]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/netfilter/xt_TCPMSS.c
netfilter: nf_tables: use WARN_ON_ONCE instead of BUG_ON in nft_do_chain()
[mirror_ubuntu-bionic-kernel.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;
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 }
de1389b1
G
65 ai = nf_get_afinfo(family);
66 if (ai != NULL)
7722e0d1 67 ai->route(net, (struct dst_entry **)&rt, &fl, false);
de1389b1
G
68
69 if (rt != NULL) {
70 mtu = dst_mtu(&rt->dst);
71 dst_release(&rt->dst);
72 }
73 return mtu;
74}
75
cdd289a2 76static int
3db05fea 77tcpmss_mangle_packet(struct sk_buff *skb,
70d19f80 78 const struct xt_action_param *par,
de1389b1 79 unsigned int family,
cdd289a2
PM
80 unsigned int tcphoff,
81 unsigned int minlen)
82{
70d19f80 83 const struct xt_tcpmss_info *info = par->targinfo;
cdd289a2 84 struct tcphdr *tcph;
71ffe9c7
PNA
85 int len, tcp_hdrlen;
86 unsigned int i;
cdd289a2
PM
87 __be16 oldval;
88 u16 newmss;
89 u8 *opt;
90
b396966c
PO
91 /* This is a fragment, no TCP header is available */
92 if (par->fragoff != 0)
1205e1fa 93 return 0;
b396966c 94
3db05fea 95 if (!skb_make_writable(skb, skb->len))
cdd289a2
PM
96 return -1;
97
71ffe9c7
PNA
98 len = skb->len - tcphoff;
99 if (len < (int)sizeof(struct tcphdr))
100 return -1;
101
3db05fea 102 tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
71ffe9c7 103 tcp_hdrlen = tcph->doff * 4;
cdd289a2 104
2638fd0f 105 if (len < tcp_hdrlen || tcp_hdrlen < sizeof(struct tcphdr))
cdd289a2 106 return -1;
cdd289a2
PM
107
108 if (info->mss == XT_TCPMSS_CLAMP_PMTU) {
613dbd95 109 struct net *net = xt_net(par);
7722e0d1 110 unsigned int in_mtu = tcpmss_reverse_mtu(net, skb, family);
50f4c7b7 111 unsigned int min_mtu = min(dst_mtu(skb_dst(skb)), in_mtu);
de1389b1 112
50f4c7b7 113 if (min_mtu <= minlen) {
e87cc472 114 net_err_ratelimited("unknown or invalid path-MTU (%u)\n",
50f4c7b7 115 min_mtu);
cdd289a2
PM
116 return -1;
117 }
50f4c7b7 118 newmss = min_mtu - minlen;
cdd289a2
PM
119 } else
120 newmss = info->mss;
121
122 opt = (u_int8_t *)tcph;
71ffe9c7
PNA
123 for (i = sizeof(struct tcphdr); i <= tcp_hdrlen - TCPOLEN_MSS; i += optlen(opt, i)) {
124 if (opt[i] == TCPOPT_MSS && opt[i+1] == TCPOLEN_MSS) {
cdd289a2
PM
125 u_int16_t oldmss;
126
127 oldmss = (opt[i+2] << 8) | opt[i+3];
128
17008064
BL
129 /* Never increase MSS, even when setting it, as
130 * doing so results in problems for hosts that rely
131 * on MSS being set correctly.
132 */
133 if (oldmss <= newmss)
cdd289a2
PM
134 return 0;
135
136 opt[i+2] = (newmss & 0xff00) >> 8;
7c4e36bc 137 opt[i+3] = newmss & 0x00ff;
cdd289a2 138
be0ea7d5
PM
139 inet_proto_csum_replace2(&tcph->check, skb,
140 htons(oldmss), htons(newmss),
4b048d6d 141 false);
cdd289a2
PM
142 return 0;
143 }
144 }
145
10a19939 146 /* There is data after the header so the option can't be added
71ffe9c7
PNA
147 * without moving it, and doing so may make the SYN packet
148 * itself too large. Accept the packet unmodified instead.
149 */
150 if (len > tcp_hdrlen)
10a19939 151 return 0;
2638fd0f
ED
152
153 /* tcph->doff has 4 bits, do not wrap it to 0 */
154 if (tcp_hdrlen >= 15 * 4)
155 return 0;
10a19939 156
cdd289a2
PM
157 /*
158 * MSS Option not found ?! add it..
