net/netfilter/nf_conntrack_proto_tcp.c

   1 /* (C) 1999-2001 Paul `Rusty' Russell
   2  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   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  * Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>:
   9  *      - Real stateful connection tracking
  10  *      - Modified state transitions table
  11  *      - Window scaling support added
  12  *      - SACK support added
  13  *
  14  * Willy Tarreau:
  15  *      - State table bugfixes
  16  *      - More robust state changes
  17  *      - Tuning timer parameters
  18  *
  19  * 27 Oct 2004: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
  20  *      - genelized Layer 3 protocol part.
  21  *
  22  * Derived from net/ipv4/netfilter/ip_conntrack_proto_tcp.c
  23  *
  24  * version 2.2
  25  */
  26
  27 #include <linux/types.h>
  28 #include <linux/timer.h>
  29 #include <linux/netfilter.h>
  30 #include <linux/module.h>
  31 #include <linux/in.h>
  32 #include <linux/tcp.h>
  33 #include <linux/spinlock.h>
  34 #include <linux/skbuff.h>
  35 #include <linux/ipv6.h>
  36 #include <net/ip6_checksum.h>
  37
  38 #include <net/tcp.h>
  39
  40 #include <linux/netfilter.h>
  41 #include <linux/netfilter_ipv4.h>
  42 #include <linux/netfilter_ipv6.h>
  43 #include <net/netfilter/nf_conntrack.h>
  44 #include <net/netfilter/nf_conntrack_l4proto.h>
  45 #include <net/netfilter/nf_conntrack_ecache.h>
  46
  47 #if 0
  48 #define DEBUGP printk
  49 #define DEBUGP_VARS
  50 #else
  51 #define DEBUGP(format, args...)
  52 #endif
  53
  54 /* Protects conntrack->proto.tcp */
  55 static DEFINE_RWLOCK(tcp_lock);
  56
  57 /* "Be conservative in what you do,
  58     be liberal in what you accept from others."
  59     If it's non-zero, we mark only out of window RST segments as INVALID. */
  60 static int nf_ct_tcp_be_liberal __read_mostly = 0;
  61
  62 /* If it is set to zero, we disable picking up already established
  63    connections. */
  64 static int nf_ct_tcp_loose __read_mostly = 1;
  65
  66 /* Max number of the retransmitted packets without receiving an (acceptable)
  67    ACK from the destination. If this number is reached, a shorter timer
  68    will be started. */
  69 static int nf_ct_tcp_max_retrans __read_mostly = 3;
  70
  71   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  72      closely.  They're more complex. --RR */
  73
  74 static const char *tcp_conntrack_names[] = {
  75         "NONE",
  76         "SYN_SENT",
  77         "SYN_RECV",
  78         "ESTABLISHED",
  79         "FIN_WAIT",
  80         "CLOSE_WAIT",
  81         "LAST_ACK",
  82         "TIME_WAIT",
  83         "CLOSE",
  84         "LISTEN"
  85 };
  86
  87 #define SECS * HZ
  88 #define MINS * 60 SECS
  89 #define HOURS * 60 MINS
  90 #define DAYS * 24 HOURS
  91
  92 static unsigned int nf_ct_tcp_timeout_syn_sent __read_mostly =      2 MINS;
  93 static unsigned int nf_ct_tcp_timeout_syn_recv __read_mostly =     60 SECS;
  94 static unsigned int nf_ct_tcp_timeout_established __read_mostly =   5 DAYS;
  95 static unsigned int nf_ct_tcp_timeout_fin_wait __read_mostly =      2 MINS;
  96 static unsigned int nf_ct_tcp_timeout_close_wait __read_mostly =   60 SECS;
  97 static unsigned int nf_ct_tcp_timeout_last_ack __read_mostly =     30 SECS;
  98 static unsigned int nf_ct_tcp_timeout_time_wait __read_mostly =     2 MINS;
  99 static unsigned int nf_ct_tcp_timeout_close __read_mostly =        10 SECS;
 100
 101 /* RFC1122 says the R2 limit should be at least 100 seconds.
