net/netfilter/xt_TCPMSS.c

   1 /*
   2  * This is a module which is used for setting the MSS option in TCP packets.
   3  *
   4  * Copyright (C) 2000 Marc Boucher <marc@mbsi.ca>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/skbuff.h>
  13 #include <linux/ip.h>
  14 #include <linux/ipv6.h>
  15 #include <linux/tcp.h>
  16 #include <net/dst.h>
  17 #include <net/flow.h>
  18 #include <net/ipv6.h>
  19 #include <net/route.h>
  20 #include <net/tcp.h>
  21
  22 #include <linux/netfilter_ipv4/ip_tables.h>
  23 #include <linux/netfilter_ipv6/ip6_tables.h>
  24 #include <linux/netfilter/x_tables.h>
  25 #include <linux/netfilter/xt_tcpudp.h>
  26 #include <linux/netfilter/xt_TCPMSS.h>
  27
  28 MODULE_LICENSE("GPL");
  29 MODULE_AUTHOR("Marc Boucher <marc@mbsi.ca>");
  30 MODULE_DESCRIPTION("Xtables: TCP Maximum Segment Size (MSS) adjustment");
  31 MODULE_ALIAS("ipt_TCPMSS");
  32 MODULE_ALIAS("ip6t_TCPMSS");
  33
  34 static inline unsigned int
  35 optlen(const u_int8_t *opt, unsigned int offset)
  36 {
  37         /* Beware zero-length options: make finite progress */
  38         if (opt[offset] <= TCPOPT_NOP || opt[offset+1] == 0)
  39                 return 1;
  40         else
  41                 return opt[offset+1];
  42 }
  43
  44 static int
  45 tcpmss_mangle_packet(struct sk_buff *skb,
  46                      const struct xt_tcpmss_info *info,
  47                      unsigned int in_mtu,
  48                      unsigned int tcphoff,
  49                      unsigned int minlen)
  50 {
  51         struct tcphdr *tcph;
  52         unsigned int tcplen, i;
  53         __be16 oldval;
  54         u16 newmss;
  55         u8 *opt;
  56
  57         if (!skb_make_writable(skb, skb->len))
  58                 return -1;
  59
  60         tcplen = skb->len - tcphoff;
  61         tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
  62
  63         /* Since it passed flags test in tcp match, we know it is is
  64            not a fragment, and has data >= tcp header length.  SYN
  65            packets should not contain data: if they did, then we risk
  66            running over MTU, sending Frag Needed and breaking things
  67            badly. --RR */
  68         if (tcplen != tcph->doff*4) {
  69                 if (net_ratelimit())
  70                         printk(KERN_ERR "xt_TCPMSS: bad length (%u bytes)\n",
  71                                skb->len);
  72                 return -1;
  73         }
  74
  75         if (info->mss == XT_TCPMSS_CLAMP_PMTU) {
  76                 if (dst_mtu(skb->dst) <= minlen) {
  77                         if (net_ratelimit())
  78                                 printk(KERN_ERR "xt_TCPMSS: "
  79                                        "unknown or invalid path-MTU (%u)\n",
  80                                        dst_mtu(skb->dst));
  81                         return -1;
  82                 }
  83                 if (in_mtu <= minlen) {
  84                         if (net_ratelimit())
  85                                 printk(KERN_ERR "xt_TCPMSS: unknown or "
  86                                        "invalid path-MTU (%u)\n", in_mtu);
  87                         return -1;
  88                 }
  89                 newmss = min(dst_mtu(skb->dst), in_mtu) - minlen;
  90         } else
  91                 newmss = info->mss;
  92
  93         opt = (u_int8_t *)tcph;
  94         for (i = sizeof(struct tcphdr); i < tcph->doff*4; i += optlen(opt, i)) {
  95                 if (opt[i] == TCPOPT_MSS && tcph->doff*4 - i >= TCPOLEN_MSS &&
  96                     opt[i+1] == TCPOLEN_MSS) {
  97                         u_int16_t oldmss;
  98
  99                         oldmss = (opt[i+2] << 8) | opt[i+3];
 100
 101                         /* Never increase MSS, even when setting it, as
 102                          * doing so results in problems for hosts that rely
 103                          * on MSS being set correctly.
 104                          */
 105                         if (oldmss <= newmss)
 106                                 return 0;
 107
 108                         opt[i+2] = (newmss & 0xff00) >> 8;
 109                         opt[i+3] = newmss & 0x00ff;
 110
 111                         inet_proto_csum_replace2(&tcph->check, skb,
 112                                                  htons(oldmss), htons(newmss),
 113                                                  0);
 114                         return 0;
 115                 }
 116         }
 117
 118         /*
 119          * MSS Option not found ?! add it..
