1 /* 2 * This is a module which is used for setting the MSS option in TCP packets. 3 * 4 * Copyright (C) 2000 Marc Boucher <[email protected]> 5 * Copyright (C) 2007 Patrick McHardy <[email protected]> 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 */ 11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 12 #include <linux/module.h> 13 #include <linux/skbuff.h> 14 #include <linux/ip.h> 15 #include <linux/gfp.h> 16 #include <linux/ipv6.h> 17 #include <linux/tcp.h> 18 #include <net/dst.h> 19 #include <net/flow.h> 20 #include <net/ipv6.h> 21 #include <net/route.h> 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 30 MODULE_LICENSE("GPL"); 31 MODULE_AUTHOR("Marc Boucher <[email protected]>"); 32 MODULE_DESCRIPTION("Xtables: TCP Maximum Segment Size (MSS) adjustment"); 33 MODULE_ALIAS("ipt_TCPMSS"); 34 MODULE_ALIAS("ip6t_TCPMSS"); 35 36 static inline unsigned int 37 optlen(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 46 static u_int32_t tcpmss_reverse_mtu(struct net *net, 47 const struct sk_buff *skb, 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) 68 ai->route(net, (struct dst_entry **)&rt, &fl, false); 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 78 static int 79 tcpmss_mangle_packet(struct sk_buff *skb, 80 const struct xt_action_param *par, 81 unsigned int family, 82 unsigned int tcphoff, 83 unsigned int minlen) 84 { 85 const struct xt_tcpmss_info *info = par->targinfo; 86 struct tcphdr *tcph; 87 int len, tcp_hdrlen; 88 unsigned int i; 89 __be16 oldval; 90 u16 newmss; 91 u8 *opt; 92 93 /* This is a fragment, no TCP header is available */ 94 if (par->fragoff != 0) 95 return 0; 96 97 if (!skb_make_writable(skb, skb->len)) 98 return -1; 99 100 len = skb->len - tcphoff; 101 if (len < (int)sizeof(struct tcphdr)) 102 return -1; 103 104 tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff); 105 tcp_hdrlen = tcph->doff * 4; 106 107 if (len < tcp_hdrlen) 108 return -1; 109 110 if (info->mss == XT_TCPMSS_CLAMP_PMTU) { 111 struct net *net = par->net; 112 unsigned int in_mtu = tcpmss_reverse_mtu(net, skb, family); 113 unsigned int min_mtu = min(dst_mtu(skb_dst(skb)), in_mtu); 114 115 if (min_mtu <= minlen) { 116 net_err_ratelimited("unknown or invalid path-MTU (%u)\n", 117 min_mtu); 118 return -1; 119 } 120 newmss = min_mtu - minlen; 121 } else 122 newmss = info->mss; 123 124 opt = (u_int8_t *)tcph; 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) { 127 u_int16_t oldmss; 128 129 oldmss = (opt[i+2] << 8) | opt[i+3]; 130 131 /* 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) 136 return 0; 137 138 opt[i+2] = (newmss & 0xff00) >> 8; 139 opt[i+3] = newmss & 0x00ff; 140 141 inet_proto_csum_replace2(&tcph->check, skb, 142 htons(oldmss), htons(newmss), 143 false); 144 return 0; 145 } 146 } 147 148 /* There is data after the header so the option can't be added 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) 153 return 0; 154 155 /* 156 * MSS Option not found ?! add it.. 157 */ 158 if (skb_tailroom(skb) < TCPOLEN_MSS) { 159 if (pskb_expand_head(skb, 0, 160 TCPOLEN_MSS - skb_tailroom(skb), 161 GFP_ATOMIC)) 162 return -1; 163 tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff); 164 } 165 166 skb_put(skb, TCPOLEN_MSS); 167 168 /* 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 174 */ 175 if (par->family == NFPROTO_IPV4) 176 newmss = min(newmss, (u16)536); 177 else 178 newmss = min(newmss, (u16)1220); 179 180 opt = (u_int8_t *)tcph + sizeof(struct tcphdr); 181 memmove(opt + TCPOLEN_MSS, opt, len - sizeof(struct tcphdr)); 182 183 inet_proto_csum_replace2(&tcph->check, skb, 184 htons(len), htons(len + TCPOLEN_MSS), true); 185 opt[0] = TCPOPT_MSS; 186 opt[1] = TCPOLEN_MSS; 187 opt[2] = (newmss & 0xff00) >> 8; 188 opt[3] = newmss & 0x00ff; 189 190 inet_proto_csum_replace4(&tcph->check, skb, 0, *((__be32 *)opt), false); 191 192 oldval = ((__be16 *)tcph)[6]; 193 tcph->doff += TCPOLEN_MSS/4; 194 inet_proto_csum_replace2(&tcph->check, skb, 195 oldval, ((__be16 *)tcph)[6], false); 196 return TCPOLEN_MSS; 197 } 198 199 static unsigned int 