Merge tag 'xfs-4.20-merge-1' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
[linux-2.6-block.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>
f229f6ce 5 * Copyright (C) 2007 Patrick McHardy <kaber@trash.net>
cdd289a2
PM
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 */
8bee4bad 11#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
cdd289a2
PM
12#include <linux/module.h>
13#include <linux/skbuff.h>
14#include <linux/ip.h>
5a0e3ad6 15#include <linux/gfp.h>
cdd289a2
PM
16#include <linux/ipv6.h>
17#include <linux/tcp.h>
37c08387
JE
18#include <net/dst.h>
19#include <net/flow.h>
cdd289a2 20#include <net/ipv6.h>
37c08387 21#include <net/route.h>
cdd289a2
PM
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>");
2ae15b64 32MODULE_DESCRIPTION("Xtables: TCP Maximum Segment Size (MSS) adjustment");
cdd289a2
PM
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
7722e0d1
G
46static u_int32_t tcpmss_reverse_mtu(struct net *net,
47 const struct sk_buff *skb,
de1389b1
G
48 unsigned int family)
49{
50 struct flowi fl;
de1389b1
G
51 struct rtable *rt = NULL;
52 u_int32_t mtu = ~0U;
53
54 if (family == PF_INET) {
55 struct flowi4 *fl4 = &fl.u.ip4;
56 memset(fl4, 0, sizeof(*fl4));
57 fl4->daddr = ip_hdr(skb)->saddr;
58 } else {
59 struct flowi6 *fl6 = &fl.u.ip6;
60
61 memset(fl6, 0, sizeof(*fl6));
62 fl6->daddr = ipv6_hdr(skb)->saddr;
63 }
de1389b1 64
3f87c08c 65 nf_route(net, (struct dst_entry **)&rt, &fl, false, family);
de1389b1
G
66 if (rt != NULL) {
67 mtu = dst_mtu(&rt->dst);
68 dst_release(&rt->dst);
69 }
70 return mtu;
71}
72
cdd289a2 73static int
3db05fea 74tcpmss_mangle_packet(struct sk_buff *skb,
70d19f80 75 const struct xt_action_param *par,
de1389b1 76 unsigned int family,
cdd289a2
PM
77 unsigned int tcphoff,
78 unsigned int minlen)
79{
70d19f80 80 const struct xt_tcpmss_info *info = par->targinfo;
cdd289a2 81 struct tcphdr *tcph;
71ffe9c7
PNA
82 int len, tcp_hdrlen;
83 unsigned int i;
cdd289a2
PM
84 __be16 oldval;
85 u16 newmss;
86 u8 *opt;
87
b396966c
PO
88 /* This is a fragment, no TCP header is available */
89 if (par->fragoff != 0)
1205e1fa 90 return 0;
b396966c 91
3db05fea 92 if (!skb_make_writable(skb, skb->len))
cdd289a2
PM
93 return -1;
94
71ffe9c7
PNA
95 len = skb->len - tcphoff;
96 if (len < (int)sizeof(struct tcphdr))
97 return -1;
98
3db05fea 99 tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
71ffe9c7 100 tcp_hdrlen = tcph->doff * 4;
cdd289a2 101
2638fd0f 102 if (len < tcp_hdrlen || tcp_hdrlen < sizeof(struct tcphdr))
cdd289a2 103 return -1;
cdd289a2
PM
104
105 if (info->mss == XT_TCPMSS_CLAMP_PMTU) {
613dbd95 106 struct net *net = xt_net(par);
7722e0d1 107 unsigned int in_mtu = tcpmss_reverse_mtu(net, skb, family);
50f4c7b7 108 unsigned int min_mtu = min(dst_mtu(skb_dst(skb)), in_mtu);
de1389b1 109
50f4c7b7 110 if (min_mtu <= minlen) {
e87cc472 111 net_err_ratelimited("unknown or invalid path-MTU (%u)\n",
50f4c7b7 112 min_mtu);
cdd289a2
PM
113 return -1;
114 }
50f4c7b7 115 newmss = min_mtu - minlen;
cdd289a2
PM
116 } else
117 newmss = info->mss;
118
119 opt = (u_int8_t *)tcph;
71ffe9c7
PNA
120 for (i = sizeof(struct tcphdr); i <= tcp_hdrlen - TCPOLEN_MSS; i += optlen(opt, i)) {
121 if (opt[i] == TCPOPT_MSS && opt[i+1] == TCPOLEN_MSS) {
cdd289a2
PM
122 u_int16_t oldmss;
123
124 oldmss = (opt[i+2] << 8) | opt[i+3];
125
17008064
BL
126 /* Never increase MSS, even when setting it, as
127 * doing so results in problems for hosts that rely
128 * on MSS being set correctly.
129 */
130 if (oldmss <= newmss)
cdd289a2
PM
131 return 0;
132
133 opt[i+2] = (newmss & 0xff00) >> 8;
7c4e36bc 134 opt[i+3] = newmss & 0x00ff;
cdd289a2 135
be0ea7d5
PM
136 inet_proto_csum_replace2(&tcph->check, skb,
137 htons(oldmss), htons(newmss),
4b048d6d 138 false);
cdd289a2
PM
139 return 0;
140 }
141 }
142
10a19939 143 /* There is data after the header so the option can't be added
71ffe9c7
PNA
144 * without moving it, and doing so may make the SYN packet
145 * itself too large. Accept the packet unmodified instead.
