Merge branch 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
[linux-2.6/linux-mips.git] / net / netfilter / xt_TCPMSS.c
blobdb7e38c08de2fe369395fd6b1ed9d980c030ad69
1 /*
2 * This is a module which is used for setting the MSS option in TCP packets.
4 * Copyright (C) 2000 Marc Boucher <marc@mbsi.ca>
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 */
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/ipv6.h>
17 #include <net/tcp.h>
19 #include <linux/netfilter_ipv4/ip_tables.h>
20 #include <linux/netfilter_ipv6/ip6_tables.h>
21 #include <linux/netfilter/x_tables.h>
22 #include <linux/netfilter/xt_tcpudp.h>
23 #include <linux/netfilter/xt_TCPMSS.h>
25 MODULE_LICENSE("GPL");
26 MODULE_AUTHOR("Marc Boucher <marc@mbsi.ca>");
27 MODULE_DESCRIPTION("x_tables TCP MSS modification module");
28 MODULE_ALIAS("ipt_TCPMSS");
29 MODULE_ALIAS("ip6t_TCPMSS");
31 static inline unsigned int
32 optlen(const u_int8_t *opt, unsigned int offset)
34 /* Beware zero-length options: make finite progress */
35 if (opt[offset] <= TCPOPT_NOP || opt[offset+1] == 0)
36 return 1;
37 else
38 return opt[offset+1];
41 static int
42 tcpmss_mangle_packet(struct sk_buff **pskb,
43 const struct xt_tcpmss_info *info,
44 unsigned int tcphoff,
45 unsigned int minlen)
47 struct tcphdr *tcph;
48 unsigned int tcplen, i;
49 __be16 oldval;
50 u16 newmss;
51 u8 *opt;
53 if (!skb_make_writable(pskb, (*pskb)->len))
54 return -1;
56 tcplen = (*pskb)->len - tcphoff;
57 tcph = (struct tcphdr *)((*pskb)->nh.raw + tcphoff);
59 /* Since it passed flags test in tcp match, we know it is is
60 not a fragment, and has data >= tcp header length. SYN
61 packets should not contain data: if they did, then we risk
62 running over MTU, sending Frag Needed and breaking things
63 badly. --RR */
64 if (tcplen != tcph->doff*4) {
65 if (net_ratelimit())
66 printk(KERN_ERR "xt_TCPMSS: bad length (%u bytes)\n",
67 (*pskb)->len);
68 return -1;
71 if (info->mss == XT_TCPMSS_CLAMP_PMTU) {
72 if (dst_mtu((*pskb)->dst) <= minlen) {
73 if (net_ratelimit())
74 printk(KERN_ERR "xt_TCPMSS: "
75 "unknown or invalid path-MTU (%u)\n",
76 dst_mtu((*pskb)->dst));
77 return -1;
79 newmss = dst_mtu((*pskb)->dst) - minlen;
80 } else
81 newmss = info->mss;
83 opt = (u_int8_t *)tcph;
84 for (i = sizeof(struct tcphdr); i < tcph->doff*4; i += optlen(opt, i)) {
85 if (opt[i] == TCPOPT_MSS && tcph->doff*4 - i >= TCPOLEN_MSS &&
86 opt[i+1] == TCPOLEN_MSS) {
87 u_int16_t oldmss;
89 oldmss = (opt[i+2] << 8) | opt[i+3];
91 if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
92 oldmss <= newmss)
93 return 0;
95 opt[i+2] = (newmss & 0xff00) >> 8;
96 opt[i+3] = (newmss & 0x00ff);
98 nf_proto_csum_replace2(&tcph->check, *pskb,
99 htons(oldmss), htons(newmss), 0);
100 return 0;
105 * MSS Option not found ?! add it..
107 if (skb_tailroom((*pskb)) < TCPOLEN_MSS) {
108 struct sk_buff *newskb;
110 newskb = skb_copy_expand(*pskb, skb_headroom(*pskb),
111 TCPOLEN_MSS, GFP_ATOMIC);
112 if (!newskb)
113 return -1;
114 kfree_skb(*pskb);
115 *pskb = newskb;
116 tcph = (struct tcphdr *)((*pskb)->nh.raw + tcphoff);
119 skb_put((*pskb), TCPOLEN_MSS);
121 opt = (u_int8_t *)tcph + sizeof(struct tcphdr);
122 memmove(opt + TCPOLEN_MSS, opt, tcplen - sizeof(struct tcphdr));
124 nf_proto_csum_replace2(&tcph->check, *pskb,
125 htons(tcplen), htons(tcplen + TCPOLEN_MSS), 1);
126 opt[0] = TCPOPT_MSS;
127 opt[1] = TCPOLEN_MSS;
128 opt[2] = (newmss & 0xff00) >> 8;
129 opt[3] = (newmss & 0x00ff);
131 nf_proto_csum_replace4(&tcph->check, *pskb, 0, *((__be32 *)opt), 0);
133 oldval = ((__be16 *)tcph)[6];
134 tcph->doff += TCPOLEN_MSS/4;
135 nf_proto_csum_replace2(&tcph->check, *pskb,
136 oldval, ((__be16 *)tcph)[6], 0);
137 return TCPOLEN_MSS;
140 static unsigned int
141 xt_tcpmss_target4(struct sk_buff **pskb,
142 const struct net_device *in,
143 const struct net_device *out,
144 unsigned int hooknum,
145 const struct xt_target *target,
