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         struct net *net = sock_net(skb->sk);
  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(const struct synproxy_net *snet,
  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         struct net *net = nf_ct_net(snet->tmpl);
  50
  51         nth->check = ~tcp_v4_check(tcp_hdr_size, niph->saddr, niph->daddr, 0);
  52         nskb->ip_summed   = CHECKSUM_PARTIAL;
  53         nskb->csum_start  = (unsigned char *)nth - nskb->head;
  54         nskb->csum_offset = offsetof(struct tcphdr, check);
  55
  56         skb_dst_set_noref(nskb, skb_dst(skb));
  57         nskb->protocol = htons(ETH_P_IP);
  58         if (ip_route_me_harder(net, nskb, RTN_UNSPEC))
  59                 goto free_nskb;
  60
  61         if (nfct) {
  62                 nskb->nfct = nfct;
  63                 nskb->nfctinfo = ctinfo;
  64                 nf_conntrack_get(nfct);
  65         }
  66
  67         ip_local_out(net, nskb->sk, nskb);
  68         return;
  69
  70 free_nskb:
  71         kfree_skb(nskb);
  72 }
  73
  74 static void
  75 synproxy_send_client_synack(const struct synproxy_net *snet,
  76                             const struct sk_buff *skb, const struct tcphdr *th,
  77                             const struct synproxy_options *opts)
  78 {
  79         struct sk_buff *nskb;
  80         struct iphdr *iph, *niph;
  81         struct tcphdr *nth;
  82         unsigned int tcp_hdr_size;
  83         u16 mss = opts->mss;
  84
  85         iph = ip_hdr(skb);
  86
  87         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
  88         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
  89                          GFP_ATOMIC);
  90         if (nskb == NULL)
  91                 return;
  92         skb_reserve(nskb, MAX_TCP_HEADER);
  93
  94         niph = synproxy_build_ip(nskb, iph->daddr, iph->saddr);
  95
  96         skb_reset_transport_header(nskb);
  97         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
  98         nth->source     = th->dest;
  99         nth->dest       = th->source;
 100         nth->seq        = htonl(__cookie_v4_init_sequence(iph, th, &mss));
 101         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
 102         tcp_flag_word(nth) = TCP_FLAG_SYN | TCP_FLAG_ACK;
 103         if (opts->options & XT_SYNPROXY_OPT_ECN)
 104                 tcp_flag_word(nth) |= TCP_FLAG_ECE;
 105         nth->doff       = tcp_hdr_size / 4;
 106         nth->window     = 0;
 107         nth->check      = 0;
 108         nth->urg_ptr    = 0;
 109
 110         synproxy_build_options(nth, opts);
 111
 112         synproxy_send_tcp(snet, skb, nskb, skb->nfct, IP_CT_ESTABLISHED_REPLY,
 113                           niph, nth, tcp_hdr_size);
 114 }
 115
 116 static void
 117 synproxy_send_server_syn(const struct synproxy_net *snet,
 118                          const struct sk_buff *skb, const struct tcphdr *th,
 119                          const struct synproxy_options *opts, u32 recv_seq)
 120 {
 121         struct sk_buff *nskb;
 122         struct iphdr *iph, *niph;
 123         struct tcphdr *nth;
 124         unsigned int tcp_hdr_size;
 125
 126         iph = ip_hdr(skb);
 127
 128         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 129         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 130                          GFP_ATOMIC);
 131         if (nskb == NULL)
 132                 return;
 133         skb_reserve(nskb, MAX_TCP_HEADER);
 134
 135         niph = synproxy_build_ip(nskb, iph->saddr, iph->daddr);
 136
 137         skb_reset_transport_header(nskb);
 138         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 139         nth->source     = th->source;
 140         nth->dest       = th->dest;
 141         nth->seq        = htonl(recv_seq - 1);
 142         /* ack_seq is used to relay our ISN to the synproxy hook to initialize
 143          * sequence number translation once a connection tracking entry exists.
