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 net *net, struct sk_buff *skb, __be32 saddr,
  22                   __be32 daddr)
  23 {
  24         struct iphdr *iph;
  25
  26         skb_reset_network_header(skb);
  27         iph = (struct iphdr *)skb_put(skb, sizeof(*iph));
  28         iph->version    = 4;
  29         iph->ihl        = sizeof(*iph) / 4;
  30         iph->tos        = 0;
  31         iph->id         = 0;
  32         iph->frag_off   = htons(IP_DF);
  33         iph->ttl        = net->ipv4.sysctl_ip_default_ttl;
  34         iph->protocol   = IPPROTO_TCP;
  35         iph->check      = 0;
  36         iph->saddr      = saddr;
  37         iph->daddr      = daddr;
  38
  39         return iph;
  40 }
  41
  42 static void
  43 synproxy_send_tcp(struct net *net,
  44                   const struct sk_buff *skb, struct sk_buff *nskb,
  45                   struct nf_conntrack *nfct, enum ip_conntrack_info ctinfo,
  46                   struct iphdr *niph, struct tcphdr *nth,
  47                   unsigned int tcp_hdr_size)
  48 {
  49         nth->check = ~tcp_v4_check(tcp_hdr_size, niph->saddr, niph->daddr, 0);
  50         nskb->ip_summed   = CHECKSUM_PARTIAL;
  51         nskb->csum_start  = (unsigned char *)nth - nskb->head;
  52         nskb->csum_offset = offsetof(struct tcphdr, check);
  53
  54         skb_dst_set_noref(nskb, skb_dst(skb));
  55         nskb->protocol = htons(ETH_P_IP);
  56         if (ip_route_me_harder(net, nskb, RTN_UNSPEC))
  57                 goto free_nskb;
  58
  59         if (nfct) {
  60                 nf_ct_set(nskb, (struct nf_conn *)nfct, ctinfo);
  61                 nf_conntrack_get(nfct);
  62         }
  63
  64         ip_local_out(net, nskb->sk, nskb);
  65         return;
  66
  67 free_nskb:
  68         kfree_skb(nskb);
  69 }
  70
  71 static void
  72 synproxy_send_client_synack(struct net *net,
  73                             const struct sk_buff *skb, const struct tcphdr *th,
  74                             const struct synproxy_options *opts)
  75 {
  76         struct sk_buff *nskb;
  77         struct iphdr *iph, *niph;
  78         struct tcphdr *nth;
  79         unsigned int tcp_hdr_size;
  80         u16 mss = opts->mss;
  81
  82         iph = ip_hdr(skb);
  83
  84         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
  85         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
  86                          GFP_ATOMIC);
  87         if (nskb == NULL)
  88                 return;
  89         skb_reserve(nskb, MAX_TCP_HEADER);
  90
  91         niph = synproxy_build_ip(net, nskb, iph->daddr, iph->saddr);
  92
  93         skb_reset_transport_header(nskb);
  94         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
  95         nth->source     = th->dest;
  96         nth->dest       = th->source;
  97         nth->seq        = htonl(__cookie_v4_init_sequence(iph, th, &mss));
  98         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
  99         tcp_flag_word(nth) = TCP_FLAG_SYN | TCP_FLAG_ACK;
 100         if (opts->options & XT_SYNPROXY_OPT_ECN)
 101                 tcp_flag_word(nth) |= TCP_FLAG_ECE;
 102         nth->doff       = tcp_hdr_size / 4;
 103         nth->window     = 0;
 104         nth->check      = 0;
 105         nth->urg_ptr    = 0;
 106
 107         synproxy_build_options(nth, opts);
 108
 109         synproxy_send_tcp(net, skb, nskb, skb_nfct(skb),
 110                           IP_CT_ESTABLISHED_REPLY, niph, nth, tcp_hdr_size);
 111 }
 112
 113 static void
 114 synproxy_send_server_syn(struct net *net,
 115                          const struct sk_buff *skb, const struct tcphdr *th,
 116                          const struct synproxy_options *opts, u32 recv_seq)
 117 {
 118         struct synproxy_net *snet = synproxy_pernet(net);
 119         struct sk_buff *nskb;
 120         struct iphdr *iph, *niph;
 121         struct tcphdr *nth;
 122         unsigned int tcp_hdr_size;
 123
 124         iph = ip_hdr(skb);
 125
 126         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 127         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 128                          GFP_ATOMIC);
 129         if (nskb == NULL)
 130                 return;
 131         skb_reserve(nskb, MAX_TCP_HEADER);
 132
 133         niph = synproxy_build_ip(net, nskb, iph->saddr, iph->daddr);
 134
 135         skb_reset_transport_header(nskb);
 136         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 137         nth->source     = th->source;
 138         nth->dest       = th->dest;
 139         nth->seq        = htonl(recv_seq - 1);
 140         /* ack_seq is used to relay our ISN to the synproxy hook to initialize
 141          * sequence number translation once a connection tracking entry exists.
