct nlattr *nla, struct nlattr *est,
  41                         struct tc_action *a, int ovr, int bind)
  42 {
  43         struct nlattr *tb[TCA_NAT_MAX + 1];
  44         struct tc_nat *parm;
  45         int ret = 0, err;
  46         struct tcf_nat *p;
  47         struct tcf_common *pc;
  48
  49         if (nla == NULL)
  50                 return -EINVAL;
  51
  52         err = nla_parse_nested(tb, TCA_NAT_MAX, nla, nat_policy);
  53         if (err < 0)
  54                 return err;
  55
  56         if (tb[TCA_NAT_PARMS] == NULL)
  57                 return -EINVAL;
  58         parm = nla_data(tb[TCA_NAT_PARMS]);
  59
  60         pc = tcf_hash_check(parm->index, a, bind);
  61         if (!pc) {
  62                 pc = tcf_hash_create(parm->index, est, a, sizeof(*p), bind);
  63                 if (IS_ERR(pc))
  64                         return PTR_ERR(pc);
  65                 ret = ACT_P_CREATED;
  66         } else {
  67                 if (bind)
  68                         return 0;
  69                 tcf_hash_release(pc, bind, a->ops->hinfo);
  70                 if (!ovr)
  71                         return -EEXIST;
  72         }
  73         p = to_tcf_nat(pc);
  74
  75         spin_lock_bh(&p->tcf_lock);
  76         p->old_addr = parm->old_addr;
  77         p->new_addr = parm->new_addr;
  78         p->mask = parm->mask;
  79         p->flags = parm->flags;
  80
  81         p->tcf_action = parm->action;
  82         spin_unlock_bh(&p->tcf_lock);
  83
  84         if (ret == ACT_P_CREATED)
  85                 tcf_hash_insert(pc, a->ops->hinfo);
  86
  87         return ret;
  88 }
  89
  90 static int tcf_nat_cleanup(struct tc_action *a, int bind)
  91 {
  92         struct tcf_nat *p = a->priv;
  93
  94         return tcf_hash_release(&p->common, bind, &nat_hash_info);
  95 }
  96
  97 static int tcf_nat(struct sk_buff *skb, const struct tc_action *a,
  98                    struct tcf_result *res)
  99 {
 100         struct tcf_nat *p = a->priv;
 101         struct iphdr *iph;
 102         __be32 old_addr;
 103         __be32 new_addr;
 104         __be32 mask;
 105         __be32 addr;
 106         int egress;
 107         int action;
 108         int ihl;
 109         int noff;
 110
 111         spin_lock(&p->tcf_lock);
 112
 113         p->tcf_tm.lastuse = jiffies;
 114         old_addr = p->old_addr;
 115         new_addr = p->new_addr;
 116         mask = p->mask;
 117         egress = p->flags & TCA_NAT_FLAG_EGRESS;
 118         action = p->tcf_action;
 119
 120         bstats_update(&p->tcf_bstats, skb);
 121
 122         spin_unlock(&p->tcf_lock);
 123
 124         if (unlikely(action == TC_ACT_SHOT))
 125                 goto drop;
 126
 127         noff = skb_network_offset(skb);
 128         if (!pskb_may_pull(skb, sizeof(*iph) + noff))
 129                 goto drop;
 130
 131         iph = ip_hdr(skb);
 132
 133         if (egress)
 134                 addr = iph->saddr;
 135         else
 136                 addr = iph->daddr;
 137
 138         if (!((old_addr ^ addr) & mask)) {
 139                 if (skb_cloned(skb) &&
 140                     !skb_clone_writable(skb, sizeof(*iph) + noff) &&
 141                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
 142                         goto drop;
 143
 144                 new_addr &= mask;
 145                 new_addr |= addr & ~mask;
 146
 147                 /* Rewrite IP header */
 148                 iph = ip_hdr(skb);
 149                 if (egress)
 150                         iph->saddr = new_addr;
 151                 else
 152                         iph->daddr = new_addr;
 153
 154                 csum_replace4(&iph->check, addr, new_addr);
 155         } else if ((iph->frag_off & htons(IP_OFFSET)) ||
 156                    iph->protocol != IPPROTO_ICMP) {
 157                 goto out;
 158         }
 159
 160         ihl = iph->ihl * 4;
 161
 162         /* It would be nice to share code with stateful NAT. */
 163         switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
 164         case IPPROTO_TCP:
 165         {
 166                 struct tcphdr *tcph;
 167
 168                 if (!pskb_may_pull(skb, ihl + sizeof(*tcph) + noff) ||
 169                     (skb_cloned(skb) &&
 170                      !skb_clone_writable(skb, ihl + sizeof(*tcph) + noff) &&
 171                      pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
 172                         goto drop;
 173
 174                 tcph = (void *)(skb_network_header(skb) + ihl);
 175                 inet_proto_csum_replace4(&tcph->check, skb, addr, new_addr, 1);
 176                 break;
 177         }
 178         case IPPROTO_UDP:
 179         {
 180                 struct udphdr *udph;
 181
 182                 if (!pskb_may_pull(skb, ihl + sizeof(*udph) + noff) ||
 183                     (skb_cloned(skb) &&
 184                      !skb_clone_writable(skb, ihl + sizeof(*udph) + noff) &&
 185                      pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
