]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/netfilter/xt_TCPMSS.c
netfilter: xtables: move extension arguments into compound structure (4/6)
[net-next-2.6.git] / net / netfilter / xt_TCPMSS.c
CommitLineData
cdd289a2
PM
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>
37c08387
JE
16#include <net/dst.h>
17#include <net/flow.h>
cdd289a2 18#include <net/ipv6.h>
37c08387 19#include <net/route.h>
cdd289a2
PM
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
28MODULE_LICENSE("GPL");
29MODULE_AUTHOR("Marc Boucher <marc@mbsi.ca>");
2ae15b64 30MODULE_DESCRIPTION("Xtables: TCP Maximum Segment Size (MSS) adjustment");
cdd289a2
PM
31MODULE_ALIAS("ipt_TCPMSS");
32MODULE_ALIAS("ip6t_TCPMSS");
33
34static inline unsigned int
35optlen(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
44static int
3db05fea 45tcpmss_mangle_packet(struct sk_buff *skb,
cdd289a2 46 const struct xt_tcpmss_info *info,
37c08387 47 unsigned int in_mtu,
cdd289a2
PM
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
3db05fea 57 if (!skb_make_writable(skb, skb->len))
cdd289a2
PM
58 return -1;
59
3db05fea
HX
60 tcplen = skb->len - tcphoff;
61 tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
cdd289a2
PM
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",
3db05fea 71 skb->len);
cdd289a2
PM
72 return -1;
73 }
74
75 if (info->mss == XT_TCPMSS_CLAMP_PMTU) {
3db05fea 76 if (dst_mtu(skb->dst) <= minlen) {
cdd289a2
PM
77 if (net_ratelimit())
78 printk(KERN_ERR "xt_TCPMSS: "
79 "unknown or invalid path-MTU (%u)\n",
3db05fea 80 dst_mtu(skb->dst));
cdd289a2
PM
81 return -1;
82 }
37c08387
JE
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;
cdd289a2
PM
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
17008064
BL
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)
cdd289a2
PM
106 return 0;
107
108 opt[i+2] = (newmss & 0xff00) >> 8;
7c4e36bc 109 opt[i+3] = newmss & 0x00ff;
cdd289a2 110
be0ea7d5
PM
111 inet_proto_csum_replace2(&tcph->check, skb,
112 htons(oldmss), htons(newmss),
113 0);
cdd289a2
PM
114 return 0;
115 }
116 }
117
118 /*
119 * MSS Option not found ?! add it..
120 */
3db05fea
HX
121 if (skb_tailroom(skb) < TCPOLEN_MSS) {
122 if (pskb_expand_head(skb, 0,
123 TCPOLEN_MSS - skb_tailroom(skb),
2ca7b0ac 124 GFP_ATOMIC))
cdd289a2 125 return -1;
3db05fea 126 tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
cdd289a2
PM
127 }
128
3db05fea 129 skb_put(skb, TCPOLEN_MSS);
cdd289a2
PM
130
131 opt = (u_int8_t *)tcph + sizeof(struct tcphdr);
132 memmove(opt + TCPOLEN_MSS, opt, tcplen - sizeof(struct tcphdr));
133
be0ea7d5
PM
134 inet_proto_csum_replace2(&tcph->check, skb,
135 htons(tcplen), htons(tcplen + TCPOLEN_MSS), 1);
cdd289a2
PM
136 opt[0] = TCPOPT_MSS;
137 opt[1] = TCPOLEN_MSS;
138 opt[2] = (newmss & 0xff00) >> 8;
7c4e36bc 139 opt[3] = newmss & 0x00ff;
cdd289a2 140
be0ea7d5 141 inet_proto_csum_replace4(&tcph->check, skb, 0, *((__be32 *)opt), 0);