146 const void *targinfo)
148 struct iphdr *iph = (*pskb)->nh.iph;
149 __be16 newlen;
150 int ret;
152 ret = tcpmss_mangle_packet(pskb, targinfo, iph->ihl * 4,
153 sizeof(*iph) + sizeof(struct tcphdr));
154 if (ret < 0)
155 return NF_DROP;
156 if (ret > 0) {
157 iph = (*pskb)->nh.iph;
158 newlen = htons(ntohs(iph->tot_len) + ret);
159 nf_csum_replace2(&iph->check, iph->tot_len, newlen);
160 iph->tot_len = newlen;
162 return XT_CONTINUE;
165 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
166 static unsigned int
167 xt_tcpmss_target6(struct sk_buff **pskb,
168 const struct net_device *in,
169 const struct net_device *out,
170 unsigned int hooknum,
171 const struct xt_target *target,
172 const void *targinfo)
174 struct ipv6hdr *ipv6h = (*pskb)->nh.ipv6h;
175 u8 nexthdr;
176 int tcphoff;
177 int ret;
179 nexthdr = ipv6h->nexthdr;
180 tcphoff = ipv6_skip_exthdr(*pskb, sizeof(*ipv6h), &nexthdr);
181 if (tcphoff < 0) {
182 WARN_ON(1);
183 return NF_DROP;
185 ret = tcpmss_mangle_packet(pskb, targinfo, tcphoff,
186 sizeof(*ipv6h) + sizeof(struct tcphdr));
187 if (ret < 0)
188 return NF_DROP;
189 if (ret > 0) {
190 ipv6h = (*pskb)->nh.ipv6h;
191 ipv6h->payload_len = htons(ntohs(ipv6h->payload_len) + ret);
193 return XT_CONTINUE;
195 #endif
197 #define TH_SYN 0x02
199 /* Must specify -p tcp --syn */
200 static inline int find_syn_match(const struct xt_entry_match *m)
202 const struct xt_tcp *tcpinfo = (const struct xt_tcp *)m->data;
204 if (strcmp(m->u.kernel.match->name, "tcp") == 0 &&
205 tcpinfo->flg_cmp & TH_SYN &&
206 !(tcpinfo->invflags & XT_TCP_INV_FLAGS))
207 return 1;
209 return 0;
212 static int
213 xt_tcpmss_checkentry4(const char *tablename,
214 const void *entry,
215 const struct xt_target *target,
216 void *targinfo,
217 unsigned int hook_mask)
219 const struct xt_tcpmss_info *info = targinfo;
220 const struct ipt_entry *e = entry;
222 if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
223 (hook_mask & ~((1 << NF_IP_FORWARD) |
224 (1 << NF_IP_LOCAL_OUT) |
225 (1 << NF_IP_POST_ROUTING))) != 0) {
226 printk("xt_TCPMSS: path-MTU clamping only supported in "
227 "FORWARD, OUTPUT and POSTROUTING hooks\n");
228 return 0;
230 if (IPT_MATCH_ITERATE(e, find_syn_match))
231 return 1;
232 printk("xt_TCPMSS: Only works on TCP SYN packets\n");
233 return 0;
236 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
237 static int
238 xt_tcpmss_checkentry6(const char *tablename,
239 const void *entry,
240 const struct xt_target *target,
241 void *targinfo,
242 unsigned int hook_mask)
244 const struct xt_tcpmss_info *info = targinfo;
245 const struct ip6t_entry *e = entry;
247 if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
248 (hook_mask & ~((1 << NF_IP6_FORWARD) |
249 (1 << NF_IP6_LOCAL_OUT) |
250 (1 << NF_IP6_POST_ROUTING))) != 0) {
251 printk("xt_TCPMSS: path-MTU clamping only supported in "
252 "FORWARD, OUTPUT and POSTROUTING hooks\n");
253 return 0;
255 if (IP6T_MATCH_ITERATE(e, find_syn_match))
256 return 1;
257 printk("xt_TCPMSS: Only works on TCP SYN packets\n");
258 return 0;
260 #endif
262 static struct xt_target xt_tcpmss_reg[] = {
264 .family = AF_INET,
265 .name = "TCPMSS",
266 .checkentry = xt_tcpmss_checkentry4,
267 .target = xt_tcpmss_target4,
268 .targetsize = sizeof(struct xt_tcpmss_info),
269 .proto = IPPROTO_TCP,
270 .me = THIS_MODULE,
272 #if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
274 .family = AF_INET6,
275 .name = "TCPMSS",
276 .checkentry = xt_tcpmss_checkentry6,
277 .target = xt_tcpmss_target6,
278 .targetsize = sizeof(struct xt_tcpmss_info),
279 .proto = IPPROTO_TCP,
280 .me = THIS_MODULE,
282 #endif
285 static int __init xt_tcpmss_init(void)
287 return xt_register_targets(xt_tcpmss_reg, ARRAY_SIZE(xt_tcpmss_reg));
290 static void __exit xt_tcpmss_fini(void)
292 xt_unregister_targets(xt_tcpmss_reg, ARRAY_SIZE(xt_tcpmss_reg));
295 module_init(xt_tcpmss_init);
296 module_exit(xt_tcpmss_fini);