net/netfilter/xt_TCPMSS.c

   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>
   5  * Copyright (C) 2007 Patrick McHardy <kaber@trash.net>
   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 <marc@mbsi.ca>");
  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         struct rtable *rt = NULL;
  52         u_int32_t mtu     = ~0U;
  53
  54         if (family == PF_INET) {
  55                 struct flowi4 *fl4 = &fl.u.ip4;
  56                 memset(fl4, 0, sizeof(*fl4));
  57                 fl4->daddr = ip_hdr(skb)->saddr;
  58         } else {
  59                 struct flowi6 *fl6 = &fl.u.ip6;
  60
  61                 memset(fl6, 0, sizeof(*fl6));
  62                 fl6->daddr = ipv6_hdr(skb)->saddr;
  63         }
  64
  65         nf_route(net, (struct dst_entry **)&rt, &fl, false, family);
  66         if (rt != NULL) {
  67                 mtu = dst_mtu(&rt->dst);
  68                 dst_release(&rt->dst);
  69         }
  70         return mtu;
  71 }
  72
  73 static int
  74 tcpmss_mangle_packet(struct sk_buff *skb,
  75                      const struct xt_action_param *par,
  76                      unsigned int family,
  77                      unsigned int tcphoff,
  78                      unsigned int minlen)
  79 {
  80         const struct xt_tcpmss_info *info = par->targinfo;
  81         struct tcphdr *tcph;
  82         int len, tcp_hdrlen;
  83         unsigned int i;
  84         __be16 oldval;
  85         u16 newmss;
  86         u8 *opt;
  87
  88         /* This is a fragment, no TCP header is available */
  89         if (par->fragoff != 0)
  90                 return 0;
  91
  92         if (!skb_make_writable(skb, skb->len))
  93                 return -1;
  94
  95         len = skb->len - tcphoff;
  96         if (len < (int)sizeof(struct tcphdr))
  97                 return -1;
  98
  99         tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
 100         tcp_hdrlen = tcph->doff * 4;
 101
 102         if (len < tcp_hdrlen || tcp_hdrlen < sizeof(struct tcphdr))
 103                 return -1;
 104
 105         if (info->mss == XT_TCPMSS_CLAMP_PMTU) {
 106                 struct net *net = xt_net(par);
 107                 unsigned int in_mtu = tcpmss_reverse_mtu(net, skb, family);
 108                 unsigned int min_mtu = min(dst_mtu(skb_dst(skb)), in_mtu);
 109
 110                 if (min_mtu <= minlen) {
 111                         net_err_ratelimited("unknown or invalid path-MTU (%u)\n",
 112                                             min_mtu);
 113                         return -1;
 114                 }
 115                 newmss = min_mtu - minlen;
 116         } else
 117                 newmss = info->mss;
 118
 119         opt = (u_int8_t *)tcph;
 120         for (i = sizeof(struct tcphdr); i <= tcp_hdrlen - TCPOLEN_MSS; i += optlen(opt, i)) {
 121                 if (opt[i] == TCPOPT_MSS && opt[i+1] == TCPOLEN_MSS) {
 122                         u_int16_t oldmss;
 123
 124                         oldmss = (opt[i+2] << 8) | opt[i+3];
 125
 126                         /* Never increase MSS, even when setting it, as
 127                          * doing so results in problems for hosts that rely
 128                          * on MSS being set correctly.
 129                          */
 130                         if (oldmss <= newmss)
 131                                 return 0;
 132
 133                         opt[i+2] = (newmss & 0xff00) >> 8;
 134                         opt[i+3] = newmss & 0x00ff;
 135
 136                         inet_proto_csum_replace2(&tcph->check, skb,
 137                                                  htons(oldmss), htons(newmss),
 138                                                  false);
 139                         return 0;
 140                 }
 141         }
 142
 143         /* There is data after the header so the option can't be added
 144          * without moving it, and doing so may make the SYN packet
 145          * itself too large. Accept the packet unmodified instead.