cdd289a2
PM
142
143 oldval = ((__be16 *)tcph)[6];
144 tcph->doff += TCPOLEN_MSS/4;
be0ea7d5
PM
145 inet_proto_csum_replace2(&tcph->check, skb,
146 oldval, ((__be16 *)tcph)[6], 0);
cdd289a2
PM
147 return TCPOLEN_MSS;
148}
149
db1a75bd
JE
150static u_int32_t tcpmss_reverse_mtu(const struct sk_buff *skb,
151 unsigned int family)
37c08387 152{
db1a75bd 153 struct flowi fl = {};
37c08387
JE
154 const struct nf_afinfo *ai;
155 struct rtable *rt = NULL;
156 u_int32_t mtu = ~0U;
157
db1a75bd
JE
158 if (family == PF_INET)
159 fl.fl4_dst = ip_hdr(skb)->saddr;
160 else
161 fl.fl6_dst = ipv6_hdr(skb)->saddr;
162
37c08387 163 rcu_read_lock();
db1a75bd 164 ai = nf_get_afinfo(family);
37c08387
JE
165 if (ai != NULL)
166 ai->route((struct dst_entry **)&rt, &fl);
167 rcu_read_unlock();
168
169 if (rt != NULL) {
170 mtu = dst_mtu(&rt->u.dst);
171 dst_release(&rt->u.dst);
172 }
173 return mtu;
174}
175
cdd289a2 176static unsigned int
7eb35586 177tcpmss_tg4(struct sk_buff *skb, const struct xt_target_param *par)
cdd289a2 178{
3db05fea 179 struct iphdr *iph = ip_hdr(skb);
cdd289a2
PM
180 __be16 newlen;
181 int ret;
182
7eb35586 183 ret = tcpmss_mangle_packet(skb, par->targinfo,
db1a75bd 184 tcpmss_reverse_mtu(skb, PF_INET),
37c08387 185 iph->ihl * 4,
cdd289a2
PM
186 sizeof(*iph) + sizeof(struct tcphdr));
187 if (ret < 0)
188 return NF_DROP;
189 if (ret > 0) {
3db05fea 190 iph = ip_hdr(skb);
cdd289a2 191 newlen = htons(ntohs(iph->tot_len) + ret);
be0ea7d5 192 csum_replace2(&iph->check, iph->tot_len, newlen);
cdd289a2
PM
193 iph->tot_len = newlen;
194 }
195 return XT_CONTINUE;
196}
197
198#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
199static unsigned int
7eb35586 200tcpmss_tg6(struct sk_buff *skb, const struct xt_target_param *par)
cdd289a2 201{
3db05fea 202 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
cdd289a2
PM
203 u8 nexthdr;
204 int tcphoff;
205 int ret;
206
207 nexthdr = ipv6h->nexthdr;
3db05fea 208 tcphoff = ipv6_skip_exthdr(skb, sizeof(*ipv6h), &nexthdr);
9dc0564e 209 if (tcphoff < 0)
cdd289a2 210 return NF_DROP;
7eb35586 211 ret = tcpmss_mangle_packet(skb, par->targinfo,
db1a75bd 212 tcpmss_reverse_mtu(skb, PF_INET6),
37c08387 213 tcphoff,
cdd289a2
PM
214 sizeof(*ipv6h) + sizeof(struct tcphdr));
215 if (ret < 0)
216 return NF_DROP;
217 if (ret > 0) {
3db05fea 218 ipv6h = ipv6_hdr(skb);
cdd289a2
PM
219 ipv6h->payload_len = htons(ntohs(ipv6h->payload_len) + ret);
220 }
221 return XT_CONTINUE;
222}
223#endif
224
225#define TH_SYN 0x02
226
227/* Must specify -p tcp --syn */
e1931b78 228static inline bool find_syn_match(const struct xt_entry_match *m)
cdd289a2
PM
229{
230 const struct xt_tcp *tcpinfo = (const struct xt_tcp *)m->data;
231
232 if (strcmp(m->u.kernel.match->name, "tcp") == 0 &&
233 tcpinfo->flg_cmp & TH_SYN &&
234 !(tcpinfo->invflags & XT_TCP_INV_FLAGS))
e1931b78 235 return true;
cdd289a2 236
