Commit | Line | Data |
---|---|---|
d2912cb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
cdd289a2 PM |
2 | /* |
3 | * This is a module which is used for setting the MSS option in TCP packets. | |
4 | * | |
5 | * Copyright (C) 2000 Marc Boucher <marc@mbsi.ca> | |
f229f6ce | 6 | * Copyright (C) 2007 Patrick McHardy <kaber@trash.net> |
cdd289a2 | 7 | */ |
8bee4bad | 8 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
cdd289a2 PM |
9 | #include <linux/module.h> |
10 | #include <linux/skbuff.h> | |
11 | #include <linux/ip.h> | |
5a0e3ad6 | 12 | #include <linux/gfp.h> |
cdd289a2 PM |
13 | #include <linux/ipv6.h> |
14 | #include <linux/tcp.h> | |
37c08387 JE |
15 | #include <net/dst.h> |
16 | #include <net/flow.h> | |
cdd289a2 | 17 | #include <net/ipv6.h> |
37c08387 | 18 | #include <net/route.h> |
cdd289a2 PM |
19 | #include <net/tcp.h> |
20 | ||
21 | #include <linux/netfilter_ipv4/ip_tables.h> | |
22 | #include <linux/netfilter_ipv6/ip6_tables.h> | |
23 | #include <linux/netfilter/x_tables.h> | |
24 | #include <linux/netfilter/xt_tcpudp.h> | |
25 | #include <linux/netfilter/xt_TCPMSS.h> | |
26 | ||
27 | MODULE_LICENSE("GPL"); | |
28 | MODULE_AUTHOR("Marc Boucher <marc@mbsi.ca>"); | |
2ae15b64 | 29 | MODULE_DESCRIPTION("Xtables: TCP Maximum Segment Size (MSS) adjustment"); |
cdd289a2 PM |
30 | MODULE_ALIAS("ipt_TCPMSS"); |
31 | MODULE_ALIAS("ip6t_TCPMSS"); | |
32 | ||
33 | static inline unsigned int | |
34 | optlen(const u_int8_t *opt, unsigned int offset) | |
35 | { | |
36 | /* Beware zero-length options: make finite progress */ | |
37 | if (opt[offset] <= TCPOPT_NOP || opt[offset+1] == 0) | |
38 | return 1; | |
39 | else | |
40 | return opt[offset+1]; | |
41 | } | |
42 | ||
7722e0d1 G |
43 | static u_int32_t tcpmss_reverse_mtu(struct net *net, |
44 | const struct sk_buff *skb, | |
de1389b1 G |
45 | unsigned int family) |
46 | { | |
47 | struct flowi fl; | |
de1389b1 G |
48 | struct rtable *rt = NULL; |
49 | u_int32_t mtu = ~0U; | |
50 | ||
51 | if (family == PF_INET) { | |
52 | struct flowi4 *fl4 = &fl.u.ip4; | |
53 | memset(fl4, 0, sizeof(*fl4)); | |
54 | fl4->daddr = ip_hdr(skb)->saddr; | |
55 | } else { | |
56 | struct flowi6 *fl6 = &fl.u.ip6; | |
57 | ||
58 | memset(fl6, 0, sizeof(*fl6)); | |
59 | fl6->daddr = ipv6_hdr(skb)->saddr; | |
60 | } | |
de1389b1 | 61 | |
3f87c08c | 62 | nf_route(net, (struct dst_entry **)&rt, &fl, false, family); |
de1389b1 G |
63 | if (rt != NULL) { |
64 | mtu = dst_mtu(&rt->dst); | |
65 | dst_release(&rt->dst); | |
66 | } | |
67 | return mtu; | |
68 | } | |
69 | ||
cdd289a2 | 70 | static int |
3db05fea | 71 | tcpmss_mangle_packet(struct sk_buff *skb, |
70d19f80 | 72 | const struct xt_action_param *par, |
de1389b1 | 73 | unsigned int family, |
