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