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 static unsigned int nat_net_id;
  33 static struct tc_action_ops act_nat_ops;
  34
  35 static const struct nla_policy nat_policy[TCA_NAT_MAX + 1] = {
  36         [TCA_NAT_PARMS] = { .len = sizeof(struct tc_nat) },
  37 };
  38
  39 static int tcf_nat_init(struct net *net, struct nlattr *nla, struct nlattr *est,
  40                         struct tc_action **a, int ovr, int bind,
  41                         bool rtnl_held, struct netlink_ext_ack *extack)
  42 {
  43         struct tc_action_net *tn = net_generic(net, nat_net_id);
  44         struct nlattr *tb[TCA_NAT_MAX + 1];
  45         struct tc_nat *parm;
  46         int ret = 0, err;
  47         struct tcf_nat *p;
  48
  49         if (nla == NULL)
  50                 return -EINVAL;
  51
  52         err = nla_parse_nested(tb, TCA_NAT_MAX, nla, nat_policy, NULL);
  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         if (!tcf_idr_check(tn, parm->index, a, bind)) {
  61                 ret = tcf_idr_create(tn, parm->index, est, a,
  62                                      &act_nat_ops, bind, false);
  63                 if (ret)
  64                         return ret;
  65                 ret = ACT_P_CREATED;
  66         } else {
  67                 if (bind)
  68                         return 0;
  69                 if (!ovr) {
  70                         tcf_idr_release(*a, bind);
  71                         return -EEXIST;
  72                 }
  73         }
  74         p = to_tcf_nat(*a);
  75
  76         spin_lock_bh(&p->tcf_lock);
  77         p->old_addr = parm->old_addr;
  78         p->new_addr = parm->new_addr;
  79         p->mask = parm->mask;
  80         p->flags = parm->flags;
  81
  82         p->tcf_action = parm->action;
  83         spin_unlock_bh(&p->tcf_lock);
  84
  85         if (ret == ACT_P_CREATED)
  86                 tcf_idr_insert(tn, *a);
  87
  88         return ret;
  89 }
  90
  91 static int tcf_nat(struct sk_buff *skb, const struct tc_action *a,
  92                    struct tcf_result *res)
  93 {
  94         struct tcf_nat *p = to_tcf_nat(a);
  95         struct iphdr *iph;
  96         __be32 old_addr;
  97         __be32 new_addr;
  98         __be32 mask;
  99         __be32 addr;
 100         int egress;
 101         int action;
 102         int ihl;
 103         int noff;
 104
 105         spin_lock(&p->tcf_lock);
 106
 107         tcf_lastuse_update(&p->tcf_tm);
 108         old_addr = p->old_addr;
 109         new_addr = p->new_addr;
 110         mask = p->mask;
 111         egress = p->flags & TCA_NAT_FLAG_EGRESS;
 112         action = p->tcf_action;
 113
 114         bstats_update(&p->tcf_bstats, skb);
 115
 116         spin_unlock(&p->tcf_lock);
 117
 118         if (unlikely(action == TC_ACT_SHOT))
 119                 goto drop;
 120
 121         noff = skb_network_offset(skb);
 122         if (!pskb_may_pull(skb, sizeof(*iph) + noff))
 123                 goto drop;
 124
 125         iph = ip_hdr(skb);
 126
 127         if (egress)
 128                 addr = iph->saddr;
 129         else
 130                 addr = iph->daddr;
 131
 132         if (!((old_addr ^ addr) & mask)) {
 133                 if (skb_try_make_writable(skb, sizeof(*iph) + noff))
 134                         goto drop;
 135
 136                 new_addr &= mask;
 137                 new_addr |= addr & ~mask;
 138
 139                 /* Rewrite IP header */
 140                 iph = ip_hdr(skb);
 141                 if (egress)
 142                         iph->saddr = new_addr;
 143                 else
 144                         iph->daddr = new_addr;
 145
 146                 csum_replace4(&iph->check, addr, new_addr);