 102    Linux uses 15 packets as limit, which corresponds
 103    to ~13-30min depending on RTO. */
 104 static unsigned int nf_ct_tcp_timeout_max_retrans __read_mostly =   5 MINS;
 105
 106 static unsigned int * tcp_timeouts[] = {
 107     NULL,                              /* TCP_CONNTRACK_NONE */
 108     &nf_ct_tcp_timeout_syn_sent,       /* TCP_CONNTRACK_SYN_SENT, */
 109     &nf_ct_tcp_timeout_syn_recv,       /* TCP_CONNTRACK_SYN_RECV, */
 110     &nf_ct_tcp_timeout_established,    /* TCP_CONNTRACK_ESTABLISHED, */
 111     &nf_ct_tcp_timeout_fin_wait,       /* TCP_CONNTRACK_FIN_WAIT, */
 112     &nf_ct_tcp_timeout_close_wait,     /* TCP_CONNTRACK_CLOSE_WAIT, */
 113     &nf_ct_tcp_timeout_last_ack,       /* TCP_CONNTRACK_LAST_ACK, */
 114     &nf_ct_tcp_timeout_time_wait,      /* TCP_CONNTRACK_TIME_WAIT, */
 115     &nf_ct_tcp_timeout_close,          /* TCP_CONNTRACK_CLOSE, */
 116     NULL,                              /* TCP_CONNTRACK_LISTEN */
 117  };
 118
 119 #define sNO TCP_CONNTRACK_NONE
 120 #define sSS TCP_CONNTRACK_SYN_SENT
 121 #define sSR TCP_CONNTRACK_SYN_RECV
 122 #define sES TCP_CONNTRACK_ESTABLISHED
 123 #define sFW TCP_CONNTRACK_FIN_WAIT
 124 #define sCW TCP_CONNTRACK_CLOSE_WAIT
 125 #define sLA TCP_CONNTRACK_LAST_ACK
 126 #define sTW TCP_CONNTRACK_TIME_WAIT
 127 #define sCL TCP_CONNTRACK_CLOSE
 128 #define sLI TCP_CONNTRACK_LISTEN
 129 #define sIV TCP_CONNTRACK_MAX
 130 #define sIG TCP_CONNTRACK_IGNORE
 131
 132 /* What TCP flags are set from RST/SYN/FIN/ACK. */
 133 enum tcp_bit_set {
 134         TCP_SYN_SET,
 135         TCP_SYNACK_SET,
 136         TCP_FIN_SET,
 137         TCP_ACK_SET,
 138         TCP_RST_SET,
 139         TCP_NONE_SET,
 140 };
 141
 142 /*
 143  * The TCP state transition table needs a few words...
 144  *
 145  * We are the man in the middle. All the packets go through us
 146  * but might get lost in transit to the destination.
 147  * It is assumed that the destinations can't receive segments
 148  * we haven't seen.
 149  *
 150  * The checked segment is in window, but our windows are *not*
 151  * equivalent with the ones of the sender/receiver. We always
 152  * try to guess the state of the current sender.
 153  *
 154  * The meaning of the states are:
 155  *
 156  * NONE:        initial state
 157  * SYN_SENT:    SYN-only packet seen
 158  * SYN_RECV:    SYN-ACK packet seen
 159  * ESTABLISHED: ACK packet seen
 160  * FIN_WAIT:    FIN packet seen
 161  * CLOSE_WAIT:  ACK seen (after FIN)
 162  * LAST_ACK:    FIN seen (after FIN)
 163  * TIME_WAIT:   last ACK seen
 164  * CLOSE:       closed connection
 165  *
 166  * LISTEN state is not used.
 167  *
 168  * Packets marked as IGNORED (sIG):
 169  *      if they may be either invalid or valid
 170  *      and the receiver may send back a connection
 171  *      closing RST or a SYN/ACK.
