net/ipv4/netfilter/ipt_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/tcp.h>
  12
  13 #include <linux/netfilter_ipv4/ip_tables.h>
  14 #include <linux/netfilter/x_tables.h>
  15 #include <linux/netfilter/xt_SYNPROXY.h>
  16 #include <net/netfilter/nf_conntrack.h>
  17 #include <net/netfilter/nf_conntrack_seqadj.h>
  18 #include <net/netfilter/nf_conntrack_synproxy.h>
  19
  20 static struct iphdr *
  21 synproxy_build_ip(struct sk_buff *skb, __be32 saddr, __be32 daddr)
  22 {
  23         struct iphdr *iph;
  24
  25         skb_reset_network_header(skb);
  26         iph = (struct iphdr *)skb_put(skb, sizeof(*iph));
  27         iph->version    = 4;
  28         iph->ihl        = sizeof(*iph) / 4;
  29         iph->tos        = 0;
  30         iph->id         = 0;
  31         iph->frag_off   = htons(IP_DF);
  32         iph->ttl        = sysctl_ip_default_ttl;
  33         iph->protocol   = IPPROTO_TCP;
  34         iph->check      = 0;
  35         iph->saddr      = saddr;
  36         iph->daddr      = daddr;
  37
  38         return iph;
  39 }
  40
  41 static void
  42 synproxy_send_tcp(const struct sk_buff *skb, struct sk_buff *nskb,
  43                   struct nf_conntrack *nfct, enum ip_conntrack_info ctinfo,
  44                   struct iphdr *niph, struct tcphdr *nth,
  45                   unsigned int tcp_hdr_size)
  46 {
  47         nth->check = ~tcp_v4_check(tcp_hdr_size, niph->saddr, niph->daddr, 0);
  48         nskb->ip_summed   = CHECKSUM_PARTIAL;
  49         nskb->csum_start  = (unsigned char *)nth - nskb->head;
  50         nskb->csum_offset = offsetof(struct tcphdr, check);
  51
  52         skb_dst_set_noref(nskb, skb_dst(skb));
  53         nskb->protocol = htons(ETH_P_IP);
  54         if (ip_route_me_harder(nskb, RTN_UNSPEC))
  55                 goto free_nskb;
  56
  57         if (nfct) {
  58                 nskb->nfct = nfct;
  59                 nskb->nfctinfo = ctinfo;
  60                 nf_conntrack_get(nfct);
  61         }
  62
  63         ip_local_out(nskb);
  64         return;
  65
  66 free_nskb:
  67         kfree_skb(nskb);
  68 }
  69
  70 static void
  71 synproxy_send_client_synack(const struct sk_buff *skb, const struct tcphdr *th,
  72                             const struct synproxy_options *opts)
  73 {
  74         struct sk_buff *nskb;
  75         struct iphdr *iph, *niph;
  76         struct tcphdr *nth;
  77         unsigned int tcp_hdr_size;
  78         u16 mss = opts->mss;
  79
  80         iph = ip_hdr(skb);
  81
  82         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
  83         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
  84                          GFP_ATOMIC);
  85         if (nskb == NULL)
  86                 return;
  87         skb_reserve(nskb, MAX_TCP_HEADER);
  88
  89         niph = synproxy_build_ip(nskb, iph->daddr, iph->saddr);
  90
  91         skb_reset_transport_header(nskb);
  92         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
  93         nth->source     = th->dest;
  94         nth->dest       = th->source;
  95         nth->seq        = htonl(__cookie_v4_init_sequence(iph, th, &mss));
  96         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
  97         tcp_flag_word(nth) = TCP_FLAG_SYN | TCP_FLAG_ACK;
  98         if (opts->options & XT_SYNPROXY_OPT_ECN)
  99                 tcp_flag_word(nth) |= TCP_FLAG_ECE;
 100         nth->doff       = tcp_hdr_size / 4;
 101         nth->window     = 0;
 102         nth->check      = 0;
 103         nth->urg_ptr    = 0;
 104
 105         synproxy_build_options(nth, opts);
 106
 107         synproxy_send_tcp(skb, nskb, skb->nfct, IP_CT_ESTABLISHED_REPLY,
 108                           niph, nth, tcp_hdr_size);
 109 }
 110
 111 static void
 112 synproxy_send_server_syn(const struct synproxy_net *snet,
 113                          const struct sk_buff *skb, const struct tcphdr *th,
 114                          const struct synproxy_options *opts, u32 recv_seq)
 115 {
 116         struct sk_buff *nskb;
 117         struct iphdr *iph, *niph;
 118         struct tcphdr *nth;
 119         unsigned int tcp_hdr_size;
 120
 121         iph = ip_hdr(skb);
 122
 123         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 124         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 125                          GFP_ATOMIC);
 126         if (nskb == NULL)
 127                 return;
 128         skb_reserve(nskb, MAX_TCP_HEADER);
 129
 130         niph = synproxy_build_ip(nskb, iph->saddr, iph->daddr);
 131
 132         skb_reset_transport_header(nskb);
 133         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 134         nth->source     = th->source;
 135         nth->dest       = th->dest;
 136         nth->seq        = htonl(recv_seq - 1);
 137         /* ack_seq is used to relay our ISN to the synproxy hook to initialize
 138          * sequence number translation once a connection tracking entry exists.
