1 | // SPDX-License-Identifier: GPL-2.0-only |
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> |
6 | * Copyright (C) 2007 Patrick McHardy <kaber@trash.net> |
7 | */ |
8 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
9 | #include <linux/module.h> |
10 | #include <linux/skbuff.h> |
11 | #include <linux/ip.h> |
12 | #include <linux/gfp.h> |
13 | #include <linux/ipv6.h> |
14 | #include <linux/tcp.h> |
15 | #include <net/dst.h> |
16 | #include <net/flow.h> |
17 | #include <net/ipv6.h> |
18 | #include <net/route.h> |
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>" ); |
29 | MODULE_DESCRIPTION("Xtables: TCP Maximum Segment Size (MSS) adjustment" ); |
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 | |
43 | static u_int32_t tcpmss_reverse_mtu(struct net *net, |
44 | const struct sk_buff *skb, |
45 | unsigned int family) |
46 | { |
47 | struct flowi fl; |
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 | } |
61 | |
62 | nf_route(net, dst: (struct dst_entry **)&rt, fl: &fl, strict: false, family); |
63 | if (rt != NULL) { |
64 | mtu = dst_mtu(dst: &rt->dst); |
65 | dst_release(dst: &rt->dst); |
66 | } |
67 | return mtu; |
68 | } |
69 | |
70 | static int |
71 | tcpmss_mangle_packet(struct sk_buff *skb, |
72 | const struct xt_action_param *par, |
73 | unsigned int family, |
74 | unsigned int tcphoff, |
75 | unsigned int minlen) |
76 | { |
77 | const struct xt_tcpmss_info *info = par->targinfo; |
78 | struct tcphdr *tcph; |
79 | int len, tcp_hdrlen; |
80 | unsigned int i; |
81 | __be16 oldval; |
82 | u16 newmss; |
83 | u8 *opt; |
84 | |
85 | /* This is a fragment, no TCP header is available */ |
86 | if (par->fragoff != 0) |
87 | return 0; |
88 | |
89 | if (skb_ensure_writable(skb, write_len: skb->len)) |
90 | return -1; |
91 | |
92 | len = skb->len - tcphoff; |
93 | if (len < (int)sizeof(struct tcphdr)) |
94 | return -1; |
95 | |
96 | tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff); |
97 | tcp_hdrlen = tcph->doff * 4; |
98 | |
99 | if (len < tcp_hdrlen || tcp_hdrlen < sizeof(struct tcphdr)) |
100 | return -1; |
101 | |
102 | if (info->mss == XT_TCPMSS_CLAMP_PMTU) { |
103 | struct net *net = xt_net(par); |
104 | unsigned int in_mtu = tcpmss_reverse_mtu(net, skb, family); |
105 | unsigned int min_mtu = min(dst_mtu(skb_dst(skb)), in_mtu); |
106 | |
107 | if (min_mtu <= minlen) { |
108 | net_err_ratelimited("unknown or invalid path-MTU (%u)\n" , |
109 | min_mtu); |
110 | return -1; |
111 | } |
112 | newmss = min_mtu - minlen; |
113 | } else |
114 | newmss = info->mss; |
115 | |
116 | opt = (u_int8_t *)tcph; |
117 | for (i = sizeof(struct tcphdr); i <= tcp_hdrlen - TCPOLEN_MSS; i += optlen(opt, offset: i)) { |
118 | if (opt[i] == TCPOPT_MSS && opt[i+1] == TCPOLEN_MSS) { |
119 | u_int16_t oldmss; |
120 | |
121 | oldmss = (opt[i+2] << 8) | opt[i+3]; |
122 | |
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) |
128 | return 0; |
129 | |
130 | opt[i+2] = (newmss & 0xff00) >> 8; |
131 | opt[i+3] = newmss & 0x00ff; |
132 | |
133 | inet_proto_csum_replace2(sum: &tcph->check, skb, |
134 | htons(oldmss), htons(newmss), |
135 | pseudohdr: false); |
136 | return 0; |
137 | } |
138 | } |
139 | |
140 | /* There is data after the header so the option can't be added |
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) |