 120          */
 121         if (skb_tailroom(skb) < TCPOLEN_MSS) {
 122                 if (pskb_expand_head(skb, 0,
 123                                      TCPOLEN_MSS - skb_tailroom(skb),
 124                                      GFP_ATOMIC))
 125                         return -1;
 126                 tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
 127         }
 128
 129         skb_put(skb, TCPOLEN_MSS);
 130
 131         opt = (u_int8_t *)tcph + sizeof(struct tcphdr);
 132         memmove(opt + TCPOLEN_MSS, opt, tcplen - sizeof(struct tcphdr));
 133
 134         inet_proto_csum_replace2(&tcph->check, skb,
 135                                  htons(tcplen), htons(tcplen + TCPOLEN_MSS), 1);
 136         opt[0] = TCPOPT_MSS;
 137         opt[1] = TCPOLEN_MSS;
 138         opt[2] = (newmss & 0xff00) >> 8;
 139         opt[3] = newmss & 0x00ff;
 140
 141         inet_proto_csum_replace4(&tcph->check, skb, 0, *((__be32 *)opt), 0);
 142
 143         oldval = ((__be16 *)tcph)[6];
 144         tcph->doff += TCPOLEN_MSS/4;
 145         inet_proto_csum_replace2(&tcph->check, skb,
 146                                  oldval, ((__be16 *)tcph)[6], 0);
 147         return TCPOLEN_MSS;
 148 }
 149
 150 static u_int32_t tcpmss_reverse_mtu4(const struct iphdr *iph)
 151 {
 152         struct flowi fl = {
 153                 .fl4_dst = iph->saddr,
 154         };
 155         const struct nf_afinfo *ai;
 156         struct rtable *rt = NULL;
 157         u_int32_t mtu     = ~0U;
 158
 159         rcu_read_lock();
 160         ai = nf_get_afinfo(AF_INET);
 161         if (ai != NULL)
 162                 ai->route((struct dst_entry **)&rt, &fl);
 163         rcu_read_unlock();
 164
 165         if (rt != NULL) {
 166                 mtu = dst_mtu(&rt->u.dst);
 167                 dst_release(&rt->u.dst);
 168         }
 169         return mtu;
 170 }
 171
 172 static unsigned int
 173 tcpmss_tg4(struct sk_buff *skb, const struct net_device *in,
 174            const struct net_device *out, unsigned int hooknum,
 175            const struct xt_target *target, const void *targinfo)
 176 {
 177         struct iphdr *iph = ip_hdr(skb);
 178         __be16 newlen;
 179         int ret;
 180
 181         ret = tcpmss_mangle_packet(skb, targinfo, tcpmss_reverse_mtu4(iph),
 182                                    iph->ihl * 4,
 183                                    sizeof(*iph) + sizeof(struct tcphdr));
 184         if (ret < 0)
 185                 return NF_DROP;
 186         if (ret > 0) {
 187                 iph = ip_hdr(skb);
 188                 newlen = htons(ntohs(iph->tot_len) + ret);
 189                 csum_replace2(&iph->check, iph->tot_len, newlen);
 190                 iph->tot_len = newlen;
 191         }
 192         return XT_CONTINUE;
 193 }
 194
 195 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
 196 static u_int32_t tcpmss_reverse_mtu6(const struct ipv6hdr *iph)
 197 {
 198         struct flowi fl = {
 199                 .fl6_dst = iph->saddr,
 200         };
 201         const struct nf_afinfo *ai;
 202         struct rtable *rt = NULL;
 203         u_int32_t mtu     = ~0U;
 204
 205         rcu_read_lock();
 206         ai = nf_get_afinfo(AF_INET6);
 207         if (ai != NULL)
 208                 ai->route((struct dst_entry **)&rt, &fl);
 209         rcu_read_unlock();
 210
 211         if (rt != NULL) {
 212                 mtu = dst_mtu(&rt->u.dst);
 213                 dst_release(&rt->u.dst);
 214         }
 215         return mtu;
 216 }
 217
 218 static unsigned int
 219 tcpmss_tg6(struct sk_buff *skb, const struct net_device *in,
 220            const struct net_device *out, unsigned int hooknum,
 221            const struct xt_target *target, const void *targinfo)
 222 {
 223         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 224         u8 nexthdr;
 225         int tcphoff;
 226         int ret;
 227
 228         nexthdr = ipv6h->nexthdr;
 229         tcphoff = ipv6_skip_exthdr(skb, sizeof(*ipv6h), &nexthdr);
 230         if (tcphoff < 0)
 231                 return NF_DROP;
 232         ret = tcpmss_mangle_packet(skb, targinfo, tcpmss_reverse_mtu6(ipv6h),