 139          */
 140         nth->ack_seq    = htonl(ntohl(th->ack_seq) - 1);
 141         tcp_flag_word(nth) = TCP_FLAG_SYN;
 142         if (opts->options & XT_SYNPROXY_OPT_ECN)
 143                 tcp_flag_word(nth) |= TCP_FLAG_ECE | TCP_FLAG_CWR;
 144         nth->doff       = tcp_hdr_size / 4;
 145         nth->window     = th->window;
 146         nth->check      = 0;
 147         nth->urg_ptr    = 0;
 148
 149         synproxy_build_options(nth, opts);
 150
 151         synproxy_send_tcp(skb, nskb, &snet->tmpl->ct_general, IP_CT_NEW,
 152                           niph, nth, tcp_hdr_size);
 153 }
 154
 155 static void
 156 synproxy_send_server_ack(const struct synproxy_net *snet,
 157                          const struct ip_ct_tcp *state,
 158                          const struct sk_buff *skb, const struct tcphdr *th,
 159                          const struct synproxy_options *opts)
 160 {
 161         struct sk_buff *nskb;
 162         struct iphdr *iph, *niph;
 163         struct tcphdr *nth;
 164         unsigned int tcp_hdr_size;
 165
 166         iph = ip_hdr(skb);
 167
 168         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 169         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 170                          GFP_ATOMIC);
 171         if (nskb == NULL)
 172                 return;
 173         skb_reserve(nskb, MAX_TCP_HEADER);
 174
 175         niph = synproxy_build_ip(nskb, iph->daddr, iph->saddr);
 176
 177         skb_reset_transport_header(nskb);
 178         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 179         nth->source     = th->dest;
 180         nth->dest       = th->source;
 181         nth->seq        = htonl(ntohl(th->ack_seq));
 182         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
 183         tcp_flag_word(nth) = TCP_FLAG_ACK;
 184         nth->doff       = tcp_hdr_size / 4;
 185         nth->window     = htons(state->seen[IP_CT_DIR_ORIGINAL].td_maxwin);
 186         nth->check      = 0;
 187         nth->urg_ptr    = 0;
 188
 189         synproxy_build_options(nth, opts);
 190
 191         synproxy_send_tcp(skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
 192 }
 193
 194 static void
 195 synproxy_send_client_ack(const struct synproxy_net *snet,
 196                          const struct sk_buff *skb, const struct tcphdr *th,
 197                          const struct synproxy_options *opts)
 198 {
 199         struct sk_buff *nskb;
 200         struct iphdr *iph, *niph;
 201         struct tcphdr *nth;
 202         unsigned int tcp_hdr_size;
 203
 204         iph = ip_hdr(skb);
 205
 206         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 207         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 208                          GFP_ATOMIC);
 209         if (nskb == NULL)
 210                 return;
 211         skb_reserve(nskb, MAX_TCP_HEADER);
 212
 213         niph = synproxy_build_ip(nskb, iph->saddr, iph->daddr);
 214
 215         skb_reset_transport_header(nskb);
 216         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 217         nth->source     = th->source;
 218         nth->dest       = th->dest;
 219         nth->seq        = htonl(ntohl(th->seq) + 1);
 220         nth->ack_seq    = th->ack_seq;
 221         tcp_flag_word(nth) = TCP_FLAG_ACK;
 222         nth->doff       = tcp_hdr_size / 4;
 223         nth->window     = htons(ntohs(th->window) >> opts->wscale);
 224         nth->check      = 0;
 225         nth->urg_ptr    = 0;
 226
 227         synproxy_build_options(nth, opts);
 228
 229         synproxy_send_tcp(skb, nskb, skb->nfct, IP_CT_ESTABLISHED_REPLY,
 230                           niph, nth, tcp_hdr_size);
 231 }
 232
 233 static bool
 234 synproxy_recv_client_ack(const struct synproxy_net *snet,
 235                          const struct sk_buff *skb, const struct tcphdr *th,
 236                          struct synproxy_options *opts, u32 recv_seq)
 237 {
 238         int mss;
 239
 240         mss = __cookie_v4_check(ip_hdr(skb), th, ntohl(th->ack_seq) - 1);
 241         if (mss == 0) {
 242                 this_cpu_inc(snet->stats->cookie_invalid);
 243                 return false;
 244         }
 245
 246         this_cpu_inc(snet->stats->cookie_valid);
 247         opts->mss = mss;
 248         opts->options |= XT_SYNPROXY_OPT_MSS;
 249
 250         if (opts->options & XT_SYNPROXY_OPT_TIMESTAMP)
 251                 synproxy_check_timestamp_cookie(opts);
 252
 253         synproxy_send_server_syn(snet, skb, th, opts, recv_seq);
