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>
5 * Copyright (C) 2007 Patrick McHardy <kaber@trash.net>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/module.h>
13 #include <linux/skbuff.h>
15 #include <linux/gfp.h>
16 #include <linux/ipv6.h>
17 #include <linux/tcp.h>
21 #include <net/route.h>
24 #include <linux/netfilter_ipv4/ip_tables.h>
25 #include <linux/netfilter_ipv6/ip6_tables.h>
26 #include <linux/netfilter/x_tables.h>
27 #include <linux/netfilter/xt_tcpudp.h>
28 #include <linux/netfilter/xt_TCPMSS.h>
30 MODULE_LICENSE("GPL");
31 MODULE_AUTHOR("Marc Boucher <marc@mbsi.ca>");
32 MODULE_DESCRIPTION("Xtables: TCP Maximum Segment Size (MSS) adjustment");
33 MODULE_ALIAS("ipt_TCPMSS");
34 MODULE_ALIAS("ip6t_TCPMSS");
36 static inline unsigned int
37 optlen(const u_int8_t
*opt
, unsigned int offset
)
39 /* Beware zero-length options: make finite progress */
40 if (opt
[offset
] <= TCPOPT_NOP
|| opt
[offset
+1] == 0)
47 tcpmss_mangle_packet(struct sk_buff
*skb
,
48 const struct xt_action_param
*par
,
53 const struct xt_tcpmss_info
*info
= par
->targinfo
;
55 unsigned int tcplen
, i
;
60 /* This is a fragment, no TCP header is available */
61 if (par
->fragoff
!= 0)
64 if (!skb_make_writable(skb
, skb
->len
))
67 tcplen
= skb
->len
- tcphoff
;
68 tcph
= (struct tcphdr
*)(skb_network_header(skb
) + tcphoff
);
70 /* Header cannot be larger than the packet */
71 if (tcplen
< tcph
->doff
*4 || tcph
->doff
*4 < sizeof(struct tcphdr
))
74 if (info
->mss
== XT_TCPMSS_CLAMP_PMTU
) {
75 if (dst_mtu(skb_dst(skb
)) <= minlen
) {
76 net_err_ratelimited("unknown or invalid path-MTU (%u)\n",
77 dst_mtu(skb_dst(skb
)));
80 if (in_mtu
<= minlen
) {
81 net_err_ratelimited("unknown or invalid path-MTU (%u)\n",
85 newmss
= min(dst_mtu(skb_dst(skb
)), in_mtu
) - minlen
;
89 opt
= (u_int8_t
*)tcph
;
90 for (i
= sizeof(struct tcphdr
); i
< tcph
->doff
*4; i
+= optlen(opt
, i
)) {
91 if (opt
[i
] == TCPOPT_MSS
&& tcph
->doff
*4 - i
>= TCPOLEN_MSS
&&
92 opt
[i
+1] == TCPOLEN_MSS
) {
95 oldmss
= (opt
[i
+2] << 8) | opt
[i
+3];
97 /* Never increase MSS, even when setting it, as
98 * doing so results in problems for hosts that rely
99 * on MSS being set correctly.
101 if (oldmss
<= newmss
)
104 opt
[i
+2] = (newmss
& 0xff00) >> 8;
105 opt
[i
+3] = newmss
& 0x00ff;
107 inet_proto_csum_replace2(&tcph
->check
, skb
,
108 htons(oldmss
), htons(newmss
),
114 /* There is data after the header so the option can't be added
115 without moving it, and doing so may make the SYN packet
116 itself too large. Accept the packet unmodified instead. */
117 if (tcplen
> tcph
->doff
*4)
120 /* tcph->doff has 4 bits, do not wrap it to 0 */
121 if (tcph
->doff
>= 15)
125 * MSS Option not found ?! add it..
127 if (skb_tailroom(skb
) < TCPOLEN_MSS
) {
128 if (pskb_expand_head(skb
, 0,
129 TCPOLEN_MSS
- skb_tailroom(skb
),
132 tcph
= (struct tcphdr
*)(skb_network_header(skb
) + tcphoff
);
135 skb_put(skb
, TCPOLEN_MSS
);
138 * IPv4: RFC 1122 states "If an MSS option is not received at
139 * connection setup, TCP MUST assume a default send MSS of 536".