 147         } else if ((iph->frag_off & htons(IP_OFFSET)) ||
 148                    iph->protocol != IPPROTO_ICMP) {
 149                 goto out;
 150         }
 151
 152         ihl = iph->ihl * 4;
 153
 154         /* It would be nice to share code with stateful NAT. */
 155         switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
 156         case IPPROTO_TCP:
 157         {
 158                 struct tcphdr *tcph;
 159
 160                 if (!pskb_may_pull(skb, ihl + sizeof(*tcph) + noff) ||
 161                     skb_try_make_writable(skb, ihl + sizeof(*tcph) + noff))
 162                         goto drop;
 163
 164                 tcph = (void *)(skb_network_header(skb) + ihl);
 165                 inet_proto_csum_replace4(&tcph->check, skb, addr, new_addr,
 166                                          true);
 167                 break;
 168         }
 169         case IPPROTO_UDP:
 170         {
 171                 struct udphdr *udph;
 172
 173                 if (!pskb_may_pull(skb, ihl + sizeof(*udph) + noff) ||
 174                     skb_try_make_writable(skb, ihl + sizeof(*udph) + noff))
 175                         goto drop;
 176
 177                 udph = (void *)(skb_network_header(skb) + ihl);
 178                 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 179                         inet_proto_csum_replace4(&udph->check, skb, addr,
 180                                                  new_addr, true);
 181                         if (!udph->check)
 182                                 udph->check = CSUM_MANGLED_0;
 183                 }
 184                 break;
 185         }
 186         case IPPROTO_ICMP:
 187         {
 188                 struct icmphdr *icmph;
 189
 190                 if (!pskb_may_pull(skb, ihl + sizeof(*icmph) + noff))
 191                         goto drop;
 192
 193                 icmph = (void *)(skb_network_header(skb) + ihl);
 194
 195                 if ((icmph->type != ICMP_DEST_UNREACH) &&
 196                     (icmph->type != ICMP_TIME_EXCEEDED) &&
 197                     (icmph->type != ICMP_PARAMETERPROB))
 198                         break;
 199
 200                 if (!pskb_may_pull(skb, ihl + sizeof(*icmph) + sizeof(*iph) +
 201                                         noff))
 202                         goto drop;
 203
 204                 icmph = (void *)(skb_network_header(skb) + ihl);
 205                 iph = (void *)(icmph + 1);
 206                 if (egress)
 207                         addr = iph->daddr;
 208                 else
 209                         addr = iph->saddr;
 210
 211                 if ((old_addr ^ addr) & mask)
 212                         break;
 213
 214                 if (skb_try_make_writable(skb, ihl + sizeof(*icmph) +
 215                                           sizeof(*iph) + noff))
 216                         goto drop;
 217
 218                 icmph = (void *)(skb_network_header(skb) + ihl);
 219                 iph = (void *)(icmph + 1);
 220
 221                 new_addr &= mask;
 222                 new_addr |= addr & ~mask;
 223
 224                 /* XXX Fix up the inner checksums. */
 225                 if (egress)
 226                         iph->daddr = new_addr;
 227                 else
 228                         iph->saddr = new_addr;
 229
 230                 inet_proto_csum_replace4(&icmph->checksum, skb, addr, new_addr,
 231                                          false);
 232                 break;
 233         }
 234         default:
 235                 break;
 236         }
 237
 238 out:
 239         return action;
 240
 241 drop:
 242         spin_lock(&p->tcf_lock);
 243         p->tcf_qstats.drops++;
