net/sched/act_nat.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Stateless NAT actions
   4  *
   5  * Copyright (c) 2007 Herbert Xu <herbert@gondor.apana.org.au>
   6  */
   7
   8 #include <linux/errno.h>
   9 #include <linux/init.h>
  10 #include <linux/kernel.h>
  11 #include <linux/module.h>
  12 #include <linux/netfilter.h>
  13 #include <linux/rtnetlink.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/slab.h>
  16 #include <linux/spinlock.h>
  17 #include <linux/string.h>
  18 #include <linux/tc_act/tc_nat.h>
  19 #include <net/act_api.h>
  20 #include <net/pkt_cls.h>
  21 #include <net/icmp.h>
  22 #include <net/ip.h>
  23 #include <net/netlink.h>
  24 #include <net/tc_act/tc_nat.h>
  25 #include <net/tcp.h>
  26 #include <net/udp.h>
  27
  28
  29 static struct tc_action_ops act_nat_ops;
  30
  31 static const struct nla_policy nat_policy[TCA_NAT_MAX + 1] = {
  32         [TCA_NAT_PARMS] = { .len = sizeof(struct tc_nat) },
  33 };
  34
  35 static int tcf_nat_init(struct net *net, struct nlattr *nla, struct nlattr *est,
  36                         struct tc_action **a, struct tcf_proto *tp,
  37                         u32 flags, struct netlink_ext_ack *extack)
  38 {
  39         struct tc_action_net *tn = net_generic(net, act_nat_ops.net_id);
  40         bool bind = flags & TCA_ACT_FLAGS_BIND;
  41         struct nlattr *tb[TCA_NAT_MAX + 1];
  42         struct tcf_chain *goto_ch = NULL;
  43         struct tc_nat *parm;
  44         int ret = 0, err;
  45         struct tcf_nat *p;
  46         u32 index;
  47
  48         if (nla == NULL)
  49                 return -EINVAL;
  50
  51         err = nla_parse_nested_deprecated(tb, TCA_NAT_MAX, nla, nat_policy,
  52                                           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         index = parm->index;
  60         err = tcf_idr_check_alloc(tn, &index, a, bind);
  61         if (!err) {
  62                 ret = tcf_idr_create(tn, index, est, a,
  63                                      &act_nat_ops, bind, false, flags);
  64                 if (ret) {
  65                         tcf_idr_cleanup(tn, index);
  66                         return ret;
  67                 }
  68                 ret = ACT_P_CREATED;
  69         } else if (err > 0) {
  70                 if (bind)
  71                         return 0;
  72                 if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
  73                         tcf_idr_release(*a, bind);
  74                         return -EEXIST;
  75                 }
  76         } else {
  77                 return err;
  78         }
  79         err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
  80         if (err < 0)
  81                 goto release_idr;
  82         p = to_tcf_nat(*a);
  83
  84         spin_lock_bh(&p->tcf_lock);
  85         p->old_addr = parm->old_addr;
  86         p->new_addr = parm->new_addr;
  87         p->mask = parm->mask;
  88         p->flags = parm->flags;
  89
  90         goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
  91         spin_unlock_bh(&p->tcf_lock);
  92         if (goto_ch)
  93                 tcf_chain_put_by_act(goto_ch);
  94
  95         return ret;
  96 release_idr:
  97         tcf_idr_release(*a, bind);
  98         return err;
  99 }
 100
 101 static int tcf_nat_act(struct sk_buff *skb, const struct tc_action *a,
 102                        struct tcf_result *res)
 103 {
 104         struct tcf_nat *p = to_tcf_nat(a);
 105         struct iphdr *iph;
 106         __be32 old_addr;
 107         __be32 new_addr;
 108         __be32 mask;
 109         __be32 addr;
 110         int egress;
 111         int action;
 112         int ihl;
 113         int noff;
 114
 115         spin_lock(&p->tcf_lock);
 116
 117         tcf_lastuse_update(&p->tcf_tm);
