kernel/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 synproxy_net *snet,
  43                   const struct sk_buff *skb, struct sk_buff *nskb,
  44                   struct nf_conntrack *nfct, enum ip_conntrack_info ctinfo,
  45                   struct iphdr *niph, struct tcphdr *nth,
  46                   unsigned int tcp_hdr_size)
  47 {
  48         struct net *net = nf_ct_net(snet->tmpl);
  49
  50         nth->check = ~tcp_v4_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         skb_dst_set_noref(nskb, skb_dst(skb));
  56         nskb->protocol = htons(ETH_P_IP);
  57         if (ip_route_me_harder(net, nskb, RTN_UNSPEC))
  58                 goto free_nskb;
  59
  60         if (nfct) {
  61                 nskb->nfct = nfct;
  62                 nskb->nfctinfo = ctinfo;
  63                 nf_conntrack_get(nfct);
  64         }
  65
  66         ip_local_out(net, nskb->sk, nskb);
  67         return;
  68
  69 free_nskb:
  70         kfree_skb(nskb);
  71 }
  72
  73 static void
  74 synproxy_send_client_synack(const struct synproxy_net *snet,
  75                             const struct sk_buff *skb, const struct tcphdr *th,
  76                             const struct synproxy_options *opts)
  77 {
  78         struct sk_buff *nskb;
  79         struct iphdr *iph, *niph;
  80         struct tcphdr *nth;
  81         unsigned int tcp_hdr_size;
  82         u16 mss = opts->mss;
  83
  84         iph = ip_hdr(skb);
  85
  86         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
  87         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
  88                          GFP_ATOMIC);
  89         if (nskb == NULL)
  90                 return;
  91         skb_reserve(nskb, MAX_TCP_HEADER);
  92
  93         niph = synproxy_build_ip(nskb, iph->daddr, iph->saddr);
  94
  95         skb_reset_transport_header(nskb);
  96         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
  97         nth->source     = th->dest;
  98         nth->dest       = th->source;
  99         nth->seq        = htonl(__cookie_v4_init_sequence(iph, th, &mss));
 100         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
 101         tcp_flag_word(nth) = TCP_FLAG_SYN | TCP_FLAG_ACK;
 102         if (opts->options & XT_SYNPROXY_OPT_ECN)
 103                 tcp_flag_word(nth) |= TCP_FLAG_ECE;
 104         nth->doff       = tcp_hdr_size / 4;
 105         nth->window     = 0;
 106         nth->check      = 0;
 107         nth->urg_ptr    = 0;
 108
 109         synproxy_build_options(nth, opts);
 110
 111         synproxy_send_tcp(snet, skb, nskb, skb->nfct, IP_CT_ESTABLISHED_REPLY,
 112                           niph, nth, tcp_hdr_size);
 113 }
 114
 115 static void
 116 synproxy_send_server_syn(const struct synproxy_net *snet,
 117                          const struct sk_buff *skb, const struct tcphdr *th,
 118                          const struct synproxy_options *opts, u32 recv_seq)
 119 {
 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(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(snet, 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(const struct synproxy_net *snet,
 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(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(snet, skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
 196 }
 197
 198 static void
 199 synproxy_send_client_ack(const struct synproxy_net *snet,
 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(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(snet, skb, nskb, skb->nfct, IP_CT_ESTABLISHED_REPLY,
 234                           niph, nth, tcp_hdr_size);
 235 }
 236
 237 static bool
 238 synproxy_recv_client_ack(const struct synproxy_net *snet,
 239                          const struct sk_buff *skb, const struct tcphdr *th,
 240                          struct synproxy_options *opts, u32 recv_seq)
 241 {
 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(snet, 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 synproxy_net *snet = synproxy_pernet(par->net);
 266         struct synproxy_options opts = {};
 267         struct tcphdr *th, _th;
 268