140 * IPv6: RFC 2460 states IPv6 has a minimum MTU of 1280 and a minimum
141 * length IPv6 header of 60, ergo the default MSS value is 1220
142 * Since no MSS was provided, we must use the default values
144 if (par
->family
== NFPROTO_IPV4
)
145 newmss
= min(newmss
, (u16
)536);
147 newmss
= min(newmss
, (u16
)1220);
149 opt
= (u_int8_t
*)tcph
+ sizeof(struct tcphdr
);
150 memmove(opt
+ TCPOLEN_MSS
, opt
, tcplen
- sizeof(struct tcphdr
));
152 inet_proto_csum_replace2(&tcph
->check
, skb
,
153 htons(tcplen
), htons(tcplen
+ TCPOLEN_MSS
), 1);
155 opt
[1] = TCPOLEN_MSS
;
156 opt
[2] = (newmss
& 0xff00) >> 8;
157 opt
[3] = newmss
& 0x00ff;
159 inet_proto_csum_replace4(&tcph
->check
, skb
, 0, *((__be32
*)opt
), 0);
161 oldval
= ((__be16
*)tcph
)[6];
162 tcph
->doff
+= TCPOLEN_MSS
/4;
163 inet_proto_csum_replace2(&tcph
->check
, skb
,
164 oldval
, ((__be16
*)tcph
)[6], 0);
168 static u_int32_t
tcpmss_reverse_mtu(const struct sk_buff
*skb
,
172 const struct nf_afinfo
*ai
;
173 struct rtable
*rt
= NULL
;
176 if (family
== PF_INET
) {
177 struct flowi4
*fl4
= &fl
.u
.ip4
;
178 memset(fl4
, 0, sizeof(*fl4
));
179 fl4
->daddr
= ip_hdr(skb
)->saddr
;
181 struct flowi6
*fl6
= &fl
.u
.ip6
;
183 memset(fl6
, 0, sizeof(*fl6
));
184 fl6
->daddr
= ipv6_hdr(skb
)->saddr
;
187 ai
= nf_get_afinfo(family
);
189 ai
->route(&init_net
, (struct dst_entry
**)&rt
, &fl
, false);
193 mtu
= dst_mtu(&rt
->dst
);
194 dst_release(&rt
->dst
);
200 tcpmss_tg4(struct sk_buff
*skb
, const struct xt_action_param
*par
)
202 struct iphdr
*iph
= ip_hdr(skb
);
206 ret
= tcpmss_mangle_packet(skb
, par
,
207 tcpmss_reverse_mtu(skb
, PF_INET
),
209 sizeof(*iph
) + sizeof(struct tcphdr
));
214 newlen
= htons(ntohs(iph
->tot_len
) + ret
);
215 csum_replace2(&iph
->check
, iph
->tot_len
, newlen
);
216 iph
->tot_len
= newlen
;
221 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
223 tcpmss_tg6(struct sk_buff
*skb
, const struct xt_action_param
*par
)
225 struct ipv6hdr
*ipv6h
= ipv6_hdr(skb
);
231 nexthdr
= ipv6h
->nexthdr
;
232 tcphoff
= ipv6_skip_exthdr(skb
, sizeof(*ipv6h
), &nexthdr
, &frag_off
);
235 ret
= tcpmss_mangle_packet(skb
, par
,
236 tcpmss_reverse_mtu(skb
, PF_INET6
),
238 sizeof(*ipv6h
) + sizeof(struct tcphdr
));
242 ipv6h
= ipv6_hdr(skb
);
243 ipv6h
->payload_len
= htons(ntohs(ipv6h
->payload_len
) + ret
);
249 /* Must specify -p tcp --syn */
250 static inline bool find_syn_match(const struct xt_entry_match
*m
)
252 const struct xt_tcp
*tcpinfo
= (const struct xt_tcp
*)m
->data
;
254 if (strcmp(m
->u
.kernel
.match
->name
, "tcp") == 0 &&
255 tcpinfo
->flg_cmp
& TCPHDR_SYN
&&
256 !(tcpinfo
->invflags
& XT_TCP_INV_FLAGS
))
262 static int tcpmss_tg4_check(const struct xt_tgchk_param
*par
)
264 const struct xt_tcpmss_info
*info
= par
->targinfo
;
265 const struct ipt_entry
*e
= par
->entryinfo
;
266 const struct xt_entry_match
*ematch
;
268 if (info
->mss
== XT_TCPMSS_CLAMP_PMTU
&&
269 (par
->hook_mask
& ~((1 << NF_INET_FORWARD
) |
270 (1 << NF_INET_LOCAL_OUT
) |
271 (1 << NF_INET_POST_ROUTING
))) != 0) {
272 pr_info("path-MTU clamping only supported in "
273 "FORWARD, OUTPUT and POSTROUTING hooks\n");
276 xt_ematch_foreach(ematch
, e
)
277 if (find_syn_match(ematch
))
279 pr_info("Only works on TCP SYN packets\n");
283 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
284 static int tcpmss_tg6_check(const struct xt_tgchk_param
*par
)
286 const struct xt_tcpmss_info
*info
= par
->targinfo
;
287 const struct ip6t_entry
*e
= par
->entryinfo
;
288 const struct xt_entry_match
*ematch
;
290 if (info
->mss
== XT_TCPMSS_CLAMP_PMTU
&&
291 (par
->hook_mask
& ~((1 << NF_INET_FORWARD
) |
292 (1 << NF_INET_LOCAL_OUT
) |
293 (1 << NF_INET_POST_ROUTING
))) != 0) {
294 pr_info("path-MTU clamping only supported in "
295 "FORWARD, OUTPUT and POSTROUTING hooks\n");
298 xt_ematch_foreach(ematch
, e
)
299 if (find_syn_match(ematch
))
301 pr_info("Only works on TCP SYN packets\n");
306 static struct xt_target tcpmss_tg_reg
[] __read_mostly
= {
308 .family
= NFPROTO_IPV4
,
310 .checkentry
= tcpmss_tg4_check
,
311 .target
= tcpmss_tg4
,
312 .targetsize
= sizeof(struct xt_tcpmss_info
),
313 .proto
= IPPROTO_TCP
,
316 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
318 .family
= NFPROTO_IPV6
,
320 .checkentry
= tcpmss_tg6_check
,
321 .target
= tcpmss_tg6
,
322 .targetsize
= sizeof(struct xt_tcpmss_info
),
323 .proto
= IPPROTO_TCP
,
329 static int __init
tcpmss_tg_init(void)
331 return xt_register_targets(tcpmss_tg_reg
, ARRAY_SIZE(tcpmss_tg_reg
));
334 static void __exit
tcpmss_tg_exit(void)
336 xt_unregister_targets(tcpmss_tg_reg
, ARRAY_SIZE(tcpmss_tg_reg
));
339 module_init(tcpmss_tg_init
);
340 module_exit(tcpmss_tg_exit
);