cdd289a2 PM |
74 | unsigned int tcphoff, |
75 | unsigned int minlen) | |
76 | { | |
70d19f80 | 77 | const struct xt_tcpmss_info *info = par->targinfo; |
cdd289a2 | 78 | struct tcphdr *tcph; |
71ffe9c7 PNA |
79 | int len, tcp_hdrlen; |
80 | unsigned int i; | |
cdd289a2 PM |
81 | __be16 oldval; |
82 | u16 newmss; | |
83 | u8 *opt; | |
84 | ||
b396966c PO |
85 | /* This is a fragment, no TCP header is available */ |
86 | if (par->fragoff != 0) | |
1205e1fa | 87 | return 0; |
b396966c | 88 | |
fb2eb1c1 | 89 | if (skb_ensure_writable(skb, skb->len)) |
cdd289a2 PM |
90 | return -1; |
91 | ||
71ffe9c7 PNA |
92 | len = skb->len - tcphoff; |
93 | if (len < (int)sizeof(struct tcphdr)) | |
94 | return -1; | |
95 | ||
3db05fea | 96 | tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff); |
71ffe9c7 | 97 | tcp_hdrlen = tcph->doff * 4; |
cdd289a2 | 98 | |
2638fd0f | 99 | if (len < tcp_hdrlen || tcp_hdrlen < sizeof(struct tcphdr)) |
cdd289a2 | 100 | return -1; |
cdd289a2 PM |
101 | |
102 | if (info->mss == XT_TCPMSS_CLAMP_PMTU) { | |
613dbd95 | 103 | struct net *net = xt_net(par); |
7722e0d1 | 104 | unsigned int in_mtu = tcpmss_reverse_mtu(net, skb, family); |
50f4c7b7 | 105 | unsigned int min_mtu = min(dst_mtu(skb_dst(skb)), in_mtu); |
de1389b1 | 106 | |
50f4c7b7 | 107 | if (min_mtu <= minlen) { |
e87cc472 | 108 | net_err_ratelimited("unknown or invalid path-MTU (%u)\n", |
50f4c7b7 | 109 | min_mtu); |
cdd289a2 PM |
110 | return -1; |
111 | } | |
50f4c7b7 | 112 | newmss = min_mtu - minlen; |
cdd289a2 PM |
113 | } else |
114 | newmss = info->mss; | |
115 | ||
116 | opt = (u_int8_t *)tcph; | |
71ffe9c7 PNA |
117 | for (i = sizeof(struct tcphdr); i <= tcp_hdrlen - TCPOLEN_MSS; i += optlen(opt, i)) { |
118 | if (opt[i] == TCPOPT_MSS && opt[i+1] == TCPOLEN_MSS) { | |
cdd289a2 PM |
119 | u_int16_t oldmss; |
120 | ||
121 | oldmss = (opt[i+2] << 8) | opt[i+3]; | |
122 | ||
17008064 BL |
123 | /* Never increase MSS, even when setting it, as |
124 | * doing so results in problems for hosts that rely | |
125 | * on MSS being set correctly. | |
126 | */ | |
127 | if (oldmss <= newmss) | |
cdd289a2 PM |
128 | return 0; |
129 | ||
130 | opt[i+2] = (newmss & 0xff00) >> 8; | |
7c4e36bc | 131 | opt[i+3] = newmss & 0x00ff; |
cdd289a2 | 132 | |
be0ea7d5 PM |
133 | inet_proto_csum_replace2(&tcph->check, skb, |
134 | htons(oldmss), htons(newmss), | |
4b048d6d | 135 | false); |
cdd289a2 PM |
136 | return 0; |
137 | } | |
138 | } | |
139 | ||
10a19939 | 140 | /* There is data after the header so the option can't be added |
71ffe9c7 PNA |
141 | * without moving it, and doing so may make the SYN packet |
142 | * itself too large. Accept the packet unmodified instead. | |
143 | */ | |
144 | if (len > tcp_hdrlen) | |