 172  *
 173  * Packets marked as INVALID (sIV):
 174  *      if they are invalid
 175  *      or we do not support the request (simultaneous open)
 176  */
 177 static enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 178         {
 179 /* ORIGINAL */
 180 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 181 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
 182 /*
 183  *      sNO -> sSS      Initialize a new connection
 184  *      sSS -> sSS      Retransmitted SYN
 185  *      sSR -> sIG      Late retransmitted SYN?
 186  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 187  *                      are errors. Receiver will reply with RST
 188  *                      and close the connection.
 189  *                      Or we are not in sync and hold a dead connection.
 190  *      sFW -> sIG
 191  *      sCW -> sIG
 192  *      sLA -> sIG
 193  *      sTW -> sSS      Reopened connection (RFC 1122).
 194  *      sCL -> sSS
 195  */
 196 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 197 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 198 /*
 199  * A SYN/ACK from the client is always invalid:
 200  *      - either it tries to set up a simultaneous open, which is
 201  *        not supported;
 202  *      - or the firewall has just been inserted between the two hosts
 203  *        during the session set-up. The SYN will be retransmitted
 204  *        by the true client (or it'll time out).
 205  */
 206 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 207 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 208 /*
 209  *      sNO -> sIV      Too late and no reason to do anything...
 210  *      sSS -> sIV      Client migth not send FIN in this state:
 211  *                      we enforce waiting for a SYN/ACK reply first.
 212  *      sSR -> sFW      Close started.
 213  *      sES -> sFW
 214  *      sFW -> sLA      FIN seen in both directions, waiting for
 215  *                      the last ACK.
 216  *                      Migth be a retransmitted FIN as well...
 217  *      sCW -> sLA
 218  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 219  *      sTW -> sTW
 220  *      sCL -> sCL
 221  */
 222 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 223 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 224 /*
 225  *      sNO -> sES      Assumed.
 226  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 227  *      sSR -> sES      Established state is reached.
 228  *      sES -> sES      :-)
 229  *      sFW -> sCW      Normal close request answered by ACK.
 230  *      sCW -> sCW
 231  *      sLA -> sTW      Last ACK detected.
 232  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 233  *      sCL -> sCL
 234  */
 235 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 236 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 237 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 238         },
 239         {
 240 /* REPLY */
 241 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 242 /*syn*/    { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 243 /*
 244  *      sNO -> sIV      Never reached.
 245  *      sSS -> sIV      Simultaneous open, not supported
 246  *      sSR -> sIV      Simultaneous open, not supported.
 247  *      sES -> sIV      Server may not initiate a connection.
 248  *      sFW -> sIV
 249  *      sCW -> sIV
 250  *      sLA -> sIV
 251  *      sTW -> sIV      Reopened connection, but server may not do it.
 252  *      sCL -> sIV
 253  */
 254 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 255 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
 256 /*
 257  *      sSS -> sSR      Standard open.
 258  *      sSR -> sSR      Retransmitted SYN/ACK.
 259  *      sES -> sIG      Late retransmitted SYN/ACK?
 260  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 261  *      sCW -> sIG
 262  *      sLA -> sIG
 263  *      sTW -> sIG
 264  *      sCL -> sIG
 265  */
 266 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 267 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 268 /*
 269  *      sSS -> sIV      Server might not send FIN in this state.
 270  *      sSR -> sFW      Close started.
 271  *      sES -> sFW
 272  *      sFW -> sLA      FIN seen in both directions.
 273  *      sCW -> sLA
 274  *      sLA -> sLA      Retransmitted FIN.
 275  *      sTW -> sTW
 276  *      sCL -> sCL
 277  */
 278 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 279 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 280 /*
 281  *      sSS -> sIG      Might be a half-open connection.
 282  *      sSR -> sSR      Might answer late resent SYN.
 283  *      sES -> sES      :-)
 284  *      sFW -> sCW      Normal close request answered by ACK.
 285  *      sCW -> sCW
 286  *      sLA -> sTW      Last ACK detected.
