net/ipv4/netfilter/ipt_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
  14 #include <linux/ip.h>
  15 #include <net/tcp.h>
  16
  17 #include <linux/netfilter_ipv4/ip_tables.h>
  18 #include <linux/netfilter_ipv4/ipt_TCPMSS.h>
  19
  20 MODULE_LICENSE("GPL");
  21 MODULE_AUTHOR("Marc Boucher <marc@mbsi.ca>");
  22 MODULE_DESCRIPTION("iptables TCP MSS modification module");
  23
  24 #if 0
  25 #define DEBUGP printk
  26 #else
  27 #define DEBUGP(format, args...)
  28 #endif
  29
  30 static u_int16_t
  31 cheat_check(u_int32_t oldvalinv, u_int32_t newval, u_int16_t oldcheck)
  32 {
  33         u_int32_t diffs[] = { oldvalinv, newval };
  34         return csum_fold(csum_partial((char *)diffs, sizeof(diffs),
  35                                       oldcheck^0xFFFF));
  36 }
  37
  38 static inline unsigned int
  39 optlen(const u_int8_t *opt, unsigned int offset)
  40 {
  41         /* Beware zero-length options: make finite progress */
  42         if (opt[offset] <= TCPOPT_NOP || opt[offset+1] == 0) return 1;
  43         else return opt[offset+1];
  44 }
  45
  46 static unsigned int
  47 ipt_tcpmss_target(struct sk_buff **pskb,
  48                   const struct net_device *in,
  49                   const struct net_device *out,
  50                   unsigned int hooknum,
  51                   const struct xt_target *target,
  52                   const void *targinfo,
  53                   void *userinfo)
  54 {
  55         const struct ipt_tcpmss_info *tcpmssinfo = targinfo;
  56         struct tcphdr *tcph;
  57         struct iphdr *iph;
  58         u_int16_t tcplen, newtotlen, oldval, newmss;
  59         unsigned int i;
  60         u_int8_t *opt;
  61
  62         if (!skb_make_writable(pskb, (*pskb)->len))
  63                 return NF_DROP;
  64
  65         if ((*pskb)->ip_summed == CHECKSUM_HW &&
  66             skb_checksum_help(*pskb, out == NULL))
  67                 return NF_DROP;
  68
  69         iph = (*pskb)->nh.iph;
  70         tcplen = (*pskb)->len - iph->ihl*4;
  71
  72         tcph = (void *)iph + iph->ihl*4;
  73
  74         /* Since it passed flags test in tcp match, we know it is is
  75            not a fragment, and has data >= tcp header length.  SYN
  76            packets should not contain data: if they did, then we risk
  77            running over MTU, sending Frag Needed and breaking things
  78            badly. --RR */
  79         if (tcplen != tcph->doff*4) {
  80                 if (net_ratelimit())
  81                         printk(KERN_ERR
  82                                "ipt_tcpmss_target: bad length (%d bytes)\n",
  83                                (*pskb)->len);
  84                 return NF_DROP;
  85         }
  86
  87         if(tcpmssinfo->mss == IPT_TCPMSS_CLAMP_PMTU) {
  88                 if(!(*pskb)->dst) {
  89                         if (net_ratelimit())
  90                                 printk(KERN_ERR
  91                                         "ipt_tcpmss_target: no dst?! can't determine path-MTU\n");
  92                         return NF_DROP; /* or IPT_CONTINUE ?? */
  93                 }
  94
  95                 if(dst_mtu((*pskb)->dst) <= (sizeof(struct iphdr) + sizeof(struct tcphdr))) {
  96                         if (net_ratelimit())
  97                                 printk(KERN_ERR
  98                                         "ipt_tcpmss_target: unknown or invalid path-MTU (%d)\n", dst_mtu((*pskb)->dst));
  99                         return NF_DROP; /* or IPT_CONTINUE ?? */
 100                 }
 101
 102                 newmss = dst_mtu((*pskb)->dst) - sizeof(struct iphdr) - sizeof(struct tcphdr);
 103         } else
 104                 newmss = tcpmssinfo->mss;
 105
 106         opt = (u_int8_t *)tcph;
 107         for (i = sizeof(struct tcphdr); i < tcph->doff*4; i += optlen(opt, i)){
 108                 if ((opt[i] == TCPOPT_MSS) &&
 109                     ((tcph->doff*4 - i) >= TCPOLEN_MSS) &&
 110                     (opt[i+1] == TCPOLEN_MSS)) {
 111                         u_int16_t oldmss;
 112
 113                         oldmss = (opt[i+2] << 8) | opt[i+3];
 114
 115                         if((tcpmssinfo->mss == IPT_TCPMSS_CLAMP_PMTU) &&
 116                                 (oldmss <= newmss))
 117                                         return IPT_CONTINUE;
 118
 119                         opt[i+2] = (newmss & 0xff00) >> 8;
 120                         opt[i+3] = (newmss & 0x00ff);
 121
 122                         tcph->check = cheat_check(htons(oldmss)^0xFFFF,
 123                                                   htons(newmss),
 124                                                   tcph->check);
 125
 126                         DEBUGP(KERN_INFO "ipt_tcpmss_target: %u.%u.%u.%u:%hu"
 127                                "->%u.%u.%u.%u:%hu changed TCP MSS option"
 128                                " (from %u to %u)\n",
 129                                NIPQUAD((*pskb)->nh.iph->saddr),
 130                                ntohs(tcph->source),
 131                                NIPQUAD((*pskb)->nh.iph->daddr),
 132                                ntohs(tcph->dest),
 133                                oldmss, newmss);
 134                         goto retmodified;
 135                 }
 136         }
 137
 138         /*
 139          * MSS Option not found ?! add it..