 144          */
 145         nth->ack_seq    = htonl(ntohl(th->ack_seq) - 1);
 146         tcp_flag_word(nth) = TCP_FLAG_SYN;
 147         if (opts->options & XT_SYNPROXY_OPT_ECN)
 148                 tcp_flag_word(nth) |= TCP_FLAG_ECE | TCP_FLAG_CWR;
 149         nth->doff       = tcp_hdr_size / 4;
 150         nth->window     = th->window;
 151         nth->check      = 0;
 152         nth->urg_ptr    = 0;
 153
 154         synproxy_build_options(nth, opts);
 155
 156         synproxy_send_tcp(snet, skb, nskb, &snet->tmpl->ct_general, IP_CT_NEW,
 157                           niph, nth, tcp_hdr_size);
 158 }
 159
 160 static void
 161 synproxy_send_server_ack(const struct synproxy_net *snet,
 162                          const struct ip_ct_tcp *state,
 163                          const struct sk_buff *skb, const struct tcphdr *th,
 164                          const struct synproxy_options *opts)
 165 {
 166         struct sk_buff *nskb;
 167         struct iphdr *iph, *niph;
 168         struct tcphdr *nth;
 169         unsigned int tcp_hdr_size;
 170
 171         iph = ip_hdr(skb);
 172
 173         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 174         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 175                          GFP_ATOMIC);
 176         if (nskb == NULL)
 177                 return;
 178         skb_reserve(nskb, MAX_TCP_HEADER);
 179
 180         niph = synproxy_build_ip(nskb, iph->daddr, iph->saddr);
 181
 182         skb_reset_transport_header(nskb);
 183         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 184         nth->source     = th->dest;
 185         nth->dest       = th->source;
 186         nth->seq        = htonl(ntohl(th->ack_seq));
 187         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
 188         tcp_flag_word(nth) = TCP_FLAG_ACK;
 189         nth->doff       = tcp_hdr_size / 4;
 190         nth->window     = htons(state->seen[IP_CT_DIR_ORIGINAL].td_maxwin);
 191         nth->check      = 0;
 192         nth->urg_ptr    = 0;
 193
 194         synproxy_build_options(nth, opts);
 195
 196         synproxy_send_tcp(snet, skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
 197 }
 198
 199 static void
 200 synproxy_send_client_ack(const struct synproxy_net *snet,
 201                          const struct sk_buff *skb, const struct tcphdr *th,
 202                          const struct synproxy_options *opts)
 203 {
 204         struct sk_buff *nskb;
 205         struct iphdr *iph, *niph;
 206         struct tcphdr *nth;
 207         unsigned int tcp_hdr_size;
 208
 209         iph = ip_hdr(skb);
 210
 211         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 212         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 213                          GFP_ATOMIC);
 214         if (nskb == NULL)
 215                 return;
 216         skb_reserve(nskb, MAX_TCP_HEADER);
 217
 218         niph = synproxy_build_ip(nskb, iph->saddr, iph->daddr);
 219
 220         skb_reset_transport_header(nskb);
 221         nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
 222         nth->source     = th->source;
 223         nth->dest       = th->dest;
 224         nth->seq        = htonl(ntohl(th->seq) + 1);
 225         nth->ack_seq    = th->ack_seq;
 226         tcp_flag_word(nth) = TCP_FLAG_ACK;
 227         nth->doff       = tcp_hdr_size / 4;
 228         nth->window     = htons(ntohs(th->window) >> opts->wscale);
 229         nth->check      = 0;
 230         nth->urg_ptr    = 0;
 231
 232         synproxy_build_options(nth, opts);
 233
 234         synproxy_send_tcp(snet, skb, nskb, skb->nfct, IP_CT_ESTABLISHED_REPLY,
 235                           niph, nth, tcp_hdr_size);
 236 }
 237
 238 static bool
 239 synproxy_recv_client_ack(const struct synproxy_net *snet,
 240                          const struct sk_buff *skb, const struct tcphdr *th,
 241                          struct synproxy_options *opts, u32 recv_seq)
 242 {
 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(snet, 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 synproxy_net *snet = synproxy_pernet(par->net);
 267         struct synproxy_options opts = {};
 268         struct tcphdr *th, _th;
 269
 270         if (nf_ip_checksum(skb, par->hooknum, 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(snet, 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(snet, 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 synproxy_net *snet = synproxy_pernet(nhs->net);
 312         enum ip_conntrack_info ctinfo;
 313         struct nf_conn *ct;
 314         struct nf_conn_synproxy *synproxy;
 315         struct synproxy_options opts = {};
 316         const struct ip_ct_tcp *state;
 317         struct tcphdr *th, _th;
 318         unsigned int thoff;
 319
 320         ct = nf_ct_get(skb, &ctinfo);
 321         if (ct == NULL)
 322                 return NF_ACCEPT;
 323
 324         synproxy = nfct_synproxy(ct);
 325         if (synproxy == NULL)
 326                 return NF_ACCEPT;
 327
 328         if (nf_is_loopback_packet(skb))
 329                 return NF_ACCEPT;
 330
 331         thoff = ip_hdrlen(skb);
 332         th = skb_header_pointer(skb, thoff, sizeof(_th), &_th);
 333         if (th == NULL)
 334                 return NF_DROP;
 335
 336         state = &ct->proto.tcp;
 337         switch (state->state) {
 338         case TCP_CONNTRACK_CLOSE:
 339                 if (th->rst && !test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 340                         nf_ct_seqadj_init(ct, ctinfo, synproxy->isn -
 341                                                       ntohl(th->seq) + 1);
 342                         break;
 343                 }
 344
 345                 if (!th->syn || th->ack ||
 346                     CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 347                         break;
 348
 349                 /* Reopened connection - reset the sequence number and timestamp
 350                  * adjustments, they will get initialized once the connection is
 351                  * reestablished.