 287  *      sTW -> sTW      Retransmitted last ACK.
 288  *      sCL -> sCL
 289  */
 290 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 291 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 292 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 293         }
 294 };
 295
 296 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
 297                             unsigned int dataoff,
 298                             struct nf_conntrack_tuple *tuple)
 299 {
 300         struct tcphdr _hdr, *hp;
 301
 302         /* Actually only need first 8 bytes. */
 303         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
 304         if (hp == NULL)
 305                 return 0;
 306
 307         tuple->src.u.tcp.port = hp->source;
 308         tuple->dst.u.tcp.port = hp->dest;
 309
 310         return 1;
 311 }
 312
 313 static int tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
 314                             const struct nf_conntrack_tuple *orig)
 315 {
 316         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
 317         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
 318         return 1;
 319 }
 320
 321 /* Print out the per-protocol part of the tuple. */
 322 static int tcp_print_tuple(struct seq_file *s,
 323                            const struct nf_conntrack_tuple *tuple)
 324 {
 325         return seq_printf(s, "sport=%hu dport=%hu ",
 326                           ntohs(tuple->src.u.tcp.port),
 327                           ntohs(tuple->dst.u.tcp.port));
 328 }
 329
 330 /* Print out the private part of the conntrack. */
 331 static int tcp_print_conntrack(struct seq_file *s,
 332                                const struct nf_conn *conntrack)
 333 {
 334         enum tcp_conntrack state;
 335
 336         read_lock_bh(&tcp_lock);
 337         state = conntrack->proto.tcp.state;
 338         read_unlock_bh(&tcp_lock);
 339
 340         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
 341 }
 342
 343 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 344 {
 345         if (tcph->rst) return TCP_RST_SET;
 346         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 347         else if (tcph->fin) return TCP_FIN_SET;
 348         else if (tcph->ack) return TCP_ACK_SET;
 349         else return TCP_NONE_SET;
 350 }
 351
 352 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 353    in IP Filter' by Guido van Rooij.
 354
 355    http://www.nluug.nl/events/sane2000/papers.html
 356    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
 357
 358    The boundaries and the conditions are changed according to RFC793:
 359    the packet must intersect the window (i.e. segments may be
 360    after the right or before the left edge) and thus receivers may ACK
 361    segments after the right edge of the window.
 362
 363         td_maxend = max(sack + max(win,1)) seen in reply packets
 364         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 365         td_maxwin += seq + len - sender.td_maxend
 366                         if seq + len > sender.td_maxend
 367         td_end    = max(seq + len) seen in sent packets
 368
 369    I.   Upper bound for valid data:     seq <= sender.td_maxend
 370    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 371    III. Upper bound for valid ack:      sack <= receiver.td_end
 372    IV.  Lower bound for valid ack:      ack >= receiver.td_end - MAXACKWINDOW
 373
 374    where sack is the highest right edge of sack block found in the packet.
 375
 376    The upper bound limit for a valid ack is not ignored -
 377    we doesn't have to deal with fragments.
 378 */
 379
 380 static inline __u32 segment_seq_plus_len(__u32 seq,
 381                                          size_t len,
 382                                          unsigned int dataoff,
 383                                          struct tcphdr *tcph)
 384 {
 385         /* XXX Should I use payload length field in IP/IPv6 header ?
