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         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         ai = nf_get_afinfo(family);
  66         if (ai != NULL)
  67                 ai->route(net, (struct dst_entry **)&rt, &fl, false);
  68
  69         if (rt != NULL) {
  70                 mtu = dst_mtu(&rt->dst);
  71                 dst_release(&rt->dst);
  72         }
  73         return mtu;
  74 }
  75
  76 static int
  77 tcpmss_mangle_packet(struct sk_buff *skb,
  78                      const struct xt_action_param *par,
  79                      unsigned int family,
  80                      unsigned int tcphoff,
  81                      unsigned int minlen)
  82 {
  83         const struct xt_tcpmss_info *info = par->targinfo;
  84         struct tcphdr *tcph;
  85         int len, tcp_hdrlen;
  86         unsigned int i;
  87         __be16 oldval;
  88         u16 newmss;
  89         u8 *opt;
  90
  91         /* This is a fragment, no TCP header is available */
  92         if (par->fragoff != 0)
  93                 return 0;
  94
  95         if (!skb_make_writable(skb, skb->len))
  96                 return -1;
  97
  98         len = skb->len - tcphoff;
  99         if (len < (int)sizeof(struct tcphdr))
 100                 return -1;
 101
 102         tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
 103         tcp_hdrlen = tcph->doff * 4;
 104
 105         if (len < tcp_hdrlen || tcp_hdrlen < sizeof(struct tcphdr))
 106                 return -1;
 107
 108         if (info->mss == XT_TCPMSS_CLAMP_PMTU) {
 109                 struct net *net = xt_net(par);
 110                 unsigned int in_mtu = tcpmss_reverse_mtu(net, skb, family);
 111                 unsigned int min_mtu = min(dst_mtu(skb_dst(skb)), in_mtu);
 112
 113                 if (min_mtu <= minlen) {
 114                         net_err_ratelimited("unknown or invalid path-MTU (%u)\n",
 115                                             min_mtu);
 116                         return -1;
 117                 }
 118                 newmss = min_mtu - minlen;
 119         } else
 120                 newmss = info->mss;
 121
 122         opt = (u_int8_t *)tcph;
 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) {
 125                         u_int16_t oldmss;
 126
 127                         oldmss = (opt[i+2] << 8) | opt[i+3];
 128
 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)
 134                                 return 0;
 135
 136                         opt[i+2] = (newmss & 0xff00) >> 8;
 137                         opt[i+3] = newmss & 0x00ff;
 138
 139                         inet_proto_csum_replace2(&tcph->check, skb,
 140                                                  htons(oldmss), htons(newmss),
 141                                                  false);
 142                         return 0;
 143                 }
 144         }
 145
 146         /* There is data after the header so the option can't be added
 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)
 151                 return 0;
 152
 153         /* tcph->doff has 4 bits, do not wrap it to 0 */
 154         if (tcp_hdrlen >= 15 * 4)
 155                 return 0;
 156
 157         /*
 158          * MSS Option not found ?! add it..
 159          */
 160         if (skb_tailroom(skb) < TCPOLEN_MSS) {
 161                 if (pskb_expand_head(skb, 0,
 162                                      TCPOLEN_MSS - skb_tailroom(skb),
 163                                      GFP_ATOMIC))
 164                         return -1;
 165                 tcph = (struct tcphdr *)(skb_network_header(skb) + tcphoff);
 166         }
 167
 168         skb_put(skb, TCPOLEN_MSS);
 169
 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
 176          */
 177         if (xt_family(par) == NFPROTO_IPV4)
 178                 newmss = min(newmss, (u16)536);
 179         else
 180                 newmss = min(newmss, (u16)1220);
 181
 182         opt = (u_int8_t *)tcph + sizeof(struct tcphdr);
 183         memmove(opt + TCPOLEN_MSS, opt, len - sizeof(struct tcphdr));
 184
 185         inet_proto_csum_replace2(&tcph->check, skb,
 186                                  htons(len), htons(len + TCPOLEN_MSS), true);
 187         opt[0] = TCPOPT_MSS;
 188         opt[1] = TCPOLEN_MSS;
 189         opt[2] = (newmss & 0xff00) >> 8;
 190         opt[3] = newmss & 0x00ff;
 191
 192         inet_proto_csum_replace4(&tcph->check, skb, 0, *((__be32 *)opt), false);
 193
 194         oldval = ((__be16 *)tcph)[6];
 195         tcph->doff += TCPOLEN_MSS/4;
 196         inet_proto_csum_replace2(&tcph->check, skb,
 197                                  oldval, ((__be16 *)tcph)[6], false);
 198         return TCPOLEN_MSS;
 199 }
 200
 201 static unsigned int
 202 tcpmss_tg4(struct sk_buff *skb, const struct xt_action_param *par)
 203 {
 204         struct iphdr *iph = ip_hdr(skb);
 205         __be16 newlen;
 206         int ret;
 207
 208         ret = tcpmss_mangle_packet(skb, par,
 209                                    PF_INET,
 210                                    iph->ihl * 4,
 211                                    sizeof(*iph) + sizeof(struct tcphdr));
 212         if (ret < 0)
