net/ipv6/netfilter/ip6t_SYNPROXY.c

   1 /*
   2  * Copyright (c) 2013 Patrick McHardy <kaber@trash.net>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  */
   8
   9 #include <linux/module.h>
  10 #include <linux/skbuff.h>
  11 #include <net/ip6_checksum.h>
  12 #include <net/ip6_route.h>
  13 #include <net/tcp.h>
  14
  15 #include <linux/netfilter_ipv6/ip6_tables.h>
  16 #include <linux/netfilter/x_tables.h>
  17 #include <linux/netfilter/xt_SYNPROXY.h>
  18 #include <net/netfilter/nf_conntrack.h>
  19 #include <net/netfilter/nf_conntrack_seqadj.h>
  20 #include <net/netfilter/nf_conntrack_synproxy.h>
  21
  22 static struct ipv6hdr *
  23 synproxy_build_ip(struct net *net, struct sk_buff *skb,
  24                   const struct in6_addr *saddr,
  25                   const struct in6_addr *daddr)
  26 {
  27         struct ipv6hdr *iph;
  28
  29         skb_reset_network_header(skb);
  30         iph = (struct ipv6hdr *)skb_put(skb, sizeof(*iph));
  31         ip6_flow_hdr(iph, 0, 0);
  32         iph->hop_limit  = net->ipv6.devconf_all->hop_limit;
  33         iph->nexthdr    = IPPROTO_TCP;
  34         iph->saddr      = *saddr;
  35         iph->daddr      = *daddr;
  36
  37         return iph;
  38 }
  39
  40 static void
  41 synproxy_send_tcp(struct net *net,
  42                   const struct sk_buff *skb, struct sk_buff *nskb,
  43                   struct nf_conntrack *nfct, enum ip_conntrack_info ctinfo,
  44                   struct ipv6hdr *niph, struct tcphdr *nth,
  45                   unsigned int tcp_hdr_size)
  46 {
  47         struct dst_entry *dst;
  48         struct flowi6 fl6;
  49
  50         nth->check = ~tcp_v6_check(tcp_hdr_size, &niph->saddr, &niph->daddr, 0);
  51         nskb->ip_summed   = CHECKSUM_PARTIAL;
  52         nskb->csum_start  = (unsigned char *)nth - nskb->head;
  53         nskb->csum_offset = offsetof(struct tcphdr, check);
  54
  55         memset(&fl6, 0, sizeof(fl6));
  56         fl6.flowi6_proto = IPPROTO_TCP;
  57         fl6.saddr = niph->saddr;
  58         fl6.daddr = niph->daddr;
  59         fl6.fl6_sport = nth->source;
  60         fl6.fl6_dport = nth->dest;
  61         security_skb_classify_flow((struct sk_buff *)skb, flowi6_to_flowi(&fl6));
  62         dst = ip6_route_output(net, NULL, &fl6);
  63         if (dst->error) {
  64                 dst_release(dst);
  65                 goto free_nskb;
  66         }
  67         dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
  68         if (IS_ERR(dst))
  69                 goto free_nskb;
  70
  71         skb_dst_set(nskb, dst);
  72
  73         if (nfct) {
  74                 nf_ct_set(nskb, (struct nf_conn *)nfct, ctinfo);
  75                 nf_conntrack_get(nfct);
  76         }
  77
  78         ip6_local_out(net, nskb->sk, nskb);
  79         return;
  80
  81 free_nskb:
  82         kfree_skb(nskb);
  83 }
  84
  85 static void
  86 synproxy_send_client_synack(struct net *net,
  87                             const struct sk_buff *skb, const struct tcphdr *th,
  88                             const struct synproxy_options *opts)
  89 {
  90         struct sk_buff *nskb;
  91         struct ipv6hdr *iph, *niph;
  92         struct tcphdr *nth;
  93         unsigned int tcp_hdr_size;
  94         u16 mss = opts->mss;
  95
  96         iph = ipv6_hdr(skb);
  97
  98         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
  99         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 100                          GFP_ATOMIC);
 101         if (nskb == NULL)
 102                 return;
 103         skb_reserve(nskb, MAX_TCP_HEADER);
 104
 105         niph = synproxy_build_ip(net, nskb, &iph->daddr, &iph->saddr);
 106
 107         skb_reset_transport_header(nskb);
 108         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 109         nth->source     = th->dest;