159 */
3db05fea
HX
160 if (skb_tailroom(skb) < TCPOLEN_MSS) {
161 if (pskb_expand_head(skb, 0,
162 TCPOLEN_MSS - skb_tailroom(skb),
2ca7b0ac 163 GFP_ATOMIC))
cdd289a2 164 return -1;
3db05fea 165 tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
cdd289a2
PM
166 }
167
3db05fea 168 skb_put(skb, TCPOLEN_MSS);
cdd289a2 169
70d19f80
PO
170 /*
171 * IPv4: RFC 1122 states "If an MSS option is not received at
172 * connection setup, TCP MUST assume a default send MSS of 536".
173 * IPv6: RFC 2460 states IPv6 has a minimum MTU of 1280 and a minimum
174 * length IPv6 header of 60, ergo the default MSS value is 1220
175 * Since no MSS was provided, we must use the default values
409b545a 176 */
613dbd95 177 if (xt_family(par) == NFPROTO_IPV4)
70d19f80
PO
178 newmss = min(newmss, (u16)536);
179 else
180 newmss = min(newmss, (u16)1220);
409b545a 181
cdd289a2 182 opt = (u_int8_t *)tcph + sizeof(struct tcphdr);
71ffe9c7 183 memmove(opt + TCPOLEN_MSS, opt, len - sizeof(struct tcphdr));
cdd289a2 184
be0ea7d5 185 inet_proto_csum_replace2(&tcph->check, skb,
4b048d6d 186 htons(len), htons(len + TCPOLEN_MSS), true);
cdd289a2
PM
187 opt[0] = TCPOPT_MSS;
188 opt[1] = TCPOLEN_MSS;
189 opt[2] = (newmss & 0xff00) >> 8;
7c4e36bc 190 opt[3] = newmss & 0x00ff;
cdd289a2 191
4b048d6d 192 inet_proto_csum_replace4(&tcph->check, skb, 0, *((__be32 *)opt), false);
cdd289a2
PM
193
194 oldval = ((__be16 *)tcph)[6];
195 tcph->doff += TCPOLEN_MSS/4;
be0ea7d5 196 inet_proto_csum_replace2(&tcph->check, skb,
4b048d6d 197 oldval, ((__be16 *)tcph)[6], false);
cdd289a2
PM
198 return TCPOLEN_MSS;
199}
200
201static unsigned int
4b560b44 202tcpmss_tg4(struct sk_buff *skb, const struct xt_action_param *par)
cdd289a2 203{
3db05fea 204 struct iphdr *iph = ip_hdr(skb);
cdd289a2
PM
205 __be16 newlen;
206 int ret;
207
70d19f80 208 ret = tcpmss_mangle_packet(skb, par,
de1389b1 209 PF_INET,
37c08387 210 iph->ihl * 4,
cdd289a2
PM
211 sizeof(*iph) + sizeof(struct tcphdr));
212 if (ret < 0)
213 return NF_DROP;
214 if (ret > 0) {
3db05fea 215 iph = ip_hdr(skb);
cdd289a2 216 newlen = htons(ntohs(iph->tot_len) + ret);
be0ea7d5 217 csum_replace2(&iph->check, iph->tot_len, newlen);
cdd289a2
PM
218 iph->tot_len = newlen;
219 }
220 return XT_CONTINUE;
221}
222
c0cd1156 223#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
cdd289a2 224static unsigned int
4b560b44 225tcpmss_tg6(struct sk_buff *skb, const struct xt_action_param *par)
cdd289a2 226{
3db05fea 227 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
cdd289a2 228 u8 nexthdr;
d6b3347b 229 __be16 frag_off, oldlen, newlen;
cdd289a2
PM
230 int tcphoff;
231 int ret;
232
233 nexthdr = ipv6h->nexthdr;
75f2811c 234 tcphoff = ipv6_skip_exthdr(skb, sizeof(*ipv6h), &nexthdr, &frag_off);
9dc0564e 235 if (tcphoff < 0)
cdd289a2 236 return NF_DROP;
70d19f80 237 ret = tcpmss_mangle_packet(skb, par,
de1389b1 238 PF_INET6,
37c08387 239 tcphoff,
cdd289a2
PM
240 sizeof(*ipv6h) + sizeof(struct tcphdr));
241 if (ret < 0)
242 return NF_DROP;
243 if (ret > 0) {
3db05fea 244 ipv6h = ipv6_hdr(skb);
d6b3347b
ED
245 oldlen = ipv6h->payload_len;
246 newlen = htons(ntohs(oldlen) + ret);
247 if (skb->ip_summed == CHECKSUM_COMPLETE)
248 skb->csum = csum_add(csum_sub(skb->csum, oldlen),
249 newlen);
250 ipv6h->payload_len = newlen;
cdd289a2
PM
251 }
252 return XT_CONTINUE;
253}
254#endif
255
cdd289a2 256/* Must specify -p tcp --syn */