 386          * - YK */
 387         return (seq + len - dataoff - tcph->doff*4
 388                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 389 }
 390
 391 /* Fixme: what about big packets? */
 392 #define MAXACKWINCONST                  66000
 393 #define MAXACKWINDOW(sender)                                            \
 394         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 395                                               : MAXACKWINCONST)
 396
 397 /*
 398  * Simplified tcp_parse_options routine from tcp_input.c
 399  */
 400 static void tcp_options(const struct sk_buff *skb,
 401                         unsigned int dataoff,
 402                         struct tcphdr *tcph,
 403                         struct ip_ct_tcp_state *state)
 404 {
 405         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 406         unsigned char *ptr;
 407         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 408
 409         if (!length)
 410                 return;
 411
 412         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 413                                  length, buff);
 414         BUG_ON(ptr == NULL);
 415
 416         state->td_scale =
 417         state->flags = 0;
 418
 419         while (length > 0) {
 420                 int opcode=*ptr++;
 421                 int opsize;
 422
 423                 switch (opcode) {
 424                 case TCPOPT_EOL:
 425                         return;
 426                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 427                         length--;
 428                         continue;
 429                 default:
 430                         opsize=*ptr++;
 431                         if (opsize < 2) /* "silly options" */
 432                                 return;
 433                         if (opsize > length)
 434                                 break;  /* don't parse partial options */
 435
 436                         if (opcode == TCPOPT_SACK_PERM
 437                             && opsize == TCPOLEN_SACK_PERM)
 438                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 439                         else if (opcode == TCPOPT_WINDOW
 440                                  && opsize == TCPOLEN_WINDOW) {
 441                                 state->td_scale = *(u_int8_t *)ptr;
 442
 443                                 if (state->td_scale > 14) {
 444                                         /* See RFC1323 */
 445                                         state->td_scale = 14;
 446                                 }
 447                                 state->flags |=
 448                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 449                         }
 450                         ptr += opsize - 2;
 451                         length -= opsize;
 452                 }
 453         }
 454 }
 455
 456 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 457                      struct tcphdr *tcph, __u32 *sack)
 458 {
 459         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 460         unsigned char *ptr;
 461         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 462         __u32 tmp;
 463
 464         if (!length)
 465                 return;
 466
 467         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 468                                  length, buff);
 469         BUG_ON(ptr == NULL);
 470
 471         /* Fast path for timestamp-only option */
 472         if (length == TCPOLEN_TSTAMP_ALIGNED*4
 473             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
 474                                        | (TCPOPT_NOP << 16)
 475                                        | (TCPOPT_TIMESTAMP << 8)
 476                                        | TCPOLEN_TIMESTAMP))
 477                 return;
 478
 479         while (length > 0) {
 480                 int opcode = *ptr++;
 481                 int opsize, i;
 482
 483                 switch (opcode) {
 484                 case TCPOPT_EOL:
 485                         return;
 486                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 487                         length--;
 488                         continue;
 489                 default:
 490                         opsize = *ptr++;
 491                         if (opsize < 2) /* "silly options" */
 492                                 return;
 493                         if (opsize > length)
 494                                 break;  /* don't parse partial options */
 495
 496                         if (opcode == TCPOPT_SACK
 497                             && opsize >= (TCPOLEN_SACK_BASE
 498                                           + TCPOLEN_SACK_PERBLOCK)
 499                             && !((opsize - TCPOLEN_SACK_BASE)
 500                                  % TCPOLEN_SACK_PERBLOCK)) {
 501                                 for (i = 0;
 502                                      i < (opsize - TCPOLEN_SACK_BASE);
 503                                      i += TCPOLEN_SACK_PERBLOCK) {
 504                                         tmp = ntohl(*((__be32 *)(ptr+i)+1));
 505
 506                                         if (after(tmp, *sack))
 507                                                 *sack = tmp;
 508                                 }
 509                                 return;
 510                         }
 511                         ptr += opsize - 2;
 512                         length -= opsize;
 513                 }
 514         }
 515 }
 516
 517 static int tcp_in_window(struct ip_ct_tcp *state,
 518                          enum ip_conntrack_dir dir,
 519                          unsigned int index,
 520                          const struct sk_buff *skb,
 521                          unsigned int dataoff,
 522                          struct tcphdr *tcph,
 523                          int pf)
 524 {
 525         struct ip_ct_tcp_state *sender = &state->seen[dir];
 526         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 527         __u32 seq, ack, sack, end, win, swin;
 528         int res;
 529
 530         /*
 531          * Get the required data from the packet.