 186                         goto drop;
 187
 188                 udph = (void *)(skb_network_header(skb) + ihl);
 189                 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 190                         inet_proto_csum_replace4(&udph->check, skb, addr,
 191                                                  new_addr, 1);
 192                         if (!udph->check)
 193                                 udph->check = CSUM_MANGLED_0;
 194                 }
 195                 break;
 196         }
 197         case IPPROTO_ICMP:
 198         {
 199                 struct icmphdr *icmph;
 200
 201                 if (!pskb_may_pull(skb, ihl + sizeof(*icmph) + noff))
 202                         goto drop;
 203
 204                 icmph = (void *)(skb_network_header(skb) + ihl);
 205
 206                 if ((icmph->type != ICMP_DEST_UNREACH) &&
 207                     (icmph->type != ICMP_TIME_EXCEEDED) &&
 208                     (icmph->type != ICMP_PARAMETERPROB))
 209                         break;
 210
 211                 if (!pskb_may_pull(skb, ihl + sizeof(*icmph) + sizeof(*iph) +
 212                                         noff))
 213                         goto drop;
 214
 215                 icmph = (void *)(skb_network_header(skb) + ihl);
 216                 iph = (void *)(icmph + 1);
 217                 if (egress)
 218                         addr = iph->daddr;
 219                 else
 220                         addr = iph->saddr;
 221
 222                 if ((old_addr ^ addr) & mask)
 223                         break;
 224
 225                 if (skb_cloned(skb) &&
 226                     !skb_clone_writable(skb, ihl + sizeof(*icmph) +
 227                                              sizeof(*iph) + noff) &&
 228                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
 229                         goto drop;
 230
 231                 icmph = (void *)(skb_network_header(skb) + ihl);
 232                 iph = (void *)(icmph + 1);
 233
 234                 new_addr &= mask;
 235                 new_addr |= addr & ~mask;
 236
 237                 /* XXX Fix up the inner checksums. */
 238                 if (egress)
 239                         iph->daddr = new_addr;
 240                 else
 241                         iph->saddr = new_addr;
 242
 243                 inet_proto_csum_replace4(&icmph->checksum, skb, addr, new_addr,
 244                                          0);
 245                 break;
 246         }
 247         default:
 248                 break;
 249         }
 250
 251 out:
 252         return action;
 253
 254 drop:
 255         spin_lock(&p->tcf_lock);
 256         p->tcf_qstats.drops++;
 257         spin_unlock(&p->tcf_lock);
 258         return TC_ACT_SHOT;
 259 }
 260
 261 static int tcf_nat_dump(struct sk_buff *skb, struct tc_action *a,
 262                         int bind, int ref)
 263 {
 264         unsigned char *b = skb_tail_pointer(skb);
 265         struct tcf_nat *p = a->priv;
 266         struct tc_nat opt = {
 267                 .old_addr = p->old_addr,
 268                 .new_addr = p->new_addr,
 269                 .mask     = p->mask,
 270                 .flags    = p->flags,
 271
 272                 .index    = p->tcf_index,
 273                 .action   = p->tcf_action,
 274                 .refcnt   = p->tcf_refcnt - ref,
 275                 .bindcnt  = p->tcf_bindcnt - bind,
 276         };
 277         struct tcf_t t;
 278
 279         if (nla_put(skb, TCA_NAT_PARMS, sizeof(opt), &opt))
 280                 goto nla_put_failure;
 281         t.install = jiffies_to_clock_t(jiffies - p->tcf_tm.install);
 282         t.lastuse = jiffies_to_clock_t(jiffies - p->tcf_tm.lastuse);
 283         t.expires = jiffies_to_clock_t(p->tcf_tm.expires);
 284         if (nla_put(skb, TCA_NAT_TM, sizeof(t), &t))
 285                 goto nla_put_failure;
 286
 287         return skb->len;
 288
 289 nla_put_failure:
 290         nlmsg_trim(skb, b);
 291         return -1;
 292 }
 293
 294 static struct tc_action_ops act_nat_ops = {
 295         .kind           =       "nat",
 296         .hinfo          =       &nat_hash_info,
 297         .type           =       TCA_ACT_NAT,
 298         .owner          =       THIS_MODULE,
 299         .act            =       tcf_nat,
 300         .dump           =       tcf_nat_dump,
 301         .cleanup        =       tcf_nat_cleanup,
 302         .init           =       tcf_nat_init,
 303 };
 304
 305 MODULE_DESCRIPTION("Stateless NAT actions");
 306 MODULE_LICENSE("GPL");
 307
 308 static int __init nat_init_module(void)
 309 {
 310         int err = tcf_hashinfo_init(&nat_hash_info, NAT_TAB_MASK);
 311         if (err)
 312                 return err;
 313         return tcf_register_action(&act_nat_ops);
 314 }
 315
 316 static void __exit nat_cleanup_module(void)
 317 {
 318         tcf_unregister_action(&act_nat_ops);
 319         tcf_hashinfo_destroy(&nat_hash_info);
 320 }
 321
 322 module_init(nat_init_module);
 323 module_exit(nat_cleanup_module);