 142          */
 143         nth->ack_seq    = htonl(ntohl(th->ack_seq) - 1);
 144         tcp_flag_word(nth) = TCP_FLAG_SYN;
 145         if (opts->options & XT_SYNPROXY_OPT_ECN)
 146                 tcp_flag_word(nth) |= TCP_FLAG_ECE | TCP_FLAG_CWR;
 147         nth->doff       = tcp_hdr_size / 4;
 148         nth->window     = th->window;
 149         nth->check      = 0;
 150         nth->urg_ptr    = 0;
 151
 152         synproxy_build_options(nth, opts);
 153
 154         synproxy_send_tcp(net, skb, nskb, &snet->tmpl->ct_general, IP_CT_NEW,
 155                           niph, nth, tcp_hdr_size);
 156 }
 157
 158 static void
 159 synproxy_send_server_ack(struct net *net,
 160                          const struct ip_ct_tcp *state,
 161                          const struct sk_buff *skb, const struct tcphdr *th,
 162                          const struct synproxy_options *opts)
 163 {
 164         struct sk_buff *nskb;
 165         struct iphdr *iph, *niph;
 166         struct tcphdr *nth;
 167         unsigned int tcp_hdr_size;
 168
 169         iph = ip_hdr(skb);
 170
 171         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 172         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 173                          GFP_ATOMIC);
 174         if (nskb == NULL)
 175                 return;
 176         skb_reserve(nskb, MAX_TCP_HEADER);
 177
 178         niph = synproxy_build_ip(net, nskb, iph->daddr, iph->saddr);
 179
 180         skb_reset_transport_header(nskb);
 181         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 182         nth->source     = th->dest;
 183         nth->dest       = th->source;
 184         nth->seq        = htonl(ntohl(th->ack_seq));
 185         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
 186         tcp_flag_word(nth) = TCP_FLAG_ACK;
 187         nth->doff       = tcp_hdr_size / 4;
 188         nth->window     = htons(state->seen[IP_CT_DIR_ORIGINAL].td_maxwin);
 189         nth->check      = 0;
 190         nth->urg_ptr    = 0;
 191
 192         synproxy_build_options(nth, opts);
 193
 194         synproxy_send_tcp(net, skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
 195 }
 196
 197 static void
 198 synproxy_send_client_ack(struct net *net,
 199                          const struct sk_buff *skb, const struct tcphdr *th,
 200                          const struct synproxy_options *opts)
 201 {
 202         struct sk_buff *nskb;
 203         struct iphdr *iph, *niph;
 204         struct tcphdr *nth;
 205         unsigned int tcp_hdr_size;
 206
 207         iph = ip_hdr(skb);
 208
 209         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 210         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 211                          GFP_ATOMIC);
 212         if (nskb == NULL)
 213                 return;
 214         skb_reserve(nskb, MAX_TCP_HEADER);
 215
 216         niph = synproxy_build_ip(net, nskb, iph->saddr, iph->daddr);
 217
 218         skb_reset_transport_header(nskb);
 219         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 220         nth->source     = th->source;
 221         nth->dest       = th->dest;
 222         nth->seq        = htonl(ntohl(th->seq) + 1);
 223         nth->ack_seq    = th->ack_seq;
 224         tcp_flag_word(nth) = TCP_FLAG_ACK;
 225         nth->doff       = tcp_hdr_size / 4;
 226         nth->window     = htons(ntohs(th->window) >> opts->wscale);
 227         nth->check      = 0;
 228         nth->urg_ptr    = 0;
 229
 230         synproxy_build_options(nth, opts);
 231
 232         synproxy_send_tcp(net, skb, nskb, skb_nfct(skb),
 233                           IP_CT_ESTABLISHED_REPLY, niph, nth, tcp_hdr_size);
 234 }
 235
 236 static bool
 237 synproxy_recv_client_ack(struct net *net,
 238                          const struct sk_buff *skb, const struct tcphdr *th,
 239                          struct synproxy_options *opts, u32 recv_seq)
 240 {
 241         struct synproxy_net *snet = synproxy_pernet(net);
 242         int mss;
 243
 244         mss = __cookie_v4_check(ip_hdr(skb), th, ntohl(th->ack_seq) - 1);
 245         if (mss == 0) {
 246                 this_cpu_inc(snet->stats->cookie_invalid);
 247                 return false;
 248         }
 249
 250         this_cpu_inc(snet->stats->cookie_valid);
 251         opts->mss = mss;
 252         opts->options |= XT_SYNPROXY_OPT_MSS;
 253
 254         if (opts->options & XT_SYNPROXY_OPT_TIMESTAMP)
 255                 synproxy_check_timestamp_cookie(opts);
 256
 257         synproxy_send_server_syn(net, skb, th, opts, recv_seq);