146 */
147 if (len > tcp_hdrlen)
10a19939 148 return 0;
2638fd0f
ED
149
150 /* tcph->doff has 4 bits, do not wrap it to 0 */
151 if (tcp_hdrlen >= 15 * 4)
152 return 0;
10a19939 153
cdd289a2
PM
154 /*
155 * MSS Option not found ?! add it..
156 */
3db05fea
HX
157 if (skb_tailroom(skb) < TCPOLEN_MSS) {
158 if (pskb_expand_head(skb, 0,
159 TCPOLEN_MSS - skb_tailroom(skb),
2ca7b0ac 160 GFP_ATOMIC))
cdd289a2 161 return -1;
3db05fea 162 tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
cdd289a2
PM
163 }
164
3db05fea 165 skb_put(skb, TCPOLEN_MSS);
cdd289a2 166
70d19f80
PO
167 /*
168 * IPv4: RFC 1122 states "If an MSS option is not received at
169 * connection setup, TCP MUST assume a default send MSS of 536".
170 * IPv6: RFC 2460 states IPv6 has a minimum MTU of 1280 and a minimum
171 * length IPv6 header of 60, ergo the default MSS value is 1220
172 * Since no MSS was provided, we must use the default values
409b545a 173 */
613dbd95 174 if (xt_family(par) == NFPROTO_IPV4)
70d19f80
PO
175 newmss = min(newmss, (u16)536);
176 else
177 newmss = min(newmss, (u16)1220);
409b545a 178
cdd289a2 179 opt = (u_int8_t *)tcph + sizeof(struct tcphdr);
71ffe9c7 180 memmove(opt + TCPOLEN_MSS, opt, len - sizeof(struct tcphdr));
cdd289a2 181
be0ea7d5 182 inet_proto_csum_replace2(&tcph->check, skb,
4b048d6d 183 htons(len), htons(len + TCPOLEN_MSS), true);
cdd289a2
PM
184 opt[0] = TCPOPT_MSS;
185 opt[1] = TCPOLEN_MSS;
186 opt[2] = (newmss & 0xff00) >> 8;
7c4e36bc 187 opt[3] = newmss & 0x00ff;
cdd289a2 188
4b048d6d 189 inet_proto_csum_replace4(&tcph->check, skb, 0, *((__be32 *)opt), false);
cdd289a2
PM
190
191 oldval = ((__be16 *)tcph)[6];
192 tcph->doff += TCPOLEN_MSS/4;
be0ea7d5 193 inet_proto_csum_replace2(&tcph->check, skb,
4b048d6d 194 oldval, ((__be16 *)tcph)[6], false);
cdd289a2
PM
195 return TCPOLEN_MSS;
196}
197
198static unsigned int
4b560b44 199tcpmss_tg4(struct sk_buff *skb, const struct xt_action_param *par)
cdd289a2 200{
3db05fea 201 struct iphdr *iph = ip_hdr(skb);
cdd289a2
PM
202 __be16 newlen;
203 int ret;
204
70d19f80 205 ret = tcpmss_mangle_packet(skb, par,
de1389b1 206 PF_INET,
37c08387 207 iph->ihl * 4,
cdd289a2
PM
208 sizeof(*iph) + sizeof(struct tcphdr));
209 if (ret < 0)
210 return NF_DROP;
211 if (ret > 0) {
3db05fea 212 iph = ip_hdr(skb);
cdd289a2 213 newlen = htons(ntohs(iph->tot_len) + ret);
be0ea7d5 214 csum_replace2(&iph->check, iph->tot_len, newlen);
cdd289a2
PM
215 iph->tot_len = newlen;
216 }
217 return XT_CONTINUE;
218}
219
c0cd1156 220#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
cdd289a2 221static unsigned int
4b560b44 222tcpmss_tg6(struct sk_buff *skb, const struct xt_action_param *par)
cdd289a2 223{
3db05fea 224 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
cdd289a2 225 u8 nexthdr;
d6b3347b 226 __be16 frag_off, oldlen, newlen;
cdd289a2
PM
227 int tcphoff;
228 int ret;
229
230 nexthdr = ipv6h->nexthdr;
75f2811c 231 tcphoff = ipv6_skip_exthdr(skb, sizeof(*ipv6h), &nexthdr, &frag_off);
9dc0564e 232 if (tcphoff < 0)
cdd289a2 233 return NF_DROP;
70d19f80 234 ret = tcpmss_mangle_packet(skb, par,
de1389b1 235 PF_INET6,
37c08387 236 tcphoff,
cdd289a2
PM
237 sizeof(*ipv6h) + sizeof(struct tcphdr));
238 if (ret < 0)
239 return NF_DROP;
240 if (ret > 0) {
3db05fea 241 ipv6h = ipv6_hdr(skb);
d6b3347b
ED
242 oldlen = ipv6h->payload_len;
243 newlen = htons(ntohs(oldlen) + ret);
244 if (skb->ip_summed == CHECKSUM_COMPLETE)
245 skb->csum = csum_add(csum_sub(skb->csum, oldlen),
246 newlen);
247 ipv6h->payload_len = newlen;
cdd289a2
