Commit | Line | Data |
---|---|---|
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 | */ | |
8bee4bad | 10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
cdd289a2 PM |
11 | #include <linux/module.h> |
12 | #include <linux/skbuff.h> | |
13 | #include <linux/ip.h> | |
5a0e3ad6 | 14 | #include <linux/gfp.h> |
cdd289a2 PM |
15 | #include <linux/ipv6.h> |
16 | #include <linux/tcp.h> | |
37c08387 JE |
17 | #include <net/dst.h> |
18 | #include <net/flow.h> | |
cdd289a2 | 19 | #include <net/ipv6.h> |
37c08387 | 20 | #include <net/route.h> |
cdd289a2 PM |
21 | #include <net/tcp.h> |
22 | ||
23 | #include <linux/netfilter_ipv4/ip_tables.h> | |
24 | #include <linux/netfilter_ipv6/ip6_tables.h> | |
25 | #include <linux/netfilter/x_tables.h> | |
26 | #include <linux/netfilter/xt_tcpudp.h> | |
27 | #include <linux/netfilter/xt_TCPMSS.h> | |
28 | ||
29 | MODULE_LICENSE("GPL"); | |
30 | MODULE_AUTHOR("Marc Boucher <marc@mbsi.ca>"); | |
2ae15b64 | 31 | MODULE_DESCRIPTION("Xtables: TCP Maximum Segment Size (MSS) adjustment"); |
cdd289a2 PM |
32 | MODULE_ALIAS("ipt_TCPMSS"); |
33 | MODULE_ALIAS("ip6t_TCPMSS"); | |
34 | ||
35 | static inline unsigned int | |
36 | optlen(const u_int8_t *opt, unsigned int offset) | |
37 | { | |
38 | /* Beware zero-length options: make finite progress */ | |
39 | if (opt[offset] <= TCPOPT_NOP || opt[offset+1] == 0) | |
40 | return 1; | |
41 | else | |
42 | return opt[offset+1]; | |
43 | } | |
44 | ||
45 | static int | |
3db05fea | 46 | tcpmss_mangle_packet(struct sk_buff *skb, |
cdd289a2 | 47 | const struct xt_tcpmss_info *info, |
37c08387 | 48 | unsigned int in_mtu, |
cdd289a2 PM |
49 | unsigned int tcphoff, |
50 | unsigned int minlen) | |
51 | { | |
52 | struct tcphdr *tcph; | |
53 | unsigned int tcplen, i; | |
54 | __be16 oldval; | |
55 | u16 newmss; | |
56 | u8 *opt; | |
57 | ||
3db05fea | 58 | if (!skb_make_writable(skb, skb->len)) |
cdd289a2 PM |
59 | return -1; |
60 | ||
3db05fea HX |
61 | tcplen = skb->len - tcphoff; |
62 | tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff); | |
cdd289a2 | 63 | |
10a19939 SA |
64 | /* Header cannot be larger than the packet */ |
65 | if (tcplen < tcph->doff*4) | |
cdd289a2 | 66 | return -1; |
cdd289a2 PM |
67 | |
68 | if (info->mss == XT_TCPMSS_CLAMP_PMTU) { | |
adf30907 | 69 | if (dst_mtu(skb_dst(skb)) <= minlen) { |
e87cc472 JP |
70 | net_err_ratelimited("unknown or invalid path-MTU (%u)\n", |
71 | dst_mtu(skb_dst(skb))); | |
cdd289a2 PM |
72 | return -1; |
73 | } | |
37c08387 | 74 | if (in_mtu <= minlen) { |
e87cc472 JP |
75 | net_err_ratelimited("unknown or invalid path-MTU (%u)\n", |
76 | in_mtu); | |
37c08387 JE |
77 | return -1; |
78 | } | |
adf30907 | 79 | newmss = min(dst_mtu(skb_dst(skb)), in_mtu) - minlen; |
cdd289a2 PM |
80 | } else |
81 | newmss = info->mss; | |
82 | ||
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; | |
88 | ||
89 | oldmss = (opt[i+2] << 8) | opt[i+3]; | |
90 | ||
17008064 BL |