 146          */
 147         if (len > tcp_hdrlen)
 148                 return 0;
 149
 150         /* tcph->doff has 4 bits, do not wrap it to 0 */
 151         if (tcp_hdrlen >= 15 * 4)
 152                 return 0;
 153
 154         /*
 155          * MSS Option not found ?! add it..
 156          */
 157         if (skb_tailroom(skb) < TCPOLEN_MSS) {
 158                 if (pskb_expand_head(skb, 0,
 159                                      TCPOLEN_MSS - skb_tailroom(skb),
 160                                      GFP_ATOMIC))
 161                         return -1;
 162                 tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
 163         }
 164
 165         skb_put(skb, TCPOLEN_MSS);
 166
 167         /*
 168          * IPv4: RFC 1122 states "If an MSS option is not received at
 169          * connection setup, TCP MUST assume a default send MSS of 536".
 170          * IPv6: RFC 2460 states IPv6 has a minimum MTU of 1280 and a minimum
 171          * length IPv6 header of 60, ergo the default MSS value is 1220
 172          * Since no MSS was provided, we must use the default values
 173          */
 174         if (xt_family(par) == NFPROTO_IPV4)
 175                 newmss = min(newmss, (u16)536);
 176         else
 177                 newmss = min(newmss, (u16)1220);
 178
 179         opt = (u_int8_t *)tcph + sizeof(struct tcphdr);
 180         memmove(opt + TCPOLEN_MSS, opt, len - sizeof(struct tcphdr));
 181
 182         inet_proto_csum_replace2(&tcph->check, skb,
 183                                  htons(len), htons(len + TCPOLEN_MSS), true);
 184         opt[0] = TCPOPT_MSS;
 185         opt[1] = TCPOLEN_MSS;
 186         opt[2] = (newmss & 0xff00) >> 8;
 187         opt[3] = newmss & 0x00ff;
 188
 189         inet_proto_csum_replace4(&tcph->check, skb, 0, *((__be32 *)opt), false);
 190
 191         oldval = ((__be16 *)tcph)[6];
 192         tcph->doff += TCPOLEN_MSS/4;
 193         inet_proto_csum_replace2(&tcph->check, skb,
 194                                  oldval, ((__be16 *)tcph)[6], false);
 195         return TCPOLEN_MSS;
 196 }
 197
 198 static unsigned int
 199 tcpmss_tg4(struct sk_buff *skb, const struct xt_action_param *par)
 200 {
 201         struct iphdr *iph = ip_hdr(skb);
 202         __be16 newlen;
 203         int ret;
 204
 205         ret = tcpmss_mangle_packet(skb, par,
 206                                    PF_INET,
 207                                    iph->ihl * 4,
 208                                    sizeof(*iph) + sizeof(struct tcphdr));
 209         if (ret < 0)
 210                 return NF_DROP;
 211         if (ret > 0) {
 212                 iph = ip_hdr(skb);
 213                 newlen = htons(ntohs(iph->tot_len) + ret);
 214                 csum_replace2(&iph->check, iph->tot_len, newlen);
 215                 iph->tot_len = newlen;
 216         }
 217         return XT_CONTINUE;
 218 }
 219
 220 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 221 static unsigned int
 222 tcpmss_tg6(struct sk_buff *skb, const struct xt_action_param *par)
 223 {
 224         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 225         u8 nexthdr;
 226         __be16 frag_off, oldlen, newlen;
 227         int tcphoff;
 228         int ret;
 229
 230         nexthdr = ipv6h->nexthdr;
 231         tcphoff = ipv6_skip_exthdr(skb, sizeof(*ipv6h), &nexthdr, &frag_off);
 232         if (tcphoff < 0)
 233                 return NF_DROP;
 234         ret = tcpmss_mangle_packet(skb, par,
 235                                    PF_INET6,
 236                                    tcphoff,
 237                                    sizeof(*ipv6h) + sizeof(struct tcphdr));
 238         if (ret < 0)
 239                 return NF_DROP;
 240         if (ret > 0) {
 241                 ipv6h = ipv6_hdr(skb);
 242                 oldlen = ipv6h->payload_len;
 243                 newlen = htons(ntohs(oldlen) + ret);
 244                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 245                         skb->csum = csum_add(csum_sub(skb->csum, oldlen),
