net/ipv4/netfilter/ipt_SYNPROXY.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2013 Patrick McHardy <kaber@trash.net>
   4  */
   5
   6 #include <linux/module.h>
   7 #include <linux/skbuff.h>
   8 #include <net/tcp.h>
   9
  10 #include <linux/netfilter_ipv4/ip_tables.h>
  11 #include <linux/netfilter/x_tables.h>
  12 #include <linux/netfilter/xt_SYNPROXY.h>
  13 #include <net/netfilter/nf_conntrack.h>
  14 #include <net/netfilter/nf_conntrack_seqadj.h>
  15 #include <net/netfilter/nf_conntrack_synproxy.h>
  16 #include <net/netfilter/nf_conntrack_ecache.h>
  17
  18 static struct iphdr *
  19 synproxy_build_ip(struct net *net, struct sk_buff *skb, __be32 saddr,
  20                   __be32 daddr)
  21 {
  22         struct iphdr *iph;
  23
  24         skb_reset_network_header(skb);
  25         iph = skb_put(skb, sizeof(*iph));
  26         iph->version    = 4;
  27         iph->ihl        = sizeof(*iph) / 4;
  28         iph->tos        = 0;
  29         iph->id         = 0;
  30         iph->frag_off   = htons(IP_DF);
  31         iph->ttl        = net->ipv4.sysctl_ip_default_ttl;
  32         iph->protocol   = IPPROTO_TCP;
  33         iph->check      = 0;
  34         iph->saddr      = saddr;
  35         iph->daddr      = daddr;
  36
  37         return iph;
  38 }
  39
  40 static void
  41 synproxy_send_tcp(struct net *net,
  42                   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(net, nskb, RTN_UNSPEC))
  55                 goto free_nskb;
  56
  57         if (nfct) {
  58                 nf_ct_set(nskb, (struct nf_conn *)nfct, ctinfo);
  59                 nf_conntrack_get(nfct);
  60         }
  61
  62         ip_local_out(net, nskb->sk, nskb);
  63         return;
  64
  65 free_nskb:
  66         kfree_skb(nskb);
  67 }
  68
  69 static void
  70 synproxy_send_client_synack(struct net *net,
  71                             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(net, nskb, iph->daddr, iph->saddr);
  90
  91         skb_reset_transport_header(nskb);
  92         nth = 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(net, skb, nskb, skb_nfct(skb),
 108                           IP_CT_ESTABLISHED_REPLY, niph, nth, tcp_hdr_size);
 109 }
 110
 111 static void
 112 synproxy_send_server_syn(struct net *net,
 113                          const struct sk_buff *skb, const struct tcphdr *th,
 114                          const struct synproxy_options *opts, u32 recv_seq)
 115 {
 116         struct synproxy_net *snet = synproxy_pernet(net);
 117         struct sk_buff *nskb;
 118         struct iphdr *iph, *niph;
 119         struct tcphdr *nth;
 120         unsigned int tcp_hdr_size;
 121
 122         iph = ip_hdr(skb);
 123
 124         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 125         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 126                          GFP_ATOMIC);
 127         if (nskb == NULL)
 128                 return;
 129         skb_reserve(nskb, MAX_TCP_HEADER);
 130
 131         niph = synproxy_build_ip(net, nskb, iph->saddr, iph->daddr);
 132
 133         skb_reset_transport_header(nskb);
 134         nth = skb_put(nskb, tcp_hdr_size);
 135         nth->source     = th->source;
 136         nth->dest       = th->dest;
 137         nth->seq        = htonl(recv_seq - 1);
 138         /* ack_seq is used to relay our ISN to the synproxy hook to initialize
 139          * sequence number translation once a connection tracking entry exists.