91 | /* Never increase MSS, even when setting it, as |
92 | * doing so results in problems for hosts that rely | |
93 | * on MSS being set correctly. | |
94 | */ | |
95 | if (oldmss <= newmss) | |
cdd289a2 PM |
96 | return 0; |
97 | ||
98 | opt[i+2] = (newmss & 0xff00) >> 8; | |
7c4e36bc | 99 | opt[i+3] = newmss & 0x00ff; |
cdd289a2 | 100 | |
be0ea7d5 PM |
101 | inet_proto_csum_replace2(&tcph->check, skb, |
102 | htons(oldmss), htons(newmss), | |
103 | 0); | |
cdd289a2 PM |
104 | return 0; |
105 | } | |
106 | } | |
107 | ||
10a19939 SA |
108 | /* There is data after the header so the option can't be added |
109 | without moving it, and doing so may make the SYN packet | |
110 | itself too large. Accept the packet unmodified instead. */ | |
111 | if (tcplen > tcph->doff*4) | |
112 | return 0; | |
113 | ||
cdd289a2 PM |
114 | /* |
115 | * MSS Option not found ?! add it.. | |
116 | */ | |
3db05fea HX |
117 | if (skb_tailroom(skb) < TCPOLEN_MSS) { |
118 | if (pskb_expand_head(skb, 0, | |
119 | TCPOLEN_MSS - skb_tailroom(skb), | |
2ca7b0ac | 120 | GFP_ATOMIC)) |
cdd289a2 | 121 | return -1; |
3db05fea | 122 | tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff); |
cdd289a2 PM |
123 | } |
124 | ||
3db05fea | 125 | skb_put(skb, TCPOLEN_MSS); |
cdd289a2 PM |
126 | |
127 | opt = (u_int8_t *)tcph + sizeof(struct tcphdr); | |
128 | memmove(opt + TCPOLEN_MSS, opt, tcplen - sizeof(struct tcphdr)); | |
129 | ||
be0ea7d5 PM |
130 | inet_proto_csum_replace2(&tcph->check, skb, |
131 | htons(tcplen), htons(tcplen + TCPOLEN_MSS), 1); | |
cdd289a2 PM |
132 | opt[0] = TCPOPT_MSS; |
133 | opt[1] = TCPOLEN_MSS; | |
134 | opt[2] = (newmss & 0xff00) >> 8; | |
7c4e36bc | 135 | opt[3] = newmss & 0x00ff; |
cdd289a2 | 136 | |
be0ea7d5 | 137 | inet_proto_csum_replace4(&tcph->check, skb, 0, *((__be32 *)opt), 0); |
cdd289a2 PM |
138 | |
139 | oldval = ((__be16 *)tcph)[6]; | |
140 | tcph->doff += TCPOLEN_MSS/4; | |
be0ea7d5 PM |
141 | inet_proto_csum_replace2(&tcph->check, skb, |
142 | oldval, ((__be16 *)tcph)[6], 0); | |
cdd289a2 PM |
143 | return TCPOLEN_MSS; |
144 | } | |
145 | ||
db1a75bd JE |
146 | static u_int32_t tcpmss_reverse_mtu(const struct sk_buff *skb, |
147 | unsigned int family) | |
37c08387 | 148 | { |
a1bbb0e6 | 149 | struct flowi fl; |
37c08387 JE |
150 | const struct nf_afinfo *ai; |
151 | struct rtable *rt = NULL; | |
152 | u_int32_t mtu = ~0U; | |
153 | ||
a1bbb0e6 DM |
154 | if (family == PF_INET) { |
155 | struct flowi4 *fl4 = &fl.u.ip4; | |
156 | memset(fl4, 0, sizeof(*fl4)); | |
157 | fl4->daddr = ip_hdr(skb)->saddr; | |
158 | } else { | |
159 | struct flowi6 *fl6 = &fl.u.ip6; | |
db1a75bd | 160 | |
a1bbb0e6 | 161 | memset(fl6, 0, sizeof(*fl6)); |
4e3fd7a0 | 162 | fl6->daddr = ipv6_hdr(skb)->saddr; |
a1bbb0e6 | 163 | } |
37c08387 | 164 | rcu_read_lock(); |
db1a75bd | 165 | ai = nf_get_afinfo(family); |
37c08387 | 166 | if (ai != NULL) |
0fae2e77 | 167 | ai->route(&init_net, (struct dst_entry **)&rt, &fl, false); |