10a19939 | 145 | return 0; |
2638fd0f ED |
146 | |
147 | /* tcph->doff has 4 bits, do not wrap it to 0 */ | |
148 | if (tcp_hdrlen >= 15 * 4) | |
149 | return 0; | |
10a19939 | 150 | |
cdd289a2 PM |
151 | /* |
152 | * MSS Option not found ?! add it.. | |
153 | */ | |
3db05fea HX |
154 | if (skb_tailroom(skb) < TCPOLEN_MSS) { |
155 | if (pskb_expand_head(skb, 0, | |
156 | TCPOLEN_MSS - skb_tailroom(skb), | |
2ca7b0ac | 157 | GFP_ATOMIC)) |
cdd289a2 | 158 | return -1; |
3db05fea | 159 | tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff); |
cdd289a2 PM |
160 | } |
161 | ||
3db05fea | 162 | skb_put(skb, TCPOLEN_MSS); |
cdd289a2 | 163 | |
70d19f80 PO |
164 | /* |
165 | * IPv4: RFC 1122 states "If an MSS option is not received at | |
166 | * connection setup, TCP MUST assume a default send MSS of 536". | |
167 | * IPv6: RFC 2460 states IPv6 has a minimum MTU of 1280 and a minimum | |
168 | * length IPv6 header of 60, ergo the default MSS value is 1220 | |
169 | * Since no MSS was provided, we must use the default values | |
409b545a | 170 | */ |
613dbd95 | 171 | if (xt_family(par) == NFPROTO_IPV4) |
70d19f80 PO |
172 | newmss = min(newmss, (u16)536); |
173 | else | |
174 | newmss = min(newmss, (u16)1220); | |
409b545a | 175 | |
cdd289a2 | 176 | opt = (u_int8_t *)tcph + sizeof(struct tcphdr); |
71ffe9c7 | 177 | memmove(opt + TCPOLEN_MSS, opt, len - sizeof(struct tcphdr)); |
cdd289a2 | 178 | |
be0ea7d5 | 179 | inet_proto_csum_replace2(&tcph->check, skb, |
4b048d6d | 180 | htons(len), htons(len + TCPOLEN_MSS), true); |
cdd289a2 PM |
181 | opt[0] = TCPOPT_MSS; |
182 | opt[1] = TCPOLEN_MSS; | |
183 | opt[2] = (newmss & 0xff00) >> 8; | |
7c4e36bc | 184 | opt[3] = newmss & 0x00ff; |
cdd289a2 | 185 | |
4b048d6d | 186 | inet_proto_csum_replace4(&tcph->check, skb, 0, *((__be32 *)opt), false); |
cdd289a2 PM |
187 | |
188 | oldval = ((__be16 *)tcph)[6]; | |
189 | tcph->doff += TCPOLEN_MSS/4; | |
be0ea7d5 | 190 | inet_proto_csum_replace2(&tcph->check, skb, |
4b048d6d | 191 | oldval, ((__be16 *)tcph)[6], false); |
cdd289a2 PM |
192 | return TCPOLEN_MSS; |
193 | } | |
194 | ||
195 | static unsigned int | |
4b560b44 | 196 | tcpmss_tg4(struct sk_buff *skb, const struct xt_action_param *par) |
cdd289a2 | 197 | { |
3db05fea | 198 | struct iphdr *iph = ip_hdr(skb); |
cdd289a2 PM |
199 | __be16 newlen; |
200 | int ret; | |
201 | ||
70d19f80 | 202 | ret = tcpmss_mangle_packet(skb, par, |
de1389b1 | 203 | PF_INET, |
37c08387 | 204 | iph->ihl * 4, |
cdd289a2 PM |
205 | sizeof(*iph) + sizeof(struct tcphdr)); |
206 | if (ret < 0) | |
207 | return NF_DROP; | |
208 | if (ret > 0) { | |
3db05fea | 209 | iph = ip_hdr(skb); |
cdd289a2 | 210 | newlen = htons(ntohs(iph->tot_len) + ret); |