 140          */
 141         nth->ack_seq    = htonl(ntohl(th->ack_seq) - 1);
 142         tcp_flag_word(nth) = TCP_FLAG_SYN;
 143         if (opts->options & XT_SYNPROXY_OPT_ECN)
 144                 tcp_flag_word(nth) |= TCP_FLAG_ECE | TCP_FLAG_CWR;
 145         nth->doff       = tcp_hdr_size / 4;
 146         nth->window     = th->window;
 147         nth->check      = 0;
 148         nth->urg_ptr    = 0;
 149
 150         synproxy_build_options(nth, opts);
 151
 152         synproxy_send_tcp(net, skb, nskb, &snet->tmpl->ct_general, IP_CT_NEW,
 153                           niph, nth, tcp_hdr_size);
 154 }
 155
 156 static void
 157 synproxy_send_server_ack(struct net *net,
 158                          const struct ip_ct_tcp *state,
 159                          const struct sk_buff *skb, const struct tcphdr *th,
 160                          const struct synproxy_options *opts)
 161 {
 162         struct sk_buff *nskb;
 163         struct iphdr *iph, *niph;
 164         struct tcphdr *nth;
 165         unsigned int tcp_hdr_size;
 166
 167         iph = ip_hdr(skb);
 168
 169         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 170         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 171                          GFP_ATOMIC);
 172         if (nskb == NULL)
 173                 return;
 174         skb_reserve(nskb, MAX_TCP_HEADER);
 175
 176         niph = synproxy_build_ip(net, nskb, iph->daddr, iph->saddr);
 177
 178         skb_reset_transport_header(nskb);
 179         nth = skb_put(nskb, tcp_hdr_size);
 180         nth->source     = th->dest;
 181         nth->dest       = th->source;
 182         nth->seq        = htonl(ntohl(th->ack_seq));
 183         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
 184         tcp_flag_word(nth) = TCP_FLAG_ACK;
 185         nth->doff       = tcp_hdr_size / 4;
 186         nth->window     = htons(state->seen[IP_CT_DIR_ORIGINAL].td_maxwin);
 187         nth->check      = 0;
 188         nth->urg_ptr    = 0;
 189
 190         synproxy_build_options(nth, opts);
 191
 192         synproxy_send_tcp(net, skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
 193 }
 194
 195 static void
 196 synproxy_send_client_ack(struct net *net,
 197                          const struct sk_buff *skb, const struct tcphdr *th,
 198                          const struct synproxy_options *opts)
 199 {
 200         struct sk_buff *nskb;
 201         struct iphdr *iph, *niph;
 202         struct tcphdr *nth;
 203         unsigned int tcp_hdr_size;
 204
 205         iph = ip_hdr(skb);
 206
 207         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
 208         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
 209                          GFP_ATOMIC);
 210         if (nskb == NULL)
 211                 return;
 212         skb_reserve(nskb, MAX_TCP_HEADER);
 213
 214         niph = synproxy_build_ip(net, nskb, iph->saddr, iph->daddr);
 215
 216         skb_reset_transport_header(nskb);
 217         nth = skb_put(nskb, tcp_hdr_size);
 218         nth->source     = th->source;
 219         nth->dest       = th->dest;
 220         nth->seq        = htonl(ntohl(th->seq) + 1);
 221         nth->ack_seq    = th->ack_seq;
 222         tcp_flag_word(nth) = TCP_FLAG_ACK;
 223         nth->doff       = tcp_hdr_size / 4;
 224         nth->window     = htons(ntohs(th->window) >> opts->wscale);
 225         nth->check      = 0;
 226         nth->urg_ptr    = 0;
 227
 228         synproxy_build_options(nth, opts);
 229
 230         synproxy_send_tcp(net, skb, nskb, skb_nfct(skb),
 231                           IP_CT_ESTABLISHED_REPLY, niph, nth, tcp_hdr_size);
 232 }
 233
 234 static bool
 235 synproxy_recv_client_ack(struct net *net,
 236                          const struct sk_buff *skb, const struct tcphdr *th,
 237                          struct synproxy_options *opts, u32 recv_seq)
 238 {
 239         struct synproxy_net *snet = synproxy_pernet(net);
 240         int mss;
 241
 242         mss = __cookie_v4_check(ip_hdr(skb), th, ntohl(th->ack_seq) - 1);
 243         if (mss == 0) {
 244                 this_cpu_inc(snet->stats->cookie_invalid);
 245                 return false;
 246         }
 247
 248         this_cpu_inc(snet->stats->cookie_valid);
 249         opts->mss = mss;
 250         opts->options |= XT_SYNPROXY_OPT_MSS;
 251
 252         if (opts->options & XT_SYNPROXY_OPT_TIMESTAMP)
 253                 synproxy_check_timestamp_cookie(opts);
 254
 255         synproxy_send_server_syn(net, skb, th, opts, recv_seq);
 256         return true;
 257 }
 258
 259 static unsigned int
 260 synproxy_tg4(struct sk_buff *skb, const struct xt_action_param *par)
 261 {