37c08387 JE |
168 | rcu_read_unlock(); |
169 | ||
170 | if (rt != NULL) { | |
d8d1f30b CG |
171 | mtu = dst_mtu(&rt->dst); |
172 | dst_release(&rt->dst); | |
37c08387 JE |
173 | } |
174 | return mtu; | |
175 | } | |
176 | ||
cdd289a2 | 177 | static unsigned int |
4b560b44 | 178 | tcpmss_tg4(struct sk_buff *skb, const struct xt_action_param *par) |
cdd289a2 | 179 | { |
3db05fea | 180 | struct iphdr *iph = ip_hdr(skb); |
cdd289a2 PM |
181 | __be16 newlen; |
182 | int ret; | |
183 | ||
7eb35586 | 184 | ret = tcpmss_mangle_packet(skb, par->targinfo, |
db1a75bd | 185 | tcpmss_reverse_mtu(skb, PF_INET), |
37c08387 | 186 | iph->ihl * 4, |
cdd289a2 PM |
187 | sizeof(*iph) + sizeof(struct tcphdr)); |
188 | if (ret < 0) | |
189 | return NF_DROP; | |
190 | if (ret > 0) { | |
3db05fea | 191 | iph = ip_hdr(skb); |
cdd289a2 | 192 | newlen = htons(ntohs(iph->tot_len) + ret); |
be0ea7d5 | 193 | csum_replace2(&iph->check, iph->tot_len, newlen); |
cdd289a2 PM |
194 | iph->tot_len = newlen; |
195 | } | |
196 | return XT_CONTINUE; | |
197 | } | |
198 | ||
c0cd1156 | 199 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
cdd289a2 | 200 | static unsigned int |
4b560b44 | 201 | tcpmss_tg6(struct sk_buff *skb, const struct xt_action_param *par) |
cdd289a2 | 202 | { |
3db05fea | 203 | struct ipv6hdr *ipv6h = ipv6_hdr(skb); |
cdd289a2 | 204 | u8 nexthdr; |
75f2811c | 205 | __be16 frag_off; |
cdd289a2 PM |
206 | int tcphoff; |
207 | int ret; | |
208 | ||
209 | nexthdr = ipv6h->nexthdr; | |
75f2811c | 210 | tcphoff = ipv6_skip_exthdr(skb, sizeof(*ipv6h), &nexthdr, &frag_off); |
9dc0564e | 211 | if (tcphoff < 0) |
cdd289a2 | 212 | return NF_DROP; |
7eb35586 | 213 | ret = tcpmss_mangle_packet(skb, par->targinfo, |
db1a75bd | 214 | tcpmss_reverse_mtu(skb, PF_INET6), |
37c08387 | 215 | tcphoff, |
cdd289a2 PM |
216 | sizeof(*ipv6h) + sizeof(struct tcphdr)); |
217 | if (ret < 0) | |
218 | return NF_DROP; | |
219 | if (ret > 0) { | |
3db05fea | 220 | ipv6h = ipv6_hdr(skb); |
cdd289a2 PM |
221 | ipv6h->payload_len = htons(ntohs(ipv6h->payload_len) + ret); |
222 | } | |
223 | return XT_CONTINUE; | |
224 | } | |
225 | #endif | |
226 | ||
cdd289a2 | 227 | /* Must specify -p tcp --syn */ |
e1931b78 | 228 | static 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 && | |
a3433f35 | 233 | tcpinfo->flg_cmp & TCPHDR_SYN && |
cdd289a2 | 234 | !(tcpinfo->invflags & XT_TCP_INV_FLAGS)) |
e1931b78 | 235 | return true; |
cdd289a2 | 236 | |
e1931b78 | 237 | return false; |
cdd289a2 PM |
238 | } |
239 | ||
135367b8 | 240 | static int tcpmss_tg4_check(const struct xt_tgchk_param *par) |
cdd289a2 | 241 | { |
af5d6dc2 JE |
242 | const struct xt_tcpmss_info *info = par->targinfo; |
243 | const struct ipt_entry *e = par->entryinfo; | |
dcea992a | 244 | const struct xt_entry_match *ematch; |
cdd289a2 PM |
245 | |
246 | if (info->mss == XT_TCPMSS_CLAMP_PMTU && | |