 110         nth->dest       = th->source;
 111         nth->seq        = htonl(__cookie_v6_init_sequence(iph, th, &mss));
 112         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
 113         tcp_flag_word(nth) = TCP_FLAG_SYN | TCP_FLAG_ACK;
 114         if (opts->options & XT_SYNPROXY_OPT_ECN)
 115                 tcp_flag_word(nth) |= TCP_FLAG_ECE;
 116         nth->doff       = tcp_hdr_size / 4;
 117         nth->window     = 0;
 118         nth->check      = 0;
 119         nth->urg_ptr    = 0;
 120
 121         synproxy_build_options(nth, opts);
 122
 123         synproxy_send_tcp(net, skb, nskb, skb_nfct(skb),
 124                           IP_CT_ESTABLISHED_REPLY, niph, nth, tcp_hdr_size);
 125 }
 126
 127 static void
 128 synproxy_send_server_syn(struct net *net,
 129                          const struct sk_buff *skb, const struct tcphdr *th,
 130                          const struct synproxy_options *opts, u32 recv_seq)
 131 {
 132         struct synproxy_net *snet = synproxy_pernet(net);
 133         struct sk_buff *nskb;
 134         struct ipv6hdr *iph, *niph;
 135         struct tcphdr *nth;
 136         unsigned int tcp_hdr_size;
 137
 138         iph = ipv6_hdr(skb);
 139
 140         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 141         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 142                          GFP_ATOMIC);
 143         if (nskb == NULL)
 144                 return;
 145         skb_reserve(nskb, MAX_TCP_HEADER);
 146
 147         niph = synproxy_build_ip(net, nskb, &iph->saddr, &iph->daddr);
 148
 149         skb_reset_transport_header(nskb);
 150         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 151         nth->source     = th->source;
 152         nth->dest       = th->dest;
 153         nth->seq        = htonl(recv_seq - 1);
 154         /* ack_seq is used to relay our ISN to the synproxy hook to initialize
 155          * sequence number translation once a connection tracking entry exists.
 156          */
 157         nth->ack_seq    = htonl(ntohl(th->ack_seq) - 1);
 158         tcp_flag_word(nth) = TCP_FLAG_SYN;
 159         if (opts->options & XT_SYNPROXY_OPT_ECN)
 160                 tcp_flag_word(nth) |= TCP_FLAG_ECE | TCP_FLAG_CWR;
 161         nth->doff       = tcp_hdr_size / 4;
 162         nth->window     = th->window;
 163         nth->check      = 0;
 164         nth->urg_ptr    = 0;
 165
 166         synproxy_build_options(nth, opts);
 167
 168         synproxy_send_tcp(net, skb, nskb, &snet->tmpl->ct_general, IP_CT_NEW,
 169                           niph, nth, tcp_hdr_size);
 170 }
 171
 172 static void
 173 synproxy_send_server_ack(struct net *net,
 174                          const struct ip_ct_tcp *state,
 175                          const struct sk_buff *skb, const struct tcphdr *th,
 176                          const struct synproxy_options *opts)
 177 {
 178         struct sk_buff *nskb;
 179         struct ipv6hdr *iph, *niph;
 180         struct tcphdr *nth;
 181         unsigned int tcp_hdr_size;
 182
 183         iph = ipv6_hdr(skb);
 184
 185         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 186         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 187                          GFP_ATOMIC);
 188         if (nskb == NULL)
 189                 return;
 190         skb_reserve(nskb, MAX_TCP_HEADER);
 191
 192         niph = synproxy_build_ip(net, nskb, &iph->daddr, &iph->saddr);
 193
 194         skb_reset_transport_header(nskb);
 195         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 196         nth->source     = th->dest;
 197         nth->dest       = th->source;
 198         nth->seq        = htonl(ntohl(th->ack_seq));
 199         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