be0ea7d5 | 211 | csum_replace2(&iph->check, iph->tot_len, newlen); |
cdd289a2 PM |
212 | iph->tot_len = newlen; |
213 | } | |
214 | return XT_CONTINUE; | |
215 | } | |
216 | ||
c0cd1156 | 217 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
cdd289a2 | 218 | static unsigned int |
4b560b44 | 219 | tcpmss_tg6(struct sk_buff *skb, const struct xt_action_param *par) |
cdd289a2 | 220 | { |
3db05fea | 221 | struct ipv6hdr *ipv6h = ipv6_hdr(skb); |
cdd289a2 | 222 | u8 nexthdr; |
d6b3347b | 223 | __be16 frag_off, oldlen, newlen; |
cdd289a2 PM |
224 | int tcphoff; |
225 | int ret; | |
226 | ||
227 | nexthdr = ipv6h->nexthdr; | |
75f2811c | 228 | tcphoff = ipv6_skip_exthdr(skb, sizeof(*ipv6h), &nexthdr, &frag_off); |
9dc0564e | 229 | if (tcphoff < 0) |
cdd289a2 | 230 | return NF_DROP; |
70d19f80 | 231 | ret = tcpmss_mangle_packet(skb, par, |
de1389b1 | 232 | PF_INET6, |
37c08387 | 233 | tcphoff, |
cdd289a2 PM |
234 | sizeof(*ipv6h) + sizeof(struct tcphdr)); |
235 | if (ret < 0) | |
236 | return NF_DROP; | |
237 | if (ret > 0) { | |
3db05fea | 238 | ipv6h = ipv6_hdr(skb); |
d6b3347b ED |
239 | oldlen = ipv6h->payload_len; |
240 | newlen = htons(ntohs(oldlen) + ret); | |
241 | if (skb->ip_summed == CHECKSUM_COMPLETE) | |
242 | skb->csum = csum_add(csum_sub(skb->csum, oldlen), | |
243 | newlen); | |
244 | ipv6h->payload_len = newlen; | |
cdd289a2 PM |
245 | } |
246 | return XT_CONTINUE; | |
247 | } | |
248 | #endif | |
249 | ||
cdd289a2 | 250 | /* Must specify -p tcp --syn */ |
e1931b78 | 251 | static inline bool find_syn_match(const struct xt_entry_match *m) |
cdd289a2 PM |
252 | { |
253 | const struct xt_tcp *tcpinfo = (const struct xt_tcp *)m->data; | |
254 | ||
255 | if (strcmp(m->u.kernel.match->name, "tcp") == 0 && | |
a3433f35 | 256 | tcpinfo->flg_cmp & TCPHDR_SYN && |
cdd289a2 | 257 | !(tcpinfo->invflags & XT_TCP_INV_FLAGS)) |
e1931b78 | 258 | return true; |
cdd289a2 | 259 | |
e1931b78 | 260 | return false; |
cdd289a2 PM |
261 | } |
262 | ||
135367b8 | 263 | static int tcpmss_tg4_check(const struct xt_tgchk_param *par) |
cdd289a2 | 264 | { |
af5d6dc2 JE |
265 | const struct xt_tcpmss_info *info = par->targinfo; |
266 | const struct ipt_entry *e = par->entryinfo; | |
dcea992a | 267 | const struct xt_entry_match *ematch; |
cdd289a2 PM |
268 | |
269 | if (info->mss == XT_TCPMSS_CLAMP_PMTU && | |
af5d6dc2 | 270 | (par->hook_mask & ~((1 << NF_INET_FORWARD) | |
6e23ae2a PM |
271 | (1 << NF_INET_LOCAL_OUT) | |
272 | (1 << NF_INET_POST_ROUTING))) != 0) { | |
b2606644 | 273 | pr_info_ratelimited("path-MTU clamping only supported in FORWARD, OUTPUT and POSTROUTING hooks\n"); |
d6b00a53 | 274 | return -EINVAL; |
cdd289a2 | 275 | } |
55917a21 PNA |
276 | if (par->nft_compat) |
277 | return 0; | |
