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 * Copyright (C) 2007 Patrick McHardy <kaber@trash.net>
6 *
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.
10 */
11#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12#include <linux/module.h>
13#include <linux/skbuff.h>
14#include <linux/ip.h>
15#include <linux/gfp.h>
16#include <linux/ipv6.h>
17#include <linux/tcp.h>
18#include <net/dst.h>
19#include <net/flow.h>
20#include <net/ipv6.h>
21#include <net/route.h>
22#include <net/tcp.h>
23
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>
29
30MODULE_LICENSE("GPL");
31MODULE_AUTHOR("Marc Boucher <marc@mbsi.ca>");
32MODULE_DESCRIPTION("Xtables: TCP Maximum Segment Size (MSS) adjustment");
33MODULE_ALIAS("ipt_TCPMSS");
34MODULE_ALIAS("ip6t_TCPMSS");
35
36static inline unsigned int
37optlen(const u_int8_t *opt, unsigned int offset)
38{
39	/* Beware zero-length options: make finite progress */
40	if (opt[offset] <= TCPOPT_NOP || opt[offset+1] == 0)
41		return 1;
42	else
43		return opt[offset+1];
44}
45
46static u_int32_t tcpmss_reverse_mtu(struct net *net,
47				    const struct sk_buff *skb,
48				    unsigned int family)
49{
50	struct flowi fl;
51	const struct nf_afinfo *ai;
52	struct rtable *rt = NULL;
53	u_int32_t mtu     = ~0U;
54
55	if (family == PF_INET) {
56		struct flowi4 *fl4 = &fl.u.ip4;
57		memset(fl4, 0, sizeof(*fl4));
58		fl4->daddr = ip_hdr(skb)->saddr;
59	} else {
60		struct flowi6 *fl6 = &fl.u.ip6;
61
62		memset(fl6, 0, sizeof(*fl6));
63		fl6->daddr = ipv6_hdr(skb)->saddr;
64	}
65	rcu_read_lock();
66	ai = nf_get_afinfo(family);
67	if (ai != NULL)
68		ai->route(net, (struct dst_entry **)&rt, &fl, false);
69	rcu_read_unlock();
70
71	if (rt != NULL) {
72		mtu = dst_mtu(&rt->dst);
73		dst_release(&rt->dst);
74	}
75	return mtu;
76}
77
78static int
79tcpmss_mangle_packet(struct sk_buff *skb,
80		     const struct xt_action_param *par,
81		     unsigned int family,
82		     unsigned int tcphoff,
83		     unsigned int minlen)
84{
85	const struct xt_tcpmss_info *info = par->targinfo;
86	struct tcphdr *tcph;
87	int len, tcp_hdrlen;
88	unsigned int i;
89	__be16 oldval;
90	u16 newmss;
91	u8 *opt;
92
93	/* This is a fragment, no TCP header is available */
94	if (par->fragoff != 0)
95		return 0;
96
97	if (!skb_make_writable(skb, skb->len))
98		return -1;
99
100	len = skb->len - tcphoff;
101	if (len < (int)sizeof(struct tcphdr))
102		return -1;
103
104	tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
105	tcp_hdrlen = tcph->doff * 4;
106
107	if (len < tcp_hdrlen)
108		return -1;
109
110	if (info->mss == XT_TCPMSS_CLAMP_PMTU) {
111		struct net *net = dev_net(par->in ? par->in : par->out);
112		unsigned int in_mtu = tcpmss_reverse_mtu(net, skb, family);
113
114		if (dst_mtu(skb_dst(skb)) <= minlen) {
115			net_err_ratelimited("unknown or invalid path-MTU (%u)\n",
116					    dst_mtu(skb_dst(skb)));
117			return -1;
118		}
119		if (in_mtu <= minlen) {
120			net_err_ratelimited("unknown or invalid path-MTU (%u)\n",
121					    in_mtu);
122			return -1;
123		}
124		newmss = min(dst_mtu(skb_dst(skb)), in_mtu) - minlen;
125	} else
126		newmss = info->mss;
127
128	opt = (u_int8_t *)tcph;
129	for (i = sizeof(struct tcphdr); i <= tcp_hdrlen - TCPOLEN_MSS; i += optlen(opt, i)) {
130		if (opt[i] == TCPOPT_MSS && opt[i+1] == TCPOLEN_MSS) {
131			u_int16_t oldmss;
132
133			oldmss = (opt[i+2] << 8) | opt[i+3];
134
135			/* Never increase MSS, even when setting it, as
136			 * doing so results in problems for hosts that rely
137			 * on MSS being set correctly.