 258         return true;
 259 }
 260
 261 static unsigned int
 262 synproxy_tg4(struct sk_buff *skb, const struct xt_action_param *par)
 263 {
 264         const struct xt_synproxy_info *info = par->targinfo;
 265         struct net *net = xt_net(par);
 266         struct synproxy_net *snet = synproxy_pernet(net);
 267         struct synproxy_options opts = {};
 268         struct tcphdr *th, _th;
 269
 270         if (nf_ip_checksum(skb, xt_hooknum(par), par->thoff, IPPROTO_TCP))
 271                 return NF_DROP;
 272
 273         th = skb_header_pointer(skb, par->thoff, sizeof(_th), &_th);
 274         if (th == NULL)
 275                 return NF_DROP;
 276
 277         if (!synproxy_parse_options(skb, par->thoff, th, &opts))
 278                 return NF_DROP;
 279
 280         if (th->syn && !(th->ack || th->fin || th->rst)) {
 281                 /* Initial SYN from client */
 282                 this_cpu_inc(snet->stats->syn_received);
 283
 284                 if (th->ece && th->cwr)
 285                         opts.options |= XT_SYNPROXY_OPT_ECN;
 286
 287                 opts.options &= info->options;
 288                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 289                         synproxy_init_timestamp_cookie(info, &opts);
 290                 else
 291                         opts.options &= ~(XT_SYNPROXY_OPT_WSCALE |
 292                                           XT_SYNPROXY_OPT_SACK_PERM |
 293                                           XT_SYNPROXY_OPT_ECN);
 294
 295                 synproxy_send_client_synack(net, skb, th, &opts);
 296                 return NF_DROP;
 297
 298         } else if (th->ack && !(th->fin || th->rst || th->syn)) {
 299                 /* ACK from client */
 300                 synproxy_recv_client_ack(net, skb, th, &opts, ntohl(th->seq));
 301                 return NF_DROP;
 302         }
 303
 304         return XT_CONTINUE;
 305 }
 306
 307 static unsigned int ipv4_synproxy_hook(void *priv,
 308                                        struct sk_buff *skb,
 309                                        const struct nf_hook_state *nhs)
 310 {
 311         struct net *net = nhs->net;
 312         struct synproxy_net *snet = synproxy_pernet(net);
 313         enum ip_conntrack_info ctinfo;
 314         struct nf_conn *ct;
 315         struct nf_conn_synproxy *synproxy;
 316         struct synproxy_options opts = {};
 317         const struct ip_ct_tcp *state;
 318         struct tcphdr *th, _th;
 319         unsigned int thoff;
 320
 321         ct = nf_ct_get(skb, &ctinfo);
 322         if (ct == NULL)
 323                 return NF_ACCEPT;
 324
 325         synproxy = nfct_synproxy(ct);
 326         if (synproxy == NULL)
 327                 return NF_ACCEPT;
 328
 329         if (nf_is_loopback_packet(skb))
 330                 return NF_ACCEPT;
 331
 332         thoff = ip_hdrlen(skb);
 333         th = skb_header_pointer(skb, thoff, sizeof(_th), &_th);
 334         if (th == NULL)
 335                 return NF_DROP;
 336
 337         state = &ct->proto.tcp;
 338         switch (state->state) {
 339         case TCP_CONNTRACK_CLOSE:
 340                 if (th->rst && !test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 341                         nf_ct_seqadj_init(ct, ctinfo, synproxy->isn -
 342                                                       ntohl(th->seq) + 1);
 343                         break;
 344                 }
 345
 346                 if (!th->syn || th->ack ||
 347                     CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 348                         break;
 349
 350                 /* Reopened connection - reset the sequence number and timestamp
 351                  * adjustments, they will get initialized once the connection is
 352                  * reestablished.
 353                  */
 354                 nf_ct_seqadj_init(ct, ctinfo, 0);
 355                 synproxy->tsoff = 0;
 356                 this_cpu_inc(snet->stats->conn_reopened);
 357
 358                 /* fall through */
 359         case TCP_CONNTRACK_SYN_SENT:
 360                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 361                         return NF_DROP;
 362
 363                 if (!th->syn && th->ack &&
 364                     CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
 365                         /* Keep-Alives are sent with SEG.SEQ = SND.NXT-1,
 366                          * therefore we need to add 1 to make the SYN sequence
 367                          * number match the one of first SYN.