 532          */
 533         seq = ntohl(tcph->seq);
 534         ack = sack = ntohl(tcph->ack_seq);
 535         win = ntohs(tcph->window);
 536         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 537
 538         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 539                 tcp_sack(skb, dataoff, tcph, &sack);
 540
 541         DEBUGP("tcp_in_window: START\n");
 542         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
 543                "seq=%u ack=%u sack=%u win=%u end=%u\n",
 544                 NIPQUAD(iph->saddr), ntohs(tcph->source),
 545                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
 546                 seq, ack, sack, win, end);
 547         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 548                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 549                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 550                 sender->td_scale,
 551                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 552                 receiver->td_scale);
 553
 554         if (sender->td_end == 0) {
 555                 /*
 556                  * Initialize sender data.
 557                  */
 558                 if (tcph->syn && tcph->ack) {
 559                         /*
 560                          * Outgoing SYN-ACK in reply to a SYN.
 561                          */
 562                         sender->td_end =
 563                         sender->td_maxend = end;
 564                         sender->td_maxwin = (win == 0 ? 1 : win);
 565
 566                         tcp_options(skb, dataoff, tcph, sender);
 567                         /*
 568                          * RFC 1323:
 569                          * Both sides must send the Window Scale option
 570                          * to enable window scaling in either direction.
 571                          */
 572                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 573                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 574                                 sender->td_scale =
 575                                 receiver->td_scale = 0;
 576                 } else {
 577                         /*
 578                          * We are in the middle of a connection,
 579                          * its history is lost for us.
 580                          * Let's try to use the data from the packet.
 581                          */
 582                         sender->td_end = end;
 583                         sender->td_maxwin = (win == 0 ? 1 : win);
 584                         sender->td_maxend = end + sender->td_maxwin;
 585                 }
 586         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 587                      && dir == IP_CT_DIR_ORIGINAL)
 588                    || (state->state == TCP_CONNTRACK_SYN_RECV
 589                      && dir == IP_CT_DIR_REPLY))
 590                    && after(end, sender->td_end)) {
 591                 /*
 592                  * RFC 793: "if a TCP is reinitialized ... then it need
 593                  * not wait at all; it must only be sure to use sequence
 594                  * numbers larger than those recently used."
 595                  */
 596                 sender->td_end =
 597                 sender->td_maxend = end;
 598                 sender->td_maxwin = (win == 0 ? 1 : win);
 599
 600                 tcp_options(skb, dataoff, tcph, sender);
 601         }
 602
 603         if (!(tcph->ack)) {
 604                 /*
 605                  * If there is no ACK, just pretend it was set and OK.
 606                  */
 607                 ack = sack = receiver->td_end;
 608         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 609                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 610                    && (ack == 0)) {
 611                 /*
 612                  * Broken TCP stacks, that set ACK in RST packets as well
 613                  * with zero ack value.
 614                  */
 615                 ack = sack = receiver->td_end;
 616         }
 617
 618         if (seq == end
 619             && (!tcph->rst
 620                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
 621                 /*
 622                  * Packets contains no data: we assume it is valid
 623                  * and check the ack value only.
 624                  * However RST segments are always validated by their
 625                  * SEQ number, except when seq == 0 (reset sent answering
 626                  * SYN.
 627                  */
 628                 seq = end = sender->td_end;
 629
 630         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
 631                "seq=%u ack=%u sack =%u win=%u end=%u\n",
 632                 NIPQUAD(iph->saddr), ntohs(tcph->source),
 633                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
 634                 seq, ack, sack, win, end);
 635         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 636                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 637                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 638                 sender->td_scale,
 639                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 640                 receiver->td_scale);
 641
 642         DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 643                 before(seq, sender->td_maxend + 1),
 644                 after(end, sender->td_end - receiver->td_maxwin - 1),
 645                 before(sack, receiver->td_end + 1),
 646                 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
 647
 648         if (before(seq, sender->td_maxend + 1) &&
 649             after(end, sender->td_end - receiver->td_maxwin - 1) &&
 650             before(sack, receiver->td_end + 1) &&
 651             after(ack, receiver->td_end - MAXACKWINDOW(sender))) {
 652                 /*
 653                  * Take into account window scaling (RFC 1323).