 352                  */
 353                 nf_ct_seqadj_init(ct, ctinfo, 0);
 354                 synproxy->tsoff = 0;
 355                 this_cpu_inc(snet->stats->conn_reopened);
 356
 357                 /* fall through */
 358         case TCP_CONNTRACK_SYN_SENT:
 359                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 360                         return NF_DROP;
 361
 362                 if (!th->syn && th->ack &&
 363                     CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
 364                         /* Keep-Alives are sent with SEG.SEQ = SND.NXT-1,
 365                          * therefore we need to add 1 to make the SYN sequence
 366                          * number match the one of first SYN.
 367                          */
 368                         if (synproxy_recv_client_ack(snet, skb, th, &opts,
 369                                                      ntohl(th->seq) + 1))
 370                                 this_cpu_inc(snet->stats->cookie_retrans);
 371
 372                         return NF_DROP;
 373                 }
 374
 375                 synproxy->isn = ntohl(th->ack_seq);
 376                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 377                         synproxy->its = opts.tsecr;
 378                 break;
 379         case TCP_CONNTRACK_SYN_RECV:
 380                 if (!th->syn || !th->ack)
 381                         break;
 382
 383                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 384                         return NF_DROP;
 385
 386                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 387                         synproxy->tsoff = opts.tsval - synproxy->its;
 388
 389                 opts.options &= ~(XT_SYNPROXY_OPT_MSS |
 390                                   XT_SYNPROXY_OPT_WSCALE |
 391                                   XT_SYNPROXY_OPT_SACK_PERM);
 392
 393                 swap(opts.tsval, opts.tsecr);
 394                 synproxy_send_server_ack(snet, state, skb, th, &opts);
 395
 396                 nf_ct_seqadj_init(ct, ctinfo, synproxy->isn - ntohl(th->seq));
 397
 398                 swap(opts.tsval, opts.tsecr);
 399                 synproxy_send_client_ack(snet, skb, th, &opts);
 400
 401                 consume_skb(skb);
 402                 return NF_STOLEN;
 403         default:
 404                 break;
 405         }
 406
 407         synproxy_tstamp_adjust(skb, thoff, th, ct, ctinfo, synproxy);
 408         return NF_ACCEPT;
 409 }
 410
 411 static int synproxy_tg4_check(const struct xt_tgchk_param *par)
 412 {
 413         const struct ipt_entry *e = par->entryinfo;
 414
 415         if (e->ip.proto != IPPROTO_TCP ||
 416             e->ip.invflags & XT_INV_PROTO)
 417                 return -EINVAL;
 418
 419         return nf_ct_l3proto_try_module_get(par->family);
 420 }
 421
 422 static void synproxy_tg4_destroy(const struct xt_tgdtor_param *par)
 423 {
 424         nf_ct_l3proto_module_put(par->family);
 425 }
 426
 427 static struct xt_target synproxy_tg4_reg __read_mostly = {
 428         .name           = "SYNPROXY",
 429         .family         = NFPROTO_IPV4,
 430         .hooks          = (1 << NF_INET_LOCAL_IN) | (1 << NF_INET_FORWARD),
 431         .target         = synproxy_tg4,
 432         .targetsize     = sizeof(struct xt_synproxy_info),
 433         .checkentry     = synproxy_tg4_check,
 434         .destroy        = synproxy_tg4_destroy,
 435         .me             = THIS_MODULE,
 436 };
 437
 438 static struct nf_hook_ops ipv4_synproxy_ops[] __read_mostly = {
 439         {
 440                 .hook           = ipv4_synproxy_hook,
 441                 .pf             = NFPROTO_IPV4,
 442                 .hooknum        = NF_INET_LOCAL_IN,
 443                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 444         },
 445         {
 446                 .hook           = ipv4_synproxy_hook,
 447                 .pf             = NFPROTO_IPV4,
 448                 .hooknum        = NF_INET_POST_ROUTING,
 449                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 450         },
 451 };
 452
 453 static int __init synproxy_tg4_init(void)
 454 {
 455         int err;
 456
 457         err = nf_register_hooks(ipv4_synproxy_ops,
 458                                 ARRAY_SIZE(ipv4_synproxy_ops));
 459         if (err < 0)
 460                 goto err1;
 461
 462         err = xt_register_target(&synproxy_tg4_reg);
 463         if (err < 0)
 464                 goto err2;
 465
 466         return 0;
 467
 468 err2:
 469         nf_unregister_hooks(ipv4_synproxy_ops, ARRAY_SIZE(ipv4_synproxy_ops));
 470 err1:
 471         return err;
 472 }
 473
 474 static void __exit synproxy_tg4_exit(void)
 475 {
 476         xt_unregister_target(&synproxy_tg4_reg);
 477         nf_unregister_hooks(ipv4_synproxy_ops, ARRAY_SIZE(ipv4_synproxy_ops));
 478 }
 479
 480 module_init(synproxy_tg4_init);
 481 module_exit(synproxy_tg4_exit);
 482
 483 MODULE_LICENSE("GPL");
 484 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");