 269         if (nf_ip_checksum(skb, par->hooknum, par->thoff, IPPROTO_TCP))
 270                 return NF_DROP;
 271
 272         th = skb_header_pointer(skb, par->thoff, sizeof(_th), &_th);
 273         if (th == NULL)
 274                 return NF_DROP;
 275
 276         if (!synproxy_parse_options(skb, par->thoff, th, &opts))
 277                 return NF_DROP;
 278
 279         if (th->syn && !(th->ack || th->fin || th->rst)) {
 280                 /* Initial SYN from client */
 281                 this_cpu_inc(snet->stats->syn_received);
 282
 283                 if (th->ece && th->cwr)
 284                         opts.options |= XT_SYNPROXY_OPT_ECN;
 285
 286                 opts.options &= info->options;
 287                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 288                         synproxy_init_timestamp_cookie(info, &opts);
 289                 else
 290                         opts.options &= ~(XT_SYNPROXY_OPT_WSCALE |
 291                                           XT_SYNPROXY_OPT_SACK_PERM |
 292                                           XT_SYNPROXY_OPT_ECN);
 293
 294                 synproxy_send_client_synack(snet, skb, th, &opts);
 295                 return NF_DROP;
 296
 297         } else if (th->ack && !(th->fin || th->rst || th->syn)) {
 298                 /* ACK from client */
 299                 synproxy_recv_client_ack(snet, skb, th, &opts, ntohl(th->seq));
 300                 return NF_DROP;
 301         }
 302
 303         return XT_CONTINUE;
 304 }
 305
 306 static unsigned int ipv4_synproxy_hook(void *priv,
 307                                        struct sk_buff *skb,
 308                                        const struct nf_hook_state *nhs)
 309 {
 310         struct synproxy_net *snet = synproxy_pernet(nhs->net);
 311         enum ip_conntrack_info ctinfo;
 312         struct nf_conn *ct;
 313         struct nf_conn_synproxy *synproxy;
 314         struct synproxy_options opts = {};
 315         const struct ip_ct_tcp *state;
 316         struct tcphdr *th, _th;
 317         unsigned int thoff;
 318
 319         ct = nf_ct_get(skb, &ctinfo);
 320         if (ct == NULL)
 321                 return NF_ACCEPT;
 322
 323         synproxy = nfct_synproxy(ct);
 324         if (synproxy == NULL)
 325                 return NF_ACCEPT;
 326
 327         if (nf_is_loopback_packet(skb))
 328                 return NF_ACCEPT;
 329
 330         thoff = ip_hdrlen(skb);
 331         th = skb_header_pointer(skb, thoff, sizeof(_th), &_th);
 332         if (th == NULL)
 333                 return NF_DROP;
 334
 335         state = &ct->proto.tcp;
 336         switch (state->state) {
 337         case TCP_CONNTRACK_CLOSE:
 338                 if (th->rst && !test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 339                         nf_ct_seqadj_init(ct, ctinfo, synproxy->isn -
 340                                                       ntohl(th->seq) + 1);
 341                         break;
 342                 }
 343
 344                 if (!th->syn || th->ack ||
 345                     CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 346                         break;
 347
 348                 /* Reopened connection - reset the sequence number and timestamp
 349                  * adjustments, they will get initialized once the connection is
 350                  * reestablished.
 351                  */
 352                 nf_ct_seqadj_init(ct, ctinfo, 0);
 353                 synproxy->tsoff = 0;
 354                 this_cpu_inc(snet->stats->conn_reopened);
 355
 356                 /* fall through */
 357         case TCP_CONNTRACK_SYN_SENT:
 358                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 359                         return NF_DROP;
 360
 361                 if (!th->syn && th->ack &&
 362                     CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
 363                         /* Keep-Alives are sent with SEG.SEQ = SND.NXT-1,
 364                          * therefore we need to add 1 to make the SYN sequence
 365                          * number match the one of first SYN.