 213                 return NF_DROP;
 214         if (ret > 0) {
 215                 iph = ip_hdr(skb);
 216                 newlen = htons(ntohs(iph->tot_len) + ret);
 217                 csum_replace2(&iph->check, iph->tot_len, newlen);
 218                 iph->tot_len = newlen;
 219         }
 220         return XT_CONTINUE;
 221 }
 222
 223 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 224 static unsigned int
 225 tcpmss_tg6(struct sk_buff *skb, const struct xt_action_param *par)
 226 {
 227         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 228         u8 nexthdr;
 229         __be16 frag_off, oldlen, newlen;
 230         int tcphoff;
 231         int ret;
 232
 233         nexthdr = ipv6h->nexthdr;
 234         tcphoff = ipv6_skip_exthdr(skb, sizeof(*ipv6h), &nexthdr, &frag_off);
 235         if (tcphoff < 0)
 236                 return NF_DROP;
 237         ret = tcpmss_mangle_packet(skb, par,
 238                                    PF_INET6,
 239                                    tcphoff,
 240                                    sizeof(*ipv6h) + sizeof(struct tcphdr));
 241         if (ret < 0)
 242                 return NF_DROP;
 243         if (ret > 0) {
 244                 ipv6h = ipv6_hdr(skb);
 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;
 251         }
 252         return XT_CONTINUE;
 253 }
 254 #endif
 255
 256 /* Must specify -p tcp --syn */
 257 static inline bool find_syn_match(const struct xt_entry_match *m)
 258 {
 259         const struct xt_tcp *tcpinfo = (const struct xt_tcp *)m->data;
 260
 261         if (strcmp(m->u.kernel.match->name, "tcp") == 0 &&
 262             tcpinfo->flg_cmp & TCPHDR_SYN &&
 263             !(tcpinfo->invflags & XT_TCP_INV_FLAGS))
 264                 return true;
 265
 266         return false;
 267 }
 268
 269 static int tcpmss_tg4_check(const struct xt_tgchk_param *par)
 270 {
 271         const struct xt_tcpmss_info *info = par->targinfo;
 272         const struct ipt_entry *e = par->entryinfo;
 273         const struct xt_entry_match *ematch;
 274
 275         if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
 276             (par->hook_mask & ~((1 << NF_INET_FORWARD) |
 277                            (1 << NF_INET_LOCAL_OUT) |
 278                            (1 << NF_INET_POST_ROUTING))) != 0) {
 279                 pr_info("path-MTU clamping only supported in "
 280                         "FORWARD, OUTPUT and POSTROUTING hooks\n");
 281                 return -EINVAL;
 282         }
 283         if (par->nft_compat)
 284                 return 0;
 285
 286         xt_ematch_foreach(ematch, e)
 287                 if (find_syn_match(ematch))
 288                         return 0;
 289         pr_info("Only works on TCP SYN packets\n");
 290         return -EINVAL;
 291 }
 292
 293 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 294 static int tcpmss_tg6_check(const struct xt_tgchk_param *par)
 295 {
 296         const struct xt_tcpmss_info *info = par->targinfo;
 297         const struct ip6t_entry *e = par->entryinfo;
 298         const struct xt_entry_match *ematch;
 299
 300         if (info->mss == XT_TCPMSS_CLAMP_PMTU &&
 301             (par->hook_mask & ~((1 << NF_INET_FORWARD) |
 302                            (1 << NF_INET_LOCAL_OUT) |
 303                            (1 << NF_INET_POST_ROUTING))) != 0) {
 304                 pr_info("path-MTU clamping only supported in "
 305                         "FORWARD, OUTPUT and POSTROUTING hooks\n");
 306                 return -EINVAL;
 307         }
 308         if (par->nft_compat)
 309                 return 0;
 310
 311         xt_ematch_foreach(ematch, e)
 312                 if (find_syn_match(ematch))
 313                         return 0;
 314         pr_info("Only works on TCP SYN packets\n");
 315         return -EINVAL;
 316 }
 317 #endif
 318
 319 static struct xt_target tcpmss_tg_reg[] __read_mostly = {
 320         {
 321                 .family         = NFPROTO_IPV4,
 322                 .name           = "TCPMSS",
 323                 .checkentry     = tcpmss_tg4_check,
 324                 .target         = tcpmss_tg4,
 325                 .targetsize     = sizeof(struct xt_tcpmss_info),
 326                 .proto          = IPPROTO_TCP,
 327                 .me             = THIS_MODULE,
 328         },
 329 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
 330         {
 331                 .family         = NFPROTO_IPV6,
 332                 .name           = "TCPMSS",
 333                 .checkentry     = tcpmss_tg6_check,
 334                 .target         = tcpmss_tg6,
 335                 .targetsize     = sizeof(struct xt_tcpmss_info),
 336                 .proto          = IPPROTO_TCP,
 337                 .me             = THIS_MODULE,
 338         },
 339 #endif
 340 };
 341
 342 static int __init tcpmss_tg_init(void)
 343 {
 344         return xt_register_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
 345 }
 346
 347 static void __exit tcpmss_tg_exit(void)
 348 {
 349         xt_unregister_targets(tcpmss_tg_reg, ARRAY_SIZE(tcpmss_tg_reg));
 350 }
 351
 352 module_init(tcpmss_tg_init);
 353 module_exit(tcpmss_tg_exit);