 254         return true;
 255 }
 256
 257 static unsigned int
 258 synproxy_tg4(struct sk_buff *skb, const struct xt_action_param *par)
 259 {
 260         const struct xt_synproxy_info *info = par->targinfo;
 261         struct synproxy_net *snet = synproxy_pernet(dev_net(par->in));
 262         struct synproxy_options opts = {};
 263         struct tcphdr *th, _th;
 264
 265         if (nf_ip_checksum(skb, par->hooknum, par->thoff, IPPROTO_TCP))
 266                 return NF_DROP;
 267
 268         th = skb_header_pointer(skb, par->thoff, sizeof(_th), &_th);
 269         if (th == NULL)
 270                 return NF_DROP;
 271
 272         if (!synproxy_parse_options(skb, par->thoff, th, &opts))
 273                 return NF_DROP;
 274
 275         if (th->syn && !(th->ack || th->fin || th->rst)) {
 276                 /* Initial SYN from client */
 277                 this_cpu_inc(snet->stats->syn_received);
 278
 279                 if (th->ece && th->cwr)
 280                         opts.options |= XT_SYNPROXY_OPT_ECN;
 281
 282                 opts.options &= info->options;
 283                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 284                         synproxy_init_timestamp_cookie(info, &opts);
 285                 else
 286                         opts.options &= ~(XT_SYNPROXY_OPT_WSCALE |
 287                                           XT_SYNPROXY_OPT_SACK_PERM |
 288                                           XT_SYNPROXY_OPT_ECN);
 289
 290                 synproxy_send_client_synack(skb, th, &opts);
 291                 return NF_DROP;
 292
 293         } else if (th->ack && !(th->fin || th->rst || th->syn)) {
 294                 /* ACK from client */
 295                 synproxy_recv_client_ack(snet, skb, th, &opts, ntohl(th->seq));
 296                 return NF_DROP;
 297         }
 298
 299         return XT_CONTINUE;
 300 }
 301
 302 static unsigned int ipv4_synproxy_hook(const struct nf_hook_ops *ops,
 303                                        struct sk_buff *skb,
 304                                        const struct nf_hook_state *nhs)
 305 {
 306         struct synproxy_net *snet = synproxy_pernet(dev_net(nhs->in ? : nhs->out));
 307         enum ip_conntrack_info ctinfo;
 308         struct nf_conn *ct;
 309         struct nf_conn_synproxy *synproxy;
 310         struct synproxy_options opts = {};
 311         const struct ip_ct_tcp *state;
 312         struct tcphdr *th, _th;
 313         unsigned int thoff;
 314
 315         ct = nf_ct_get(skb, &ctinfo);
 316         if (ct == NULL)
 317                 return NF_ACCEPT;
 318
 319         synproxy = nfct_synproxy(ct);
 320         if (synproxy == NULL)
 321                 return NF_ACCEPT;
 322
 323         if (nf_is_loopback_packet(skb))
 324                 return NF_ACCEPT;
 325
 326         thoff = ip_hdrlen(skb);
 327         th = skb_header_pointer(skb, thoff, sizeof(_th), &_th);
 328         if (th == NULL)
 329                 return NF_DROP;
 330
 331         state = &ct->proto.tcp;
 332         switch (state->state) {
 333         case TCP_CONNTRACK_CLOSE:
 334                 if (th->rst && !test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 335                         nf_ct_seqadj_init(ct, ctinfo, synproxy->isn -
 336                                                       ntohl(th->seq) + 1);
 337                         break;
 338                 }
 339
 340                 if (!th->syn || th->ack ||
 341                     CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 342                         break;
 343
 344                 /* Reopened connection - reset the sequence number and timestamp
 345                  * adjustments, they will get initialized once the connection is
 346                  * reestablished.
 347                  */
 348                 nf_ct_seqadj_init(ct, ctinfo, 0);
 349                 synproxy->tsoff = 0;
 350                 this_cpu_inc(snet->stats->conn_reopened);
 351
 352                 /* fall through */
 353         case TCP_CONNTRACK_SYN_SENT:
 354                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 355                         return NF_DROP;
 356
 357                 if (!th->syn && th->ack &&
 358                     CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
 359                         /* Keep-Alives are sent with SEG.SEQ = SND.NXT-1,
 360                          * therefore we need to add 1 to make the SYN sequence
 361                          * number match the one of first SYN.
