net/ipv6/netfilter/ip6t_SYNPROXY.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2013 Patrick McHardy <kaber@trash.net>
   4  */
   5
   6 #include <linux/module.h>
   7 #include <linux/skbuff.h>
   8 #include <net/ip6_checksum.h>
   9 #include <net/ip6_route.h>
  10 #include <net/tcp.h>
  11
  12 #include <linux/netfilter_ipv6/ip6_tables.h>
  13 #include <linux/netfilter/x_tables.h>
  14 #include <linux/netfilter/xt_SYNPROXY.h>
  15 #include <net/netfilter/nf_conntrack.h>
  16 #include <net/netfilter/nf_conntrack_seqadj.h>
  17 #include <net/netfilter/nf_conntrack_synproxy.h>
  18 #include <net/netfilter/nf_conntrack_ecache.h>
  19
  20 static struct ipv6hdr *
  21 synproxy_build_ip(struct net *net, struct sk_buff *skb,
  22                   const struct in6_addr *saddr,
  23                   const struct in6_addr *daddr)
  24 {
  25         struct ipv6hdr *iph;
  26
  27         skb_reset_network_header(skb);
  28         iph = skb_put(skb, sizeof(*iph));
  29         ip6_flow_hdr(iph, 0, 0);
  30         iph->hop_limit  = net->ipv6.devconf_all->hop_limit;
  31         iph->nexthdr    = IPPROTO_TCP;
  32         iph->saddr      = *saddr;
  33         iph->daddr      = *daddr;
  34
  35         return iph;
  36 }
  37
  38 static void
  39 synproxy_send_tcp(struct net *net,
  40                   const struct sk_buff *skb, struct sk_buff *nskb,
  41                   struct nf_conntrack *nfct, enum ip_conntrack_info ctinfo,
  42                   struct ipv6hdr *niph, struct tcphdr *nth,
  43                   unsigned int tcp_hdr_size)
  44 {
  45         struct dst_entry *dst;
  46         struct flowi6 fl6;
  47
  48         nth->check = ~tcp_v6_check(tcp_hdr_size, &niph->saddr, &niph->daddr, 0);
  49         nskb->ip_summed   = CHECKSUM_PARTIAL;
  50         nskb->csum_start  = (unsigned char *)nth - nskb->head;
  51         nskb->csum_offset = offsetof(struct tcphdr, check);
  52
  53         memset(&fl6, 0, sizeof(fl6));
  54         fl6.flowi6_proto = IPPROTO_TCP;
  55         fl6.saddr = niph->saddr;
  56         fl6.daddr = niph->daddr;
  57         fl6.fl6_sport = nth->source;
  58         fl6.fl6_dport = nth->dest;
  59         security_skb_classify_flow((struct sk_buff *)skb, flowi6_to_flowi(&fl6));
  60         dst = ip6_route_output(net, NULL, &fl6);
  61         if (dst->error) {
  62                 dst_release(dst);
  63                 goto free_nskb;
  64         }
  65         dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
  66         if (IS_ERR(dst))
  67                 goto free_nskb;
  68
  69         skb_dst_set(nskb, dst);
  70
  71         if (nfct) {
  72                 nf_ct_set(nskb, (struct nf_conn *)nfct, ctinfo);
  73                 nf_conntrack_get(nfct);
  74         }
  75
  76         ip6_local_out(net, nskb->sk, nskb);
  77         return;
  78
  79 free_nskb:
  80         kfree_skb(nskb);
  81 }
  82
  83 static void
  84 synproxy_send_client_synack(struct net *net,
  85                             const struct sk_buff *skb, const struct tcphdr *th,
  86                             const struct synproxy_options *opts)
  87 {
  88         struct sk_buff *nskb;
  89         struct ipv6hdr *iph, *niph;
  90         struct tcphdr *nth;
  91         unsigned int tcp_hdr_size;
  92         u16 mss = opts->mss;
  93
  94         iph = ipv6_hdr(skb);
  95
  96         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
  97         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
  98                          GFP_ATOMIC);
  99         if (nskb == NULL)
 100                 return;
 101         skb_reserve(nskb, MAX_TCP_HEADER);
 102
 103         niph = synproxy_build_ip(net, nskb, &iph->daddr, &iph->saddr);
 104
 105         skb_reset_transport_header(nskb);
 106         nth = skb_put(nskb, tcp_hdr_size);
 107         nth->source     = th->dest;