 654                  */
 655                 if (!tcph->syn)
 656                         win <<= sender->td_scale;
 657
 658                 /*
 659                  * Update sender data.
 660                  */
 661                 swin = win + (sack - ack);
 662                 if (sender->td_maxwin < swin)
 663                         sender->td_maxwin = swin;
 664                 if (after(end, sender->td_end))
 665                         sender->td_end = end;
 666                 /*
 667                  * Update receiver data.
 668                  */
 669                 if (after(end, sender->td_maxend))
 670                         receiver->td_maxwin += end - sender->td_maxend;
 671                 if (after(sack + win, receiver->td_maxend - 1)) {
 672                         receiver->td_maxend = sack + win;
 673                         if (win == 0)
 674                                 receiver->td_maxend++;
 675                 }
 676
 677                 /*
 678                  * Check retransmissions.
 679                  */
 680                 if (index == TCP_ACK_SET) {
 681                         if (state->last_dir == dir
 682                             && state->last_seq == seq
 683                             && state->last_ack == ack
 684                             && state->last_end == end
 685                             && state->last_win == win)
 686                                 state->retrans++;
 687                         else {
 688                                 state->last_dir = dir;
 689                                 state->last_seq = seq;
 690                                 state->last_ack = ack;
 691                                 state->last_end = end;
 692                                 state->last_win = win;
 693                                 state->retrans = 0;
 694                         }
 695                 }
 696                 res = 1;
 697         } else {
 698                 res = 0;
 699                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 700                     nf_ct_tcp_be_liberal)
 701                         res = 1;
 702                 if (!res && LOG_INVALID(IPPROTO_TCP))
 703                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 704                         "nf_ct_tcp: %s ",
 705                         before(seq, sender->td_maxend + 1) ?
 706                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
 707                         before(sack, receiver->td_end + 1) ?
 708                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
 709                         : "ACK is under the lower bound (possible overly delayed ACK)"
 710                         : "ACK is over the upper bound (ACKed data not seen yet)"
 711                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 712                         : "SEQ is over the upper bound (over the window of the receiver)");
 713         }
 714
 715         DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
 716                "receiver end=%u maxend=%u maxwin=%u\n",
 717                 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 718                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 719
 720         return res;
 721 }
 722
 723 #ifdef CONFIG_NF_NAT_NEEDED
 724 /* Update sender->td_end after NAT successfully mangled the packet */
 725 /* Caller must linearize skb at tcp header. */
 726 void nf_conntrack_tcp_update(struct sk_buff *skb,
 727                              unsigned int dataoff,
 728                              struct nf_conn *conntrack,
 729                              int dir)
 730 {
 731         struct tcphdr *tcph = (void *)skb->data + dataoff;
 732         __u32 end;
 733 #ifdef DEBUGP_VARS
 734         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
 735         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
 736 #endif
 737
 738         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
 739
 740         write_lock_bh(&tcp_lock);
 741         /*
 742          * We have to worry for the ack in the reply packet only...