 118         old_addr = p->old_addr;
 119         new_addr = p->new_addr;
 120         mask = p->mask;
 121         egress = p->flags & TCA_NAT_FLAG_EGRESS;
 122         action = p->tcf_action;
 123
 124         bstats_update(&p->tcf_bstats, skb);
 125
 126         spin_unlock(&p->tcf_lock);
 127
 128         if (unlikely(action == TC_ACT_SHOT))
 129                 goto drop;
 130
 131         noff = skb_network_offset(skb);
 132         if (!pskb_may_pull(skb, sizeof(*iph) + noff))
 133                 goto drop;
 134
 135         iph = ip_hdr(skb);
 136
 137         if (egress)
 138                 addr = iph->saddr;
 139         else
 140                 addr = iph->daddr;
 141
 142         if (!((old_addr ^ addr) & mask)) {
 143                 if (skb_try_make_writable(skb, sizeof(*iph) + noff))
 144                         goto drop;
 145
 146                 new_addr &= mask;
 147                 new_addr |= addr & ~mask;
 148
 149                 /* Rewrite IP header */
 150                 iph = ip_hdr(skb);
 151                 if (egress)
 152                         iph->saddr = new_addr;
 153                 else
 154                         iph->daddr = new_addr;
 155
 156                 csum_replace4(&iph->check, addr, new_addr);
 157         } else if ((iph->frag_off & htons(IP_OFFSET)) ||
 158                    iph->protocol != IPPROTO_ICMP) {
 159                 goto out;
 160         }
 161
 162         ihl = iph->ihl * 4;
 163
 164         /* It would be nice to share code with stateful NAT. */
 165         switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
 166         case IPPROTO_TCP:
 167         {
 168                 struct tcphdr *tcph;
 169
 170                 if (!pskb_may_pull(skb, ihl + sizeof(*tcph) + noff) ||
 171                     skb_try_make_writable(skb, ihl + sizeof(*tcph) + noff))
 172                         goto drop;
 173
 174                 tcph = (void *)(skb_network_header(skb) + ihl);
 175                 inet_proto_csum_replace4(&tcph->check, skb, addr, new_addr,
 176                                          true);
 177                 break;
 178         }
 179         case IPPROTO_UDP:
 180         {
 181                 struct udphdr *udph;
 182
 183                 if (!pskb_may_pull(skb, ihl + sizeof(*udph) + noff) ||
 184                     skb_try_make_writable(skb, ihl + sizeof(*udph) + noff))
 185                         goto drop;
 186
 187                 udph = (void *)(skb_network_header(skb) + ihl);
 188                 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 189                         inet_proto_csum_replace4(&udph->check, skb, addr,
 190                                                  new_addr, true);
 191                         if (!udph->check)
 192                                 udph->check = CSUM_MANGLED_0;
 193                 }
 194                 break;
 195         }
 196         case IPPROTO_ICMP:
 197         {
 198                 struct icmphdr *icmph;
 199
 200                 if (!pskb_may_pull(skb, ihl + sizeof(*icmph) + noff))
 201                         goto drop;
 202
 203                 icmph = (void *)(skb_network_header(skb) + ihl);
 204
 205                 if (!icmp_is_err(icmph->type))
 206                         break;
 207
 208                 if (!pskb_may_pull(skb, ihl + sizeof(*icmph) + sizeof(*iph) +
 209                                         noff))
 210                         goto drop;
 211
 212                 icmph = (void *)(skb_network_header(skb) + ihl);
 213                 iph = (void *)(icmph + 1);
 214                 if (egress)
 215                         addr = iph->daddr;
 216                 else
 217                         addr = iph->saddr;
 218
 219                 if ((old_addr ^ addr) & mask)