145 | return 0; |
146 | |
147 | /* tcph->doff has 4 bits, do not wrap it to 0 */ |
148 | if (tcp_hdrlen >= 15 * 4) |
149 | return 0; |
150 | |
151 | /* |
152 | * MSS Option not found ?! add it.. |
153 | */ |
154 | if (skb_tailroom(skb) < TCPOLEN_MSS) { |
155 | if (pskb_expand_head(skb, nhead: 0, |
156 | TCPOLEN_MSS - skb_tailroom(skb), |
157 | GFP_ATOMIC)) |
158 | return -1; |
159 | tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff); |
160 | } |
161 | |
162 | skb_put(skb, TCPOLEN_MSS); |
163 | |
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 |
170 | */ |
171 | if (xt_family(par) == NFPROTO_IPV4) |
172 | newmss = min(newmss, (u16)536); |
173 | else |
174 | newmss = min(newmss, (u16)1220); |
175 | |
176 | opt = (u_int8_t *)tcph + sizeof(struct tcphdr); |
177 | memmove(opt + TCPOLEN_MSS, opt, len - sizeof(struct tcphdr)); |
178 | |
179 | inet_proto_csum_replace2(sum: &tcph->check, skb, |
180 | htons(len), htons(len + TCPOLEN_MSS), pseudohdr: true); |
181 | opt[0] = TCPOPT_MSS; |
182 | opt[1] = TCPOLEN_MSS; |
183 | opt[2] = (newmss & 0xff00) >> 8; |
184 | opt[3] = newmss & 0x00ff; |
185 | |
186 | inet_proto_csum_replace4(sum: &tcph->check, skb, from: 0, to: *((__be32 *)opt), pseudohdr: false); |
187 | |
188 | oldval = ((__be16 *)tcph)[6]; |
189 | tcph->doff += TCPOLEN_MSS/4; |
190 | inet_proto_csum_replace2(sum: &tcph->check, skb, |
191 | from: oldval, to: ((__be16 *)tcph)[6], pseudohdr: false); |
192 | return TCPOLEN_MSS; |
193 | } |
194 | |
195 | static unsigned int |
196 | tcpmss_tg4(struct sk_buff *skb, const struct xt_action_param *par) |
197 | { |
198 | struct iphdr *iph = ip_hdr(skb); |
199 | __be16 newlen; |
200 | int ret; |
201 | |
202 | ret = tcpmss_mangle_packet(skb, par, |
203 | PF_INET, |
204 | tcphoff: iph->ihl * 4, |
205 | minlen: sizeof(*iph) + sizeof(struct tcphdr)); |
206 | if (ret < 0) |
207 | return NF_DROP; |
208 | if (ret > 0) { |
209 | iph = ip_hdr(skb); |
210 | newlen = htons(ntohs(iph->tot_len) + ret); |
211 | csum_replace2(sum: &iph->check, old: iph->tot_len, new: newlen); |
212 | iph->tot_len = newlen; |
213 | } |
214 | return XT_CONTINUE; |
215 | } |
216 | |
217 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
218 | static unsigned int |
219 | tcpmss_tg6(struct sk_buff *skb, const struct xt_action_param *par) |
220 | { |
221 | struct ipv6hdr *ipv6h = ipv6_hdr(skb); |
222 | u8 nexthdr; |
223 | __be16 frag_off, oldlen, newlen; |
224 | int tcphoff; |
225 | int ret; |
226 | |
227 | nexthdr = ipv6h->nexthdr; |
228 | tcphoff = ipv6_skip_exthdr(skb, start: sizeof(*ipv6h), nexthdrp: &nexthdr, frag_offp: &frag_off); |
229 | if (tcphoff < 0) |
230 | return NF_DROP; |
231 | ret = tcpmss_mangle_packet(skb, par, |
232 | PF_INET6, |
233 | tcphoff, |
234 | minlen: sizeof(*ipv6h) + sizeof(struct tcphdr)); |
235 | if (ret < 0) |
236 | return NF_DROP; |
237 | if (ret > 0) { |
238 | ipv6h = ipv6_hdr(skb); |
239 | oldlen = ipv6h->payload_len; |
240 | newlen = htons(ntohs(oldlen) + ret); |
241 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
242 | skb->csum = csum_add(csum: csum_sub(csum: skb->csum, addend: (__force __wsum)oldlen), |
243 | addend: (__force __wsum)newlen); |
244 | ipv6h->payload_len = newlen; |
245 | } |
246 | return XT_CONTINUE; |
247 | } |
248 | #endif |