 140          */
 141         if (skb_tailroom((*pskb)) < TCPOLEN_MSS) {
 142                 struct sk_buff *newskb;
 143
 144                 newskb = skb_copy_expand(*pskb, skb_headroom(*pskb),
 145                                          TCPOLEN_MSS, GFP_ATOMIC);
 146                 if (!newskb) {
 147                         if (net_ratelimit())
 148                                 printk(KERN_ERR "ipt_tcpmss_target:"
 149                                        " unable to allocate larger skb\n");
 150                         return NF_DROP;
 151                 }
 152
 153                 kfree_skb(*pskb);
 154                 *pskb = newskb;
 155                 iph = (*pskb)->nh.iph;
 156                 tcph = (void *)iph + iph->ihl*4;
 157         }
 158
 159         skb_put((*pskb), TCPOLEN_MSS);
 160
 161         opt = (u_int8_t *)tcph + sizeof(struct tcphdr);
 162         memmove(opt + TCPOLEN_MSS, opt, tcplen - sizeof(struct tcphdr));
 163
 164         tcph->check = cheat_check(htons(tcplen) ^ 0xFFFF,
 165                                   htons(tcplen + TCPOLEN_MSS), tcph->check);
 166         tcplen += TCPOLEN_MSS;
 167
 168         opt[0] = TCPOPT_MSS;
 169         opt[1] = TCPOLEN_MSS;
 170         opt[2] = (newmss & 0xff00) >> 8;
 171         opt[3] = (newmss & 0x00ff);
 172
 173         tcph->check = cheat_check(~0, *((u_int32_t *)opt), tcph->check);
 174
 175         oldval = ((u_int16_t *)tcph)[6];
 176         tcph->doff += TCPOLEN_MSS/4;
 177         tcph->check = cheat_check(oldval ^ 0xFFFF,
 178                                   ((u_int16_t *)tcph)[6], tcph->check);
 179
 180         newtotlen = htons(ntohs(iph->tot_len) + TCPOLEN_MSS);
 181         iph->check = cheat_check(iph->tot_len ^ 0xFFFF,
 182                                  newtotlen, iph->check);
 183         iph->tot_len = newtotlen;
 184
 185         DEBUGP(KERN_INFO "ipt_tcpmss_target: %u.%u.%u.%u:%hu"
 186                "->%u.%u.%u.%u:%hu added TCP MSS option (%u)\n",
 187                NIPQUAD((*pskb)->nh.iph->saddr),
 188                ntohs(tcph->source),
 189                NIPQUAD((*pskb)->nh.iph->daddr),
 190                ntohs(tcph->dest),
 191                newmss);
 192
 193  retmodified:
 194         return IPT_CONTINUE;
 195 }
 196
 197 #define TH_SYN 0x02
 198
 199 static inline int find_syn_match(const struct ipt_entry_match *m)
 200 {
 201         const struct ipt_tcp *tcpinfo = (const struct ipt_tcp *)m->data;
 202
 203         if (strcmp(m->u.kernel.match->name, "tcp") == 0
 204             && (tcpinfo->flg_cmp & TH_SYN)
 205             && !(tcpinfo->invflags & IPT_TCP_INV_FLAGS))
 206                 return 1;
 207
 208         return 0;
 209 }
 210
 211 /* Must specify -p tcp --syn/--tcp-flags SYN */
 212 static int
 213 ipt_tcpmss_checkentry(const char *tablename,
 214                       const void *e_void,
 215                       const struct xt_target *target,
 216                       void *targinfo,
 217                       unsigned int targinfosize,
 218                       unsigned int hook_mask)
 219 {
 220         const struct ipt_tcpmss_info *tcpmssinfo = targinfo;
 221         const struct ipt_entry *e = e_void;
 222
 223         if((tcpmssinfo->mss == IPT_TCPMSS_CLAMP_PMTU) &&
 224                         ((hook_mask & ~((1 << NF_IP_FORWARD)
 225                                 | (1 << NF_IP_LOCAL_OUT)
 226                                 | (1 << NF_IP_POST_ROUTING))) != 0)) {
 227                 printk("TCPMSS: path-MTU clamping only supported in FORWARD, OUTPUT and POSTROUTING hooks\n");
 228                 return 0;
 229         }
 230
 231         if (IPT_MATCH_ITERATE(e, find_syn_match))
 232                 return 1;
 233         printk("TCPMSS: Only works on TCP SYN packets\n");
 234         return 0;
 235 }
 236
 237 static struct ipt_target ipt_tcpmss_reg = {
 238         .name           = "TCPMSS",
 239         .target         = ipt_tcpmss_target,
 240         .targetsize     = sizeof(struct ipt_tcpmss_info),
 241         .proto          = IPPROTO_TCP,
 242         .checkentry     = ipt_tcpmss_checkentry,
 243         .me             = THIS_MODULE,
 244 };
 245
 246 static int __init init(void)
 247 {
 248         return ipt_register_target(&ipt_tcpmss_reg);
 249 }
 250
 251 static void __exit fini(void)
 252 {
 253         ipt_unregister_target(&ipt_tcpmss_reg);
 254 }
 255
 256 module_init(init);
 257 module_exit(fini);