 200         tcp_flag_word(nth) = TCP_FLAG_ACK;
 201         nth->doff       = tcp_hdr_size / 4;
 202         nth->window     = htons(state->seen[IP_CT_DIR_ORIGINAL].td_maxwin);
 203         nth->check      = 0;
 204         nth->urg_ptr    = 0;
 205
 206         synproxy_build_options(nth, opts);
 207
 208         synproxy_send_tcp(net, skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
 209 }
 210
 211 static void
 212 synproxy_send_client_ack(struct net *net,
 213                          const struct sk_buff *skb, const struct tcphdr *th,
 214                          const struct synproxy_options *opts)
 215 {
 216         struct sk_buff *nskb;
 217         struct ipv6hdr *iph, *niph;
 218         struct tcphdr *nth;
 219         unsigned int tcp_hdr_size;
 220
 221         iph = ipv6_hdr(skb);
 222
 223         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 224         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 225                          GFP_ATOMIC);
 226         if (nskb == NULL)
 227                 return;
 228         skb_reserve(nskb, MAX_TCP_HEADER);
 229
 230         niph = synproxy_build_ip(net, nskb, &iph->saddr, &iph->daddr);
 231
 232         skb_reset_transport_header(nskb);
 233         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 234         nth->source     = th->source;
 235         nth->dest       = th->dest;
 236         nth->seq        = htonl(ntohl(th->seq) + 1);
 237         nth->ack_seq    = th->ack_seq;
 238         tcp_flag_word(nth) = TCP_FLAG_ACK;
 239         nth->doff       = tcp_hdr_size / 4;
 240         nth->window     = htons(ntohs(th->window) >> opts->wscale);
 241         nth->check      = 0;
 242         nth->urg_ptr    = 0;
 243
 244         synproxy_build_options(nth, opts);
 245
 246         synproxy_send_tcp(net, skb, nskb, skb_nfct(skb),
 247                           IP_CT_ESTABLISHED_REPLY, niph, nth, tcp_hdr_size);
 248 }
 249
 250 static bool
 251 synproxy_recv_client_ack(struct net *net,
 252                          const struct sk_buff *skb, const struct tcphdr *th,
 253                          struct synproxy_options *opts, u32 recv_seq)
 254 {
 255         struct synproxy_net *snet = synproxy_pernet(net);
 256         int mss;
 257
 258         mss = __cookie_v6_check(ipv6_hdr(skb), th, ntohl(th->ack_seq) - 1);
 259         if (mss == 0) {
 260                 this_cpu_inc(snet->stats->cookie_invalid);
 261                 return false;
 262         }
 263
 264         this_cpu_inc(snet->stats->cookie_valid);
 265         opts->mss = mss;
 266         opts->options |= XT_SYNPROXY_OPT_MSS;
 267
 268         if (opts->options & XT_SYNPROXY_OPT_TIMESTAMP)
 269                 synproxy_check_timestamp_cookie(opts);
 270
 271         synproxy_send_server_syn(net, skb, th, opts, recv_seq);
 272         return true;
 273 }
 274
 275 static unsigned int
 276 synproxy_tg6(struct sk_buff *skb, const struct xt_action_param *par)
 277 {
 278         const struct xt_synproxy_info *info = par->targinfo;
 279         struct net *net = xt_net(par);
 280         struct synproxy_net *snet = synproxy_pernet(net);
 281         struct synproxy_options opts = {};
 282         struct tcphdr *th, _th;
 283
 284         if (nf_ip6_checksum(skb, xt_hooknum(par), par->thoff, IPPROTO_TCP))
 285                 return NF_DROP;
 286
 287         th = skb_header_pointer(skb, par->thoff, sizeof(_th), &_th);
 288         if (th == NULL)
 289                 return NF_DROP;
 290
 291         if (!synproxy_parse_options(skb, par->thoff, th, &opts))
 292                 return NF_DROP;
 293
 294         if (th->syn && !(th->ack || th->fin || th->rst)) {
 295                 /* Initial SYN from client */
 296                 this_cpu_inc(snet->stats->syn_received);
 297
 298                 if (th->ece && th->cwr)