138			 */
139			if (oldmss <= newmss)
140				return 0;
141
142			opt[i+2] = (newmss & 0xff00) >> 8;
143			opt[i+3] = newmss & 0x00ff;
144
145			inet_proto_csum_replace2(&tcph->check, skb,
146						 htons(oldmss), htons(newmss),
147						 0);
148			return 0;
149		}
150	}
151
152	/* There is data after the header so the option can't be added
153	 * without moving it, and doing so may make the SYN packet
154	 * itself too large. Accept the packet unmodified instead.
155	 */
156	if (len > tcp_hdrlen)
157		return 0;
158
159	/*
160	 * MSS Option not found ?! add it..
161	 */
162	if (skb_tailroom(skb) < TCPOLEN_MSS) {
163		if (pskb_expand_head(skb, 0,
164				     TCPOLEN_MSS - skb_tailroom(skb),
165				     GFP_ATOMIC))
166			return -1;
167		tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
168	}
169
170	skb_put(skb, TCPOLEN_MSS);
171
172	/*
173	 * IPv4: RFC 1122 states "If an MSS option is not received at
174	 * connection setup, TCP MUST assume a default send MSS of 536".
175	 * IPv6: RFC 2460 states IPv6 has a minimum MTU of 1280 and a minimum
176	 * length IPv6 header of 60, ergo the default MSS value is 1220
177	 * Since no MSS was provided, we must use the default values
178	 */
179	if (par->family == NFPROTO_IPV4)
180		newmss = min(newmss, (u16)536);
181	else
182		newmss = min(newmss, (u16)1220);
183
184	opt = (u_int8_t *)tcph + sizeof(struct tcphdr);
185	memmove(opt + TCPOLEN_MSS, opt, len - sizeof(struct tcphdr));
186
187	inet_proto_csum_replace2(&tcph->check, skb,
188				 htons(len), htons(len + TCPOLEN_MSS), 1);
189	opt[0] = TCPOPT_MSS;
190	opt[1] = TCPOLEN_MSS;
191	opt[2] = (newmss & 0xff00) >> 8;
192	opt[3] = newmss & 0x00ff;
193
194	inet_proto_csum_replace4(&tcph->check, skb, 0, *((__be32 *)opt), 0);
195
196	oldval = ((__be16 *)tcph)[6];
197	tcph->doff += TCPOLEN_MSS/4;
198	inet_proto_csum_replace2(&tcph->check, skb,
199				 oldval, ((__be16 *)tcph)[6], 0);
200	return TCPOLEN_MSS;
201}
202
203static unsigned int
204tcpmss_tg4(struct sk_buff *skb, const struct xt_action_param *par)
205{
206	struct iphdr *iph = ip_hdr(skb);
207	__be16 newlen;
208	int ret;
209
210	ret = tcpmss_mangle_packet(skb, par,
211				   PF_INET,
212				   iph->ihl * 4,
213				   sizeof(*iph) + sizeof(struct tcphdr));
214	if (ret < 0)
215		return NF_DROP;
216	if (ret > 0) {
217		iph = ip_hdr(skb);
218		newlen = htons(ntohs(iph->tot_len) + ret);
219		csum_replace2(&iph->check, iph->tot_len, newlen);
220		iph->tot_len = newlen;
221	}
222	return XT_CONTINUE;
223}
224
225#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
226static unsigned int
227tcpmss_tg6(struct sk_buff *skb, const struct xt_action_param *par)
228{
229	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
230	u8 nexthdr;
231	__be16 frag_off;
232	int tcphoff;
233	int ret;
234
235	nexthdr = ipv6h->nexthdr;
236	tcphoff = ipv6_skip_exthdr(skb, sizeof(*ipv6h), &nexthdr, &frag_off);
237	if (tcphoff < 0)
238		return NF_DROP;
239	ret = tcpmss_mangle_packet(skb, par,
240				   PF_INET6,
241				   tcphoff,