 368                          */
 369                         if (synproxy_recv_client_ack(net, skb, th, &opts,
 370                                                      ntohl(th->seq) + 1))
 371                                 this_cpu_inc(snet->stats->cookie_retrans);
 372
 373                         return NF_DROP;
 374                 }
 375
 376                 synproxy->isn = ntohl(th->ack_seq);
 377                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 378                         synproxy->its = opts.tsecr;
 379                 break;
 380         case TCP_CONNTRACK_SYN_RECV:
 381                 if (!th->syn || !th->ack)
 382                         break;
 383
 384                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 385                         return NF_DROP;
 386
 387                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 388                         synproxy->tsoff = opts.tsval - synproxy->its;
 389
 390                 opts.options &= ~(XT_SYNPROXY_OPT_MSS |
 391                                   XT_SYNPROXY_OPT_WSCALE |
 392                                   XT_SYNPROXY_OPT_SACK_PERM);
 393
 394                 swap(opts.tsval, opts.tsecr);
 395                 synproxy_send_server_ack(net, state, skb, th, &opts);
 396
 397                 nf_ct_seqadj_init(ct, ctinfo, synproxy->isn - ntohl(th->seq));
 398
 399                 swap(opts.tsval, opts.tsecr);
 400                 synproxy_send_client_ack(net, skb, th, &opts);
 401
 402                 consume_skb(skb);
 403                 return NF_STOLEN;
 404         default:
 405                 break;
 406         }
 407
 408         synproxy_tstamp_adjust(skb, thoff, th, ct, ctinfo, synproxy);
 409         return NF_ACCEPT;
 410 }
 411
 412 static int synproxy_tg4_check(const struct xt_tgchk_param *par)
 413 {
 414         const struct ipt_entry *e = par->entryinfo;
 415
 416         if (e->ip.proto != IPPROTO_TCP ||
 417             e->ip.invflags & XT_INV_PROTO)
 418                 return -EINVAL;
 419
 420         return nf_ct_netns_get(par->net, par->family);
 421 }
 422
 423 static void synproxy_tg4_destroy(const struct xt_tgdtor_param *par)
 424 {
 425         nf_ct_netns_put(par->net, par->family);
 426 }
 427
 428 static struct xt_target synproxy_tg4_reg __read_mostly = {
 429         .name           = "SYNPROXY",
 430         .family         = NFPROTO_IPV4,
 431         .hooks          = (1 << NF_INET_LOCAL_IN) | (1 << NF_INET_FORWARD),
 432         .target         = synproxy_tg4,
 433         .targetsize     = sizeof(struct xt_synproxy_info),
 434         .checkentry     = synproxy_tg4_check,
 435         .destroy        = synproxy_tg4_destroy,
 436         .me             = THIS_MODULE,
 437 };
 438
 439 static struct nf_hook_ops ipv4_synproxy_ops[] __read_mostly = {
 440         {
 441                 .hook           = ipv4_synproxy_hook,
 442                 .pf             = NFPROTO_IPV4,
 443                 .hooknum        = NF_INET_LOCAL_IN,
 444                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 445         },
 446         {
 447                 .hook           = ipv4_synproxy_hook,
 448                 .pf             = NFPROTO_IPV4,
 449                 .hooknum        = NF_INET_POST_ROUTING,
 450                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 451         },
 452 };
 453
 454 static int __init synproxy_tg4_init(void)
 455 {
 456         int err;
 457
 458         err = nf_register_hooks(ipv4_synproxy_ops,
 459                                 ARRAY_SIZE(ipv4_synproxy_ops));
 460         if (err < 0)
 461                 goto err1;
 462
 463         err = xt_register_target(&synproxy_tg4_reg);
 464         if (err < 0)
 465                 goto err2;
 466
 467         return 0;
 468
 469 err2:
 470         nf_unregister_hooks(ipv4_synproxy_ops, ARRAY_SIZE(ipv4_synproxy_ops));
 471 err1:
 472         return err;
 473 }
 474
 475 static void __exit synproxy_tg4_exit(void)
 476 {
 477         xt_unregister_target(&synproxy_tg4_reg);
 478         nf_unregister_hooks(ipv4_synproxy_ops, ARRAY_SIZE(ipv4_synproxy_ops));
 479 }
 480
 481 module_init(synproxy_tg4_init);
 482 module_exit(synproxy_tg4_exit);
 483
 484 MODULE_LICENSE("GPL");
 485 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");