 262         const struct xt_synproxy_info *info = par->targinfo;
 263         struct net *net = xt_net(par);
 264         struct synproxy_net *snet = synproxy_pernet(net);
 265         struct synproxy_options opts = {};
 266         struct tcphdr *th, _th;
 267
 268         if (nf_ip_checksum(skb, xt_hooknum(par), par->thoff, IPPROTO_TCP))
 269                 return NF_DROP;
 270
 271         th = skb_header_pointer(skb, par->thoff, sizeof(_th), &_th);
 272         if (th == NULL)
 273                 return NF_DROP;
 274
 275         if (!synproxy_parse_options(skb, par->thoff, th, &opts))
 276                 return NF_DROP;
 277
 278         if (th->syn && !(th->ack || th->fin || th->rst)) {
 279                 /* Initial SYN from client */
 280                 this_cpu_inc(snet->stats->syn_received);
 281
 282                 if (th->ece && th->cwr)
 283                         opts.options |= XT_SYNPROXY_OPT_ECN;
 284
 285                 opts.options &= info->options;
 286                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 287                         synproxy_init_timestamp_cookie(info, &opts);
 288                 else
 289                         opts.options &= ~(XT_SYNPROXY_OPT_WSCALE |
 290                                           XT_SYNPROXY_OPT_SACK_PERM |
 291                                           XT_SYNPROXY_OPT_ECN);
 292
 293                 synproxy_send_client_synack(net, skb, th, &opts);
 294                 consume_skb(skb);
 295                 return NF_STOLEN;
 296         } else if (th->ack && !(th->fin || th->rst || th->syn)) {
 297                 /* ACK from client */
 298                 if (synproxy_recv_client_ack(net, skb, th, &opts, ntohl(th->seq))) {
 299                         consume_skb(skb);
 300                         return NF_STOLEN;
 301                 } else {
 302                         return NF_DROP;
 303                 }
 304         }
 305
 306         return XT_CONTINUE;
 307 }
 308
 309 static unsigned int ipv4_synproxy_hook(void *priv,
 310                                        struct sk_buff *skb,
 311                                        const struct nf_hook_state *nhs)
 312 {
 313         struct net *net = nhs->net;
 314         struct synproxy_net *snet = synproxy_pernet(net);
 315         enum ip_conntrack_info ctinfo;
 316         struct nf_conn *ct;
 317         struct nf_conn_synproxy *synproxy;
 318         struct synproxy_options opts = {};
 319         const struct ip_ct_tcp *state;
 320         struct tcphdr *th, _th;
 321         unsigned int thoff;
 322
 323         ct = nf_ct_get(skb, &ctinfo);
 324         if (ct == NULL)
 325                 return NF_ACCEPT;
 326
 327         synproxy = nfct_synproxy(ct);
 328         if (synproxy == NULL)
 329                 return NF_ACCEPT;
 330
 331         if (nf_is_loopback_packet(skb) ||
 332             ip_hdr(skb)->protocol != IPPROTO_TCP)
 333                 return NF_ACCEPT;
 334
 335         thoff = ip_hdrlen(skb);
 336         th = skb_header_pointer(skb, thoff, sizeof(_th), &_th);
 337         if (th == NULL)
 338                 return NF_DROP;
 339
 340         state = &ct->proto.tcp;
 341         switch (state->state) {
 342         case TCP_CONNTRACK_CLOSE:
 343                 if (th->rst && !test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 344                         nf_ct_seqadj_init(ct, ctinfo, synproxy->isn -
 345                                                       ntohl(th->seq) + 1);
 346                         break;
 347                 }
 348
 349                 if (!th->syn || th->ack ||
 350                     CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 351                         break;
 352
 353                 /* Reopened connection - reset the sequence number and timestamp
 354                  * adjustments, they will get initialized once the connection is
 355                  * reestablished.
 356                  */
 357                 nf_ct_seqadj_init(ct, ctinfo, 0);
 358                 synproxy->tsoff = 0;
 359                 this_cpu_inc(snet->stats->conn_reopened);
 360
 361                 /* fall through */
 362         case TCP_CONNTRACK_SYN_SENT:
 363                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 364                         return NF_DROP;
 365
 366                 if (!th->syn && th->ack &&
 367                     CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
 368                         /* Keep-Alives are sent with SEG.SEQ = SND.NXT-1,
 369                          * therefore we need to add 1 to make the SYN sequence
 370                          * number match the one of first SYN.