 108         nth->dest       = th->source;
 109         nth->seq        = htonl(__cookie_v6_init_sequence(iph, th, &mss));
 110         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
 111         tcp_flag_word(nth) = TCP_FLAG_SYN | TCP_FLAG_ACK;
 112         if (opts->options & XT_SYNPROXY_OPT_ECN)
 113                 tcp_flag_word(nth) |= TCP_FLAG_ECE;
 114         nth->doff       = tcp_hdr_size / 4;
 115         nth->window     = 0;
 116         nth->check      = 0;
 117         nth->urg_ptr    = 0;
 118
 119         synproxy_build_options(nth, opts);
 120
 121         synproxy_send_tcp(net, skb, nskb, skb_nfct(skb),
 122                           IP_CT_ESTABLISHED_REPLY, niph, nth, tcp_hdr_size);
 123 }
 124
 125 static void
 126 synproxy_send_server_syn(struct net *net,
 127                          const struct sk_buff *skb, const struct tcphdr *th,
 128                          const struct synproxy_options *opts, u32 recv_seq)
 129 {
 130         struct synproxy_net *snet = synproxy_pernet(net);
 131         struct sk_buff *nskb;
 132         struct ipv6hdr *iph, *niph;
 133         struct tcphdr *nth;
 134         unsigned int tcp_hdr_size;
 135
 136         iph = ipv6_hdr(skb);
 137
 138         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 139         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 140                          GFP_ATOMIC);
 141         if (nskb == NULL)
 142                 return;
 143         skb_reserve(nskb, MAX_TCP_HEADER);
 144
 145         niph = synproxy_build_ip(net, nskb, &iph->saddr, &iph->daddr);
 146
 147         skb_reset_transport_header(nskb);
 148         nth = skb_put(nskb, tcp_hdr_size);
 149         nth->source     = th->source;
 150         nth->dest       = th->dest;
 151         nth->seq        = htonl(recv_seq - 1);
 152         /* ack_seq is used to relay our ISN to the synproxy hook to initialize
 153          * sequence number translation once a connection tracking entry exists.
 154          */
 155         nth->ack_seq    = htonl(ntohl(th->ack_seq) - 1);
 156         tcp_flag_word(nth) = TCP_FLAG_SYN;
 157         if (opts->options & XT_SYNPROXY_OPT_ECN)
 158                 tcp_flag_word(nth) |= TCP_FLAG_ECE | TCP_FLAG_CWR;
 159         nth->doff       = tcp_hdr_size / 4;
 160         nth->window     = th->window;
 161         nth->check      = 0;
 162         nth->urg_ptr    = 0;
 163
 164         synproxy_build_options(nth, opts);
 165
 166         synproxy_send_tcp(net, skb, nskb, &snet->tmpl->ct_general, IP_CT_NEW,
 167                           niph, nth, tcp_hdr_size);
 168 }
 169
 170 static void
 171 synproxy_send_server_ack(struct net *net,
 172                          const struct ip_ct_tcp *state,
 173                          const struct sk_buff *skb, const struct tcphdr *th,
 174                          const struct synproxy_options *opts)
 175 {
 176         struct sk_buff *nskb;
 177         struct ipv6hdr *iph, *niph;
 178         struct tcphdr *nth;
 179         unsigned int tcp_hdr_size;
 180
 181         iph = ipv6_hdr(skb);
 182
 183         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 184         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 185                          GFP_ATOMIC);
 186         if (nskb == NULL)
 187                 return;
 188         skb_reserve(nskb, MAX_TCP_HEADER);
 189
 190         niph = synproxy_build_ip(net, nskb, &iph->daddr, &iph->saddr);
 191
 192         skb_reset_transport_header(nskb);
 193         nth = skb_put(nskb, tcp_hdr_size);
 194         nth->source     = th->dest;
 195         nth->dest       = th->source;
 196         nth->seq        = htonl(ntohl(th->ack_seq));
 197         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
 198         tcp_flag_word(nth) = TCP_FLAG_ACK;
 199         nth->doff       = tcp_hdr_size / 4;