 246                                              newlen);
 247                 ipv6h->payload_len = newlen;
 248         }
 249         return XT_CONTINUE;
 250 }
 251 #endif
 252
 253 /* Must specify -p tcp --syn */
 254 static inline bool find_syn_match(const struct xt_entry_match *m)
 255 {
 256         const struct xt_tcp *tcpinfo = (const struct xt_tcp *)m->data;
 257
 258         if (strcmp(m->u.kernel.match->name, "tcp") == 0 &&
 259             tcpinfo->flg_cmp & TCPHDR_SYN &&
 260             !(tcpinfo->invflags & XT_TCP_INV_FLAGS))
 261                 return true;
 262
 263         return false;
 264 }
 265
 266 static int tcpmss_tg4_check(const struct xt_tgchk_param *par)
 267 {
 268         const struct xt_tcpmss_info *info = par->targinfo;
 269         const struct ipt_entry *e = par->entryinfo;
 270         const struct xt_entry_match *ematch;
 271
 272         if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
 273             (par->hook_mask & ~((1 << NF_INET_FORWARD) |
 274                            (1 << NF_INET_LOCAL_OUT) |
 275                            (1 << NF_INET_POST_ROUTING))) != 0) {
 276                 pr_info_ratelimited("path-MTU clamping only supported in FORWARD, OUTPUT and POSTROUTING hooks\n");
 277                 return -EINVAL;
 278         }
 279         if (par->nft_compat)
 280                 return 0;
 281
 282         xt_ematch_foreach(ematch, e)
 283                 if (find_syn_match(ematch))
 284                         return 0;
 285         pr_info_ratelimited("Only works on TCP SYN packets\n");
 286         return -EINVAL;
 287 }
 288
 289 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 290 static int tcpmss_tg6_check(const struct xt_tgchk_param *par)
 291 {
 292         const struct xt_tcpmss_info *info = par->targinfo;
 293         const struct ip6t_entry *e = par->entryinfo;
 294         const struct xt_entry_match *ematch;
 295
 296         if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
 297             (par->hook_mask & ~((1 << NF_INET_FORWARD) |
 298                            (1 << NF_INET_LOCAL_OUT) |
 299                            (1 << NF_INET_POST_ROUTING))) != 0) {
 300                 pr_info_ratelimited("path-MTU clamping only supported in FORWARD, OUTPUT and POSTROUTING hooks\n");
 301                 return -EINVAL;
 302         }
 303         if (par->nft_compat)
 304                 return 0;
 305
 306         xt_ematch_foreach(ematch, e)
 307                 if (find_syn_match(ematch))
 308                         return 0;
 309         pr_info_ratelimited("Only works on TCP SYN packets\n");
 310         return -EINVAL;
 311 }
 312 #endif
 313
 314 static struct xt_target tcpmss_tg_reg[] __read_mostly = {
 315         {
 316                 .family         = NFPROTO_IPV4,
 317                 .name           = "TCPMSS",
 318                 .checkentry     = tcpmss_tg4_check,
 319                 .target         = tcpmss_tg4,
 320                 .targetsize     = sizeof(struct xt_tcpmss_info),
 321                 .proto          = IPPROTO_TCP,
 322                 .me             = THIS_MODULE,
 323         },
 324 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 325         {
 326                 .family         = NFPROTO_IPV6,
 327                 .name           = "TCPMSS",
 328                 .checkentry     = tcpmss_tg6_check,
 329                 .target         = tcpmss_tg6,
 330                 .targetsize     = sizeof(struct xt_tcpmss_info),
 331                 .proto          = IPPROTO_TCP,
 332                 .me             = THIS_MODULE,
 333         },
 334 #endif
 335 };
 336
 337 static int __init tcpmss_tg_init(void)
 338 {
 339         return xt_register_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
 340 }
 341
 342 static void __exit tcpmss_tg_exit(void)
 343 {
 344         xt_unregister_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
 345 }
 346
 347 module_init(tcpmss_tg_init);
 348 module_exit(tcpmss_tg_exit);