 371                          */
 372                         if (synproxy_recv_client_ack(net, skb, th, &opts,
 373                                                      ntohl(th->seq) + 1)) {
 374                                 this_cpu_inc(snet->stats->cookie_retrans);
 375                                 consume_skb(skb);
 376                                 return NF_STOLEN;
 377                         } else {
 378                                 return NF_DROP;
 379                         }
 380                 }
 381
 382                 synproxy->isn = ntohl(th->ack_seq);
 383                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
 384                         synproxy->its = opts.tsecr;
 385
 386                 nf_conntrack_event_cache(IPCT_SYNPROXY, ct);
 387                 break;
 388         case TCP_CONNTRACK_SYN_RECV:
 389                 if (!th->syn || !th->ack)
 390                         break;
 391
 392                 if (!synproxy_parse_options(skb, thoff, th, &opts))
 393                         return NF_DROP;
 394
 395                 if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP) {
 396                         synproxy->tsoff = opts.tsval - synproxy->its;
 397                         nf_conntrack_event_cache(IPCT_SYNPROXY, ct);
 398                 }
 399
 400                 opts.options &= ~(XT_SYNPROXY_OPT_MSS |
 401                                   XT_SYNPROXY_OPT_WSCALE |
 402                                   XT_SYNPROXY_OPT_SACK_PERM);
 403
 404                 swap(opts.tsval, opts.tsecr);
 405                 synproxy_send_server_ack(net, state, skb, th, &opts);
 406
 407                 nf_ct_seqadj_init(ct, ctinfo, synproxy->isn - ntohl(th->seq));
 408                 nf_conntrack_event_cache(IPCT_SEQADJ, ct);
 409
 410                 swap(opts.tsval, opts.tsecr);
 411                 synproxy_send_client_ack(net, skb, th, &opts);
 412
 413                 consume_skb(skb);
 414                 return NF_STOLEN;
 415         default:
 416                 break;
 417         }
 418
 419         synproxy_tstamp_adjust(skb, thoff, th, ct, ctinfo, synproxy);
 420         return NF_ACCEPT;
 421 }
 422
 423 static const struct nf_hook_ops ipv4_synproxy_ops[] = {
 424         {
 425                 .hook           = ipv4_synproxy_hook,
 426                 .pf             = NFPROTO_IPV4,
 427                 .hooknum        = NF_INET_LOCAL_IN,
 428                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 429         },
 430         {
 431                 .hook           = ipv4_synproxy_hook,
 432                 .pf             = NFPROTO_IPV4,
 433                 .hooknum        = NF_INET_POST_ROUTING,
 434                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
 435         },
 436 };
 437
 438 static int synproxy_tg4_check(const struct xt_tgchk_param *par)
 439 {
 440         struct synproxy_net *snet = synproxy_pernet(par->net);
 441         const struct ipt_entry *e = par->entryinfo;
 442         int err;
 443
 444         if (e->ip.proto != IPPROTO_TCP ||
 445             e->ip.invflags & XT_INV_PROTO)
 446                 return -EINVAL;
 447
 448         err = nf_ct_netns_get(par->net, par->family);
 449         if (err)
 450                 return err;
 451
 452         if (snet->hook_ref4 == 0) {
 453                 err = nf_register_net_hooks(par->net, ipv4_synproxy_ops,
 454                                             ARRAY_SIZE(ipv4_synproxy_ops));
 455                 if (err) {
 456                         nf_ct_netns_put(par->net, par->family);
 457                         return err;
 458                 }
 459         }
 460
 461         snet->hook_ref4++;
 462         return err;
 463 }
 464
 465 static void synproxy_tg4_destroy(const struct xt_tgdtor_param *par)
 466 {
 467         struct synproxy_net *snet = synproxy_pernet(par->net);
 468
 469         snet->hook_ref4--;
 470         if (snet->hook_ref4 == 0)
 471                 nf_unregister_net_hooks(par->net, ipv4_synproxy_ops,
 472                                         ARRAY_SIZE(ipv4_synproxy_ops));
 473         nf_ct_netns_put(par->net, par->family);
 474 }
 475
 476 static struct xt_target synproxy_tg4_reg __read_mostly = {
 477         .name           = "SYNPROXY",
 478         .family         = NFPROTO_IPV4,
 479         .hooks          = (1 << NF_INET_LOCAL_IN) | (1 << NF_INET_FORWARD),
 480         .target         = synproxy_tg4,
 481         .targetsize     = sizeof(struct xt_synproxy_info),
 482         .checkentry     = synproxy_tg4_check,
 483         .destroy        = synproxy_tg4_destroy,
 484         .me             = THIS_MODULE,
 485 };
 486
 487 static int __init synproxy_tg4_init(void)
 488 {
 489         return xt_register_target(&synproxy_tg4_reg);
 490 }
 491
 492 static void __exit synproxy_tg4_exit(void)
 493 {
 494         xt_unregister_target(&synproxy_tg4_reg);
 495 }
 496
 497 module_init(synproxy_tg4_init);
 498 module_exit(synproxy_tg4_exit);
 499
 500 MODULE_LICENSE("GPL");
 501 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");