 200         nth->window     = htons(state->seen[IP_CT_DIR_ORIGINAL].td_maxwin);
 201         nth->check      = 0;
 202         nth->urg_ptr    = 0;
 203
 204         synproxy_build_options(nth, opts);
 205
 206         synproxy_send_tcp(net, skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
 207 }
 208
 209 static void
 210 synproxy_send_client_ack(struct net *net,
 211                          const struct sk_buff *skb, const struct tcphdr *th,
 212                          const struct synproxy_options *opts)
 213 {
 214         struct sk_buff *nskb;
 215         struct ipv6hdr *iph, *niph;
 216         struct tcphdr *nth;
 217         unsigned int tcp_hdr_size;
 218
 219         iph = ipv6_hdr(skb);
 220
 221         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 222         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 223                          GFP_ATOMIC);
 224         if (nskb == NULL)
 225                 return;
 226         skb_reserve(nskb, MAX_TCP_HEADER);
 227
 228         niph = synproxy_build_ip(net, nskb, &iph->saddr, &iph->daddr);
 229
 230         skb_reset_transport_header(nskb);
 231         nth = skb_put(nskb, tcp_hdr_size);
 232         nth->source     = th->source;
 233         nth->dest       = th->dest;
 234         nth->seq        = htonl(ntohl(th->seq) + 1);
 235         nth->ack_seq    = th->ack_seq;
 236         tcp_flag_word(nth) = TCP_FLAG_ACK;
 237         nth->doff       = tcp_hdr_size / 4;
 238         nth->window     = htons(ntohs(th->window) >> opts->wscale);
 239         nth->check      = 0;
 240         nth->urg_ptr    = 0;
 241
 242         synproxy_build_options(nth, opts);
 243
 244         synproxy_send_tcp(net, skb, nskb, skb_nfct(skb),
 245                           IP_CT_ESTABLISHED_REPLY, niph, nth, tcp_hdr_size);
 246 }
 247
 248 static bool
 249 synproxy_recv_client_ack(struct net *net,
 250                          const struct sk_buff *skb, const struct tcphdr *th,
 251                          struct synproxy_options *opts, u32 recv_seq)
 252 {
 253         struct synproxy_net *snet = synproxy_pernet(net);
 254         int mss;
 255
 256         mss = __cookie_v6_check(ipv6_hdr(skb), th, ntohl(th->ack_seq) - 1);
 257         if (mss == 0) {
 258                 this_cpu_inc(snet->stats->cookie_invalid);
 259                 return false;
 260         }
 261
 262         this_cpu_inc(snet->stats->cookie_valid);
 263         opts->mss = mss;
 264         opts->options |= XT_SYNPROXY_OPT_MSS;
 265
 266         if (opts->options & XT_SYNPROXY_OPT_TIMESTAMP)
 267                 synproxy_check_timestamp_cookie(opts);
 268
 269         synproxy_send_server_syn(net, skb, th, opts, recv_seq);
 270         return true;
 271 }
 272
 273 static unsigned int
 274 synproxy_tg6(struct sk_buff *skb, const struct xt_action_param *par)
 275 {
 276         const struct xt_synproxy_info *info = par->targinfo;
 277         struct net *net = xt_net(par);
 278         struct synproxy_net *snet = synproxy_pernet(net);
 279         struct synproxy_options opts = {};
 280         struct tcphdr *th, _th;
 281
 282         if (nf_ip6_checksum(skb, xt_hooknum(par), par->thoff, IPPROTO_TCP))
 283                 return NF_DROP;
 284
 285         th = skb_header_pointer(skb, par->thoff, sizeof(_th), &_th);
 286         if (th == NULL)
 287                 return NF_DROP;
 288
 289         if (!synproxy_parse_options(skb, par->thoff, th, &opts))
 290                 return NF_DROP;
 291
 292         if (th->syn && !(th->ack || th->fin || th->rst)) {
 293                 /* Initial SYN from client */
 294                 this_cpu_inc(snet->stats->syn_received);
 295
 296                 if (th->ece && th->cwr)
 297                         opts.options |= XT_SYNPROXY_OPT_ECN;
 298
 299                 opts.options &= info->options;