 233                                    tcphoff,
 234                                    sizeof(*ipv6h) + sizeof(struct tcphdr));
 235         if (ret < 0)
 236                 return NF_DROP;
 237         if (ret > 0) {
 238                 ipv6h = ipv6_hdr(skb);
 239                 ipv6h->payload_len = htons(ntohs(ipv6h->payload_len) + ret);
 240         }
 241         return XT_CONTINUE;
 242 }
 243 #endif
 244
 245 #define TH_SYN 0x02
 246
 247 /* Must specify -p tcp --syn */
 248 static inline bool find_syn_match(const struct xt_entry_match *m)
 249 {
 250         const struct xt_tcp *tcpinfo = (const struct xt_tcp *)m->data;
 251
 252         if (strcmp(m->u.kernel.match->name, "tcp") == 0 &&
 253             tcpinfo->flg_cmp & TH_SYN &&
 254             !(tcpinfo->invflags & XT_TCP_INV_FLAGS))
 255                 return true;
 256
 257         return false;
 258 }
 259
 260 static bool
 261 tcpmss_tg4_check(const char *tablename, const void *entry,
 262                  const struct xt_target *target, void *targinfo,
 263                  unsigned int hook_mask)
 264 {
 265         const struct xt_tcpmss_info *info = targinfo;
 266         const struct ipt_entry *e = entry;
 267
 268         if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
 269             (hook_mask & ~((1 << NF_INET_FORWARD) |
 270                            (1 << NF_INET_LOCAL_OUT) |
 271                            (1 << NF_INET_POST_ROUTING))) != 0) {
 272                 printk("xt_TCPMSS: path-MTU clamping only supported in "
 273                        "FORWARD, OUTPUT and POSTROUTING hooks\n");
 274                 return false;
 275         }
 276         if (IPT_MATCH_ITERATE(e, find_syn_match))
 277                 return true;
 278         printk("xt_TCPMSS: Only works on TCP SYN packets\n");
 279         return false;
 280 }
 281
 282 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
 283 static bool
 284 tcpmss_tg6_check(const char *tablename, const void *entry,
 285                  const struct xt_target *target, void *targinfo,
 286                  unsigned int hook_mask)
 287 {
 288         const struct xt_tcpmss_info *info = targinfo;
 289         const struct ip6t_entry *e = entry;
 290
 291         if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
 292             (hook_mask & ~((1 << NF_INET_FORWARD) |
 293                            (1 << NF_INET_LOCAL_OUT) |
 294                            (1 << NF_INET_POST_ROUTING))) != 0) {
 295                 printk("xt_TCPMSS: path-MTU clamping only supported in "
 296                        "FORWARD, OUTPUT and POSTROUTING hooks\n");
 297                 return false;
 298         }
 299         if (IP6T_MATCH_ITERATE(e, find_syn_match))
 300                 return true;
 301         printk("xt_TCPMSS: Only works on TCP SYN packets\n");
 302         return false;
 303 }
 304 #endif
 305
 306 static struct xt_target tcpmss_tg_reg[] __read_mostly = {
 307         {
 308                 .family         = AF_INET,
 309                 .name           = "TCPMSS",
 310                 .checkentry     = tcpmss_tg4_check,
 311                 .target         = tcpmss_tg4,
 312                 .targetsize     = sizeof(struct xt_tcpmss_info),
 313                 .proto          = IPPROTO_TCP,
 314                 .me             = THIS_MODULE,
 315         },
 316 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
 317         {
 318                 .family         = AF_INET6,
 319                 .name           = "TCPMSS",
 320                 .checkentry     = tcpmss_tg6_check,
 321                 .target         = tcpmss_tg6,
 322                 .targetsize     = sizeof(struct xt_tcpmss_info),
 323                 .proto          = IPPROTO_TCP,
 324                 .me             = THIS_MODULE,
 325         },
 326 #endif
 327 };
 328
 329 static int __init tcpmss_tg_init(void)
 330 {
 331         return xt_register_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
 332 }
 333
 334 static void __exit tcpmss_tg_exit(void)
 335 {
 336         xt_unregister_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
 337 }
 338
 339 module_init(tcpmss_tg_init);
 340 module_exit(tcpmss_tg_exit);