 366                          */
 367                         if (synproxy_recv_client_ack(snet, skb, th, &opts,
 368                                                      ntohl(th->seq) + 1))
 369                                 this_cpu_inc(snet->stats->cookie_retrans);
 370
 371                         return NF_DROP;
 372                 }
 373
 374                 synproxy->isn = ntohl(th->ack_seq);
 375                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 376                         synproxy->its = opts.tsecr;
 377                 break;
 378         case TCP_CONNTRACK_SYN_RECV:
 379                 if (!th->syn || !th->ack)
 380                         break;
 381
 382                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 383                         return NF_DROP;
 384
 385                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 386                         synproxy->tsoff = opts.tsval - synproxy->its;
 387
 388                 opts.options &= ~(XT_SYNPROXY_OPT_MSS |
 389                                   XT_SYNPROXY_OPT_WSCALE |
 390                                   XT_SYNPROXY_OPT_SACK_PERM);
 391
 392                 swap(opts.tsval, opts.tsecr);
 393                 synproxy_send_server_ack(snet, state, skb, th, &opts);
 394
 395                 nf_ct_seqadj_init(ct, ctinfo, synproxy->isn - ntohl(th->seq));
 396
 397                 swap(opts.tsval, opts.tsecr);
 398                 synproxy_send_client_ack(snet, skb, th, &opts);
 399
 400                 consume_skb(skb);
 401                 return NF_STOLEN;
 402         default:
 403                 break;
 404         }
 405
 406         synproxy_tstamp_adjust(skb, thoff, th, ct, ctinfo, synproxy);
 407         return NF_ACCEPT;
 408 }
 409
 410 static int synproxy_tg4_check(const struct xt_tgchk_param *par)
 411 {
 412         const struct ipt_entry *e = par->entryinfo;
 413
 414         if (e->ip.proto != IPPROTO_TCP ||
 415             e->ip.invflags & XT_INV_PROTO)
 416                 return -EINVAL;
 417
 418         return nf_ct_l3proto_try_module_get(par->family);
 419 }
 420
 421 static void synproxy_tg4_destroy(const struct xt_tgdtor_param *par)
 422 {
 423         nf_ct_l3proto_module_put(par->family);
 424 }
 425
 426 static struct xt_target synproxy_tg4_reg __read_mostly = {
 427         .name           = "SYNPROXY",
 428         .family         = NFPROTO_IPV4,
 429         .hooks          = (1 << NF_INET_LOCAL_IN) | (1 << NF_INET_FORWARD),
 430         .target         = synproxy_tg4,
 431         .targetsize     = sizeof(struct xt_synproxy_info),
 432         .checkentry     = synproxy_tg4_check,
 433         .destroy        = synproxy_tg4_destroy,
 434         .me             = THIS_MODULE,
 435 };
 436
 437 static struct nf_hook_ops ipv4_synproxy_ops[] __read_mostly = {
 438         {
 439                 .hook           = ipv4_synproxy_hook,
 440                 .pf             = NFPROTO_IPV4,
 441                 .hooknum        = NF_INET_LOCAL_IN,
 442                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 443         },
 444         {
 445                 .hook           = ipv4_synproxy_hook,
 446                 .pf             = NFPROTO_IPV4,
 447                 .hooknum        = NF_INET_POST_ROUTING,
 448                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 449         },
 450 };
 451
 452 static int __init synproxy_tg4_init(void)
 453 {
 454         int err;
 455
 456         err = nf_register_hooks(ipv4_synproxy_ops,
 457                                 ARRAY_SIZE(ipv4_synproxy_ops));
 458         if (err < 0)
 459                 goto err1;
 460
 461         err = xt_register_target(&synproxy_tg4_reg);
 462         if (err < 0)
 463                 goto err2;
 464
 465         return 0;
 466
 467 err2:
 468         nf_unregister_hooks(ipv4_synproxy_ops, ARRAY_SIZE(ipv4_synproxy_ops));
 469 err1:
 470         return err;
 471 }
 472
 473 static void __exit synproxy_tg4_exit(void)
 474 {
 475         xt_unregister_target(&synproxy_tg4_reg);
 476         nf_unregister_hooks(ipv4_synproxy_ops, ARRAY_SIZE(ipv4_synproxy_ops));
 477 }
 478
 479 module_init(synproxy_tg4_init);
 480 module_exit(synproxy_tg4_exit);
 481
 482 MODULE_LICENSE("GPL");
 483 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");