 299                         opts.options |= XT_SYNPROXY_OPT_ECN;
 300
 301                 opts.options &= info->options;
 302                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 303                         synproxy_init_timestamp_cookie(info, &opts);
 304                 else
 305                         opts.options &= ~(XT_SYNPROXY_OPT_WSCALE |
 306                                           XT_SYNPROXY_OPT_SACK_PERM |
 307                                           XT_SYNPROXY_OPT_ECN);
 308
 309                 synproxy_send_client_synack(net, skb, th, &opts);
 310                 return NF_DROP;
 311
 312         } else if (th->ack && !(th->fin || th->rst || th->syn)) {
 313                 /* ACK from client */
 314                 synproxy_recv_client_ack(net, skb, th, &opts, ntohl(th->seq));
 315                 return NF_DROP;
 316         }
 317
 318         return XT_CONTINUE;
 319 }
 320
 321 static unsigned int ipv6_synproxy_hook(void *priv,
 322                                        struct sk_buff *skb,
 323                                        const struct nf_hook_state *nhs)
 324 {
 325         struct net *net = nhs->net;
 326         struct synproxy_net *snet = synproxy_pernet(net);
 327         enum ip_conntrack_info ctinfo;
 328         struct nf_conn *ct;
 329         struct nf_conn_synproxy *synproxy;
 330         struct synproxy_options opts = {};
 331         const struct ip_ct_tcp *state;
 332         struct tcphdr *th, _th;
 333         __be16 frag_off;
 334         u8 nexthdr;
 335         int thoff;
 336
 337         ct = nf_ct_get(skb, &ctinfo);
 338         if (ct == NULL)
 339                 return NF_ACCEPT;
 340
 341         synproxy = nfct_synproxy(ct);
 342         if (synproxy == NULL)
 343                 return NF_ACCEPT;
 344
 345         if (nf_is_loopback_packet(skb))
 346                 return NF_ACCEPT;
 347
 348         nexthdr = ipv6_hdr(skb)->nexthdr;
 349         thoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 350                                  &frag_off);
 351         if (thoff < 0)
 352                 return NF_ACCEPT;
 353
 354         th = skb_header_pointer(skb, thoff, sizeof(_th), &_th);
 355         if (th == NULL)
 356                 return NF_DROP;
 357
 358         state = &ct->proto.tcp;
 359         switch (state->state) {
 360         case TCP_CONNTRACK_CLOSE:
 361                 if (th->rst && !test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 362                         nf_ct_seqadj_init(ct, ctinfo, synproxy->isn -
 363                                                       ntohl(th->seq) + 1);
 364                         break;
 365                 }
 366
 367                 if (!th->syn || th->ack ||
 368                     CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 369                         break;
 370
 371                 /* Reopened connection - reset the sequence number and timestamp
 372                  * adjustments, they will get initialized once the connection is
 373                  * reestablished.
 374                  */
 375                 nf_ct_seqadj_init(ct, ctinfo, 0);
 376                 synproxy->tsoff = 0;
 377                 this_cpu_inc(snet->stats->conn_reopened);
 378
 379                 /* fall through */
 380         case TCP_CONNTRACK_SYN_SENT:
 381                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 382                         return NF_DROP;
 383
 384                 if (!th->syn && th->ack &&
 385                     CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
 386                         /* Keep-Alives are sent with SEG.SEQ = SND.NXT-1,
 387                          * therefore we need to add 1 to make the SYN sequence
 388                          * number match the one of first SYN.