200 tcpmss_tg4(struct sk_buff *skb, const struct xt_action_param *par) 201 { 202 struct iphdr *iph = ip_hdr(skb); 203 __be16 newlen; 204 int ret; 205 206 ret = tcpmss_mangle_packet(skb, par, 207 PF_INET, 208 iph->ihl * 4, 209 sizeof(*iph) + sizeof(struct tcphdr)); 210 if (ret < 0) 211 return NF_DROP; 212 if (ret > 0) { 213 iph = ip_hdr(skb); 214 newlen = htons(ntohs(iph->tot_len) + ret); 215 csum_replace2(&iph->check, iph->tot_len, newlen); 216 iph->tot_len = newlen; 217 } 218 return XT_CONTINUE; 219 } 220 221 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) 222 static unsigned int 223 tcpmss_tg6(struct sk_buff *skb, const struct xt_action_param *par) 224 { 225 struct ipv6hdr *ipv6h = ipv6_hdr(skb); 226 u8 nexthdr; 227 __be16 frag_off, oldlen, newlen; 228 int tcphoff; 229 int ret; 230 231 nexthdr = ipv6h->nexthdr; 232 tcphoff = ipv6_skip_exthdr(skb, sizeof(*ipv6h), &nexthdr, &frag_off); 233 if (tcphoff < 0) 234 return NF_DROP; 235 ret = tcpmss_mangle_packet(skb, par, 236 PF_INET6, 237 tcphoff, 238 sizeof(*ipv6h) + sizeof(struct tcphdr)); 239 if (ret < 0) 240 return NF_DROP; 241 if (ret > 0) { 242 ipv6h = ipv6_hdr(skb); 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; 249 } 250 return XT_CONTINUE; 251 } 252 #endif 253 254 /* Must specify -p tcp --syn */ 255 static inline bool find_syn_match(const struct xt_entry_match *m) 256 { 257 const struct xt_tcp *tcpinfo = (const struct xt_tcp *)m->data; 258 259 if (strcmp(m->u.kernel.match->name, "tcp") == 0 && 260 tcpinfo->flg_cmp & TCPHDR_SYN && 261 !(tcpinfo->invflags & XT_TCP_INV_FLAGS)) 262 return true; 263 264 return false; 265 } 266 267 static int tcpmss_tg4_check(const struct xt_tgchk_param *par) 268 { 269 const struct xt_tcpmss_info *info = par->targinfo; 270 const struct ipt_entry *e = par->entryinfo; 271 const struct xt_entry_match *ematch; 272 273 if (info->mss == XT_TCPMSS_CLAMP_PMTU && 274 (par->hook_mask & ~((1 << NF_INET_FORWARD) | 275 (1 << NF_INET_LOCAL_OUT) | 276 (1 << NF_INET_POST_ROUTING))) != 0) { 277 pr_info("path-MTU clamping only supported in " 278 "FORWARD, OUTPUT and POSTROUTING hooks\n"); 279 return -EINVAL; 280 } 281 if (par->nft_compat) 282 return 0; 283 284 xt_ematch_foreach(ematch, e) 285 if (find_syn_match(ematch)) 286 return 0; 287 pr_info("Only works on TCP SYN packets\n"); 288 return -EINVAL; 289 } 290 291 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) 292 static int tcpmss_tg6_check(const struct xt_tgchk_param *par) 293 { 294 const struct xt_tcpmss_info *info = par->targinfo; 295 const struct ip6t_entry *e = par->entryinfo; 296 const struct xt_entry_match *ematch; 297 298 if (info->mss == XT_TCPMSS_CLAMP_PMTU && 299 (par->hook_mask & ~((1 << NF_INET_FORWARD) | 300 (1 << NF_INET_LOCAL_OUT) | 301 (1 << NF_INET_POST_ROUTING))) != 0) { 302 pr_info("path-MTU clamping only supported in " 303 "FORWARD, OUTPUT and POSTROUTING hooks\n"); 304 return -EINVAL; 305 } 306 if (par->nft_compat) 307 return 0; 308 309 xt_ematch_foreach(ematch, e) 310 if (find_syn_match(ematch)) 311 return 0; 312 pr_info("Only works on TCP SYN packets\n"); 313 return -EINVAL; 314 } 315 #endif 316 317 static struct xt_target tcpmss_tg_reg[] __read_mostly = { 318 { 319 .family = NFPROTO_IPV4, 320 .name = "TCPMSS", 321 .checkentry = tcpmss_tg4_check, 322 .target = tcpmss_tg4, 323 .targetsize = sizeof(struct xt_tcpmss_info), 324 .proto = IPPROTO_TCP, 325 .me = THIS_MODULE, 326 }, 327 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) 328 { 329 .family = NFPROTO_IPV6, 330 .name = "TCPMSS", 331 .checkentry = tcpmss_tg6_check, 332 .target = tcpmss_tg6, 333 .targetsize = sizeof(struct xt_tcpmss_info), 334 .proto = IPPROTO_TCP, 335 .me = THIS_MODULE, 336 }, 337 #endif 338 }; 339 340 static int __init tcpmss_tg_init(void) 341 { 342 return xt_register_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg)); 343 } 344 345 static void __exit tcpmss_tg_exit(void) 346 { 347 xt_unregister_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg)); 348 } 349 350 module_init(tcpmss_tg_init); 351 module_exit(tcpmss_tg_exit); 352