249 | |
250 | /* Must specify -p tcp --syn */ |
251 | static inline bool find_syn_match(const struct xt_entry_match *m) |
252 | { |
253 | const struct xt_tcp *tcpinfo = (const struct xt_tcp *)m->data; |
254 | |
255 | if (strcmp(m->u.kernel.match->name, "tcp" ) == 0 && |
256 | tcpinfo->flg_cmp & TCPHDR_SYN && |
257 | !(tcpinfo->invflags & XT_TCP_INV_FLAGS)) |
258 | return true; |
259 | |
260 | return false; |
261 | } |
262 | |
263 | static int tcpmss_tg4_check(const struct xt_tgchk_param *par) |
264 | { |
265 | const struct xt_tcpmss_info *info = par->targinfo; |
266 | const struct ipt_entry *e = par->entryinfo; |
267 | const struct xt_entry_match *ematch; |
268 | |
269 | if (info->mss == XT_TCPMSS_CLAMP_PMTU && |
270 | (par->hook_mask & ~((1 << NF_INET_FORWARD) | |
271 | (1 << NF_INET_LOCAL_OUT) | |
272 | (1 << NF_INET_POST_ROUTING))) != 0) { |
273 | pr_info_ratelimited("path-MTU clamping only supported in FORWARD, OUTPUT and POSTROUTING hooks\n" ); |
274 | return -EINVAL; |
275 | } |
276 | if (par->nft_compat) |
277 | return 0; |
278 | |
279 | xt_ematch_foreach(ematch, e) |
280 | if (find_syn_match(m: ematch)) |
281 | return 0; |
282 | pr_info_ratelimited("Only works on TCP SYN packets\n" ); |
283 | return -EINVAL; |
284 | } |
285 | |
286 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
287 | static int tcpmss_tg6_check(const struct xt_tgchk_param *par) |
288 | { |
289 | const struct xt_tcpmss_info *info = par->targinfo; |
290 | const struct ip6t_entry *e = par->entryinfo; |
291 | const struct xt_entry_match *ematch; |
292 | |
293 | if (info->mss == XT_TCPMSS_CLAMP_PMTU && |
294 | (par->hook_mask & ~((1 << NF_INET_FORWARD) | |
295 | (1 << NF_INET_LOCAL_OUT) | |
296 | (1 << NF_INET_POST_ROUTING))) != 0) { |
297 | pr_info_ratelimited("path-MTU clamping only supported in FORWARD, OUTPUT and POSTROUTING hooks\n" ); |
298 | return -EINVAL; |
299 | } |
300 | if (par->nft_compat) |
301 | return 0; |
302 | |
303 | xt_ematch_foreach(ematch, e) |
304 | if (find_syn_match(m: ematch)) |
305 | return 0; |
306 | pr_info_ratelimited("Only works on TCP SYN packets\n" ); |
307 | return -EINVAL; |
308 | } |
309 | #endif |
310 | |
311 | static struct xt_target tcpmss_tg_reg[] __read_mostly = { |
312 | { |
313 | .family = NFPROTO_IPV4, |
314 | .name = "TCPMSS" , |
315 | .checkentry = tcpmss_tg4_check, |
316 | .target = tcpmss_tg4, |
317 | .targetsize = sizeof(struct xt_tcpmss_info), |
318 | .proto = IPPROTO_TCP, |
319 | .me = THIS_MODULE, |
320 | }, |
321 | #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) |
322 | { |
323 | .family = NFPROTO_IPV6, |
324 | .name = "TCPMSS" , |
325 | .checkentry = tcpmss_tg6_check, |
326 | .target = tcpmss_tg6, |
327 | .targetsize = sizeof(struct xt_tcpmss_info), |
328 | .proto = IPPROTO_TCP, |
329 | .me = THIS_MODULE, |
330 | }, |
331 | #endif |
332 | }; |
333 | |
334 | static int __init tcpmss_tg_init(void) |
335 | { |
336 | return xt_register_targets(target: tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg)); |
337 | } |
338 | |
339 | static void __exit tcpmss_tg_exit(void) |
340 | { |
341 | xt_unregister_targets(target: tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg)); |
342 | } |
343 | |
344 | module_init(tcpmss_tg_init); |
345 | module_exit(tcpmss_tg_exit); |
346 | |