PM
248 }
249 return XT_CONTINUE;
250}
251#endif
252
cdd289a2 253/* Must specify -p tcp --syn */
e1931b78 254static inline bool find_syn_match(const struct xt_entry_match *m)
cdd289a2
PM
255{
256 const struct xt_tcp *tcpinfo = (const struct xt_tcp *)m->data;
257
258 if (strcmp(m->u.kernel.match->name, "tcp") == 0 &&
a3433f35 259 tcpinfo->flg_cmp & TCPHDR_SYN &&
cdd289a2 260 !(tcpinfo->invflags & XT_TCP_INV_FLAGS))
e1931b78 261 return true;
cdd289a2 262
e1931b78 263 return false;
cdd289a2
PM
264}
265
135367b8 266static int tcpmss_tg4_check(const struct xt_tgchk_param *par)
cdd289a2 267{
af5d6dc2
JE
268 const struct xt_tcpmss_info *info = par->targinfo;
269 const struct ipt_entry *e = par->entryinfo;
dcea992a 270 const struct xt_entry_match *ematch;
cdd289a2
PM
271
272 if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
af5d6dc2 273 (par->hook_mask & ~((1 << NF_INET_FORWARD) |
6e23ae2a
PM
274 (1 << NF_INET_LOCAL_OUT) |
275 (1 << NF_INET_POST_ROUTING))) != 0) {
b2606644 276 pr_info_ratelimited("path-MTU clamping only supported in FORWARD, OUTPUT and POSTROUTING hooks\n");
d6b00a53 277 return -EINVAL;
cdd289a2 278 }
55917a21
PNA
279 if (par->nft_compat)
280 return 0;
281
dcea992a
JE
282 xt_ematch_foreach(ematch, e)
283 if (find_syn_match(ematch))
d6b00a53 284 return 0;
b2606644 285 pr_info_ratelimited("Only works on TCP SYN packets\n");
d6b00a53 286 return -EINVAL;
cdd289a2
PM
287}
288
c0cd1156 289#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
135367b8 290static int tcpmss_tg6_check(const struct xt_tgchk_param *par)
cdd289a2 291{
af5d6dc2
JE
292 const struct xt_tcpmss_info *info = par->targinfo;
293 const struct ip6t_entry *e = par->entryinfo;
dcea992a 294 const struct xt_entry_match *ematch;
cdd289a2
PM
295
296 if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
af5d6dc2 297 (par->hook_mask & ~((1 << NF_INET_FORWARD) |
6e23ae2a
PM
298 (1 << NF_INET_LOCAL_OUT) |
299 (1 << NF_INET_POST_ROUTING))) != 0) {
b2606644 300 pr_info_ratelimited("path-MTU clamping only supported in FORWARD, OUTPUT and POSTROUTING hooks\n");
d6b00a53 301 return -EINVAL;
cdd289a2 302 }
55917a21
PNA
303 if (par->nft_compat)
304 return 0;
305
dcea992a
JE
306 xt_ematch_foreach(ematch, e)
307 if (find_syn_match(ematch))
d6b00a53 308 return 0;
b2606644 309 pr_info_ratelimited("Only works on TCP SYN packets\n");
d6b00a53 310 return -EINVAL;
cdd289a2
PM
311}
312#endif
313
d3c5ee6d 314static struct xt_target tcpmss_tg_reg[] __read_mostly = {
cdd289a2 315 {
ee999d8b 316 .family = NFPROTO_IPV4,
cdd289a2 317 .name = "TCPMSS",
d3c5ee6d
JE
318 .checkentry = tcpmss_tg4_check,
319 .target = tcpmss_tg4,
cdd289a2
PM
320 .targetsize = sizeof(struct xt_tcpmss_info),
321 .proto = IPPROTO_TCP,
322 .me = THIS_MODULE,
323 },
c0cd1156 324#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
cdd289a2 325 {
ee999d8b 326 .family = NFPROTO_IPV6,
cdd289a2 327 .name = "TCPMSS",
d3c5ee6d
JE
328 .checkentry = tcpmss_tg6_check,
329 .target = tcpmss_tg6,
cdd289a2
PM
330 .targetsize = sizeof(struct xt_tcpmss_info),
331 .proto = IPPROTO_TCP,
332 .me = THIS_MODULE,
333 },
334#endif
335};
336
d3c5ee6d 337static int __init tcpmss_tg_init(void)
cdd289a2 338{
d3c5ee6d 339 return xt_register_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
cdd289a2
PM
340}
341
d3c5ee6d 342static void __exit tcpmss_tg_exit(void)
cdd289a2 343{
d3c5ee6d 344 xt_unregister_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
cdd289a2
PM
345}
346
d3c5ee6d
JE
347module_init(tcpmss_tg_init);
348module_exit(tcpmss_tg_exit);