 300                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 301                         synproxy_init_timestamp_cookie(info, &opts);
 302                 else
 303                         opts.options &= ~(XT_SYNPROXY_OPT_WSCALE |
 304                                           XT_SYNPROXY_OPT_SACK_PERM |
 305                                           XT_SYNPROXY_OPT_ECN);
 306
 307                 synproxy_send_client_synack(net, skb, th, &opts);
 308                 consume_skb(skb);
 309                 return NF_STOLEN;
 310
 311         } else if (th->ack && !(th->fin || th->rst || th->syn)) {
 312                 /* ACK from client */
 313                 if (synproxy_recv_client_ack(net, skb, th, &opts, ntohl(th->seq))) {
 314                         consume_skb(skb);
 315                         return NF_STOLEN;
 316                 } else {
 317                         return NF_DROP;
 318                 }
 319         }
 320
 321         return XT_CONTINUE;
 322 }
 323
 324 static unsigned int ipv6_synproxy_hook(void *priv,
 325                                        struct sk_buff *skb,
 326                                        const struct nf_hook_state *nhs)
 327 {
 328         struct net *net = nhs->net;
 329         struct synproxy_net *snet = synproxy_pernet(net);
 330         enum ip_conntrack_info ctinfo;
 331         struct nf_conn *ct;
 332         struct nf_conn_synproxy *synproxy;
 333         struct synproxy_options opts = {};
 334         const struct ip_ct_tcp *state;
 335         struct tcphdr *th, _th;
 336         __be16 frag_off;
 337         u8 nexthdr;
 338         int thoff;
 339
 340         ct = nf_ct_get(skb, &ctinfo);
 341         if (ct == NULL)
 342                 return NF_ACCEPT;
 343
 344         synproxy = nfct_synproxy(ct);
 345         if (synproxy == NULL)
 346                 return NF_ACCEPT;
 347
 348         if (nf_is_loopback_packet(skb))
 349                 return NF_ACCEPT;
 350
 351         nexthdr = ipv6_hdr(skb)->nexthdr;
 352         thoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
 353                                  &frag_off);
 354         if (thoff < 0 || nexthdr != IPPROTO_TCP)
 355                 return NF_ACCEPT;
 356
 357         th = skb_header_pointer(skb, thoff, sizeof(_th), &_th);
 358         if (th == NULL)
 359                 return NF_DROP;
 360
 361         state = &ct->proto.tcp;
 362         switch (state->state) {
 363         case TCP_CONNTRACK_CLOSE:
 364                 if (th->rst && !test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 365                         nf_ct_seqadj_init(ct, ctinfo, synproxy->isn -
 366                                                       ntohl(th->seq) + 1);
 367                         break;
 368                 }
 369
 370                 if (!th->syn || th->ack ||
 371                     CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 372                         break;
 373
 374                 /* Reopened connection - reset the sequence number and timestamp
 375                  * adjustments, they will get initialized once the connection is
 376                  * reestablished.
 377                  */
 378                 nf_ct_seqadj_init(ct, ctinfo, 0);
 379                 synproxy->tsoff = 0;
 380                 this_cpu_inc(snet->stats->conn_reopened);
 381
 382                 /* fall through */
 383         case TCP_CONNTRACK_SYN_SENT:
 384                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 385                         return NF_DROP;
 386
 387                 if (!th->syn && th->ack &&
 388                     CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
 389                         /* Keep-Alives are sent with SEG.SEQ = SND.NXT-1,
 390                          * therefore we need to add 1 to make the SYN sequence
 391                          * number match the one of first SYN.