 220                         break;
 221
 222                 if (skb_try_make_writable(skb, ihl + sizeof(*icmph) +
 223                                           sizeof(*iph) + noff))
 224                         goto drop;
 225
 226                 icmph = (void *)(skb_network_header(skb) + ihl);
 227                 iph = (void *)(icmph + 1);
 228
 229                 new_addr &= mask;
 230                 new_addr |= addr & ~mask;
 231
 232                 /* XXX Fix up the inner checksums. */
 233                 if (egress)
 234                         iph->daddr = new_addr;
 235                 else
 236                         iph->saddr = new_addr;
 237
 238                 inet_proto_csum_replace4(&icmph->checksum, skb, addr, new_addr,
 239                                          false);
 240                 break;
 241         }
 242         default:
 243                 break;
 244         }
 245
 246 out:
 247         return action;
 248
 249 drop:
 250         spin_lock(&p->tcf_lock);
 251         p->tcf_qstats.drops++;
 252         spin_unlock(&p->tcf_lock);
 253         return TC_ACT_SHOT;
 254 }
 255
 256 static int tcf_nat_dump(struct sk_buff *skb, struct tc_action *a,
 257                         int bind, int ref)
 258 {
 259         unsigned char *b = skb_tail_pointer(skb);
 260         struct tcf_nat *p = to_tcf_nat(a);
 261         struct tc_nat opt = {
 262                 .index    = p->tcf_index,
 263                 .refcnt   = refcount_read(&p->tcf_refcnt) - ref,
 264                 .bindcnt  = atomic_read(&p->tcf_bindcnt) - bind,
 265         };
 266         struct tcf_t t;
 267
 268         spin_lock_bh(&p->tcf_lock);
 269         opt.old_addr = p->old_addr;
 270         opt.new_addr = p->new_addr;
 271         opt.mask = p->mask;
 272         opt.flags = p->flags;
 273         opt.action = p->tcf_action;
 274
 275         if (nla_put(skb, TCA_NAT_PARMS, sizeof(opt), &opt))
 276                 goto nla_put_failure;
 277
 278         tcf_tm_dump(&t, &p->tcf_tm);
 279         if (nla_put_64bit(skb, TCA_NAT_TM, sizeof(t), &t, TCA_NAT_PAD))
 280                 goto nla_put_failure;
 281         spin_unlock_bh(&p->tcf_lock);
 282
 283         return skb->len;
 284
 285 nla_put_failure:
 286         spin_unlock_bh(&p->tcf_lock);
 287         nlmsg_trim(skb, b);
 288         return -1;
 289 }
 290
 291 static struct tc_action_ops act_nat_ops = {
 292         .kind           =       "nat",
 293         .id             =       TCA_ID_NAT,
 294         .owner          =       THIS_MODULE,
 295         .act            =       tcf_nat_act,
 296         .dump           =       tcf_nat_dump,
 297         .init           =       tcf_nat_init,
 298         .size           =       sizeof(struct tcf_nat),
 299 };
 300
 301 static __net_init int nat_init_net(struct net *net)
 302 {
 303         struct tc_action_net *tn = net_generic(net, act_nat_ops.net_id);
 304
 305         return tc_action_net_init(net, tn, &act_nat_ops);
 306 }
 307
 308 static void __net_exit nat_exit_net(struct list_head *net_list)
 309 {
 310         tc_action_net_exit(net_list, act_nat_ops.net_id);
 311 }
 312
 313 static struct pernet_operations nat_net_ops = {
 314         .init = nat_init_net,
 315         .exit_batch = nat_exit_net,
 316         .id   = &act_nat_ops.net_id,
 317         .size = sizeof(struct tc_action_net),
 318 };
 319
 320 MODULE_DESCRIPTION("Stateless NAT actions");
 321 MODULE_LICENSE("GPL");
 322
 323 static int __init nat_init_module(void)
 324 {
 325         return tcf_register_action(&act_nat_ops, &nat_net_ops);
 326 }
 327
 328 static void __exit nat_cleanup_module(void)
 329 {
 330         tcf_unregister_action(&act_nat_ops, &nat_net_ops);
 331 }
 332
 333 module_init(nat_init_module);
 334 module_exit(nat_cleanup_module);