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