 392                          */
 393                         if (synproxy_recv_client_ack(net, skb, th, &opts,
 394                                                      ntohl(th->seq) + 1)) {
 395                                 this_cpu_inc(snet->stats->cookie_retrans);
 396                                 consume_skb(skb);
 397                                 return NF_STOLEN;
 398                         } else {
 399                                 return NF_DROP;
 400                         }
 401                 }
 402
 403                 synproxy->isn = ntohl(th->ack_seq);
 404                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 405                         synproxy->its = opts.tsecr;
 406
 407                 nf_conntrack_event_cache(IPCT_SYNPROXY, ct);
 408                 break;
 409         case TCP_CONNTRACK_SYN_RECV:
 410                 if (!th->syn || !th->ack)
 411                         break;
 412
 413                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 414                         return NF_DROP;
 415
 416                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP) {
 417                         synproxy->tsoff = opts.tsval - synproxy->its;
 418                         nf_conntrack_event_cache(IPCT_SYNPROXY, ct);
 419                 }
 420
 421                 opts.options &= ~(XT_SYNPROXY_OPT_MSS |
 422                                   XT_SYNPROXY_OPT_WSCALE |
 423                                   XT_SYNPROXY_OPT_SACK_PERM);
 424
 425                 swap(opts.tsval, opts.tsecr);
 426                 synproxy_send_server_ack(net, state, skb, th, &opts);
 427
 428                 nf_ct_seqadj_init(ct, ctinfo, synproxy->isn - ntohl(th->seq));
 429                 nf_conntrack_event_cache(IPCT_SEQADJ, ct);
 430
 431                 swap(opts.tsval, opts.tsecr);
 432                 synproxy_send_client_ack(net, skb, th, &opts);
 433
 434                 consume_skb(skb);
 435                 return NF_STOLEN;
 436         default:
 437                 break;
 438         }
 439
 440         synproxy_tstamp_adjust(skb, thoff, th, ct, ctinfo, synproxy);
 441         return NF_ACCEPT;
 442 }
 443
 444 static const struct nf_hook_ops ipv6_synproxy_ops[] = {
 445         {
 446                 .hook           = ipv6_synproxy_hook,
 447                 .pf             = NFPROTO_IPV6,
 448                 .hooknum        = NF_INET_LOCAL_IN,
 449                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 450         },
 451         {
 452                 .hook           = ipv6_synproxy_hook,
 453                 .pf             = NFPROTO_IPV6,
 454                 .hooknum        = NF_INET_POST_ROUTING,
 455                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 456         },
 457 };
 458
 459 static int synproxy_tg6_check(const struct xt_tgchk_param *par)
 460 {
 461         struct synproxy_net *snet = synproxy_pernet(par->net);
 462         const struct ip6t_entry *e = par->entryinfo;
 463         int err;
 464
 465         if (!(e->ipv6.flags & IP6T_F_PROTO) ||
 466             e->ipv6.proto != IPPROTO_TCP ||
 467             e->ipv6.invflags & XT_INV_PROTO)
 468                 return -EINVAL;
 469
 470         err = nf_ct_netns_get(par->net, par->family);
 471         if (err)
 472                 return err;
 473
 474         if (snet->hook_ref6 == 0) {
 475                 err = nf_register_net_hooks(par->net, ipv6_synproxy_ops,
 476                                             ARRAY_SIZE(ipv6_synproxy_ops));
 477                 if (err) {
 478                         nf_ct_netns_put(par->net, par->family);
 479                         return err;
 480                 }
 481         }
 482
 483         snet->hook_ref6++;
 484         return err;
 485 }
 486
 487 static void synproxy_tg6_destroy(const struct xt_tgdtor_param *par)
 488 {
 489         struct synproxy_net *snet = synproxy_pernet(par->net);
 490
 491         snet->hook_ref6--;
 492         if (snet->hook_ref6 == 0)
 493                 nf_unregister_net_hooks(par->net, ipv6_synproxy_ops,
 494                                         ARRAY_SIZE(ipv6_synproxy_ops));
 495         nf_ct_netns_put(par->net, par->family);
 496 }
 497
 498 static struct xt_target synproxy_tg6_reg __read_mostly = {
 499         .name           = "SYNPROXY",
 500         .family         = NFPROTO_IPV6,
 501         .hooks          = (1 << NF_INET_LOCAL_IN) | (1 << NF_INET_FORWARD),
 502         .target         = synproxy_tg6,
 503         .targetsize     = sizeof(struct xt_synproxy_info),
 504         .checkentry     = synproxy_tg6_check,
 505         .destroy        = synproxy_tg6_destroy,
 506         .me             = THIS_MODULE,
 507 };
 508
 509 static int __init synproxy_tg6_init(void)
 510 {
 511         return xt_register_target(&synproxy_tg6_reg);
 512 }
 513
 514 static void __exit synproxy_tg6_exit(void)
 515 {
 516         xt_unregister_target(&synproxy_tg6_reg);
 517 }
 518
 519 module_init(synproxy_tg6_init);
 520 module_exit(synproxy_tg6_exit);
 521
 522 MODULE_LICENSE("GPL");
 523 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");