278 | ||
dcea992a JE |
279 | xt_ematch_foreach(ematch, e) |
280 | if (find_syn_match(ematch)) | |
d6b00a53 | 281 | return 0; |
b2606644 | 282 | pr_info_ratelimited("Only works on TCP SYN packets\n"); |
d6b00a53 | 283 | return -EINVAL; |
cdd289a2 PM |
284 | } |
285 | ||
c0cd1156 | 286 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
135367b8 | 287 | static int tcpmss_tg6_check(const struct xt_tgchk_param *par) |
cdd289a2 | 288 | { |
af5d6dc2 JE |
289 | const struct xt_tcpmss_info *info = par->targinfo; |
290 | const struct ip6t_entry *e = par->entryinfo; | |
dcea992a | 291 | const struct xt_entry_match *ematch; |
cdd289a2 PM |
292 | |
293 | if (info->mss == XT_TCPMSS_CLAMP_PMTU && | |
af5d6dc2 | 294 | (par->hook_mask & ~((1 << NF_INET_FORWARD) | |
6e23ae2a PM |
295 | (1 << NF_INET_LOCAL_OUT) | |
296 | (1 << NF_INET_POST_ROUTING))) != 0) { | |
b2606644 | 297 | pr_info_ratelimited("path-MTU clamping only supported in FORWARD, OUTPUT and POSTROUTING hooks\n"); |
d6b00a53 | 298 | return -EINVAL; |
cdd289a2 | 299 | } |
55917a21 PNA |
300 | if (par->nft_compat) |
301 | return 0; | |
302 | ||
dcea992a JE |
303 | xt_ematch_foreach(ematch, e) |
304 | if (find_syn_match(ematch)) | |
d6b00a53 | 305 | return 0; |
b2606644 | 306 | pr_info_ratelimited("Only works on TCP SYN packets\n"); |
d6b00a53 | 307 | return -EINVAL; |
cdd289a2 PM |
308 | } |
309 | #endif | |
310 | ||
d3c5ee6d | 311 | static struct xt_target tcpmss_tg_reg[] __read_mostly = { |
cdd289a2 | 312 | { |
ee999d8b | 313 | .family = NFPROTO_IPV4, |
cdd289a2 | 314 | .name = "TCPMSS", |
d3c5ee6d JE |
315 | .checkentry = tcpmss_tg4_check, |
316 | .target = tcpmss_tg4, | |
cdd289a2 PM |
317 | .targetsize = sizeof(struct xt_tcpmss_info), |
318 | .proto = IPPROTO_TCP, | |
319 | .me = THIS_MODULE, | |
320 | }, | |
c0cd1156 | 321 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
cdd289a2 | 322 | { |
ee999d8b | 323 | .family = NFPROTO_IPV6, |
cdd289a2 | 324 | .name = "TCPMSS", |
d3c5ee6d JE |
325 | .checkentry = tcpmss_tg6_check, |
326 | .target = tcpmss_tg6, | |
cdd289a2 PM |
327 | .targetsize = sizeof(struct xt_tcpmss_info), |
328 | .proto = IPPROTO_TCP, | |
329 | .me = THIS_MODULE, | |
330 | }, | |
331 | #endif | |
332 | }; | |
333 | ||
d3c5ee6d | 334 | static int __init tcpmss_tg_init(void) |
cdd289a2 | 335 | { |
d3c5ee6d | 336 | return xt_register_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg)); |
cdd289a2 PM |
337 | } |
338 | ||
d3c5ee6d | 339 | static void __exit tcpmss_tg_exit(void) |
cdd289a2 | 340 | { |
d3c5ee6d | 341 | xt_unregister_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg)); |
cdd289a2 PM |
342 | } |
343 | ||
d3c5ee6d JE |
344 | module_init(tcpmss_tg_init); |
345 | module_exit(tcpmss_tg_exit); |