242				   sizeof(*ipv6h) + sizeof(struct tcphdr));
243	if (ret < 0)
244		return NF_DROP;
245	if (ret > 0) {
246		ipv6h = ipv6_hdr(skb);
247		ipv6h->payload_len = htons(ntohs(ipv6h->payload_len) + ret);
248	}
249	return XT_CONTINUE;
250}
251#endif
252
253/* Must specify -p tcp --syn */
254static inline bool find_syn_match(const struct xt_entry_match *m)
255{
256	const struct xt_tcp *tcpinfo = (const struct xt_tcp *)m->data;
257
258	if (strcmp(m->u.kernel.match->name, "tcp") == 0 &&
259	    tcpinfo->flg_cmp & TCPHDR_SYN &&
260	    !(tcpinfo->invflags & XT_TCP_INV_FLAGS))
261		return true;
262
263	return false;
264}
265
266static int tcpmss_tg4_check(const struct xt_tgchk_param *par)
267{
268	const struct xt_tcpmss_info *info = par->targinfo;
269	const struct ipt_entry *e = par->entryinfo;
270	const struct xt_entry_match *ematch;
271
272	if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
273	    (par->hook_mask & ~((1 << NF_INET_FORWARD) |
274			   (1 << NF_INET_LOCAL_OUT) |
275			   (1 << NF_INET_POST_ROUTING))) != 0) {
276		pr_info("path-MTU clamping only supported in "
277			"FORWARD, OUTPUT and POSTROUTING hooks\n");
278		return -EINVAL;
279	}
280	xt_ematch_foreach(ematch, e)
281		if (find_syn_match(ematch))
282			return 0;
283	pr_info("Only works on TCP SYN packets\n");
284	return -EINVAL;
285}
286
287#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
288static int tcpmss_tg6_check(const struct xt_tgchk_param *par)
289{
290	const struct xt_tcpmss_info *info = par->targinfo;
291	const struct ip6t_entry *e = par->entryinfo;
292	const struct xt_entry_match *ematch;
293
294	if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
295	    (par->hook_mask & ~((1 << NF_INET_FORWARD) |
296			   (1 << NF_INET_LOCAL_OUT) |
297			   (1 << NF_INET_POST_ROUTING))) != 0) {
298		pr_info("path-MTU clamping only supported in "
299			"FORWARD, OUTPUT and POSTROUTING hooks\n");
300		return -EINVAL;
301	}
302	xt_ematch_foreach(ematch, e)
303		if (find_syn_match(ematch))
304			return 0;
305	pr_info("Only works on TCP SYN packets\n");
306	return -EINVAL;
307}
308#endif
309
310static struct xt_target tcpmss_tg_reg[] __read_mostly = {
311	{
312		.family		= NFPROTO_IPV4,
313		.name		= "TCPMSS",
314		.checkentry	= tcpmss_tg4_check,
315		.target		= tcpmss_tg4,
316		.targetsize	= sizeof(struct xt_tcpmss_info),
317		.proto		= IPPROTO_TCP,
318		.me		= THIS_MODULE,
319	},
320#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
321	{
322		.family		= NFPROTO_IPV6,
323		.name		= "TCPMSS",
324		.checkentry	= tcpmss_tg6_check,
325		.target		= tcpmss_tg6,
326		.targetsize	= sizeof(struct xt_tcpmss_info),
327		.proto		= IPPROTO_TCP,
328		.me		= THIS_MODULE,
329	},
330#endif
331};
332
333static int __init tcpmss_tg_init(void)
334{
335	return xt_register_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
336}
337
338static void __exit tcpmss_tg_exit(void)
339{
340	xt_unregister_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
341}
342
343module_init(tcpmss_tg_init);
344module_exit(tcpmss_tg_exit);
345