 244         spin_unlock(&p->tcf_lock);
 245         return TC_ACT_SHOT;
 246 }
 247
 248 static int tcf_nat_dump(struct sk_buff *skb, struct tc_action *a,
 249                         int bind, int ref)
 250 {
 251         unsigned char *b = skb_tail_pointer(skb);
 252         struct tcf_nat *p = to_tcf_nat(a);
 253         struct tc_nat opt = {
 254                 .old_addr = p->old_addr,
 255                 .new_addr = p->new_addr,
 256                 .mask     = p->mask,
 257                 .flags    = p->flags,
 258
 259                 .index    = p->tcf_index,
 260                 .action   = p->tcf_action,
 261                 .refcnt   = refcount_read(&p->tcf_refcnt) - ref,
 262                 .bindcnt  = atomic_read(&p->tcf_bindcnt) - bind,
 263         };
 264         struct tcf_t t;
 265
 266         if (nla_put(skb, TCA_NAT_PARMS, sizeof(opt), &opt))
 267                 goto nla_put_failure;
 268
 269         tcf_tm_dump(&t, &p->tcf_tm);
 270         if (nla_put_64bit(skb, TCA_NAT_TM, sizeof(t), &t, TCA_NAT_PAD))
 271                 goto nla_put_failure;
 272
 273         return skb->len;
 274
 275 nla_put_failure:
 276         nlmsg_trim(skb, b);
 277         return -1;
 278 }
 279
 280 static int tcf_nat_walker(struct net *net, struct sk_buff *skb,
 281                           struct netlink_callback *cb, int type,
 282                           const struct tc_action_ops *ops,
 283                           struct netlink_ext_ack *extack)
 284 {
 285         struct tc_action_net *tn = net_generic(net, nat_net_id);
 286
 287         return tcf_generic_walker(tn, skb, cb, type, ops, extack);
 288 }
 289
 290 static int tcf_nat_search(struct net *net, struct tc_action **a, u32 index,
 291                           struct netlink_ext_ack *extack)
 292 {
 293         struct tc_action_net *tn = net_generic(net, nat_net_id);
 294
 295         return tcf_idr_search(tn, a, index);
 296 }
 297
 298 static int tcf_nat_delete(struct net *net, u32 index)
 299 {
 300         struct tc_action_net *tn = net_generic(net, nat_net_id);
 301
 302         return tcf_idr_delete_index(tn, index);
 303 }
 304
 305 static struct tc_action_ops act_nat_ops = {
 306         .kind           =       "nat",
 307         .type           =       TCA_ACT_NAT,
 308         .owner          =       THIS_MODULE,
 309         .act            =       tcf_nat,
 310         .dump           =       tcf_nat_dump,
 311         .init           =       tcf_nat_init,
 312         .walk           =       tcf_nat_walker,
 313         .lookup         =       tcf_nat_search,
 314         .delete         =       tcf_nat_delete,
 315         .size           =       sizeof(struct tcf_nat),
 316 };
 317
 318 static __net_init int nat_init_net(struct net *net)
 319 {
 320         struct tc_action_net *tn = net_generic(net, nat_net_id);
 321
 322         return tc_action_net_init(tn, &act_nat_ops);
 323 }
 324
 325 static void __net_exit nat_exit_net(struct list_head *net_list)
 326 {
 327         tc_action_net_exit(net_list, nat_net_id);
 328 }
 329
 330 static struct pernet_operations nat_net_ops = {
 331         .init = nat_init_net,
 332         .exit_batch = nat_exit_net,
 333         .id   = &nat_net_id,
 334         .size = sizeof(struct tc_action_net),
 335 };
 336
 337 MODULE_DESCRIPTION("Stateless NAT actions");
 338 MODULE_LICENSE("GPL");
 339
 340 static int __init nat_init_module(void)
 341 {
 342         return tcf_register_action(&act_nat_ops, &nat_net_ops);
 343 }
 344
 345 static void __exit nat_cleanup_module(void)
 346 {
 347         tcf_unregister_action(&act_nat_ops, &nat_net_ops);
 348 }
 349
 350 module_init(nat_init_module);
 351 module_exit(nat_cleanup_module);