af5d6dc2 | 247 | (par->hook_mask & ~((1 << NF_INET_FORWARD) | |
6e23ae2a PM |
248 | (1 << NF_INET_LOCAL_OUT) | |
249 | (1 << NF_INET_POST_ROUTING))) != 0) { | |
8bee4bad JE |
250 | pr_info("path-MTU clamping only supported in " |
251 | "FORWARD, OUTPUT and POSTROUTING hooks\n"); | |
d6b00a53 | 252 | return -EINVAL; |
cdd289a2 | 253 | } |
dcea992a JE |
254 | xt_ematch_foreach(ematch, e) |
255 | if (find_syn_match(ematch)) | |
d6b00a53 | 256 | return 0; |
8bee4bad | 257 | pr_info("Only works on TCP SYN packets\n"); |
d6b00a53 | 258 | return -EINVAL; |
cdd289a2 PM |
259 | } |
260 | ||
c0cd1156 | 261 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
135367b8 | 262 | static int tcpmss_tg6_check(const struct xt_tgchk_param *par) |
cdd289a2 | 263 | { |
af5d6dc2 JE |
264 | const struct xt_tcpmss_info *info = par->targinfo; |
265 | const struct ip6t_entry *e = par->entryinfo; | |
dcea992a | 266 | const struct xt_entry_match *ematch; |
cdd289a2 PM |
267 | |
268 | if (info->mss == XT_TCPMSS_CLAMP_PMTU && | |
af5d6dc2 | 269 | (par->hook_mask & ~((1 << NF_INET_FORWARD) | |
6e23ae2a PM |
270 | (1 << NF_INET_LOCAL_OUT) | |
271 | (1 << NF_INET_POST_ROUTING))) != 0) { | |
8bee4bad JE |
272 | pr_info("path-MTU clamping only supported in " |
273 | "FORWARD, OUTPUT and POSTROUTING hooks\n"); | |
d6b00a53 | 274 | return -EINVAL; |
cdd289a2 | 275 | } |
dcea992a JE |
276 | xt_ematch_foreach(ematch, e) |
277 | if (find_syn_match(ematch)) | |
d6b00a53 | 278 | return 0; |
8bee4bad | 279 | pr_info("Only works on TCP SYN packets\n"); |
d6b00a53 | 280 | return -EINVAL; |
cdd289a2 PM |
281 | } |
282 | #endif | |
283 | ||
d3c5ee6d | 284 | static struct xt_target tcpmss_tg_reg[] __read_mostly = { |
cdd289a2 | 285 | { |
ee999d8b | 286 | .family = NFPROTO_IPV4, |
cdd289a2 | 287 | .name = "TCPMSS", |
d3c5ee6d JE |
288 | .checkentry = tcpmss_tg4_check, |
289 | .target = tcpmss_tg4, | |
cdd289a2 PM |
290 | .targetsize = sizeof(struct xt_tcpmss_info), |
291 | .proto = IPPROTO_TCP, | |
292 | .me = THIS_MODULE, | |
293 | }, | |
c0cd1156 | 294 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
cdd289a2 | 295 | { |
ee999d8b | 296 | .family = NFPROTO_IPV6, |
cdd289a2 | 297 | .name = "TCPMSS", |
d3c5ee6d JE |
298 | .checkentry = tcpmss_tg6_check, |
299 | .target = tcpmss_tg6, | |
cdd289a2 PM |
300 | .targetsize = sizeof(struct xt_tcpmss_info), |
301 | .proto = IPPROTO_TCP, | |
302 | .me = THIS_MODULE, | |
303 | }, | |
304 | #endif | |
305 | }; | |
306 | ||
d3c5ee6d | 307 | static int __init tcpmss_tg_init(void) |
cdd289a2 | 308 | { |
d3c5ee6d | 309 | return xt_register_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg)); |
cdd289a2 PM |
310 | } |
311 | ||
d3c5ee6d | 312 | static void __exit tcpmss_tg_exit(void) |
cdd289a2 | 313 | { |
d3c5ee6d | 314 | xt_unregister_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg)); |
cdd289a2 PM |
315 | } |
316 | ||
d3c5ee6d JE |
317 | module_init(tcpmss_tg_init); |
318 | module_exit(tcpmss_tg_exit); |