 389                          */
 390                         if (synproxy_recv_client_ack(net, skb, th, &opts,
 391                                                      ntohl(th->seq) + 1))
 392                                 this_cpu_inc(snet->stats->cookie_retrans);
 393
 394                         return NF_DROP;
 395                 }
 396
 397                 synproxy->isn = ntohl(th->ack_seq);
 398                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 399                         synproxy->its = opts.tsecr;
 400                 break;
 401         case TCP_CONNTRACK_SYN_RECV:
 402                 if (!th->syn || !th->ack)
 403                         break;
 404
 405                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 406                         return NF_DROP;
 407
 408                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 409                         synproxy->tsoff = opts.tsval - synproxy->its;
 410
 411                 opts.options &= ~(XT_SYNPROXY_OPT_MSS |
 412                                   XT_SYNPROXY_OPT_WSCALE |
 413                                   XT_SYNPROXY_OPT_SACK_PERM);
 414
 415                 swap(opts.tsval, opts.tsecr);
 416                 synproxy_send_server_ack(net, state, skb, th, &opts);
 417
 418                 nf_ct_seqadj_init(ct, ctinfo, synproxy->isn - ntohl(th->seq));
 419
 420                 swap(opts.tsval, opts.tsecr);
 421                 synproxy_send_client_ack(net, skb, th, &opts);
 422
 423                 consume_skb(skb);
 424                 return NF_STOLEN;
 425         default:
 426                 break;
 427         }
 428
 429         synproxy_tstamp_adjust(skb, thoff, th, ct, ctinfo, synproxy);
 430         return NF_ACCEPT;
 431 }
 432
 433 static struct nf_hook_ops ipv6_synproxy_ops[] __read_mostly = {
 434         {
 435                 .hook           = ipv6_synproxy_hook,
 436                 .pf             = NFPROTO_IPV6,
 437                 .hooknum        = NF_INET_LOCAL_IN,
 438                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 439         },
 440         {
 441                 .hook           = ipv6_synproxy_hook,
 442                 .pf             = NFPROTO_IPV6,
 443                 .hooknum        = NF_INET_POST_ROUTING,
 444                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 445         },
 446 };
 447
 448 static int synproxy_tg6_check(const struct xt_tgchk_param *par)
 449 {
 450         struct synproxy_net *snet = synproxy_pernet(par->net);
 451         const struct ip6t_entry *e = par->entryinfo;
 452         int err;
 453
 454         if (!(e->ipv6.flags & IP6T_F_PROTO) ||
 455             e->ipv6.proto != IPPROTO_TCP ||
 456             e->ipv6.invflags & XT_INV_PROTO)
 457                 return -EINVAL;
 458
 459         err = nf_ct_netns_get(par->net, par->family);
 460         if (err)
 461                 return err;
 462
 463         if (snet->hook_ref6 == 0) {
 464                 err = nf_register_net_hooks(par->net, ipv6_synproxy_ops,
 465                                             ARRAY_SIZE(ipv6_synproxy_ops));
 466                 if (err) {
 467                         nf_ct_netns_put(par->net, par->family);
 468                         return err;
 469                 }
 470         }
 471
 472         snet->hook_ref6++;
 473         return err;
 474 }
 475
 476 static void synproxy_tg6_destroy(const struct xt_tgdtor_param *par)
 477 {
 478         struct synproxy_net *snet = synproxy_pernet(par->net);
 479
 480         snet->hook_ref6--;
 481         if (snet->hook_ref6 == 0)
 482                 nf_unregister_net_hooks(par->net, ipv6_synproxy_ops,
 483                                         ARRAY_SIZE(ipv6_synproxy_ops));
 484         nf_ct_netns_put(par->net, par->family);
 485 }
 486
 487 static struct xt_target synproxy_tg6_reg __read_mostly = {
 488         .name           = "SYNPROXY",
 489         .family         = NFPROTO_IPV6,
 490         .hooks          = (1 << NF_INET_LOCAL_IN) | (1 << NF_INET_FORWARD),
 491         .target         = synproxy_tg6,
 492         .targetsize     = sizeof(struct xt_synproxy_info),
 493         .checkentry     = synproxy_tg6_check,
 494         .destroy        = synproxy_tg6_destroy,
 495         .me             = THIS_MODULE,
 496 };
 497
 498 static int __init synproxy_tg6_init(void)
 499 {
 500         return xt_register_target(&synproxy_tg6_reg);
 501 }
 502
 503 static void __exit synproxy_tg6_exit(void)
 504 {
 505         xt_unregister_target(&synproxy_tg6_reg);
 506 }
 507
 508 module_init(synproxy_tg6_init);
 509 module_exit(synproxy_tg6_exit);
 510
 511 MODULE_LICENSE("GPL");
 512 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");