e1931b78 257static inline bool find_syn_match(const struct xt_entry_match *m)
cdd289a2
PM
258{
259 const struct xt_tcp *tcpinfo = (const struct xt_tcp *)m->data;
260
261 if (strcmp(m->u.kernel.match->name, "tcp") == 0 &&
a3433f35 262 tcpinfo->flg_cmp & TCPHDR_SYN &&
cdd289a2 263 !(tcpinfo->invflags & XT_TCP_INV_FLAGS))
e1931b78 264 return true;
cdd289a2 265
e1931b78 266 return false;
cdd289a2
PM
267}
268
135367b8 269static int tcpmss_tg4_check(const struct xt_tgchk_param *par)
cdd289a2 270{
af5d6dc2
JE
271 const struct xt_tcpmss_info *info = par->targinfo;
272 const struct ipt_entry *e = par->entryinfo;
dcea992a 273 const struct xt_entry_match *ematch;
cdd289a2
PM
274
275 if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
af5d6dc2 276 (par->hook_mask & ~((1 << NF_INET_FORWARD) |
6e23ae2a
PM
277 (1 << NF_INET_LOCAL_OUT) |
278 (1 << NF_INET_POST_ROUTING))) != 0) {
8bee4bad
JE
279 pr_info("path-MTU clamping only supported in "
280 "FORWARD, OUTPUT and POSTROUTING hooks\n");
d6b00a53 281 return -EINVAL;
cdd289a2 282 }
55917a21
PNA
283 if (par->nft_compat)
284 return 0;
285
dcea992a
JE
286 xt_ematch_foreach(ematch, e)
287 if (find_syn_match(ematch))
d6b00a53 288 return 0;
8bee4bad 289 pr_info("Only works on TCP SYN packets\n");
d6b00a53 290 return -EINVAL;
cdd289a2
PM
291}
292
c0cd1156 293#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
135367b8 294static int tcpmss_tg6_check(const struct xt_tgchk_param *par)
cdd289a2 295{
af5d6dc2
JE
296 const struct xt_tcpmss_info *info = par->targinfo;
297 const struct ip6t_entry *e = par->entryinfo;
dcea992a 298 const struct xt_entry_match *ematch;
cdd289a2
PM
299
300 if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
af5d6dc2 301 (par->hook_mask & ~((1 << NF_INET_FORWARD) |
6e23ae2a
PM
302 (1 << NF_INET_LOCAL_OUT) |
303 (1 << NF_INET_POST_ROUTING))) != 0) {
8bee4bad
JE
304 pr_info("path-MTU clamping only supported in "
305 "FORWARD, OUTPUT and POSTROUTING hooks\n");
d6b00a53 306 return -EINVAL;
cdd289a2 307 }
55917a21
PNA
308 if (par->nft_compat)
309 return 0;
310
dcea992a
JE
311 xt_ematch_foreach(ematch, e)
312 if (find_syn_match(ematch))
d6b00a53 313 return 0;
8bee4bad 314 pr_info("Only works on TCP SYN packets\n");
d6b00a53 315 return -EINVAL;
cdd289a2
PM
316}
317#endif
318
d3c5ee6d 319static struct xt_target tcpmss_tg_reg[] __read_mostly = {
cdd289a2 320 {
ee999d8b 321 .family = NFPROTO_IPV4,
cdd289a2 322 .name = "TCPMSS",
d3c5ee6d
JE
323 .checkentry = tcpmss_tg4_check,
324 .target = tcpmss_tg4,
cdd289a2
PM
325 .targetsize = sizeof(struct xt_tcpmss_info),
326 .proto = IPPROTO_TCP,
327 .me = THIS_MODULE,
328 },
c0cd1156 329#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
cdd289a2 330 {
ee999d8b 331 .family = NFPROTO_IPV6,
cdd289a2 332 .name = "TCPMSS",
d3c5ee6d
JE
333 .checkentry = tcpmss_tg6_check,
334 .target = tcpmss_tg6,
cdd289a2
PM
335 .targetsize = sizeof(struct xt_tcpmss_info),
336 .proto = IPPROTO_TCP,
337 .me = THIS_MODULE,
338 },
339#endif
340};
341
d3c5ee6d 342static int __init tcpmss_tg_init(void)
cdd289a2 343{
d3c5ee6d 344 return xt_register_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
cdd289a2
PM
345}
346
d3c5ee6d 347static void __exit tcpmss_tg_exit(void)
cdd289a2 348{
d3c5ee6d 349 xt_unregister_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
cdd289a2
PM
350}
351
d3c5ee6d
JE
352module_init(tcpmss_tg_init);
353module_exit(tcpmss_tg_exit);