e1931b78 237 return false;
cdd289a2
PM
238}
239
e1931b78 240static bool
d3c5ee6d
JE
241tcpmss_tg4_check(const char *tablename, const void *entry,
242 const struct xt_target *target, void *targinfo,
243 unsigned int hook_mask)
cdd289a2
PM
244{
245 const struct xt_tcpmss_info *info = targinfo;
246 const struct ipt_entry *e = entry;
247
248 if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
6e23ae2a
PM
249 (hook_mask & ~((1 << NF_INET_FORWARD) |
250 (1 << NF_INET_LOCAL_OUT) |
251 (1 << NF_INET_POST_ROUTING))) != 0) {
cdd289a2
PM
252 printk("xt_TCPMSS: path-MTU clamping only supported in "
253 "FORWARD, OUTPUT and POSTROUTING hooks\n");
e1931b78 254 return false;
cdd289a2
PM
255 }
256 if (IPT_MATCH_ITERATE(e, find_syn_match))
e1931b78 257 return true;
cdd289a2 258 printk("xt_TCPMSS: Only works on TCP SYN packets\n");
e1931b78 259 return false;
cdd289a2
PM
260}
261
262#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
e1931b78 263static bool
d3c5ee6d
JE
264tcpmss_tg6_check(const char *tablename, const void *entry,
265 const struct xt_target *target, void *targinfo,
266 unsigned int hook_mask)
cdd289a2
PM
267{
268 const struct xt_tcpmss_info *info = targinfo;
269 const struct ip6t_entry *e = entry;
270
271 if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
6e23ae2a
PM
272 (hook_mask & ~((1 << NF_INET_FORWARD) |
273 (1 << NF_INET_LOCAL_OUT) |
274 (1 << NF_INET_POST_ROUTING))) != 0) {
cdd289a2
PM
275 printk("xt_TCPMSS: path-MTU clamping only supported in "
276 "FORWARD, OUTPUT and POSTROUTING hooks\n");
e1931b78 277 return false;
cdd289a2
PM
278 }
279 if (IP6T_MATCH_ITERATE(e, find_syn_match))
e1931b78 280 return true;
cdd289a2 281 printk("xt_TCPMSS: Only works on TCP SYN packets\n");
e1931b78 282 return false;
cdd289a2
PM
283}
284#endif
285
d3c5ee6d 286static struct xt_target tcpmss_tg_reg[] __read_mostly = {
cdd289a2 287 {
ee999d8b 288 .family = NFPROTO_IPV4,
cdd289a2 289 .name = "TCPMSS",
d3c5ee6d
JE
290 .checkentry = tcpmss_tg4_check,
291 .target = tcpmss_tg4,
cdd289a2
PM
292 .targetsize = sizeof(struct xt_tcpmss_info),
293 .proto = IPPROTO_TCP,
294 .me = THIS_MODULE,
295 },
296#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
297 {
ee999d8b 298 .family = NFPROTO_IPV6,
cdd289a2 299 .name = "TCPMSS",
d3c5ee6d
JE
300 .checkentry = tcpmss_tg6_check,
301 .target = tcpmss_tg6,
cdd289a2
PM
302 .targetsize = sizeof(struct xt_tcpmss_info),
303 .proto = IPPROTO_TCP,
304 .me = THIS_MODULE,
305 },
306#endif
307};
308
d3c5ee6d 309static int __init tcpmss_tg_init(void)
cdd289a2 310{
d3c5ee6d 311 return xt_register_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
cdd289a2
PM
312}
313
d3c5ee6d 314static void __exit tcpmss_tg_exit(void)
cdd289a2 315{
d3c5ee6d 316 xt_unregister_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
cdd289a2
PM
317}
318
d3c5ee6d
JE
319module_init(tcpmss_tg_init);
320module_exit(tcpmss_tg_exit);