 362                          */
 363                         if (synproxy_recv_client_ack(snet, skb, th, &opts,
 364                                                      ntohl(th->seq) + 1))
 365                                 this_cpu_inc(snet->stats->cookie_retrans);
 366
 367                         return NF_DROP;
 368                 }
 369
 370                 synproxy->isn = ntohl(th->ack_seq);
 371                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 372                         synproxy->its = opts.tsecr;
 373                 break;
 374         case TCP_CONNTRACK_SYN_RECV:
 375                 if (!th->syn || !th->ack)
 376                         break;
 377
 378                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 379                         return NF_DROP;
 380
 381                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 382                         synproxy->tsoff = opts.tsval - synproxy->its;
 383
 384                 opts.options &= ~(XT_SYNPROXY_OPT_MSS |
 385                                   XT_SYNPROXY_OPT_WSCALE |
 386                                   XT_SYNPROXY_OPT_SACK_PERM);
 387
 388                 swap(opts.tsval, opts.tsecr);
 389                 synproxy_send_server_ack(snet, state, skb, th, &opts);
 390
 391                 nf_ct_seqadj_init(ct, ctinfo, synproxy->isn - ntohl(th->seq));
 392
 393                 swap(opts.tsval, opts.tsecr);
 394                 synproxy_send_client_ack(snet, skb, th, &opts);
 395
 396                 consume_skb(skb);
 397                 return NF_STOLEN;
 398         default:
 399                 break;
 400         }
 401
 402         synproxy_tstamp_adjust(skb, thoff, th, ct, ctinfo, synproxy);
 403         return NF_ACCEPT;
 404 }
 405
 406 static int synproxy_tg4_check(const struct xt_tgchk_param *par)
 407 {
 408         const struct ipt_entry *e = par->entryinfo;
 409
 410         if (e->ip.proto != IPPROTO_TCP ||
 411             e->ip.invflags & XT_INV_PROTO)
 412                 return -EINVAL;
 413
 414         return nf_ct_l3proto_try_module_get(par->family);
 415 }
 416
 417 static void synproxy_tg4_destroy(const struct xt_tgdtor_param *par)
 418 {
 419         nf_ct_l3proto_module_put(par->family);
 420 }
 421
 422 static struct xt_target synproxy_tg4_reg __read_mostly = {
 423         .name           = "SYNPROXY",
 424         .family         = NFPROTO_IPV4,
 425         .hooks          = (1 << NF_INET_LOCAL_IN) | (1 << NF_INET_FORWARD),
 426         .target         = synproxy_tg4,
 427         .targetsize     = sizeof(struct xt_synproxy_info),
 428         .checkentry     = synproxy_tg4_check,
 429         .destroy        = synproxy_tg4_destroy,
 430         .me             = THIS_MODULE,
 431 };
 432
 433 static struct nf_hook_ops ipv4_synproxy_ops[] __read_mostly = {
 434         {
 435                 .hook           = ipv4_synproxy_hook,
 436                 .owner          = THIS_MODULE,
 437                 .pf             = NFPROTO_IPV4,
 438                 .hooknum        = NF_INET_LOCAL_IN,
 439                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 440         },
 441         {
 442                 .hook           = ipv4_synproxy_hook,
 443                 .owner          = THIS_MODULE,
 444                 .pf             = NFPROTO_IPV4,
 445                 .hooknum        = NF_INET_POST_ROUTING,
 446                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 447         },
 448 };
 449
 450 static int __init synproxy_tg4_init(void)
 451 {
 452         int err;
 453
 454         err = nf_register_hooks(ipv4_synproxy_ops,
 455                                 ARRAY_SIZE(ipv4_synproxy_ops));
 456         if (err < 0)
 457                 goto err1;
 458
 459         err = xt_register_target(&synproxy_tg4_reg);
 460         if (err < 0)
 461                 goto err2;
 462
 463         return 0;
 464
 465 err2:
 466         nf_unregister_hooks(ipv4_synproxy_ops, ARRAY_SIZE(ipv4_synproxy_ops));
 467 err1:
 468         return err;
 469 }
 470
 471 static void __exit synproxy_tg4_exit(void)
 472 {
 473         xt_unregister_target(&synproxy_tg4_reg);
 474         nf_unregister_hooks(ipv4_synproxy_ops, ARRAY_SIZE(ipv4_synproxy_ops));
 475 }
 476
 477 module_init(synproxy_tg4_init);
 478 module_exit(synproxy_tg4_exit);
 479
 480 MODULE_LICENSE("GPL");
 481 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");