 743          */
 744         if (after(end, conntrack->proto.tcp.seen[dir].td_end))
 745                 conntrack->proto.tcp.seen[dir].td_end = end;
 746         conntrack->proto.tcp.last_end = end;
 747         write_unlock_bh(&tcp_lock);
 748         DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
 749                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 750                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 751                 sender->td_scale,
 752                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 753                 receiver->td_scale);
 754 }
 755 EXPORT_SYMBOL_GPL(nf_conntrack_tcp_update);
 756 #endif
 757
 758 #define TH_FIN  0x01
 759 #define TH_SYN  0x02
 760 #define TH_RST  0x04
 761 #define TH_PUSH 0x08
 762 #define TH_ACK  0x10
 763 #define TH_URG  0x20
 764 #define TH_ECE  0x40
 765 #define TH_CWR  0x80
 766
 767 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 768 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_ACK|TH_URG) + 1] =
 769 {
 770         [TH_SYN]                        = 1,
 771         [TH_SYN|TH_URG]                 = 1,
 772         [TH_SYN|TH_ACK]                 = 1,
 773         [TH_RST]                        = 1,
 774         [TH_RST|TH_ACK]                 = 1,
 775         [TH_FIN|TH_ACK]                 = 1,
 776         [TH_FIN|TH_ACK|TH_URG]          = 1,
 777         [TH_ACK]                        = 1,
 778         [TH_ACK|TH_URG]                 = 1,
 779 };
 780
 781 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 782 static int tcp_error(struct sk_buff *skb,
 783                      unsigned int dataoff,
 784                      enum ip_conntrack_info *ctinfo,
 785                      int pf,
 786                      unsigned int hooknum)
 787 {
 788         struct tcphdr _tcph, *th;
 789         unsigned int tcplen = skb->len - dataoff;
 790         u_int8_t tcpflags;
 791
 792         /* Smaller that minimal TCP header? */
 793         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 794         if (th == NULL) {
 795                 if (LOG_INVALID(IPPROTO_TCP))
 796                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 797                                 "nf_ct_tcp: short packet ");
 798                 return -NF_ACCEPT;
 799         }
 800
 801         /* Not whole TCP header or malformed packet */
 802         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 803                 if (LOG_INVALID(IPPROTO_TCP))
 804                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 805                                 "nf_ct_tcp: truncated/malformed packet ");
 806                 return -NF_ACCEPT;
 807         }
 808
 809         /* Checksum invalid? Ignore.
 810          * We skip checking packets on the outgoing path
 811          * because the checksum is assumed to be correct.
 812          */
 813         /* FIXME: Source route IP option packets --RR */
 814         if (nf_conntrack_checksum &&
 815             ((pf == PF_INET && hooknum == NF_IP_PRE_ROUTING) ||
 816              (pf == PF_INET6 && hooknum == NF_IP6_PRE_ROUTING)) &&
 817             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
 818                 if (LOG_INVALID(IPPROTO_TCP))
 819                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 820                                   "nf_ct_tcp: bad TCP checksum ");
 821                 return -NF_ACCEPT;
 822         }
 823
 824         /* Check TCP flags. */
 825         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR|TH_PUSH));
 826         if (!tcp_valid_flags[tcpflags]) {
 827                 if (LOG_INVALID(IPPROTO_TCP))
 828                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 829                                   "nf_ct_tcp: invalid TCP flag combination ");
 830                 return -NF_ACCEPT;
 831         }
 832
 833         return NF_ACCEPT;
 834 }
 835
 836 /* Returns verdict for packet, or -1 for invalid. */
 837 static int tcp_packet(struct nf_conn *conntrack,
 838                       const struct sk_buff *skb,
 839                       unsigned int dataoff,
 840                       enum ip_conntrack_info ctinfo,
 841                       int pf,
 842                       unsigned int hooknum)
 843 {
 844         enum tcp_conntrack new_state, old_state;
 845         enum ip_conntrack_dir dir;
 846         struct tcphdr *th, _tcph;
 847         unsigned long timeout;
 848         unsigned int index;
 849
 850         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 851         BUG_ON(th == NULL);
 852
 853         write_lock_bh(&tcp_lock);
 854         old_state = conntrack->proto.tcp.state;
 855         dir = CTINFO2DIR(ctinfo);
 856         index = get_conntrack_index(th);
 857         new_state = tcp_conntracks[dir][index][old_state];
 858
 859         switch (new_state) {
 860         case TCP_CONNTRACK_IGNORE:
 861                 /* Ignored packets:
 862                  *
 863                  * a) SYN in ORIGINAL
 864                  * b) SYN/ACK in REPLY
 865                  * c) ACK in reply direction after initial SYN in original.
 866                  */
 867                 if (index == TCP_SYNACK_SET
 868                     && conntrack->proto.tcp.last_index == TCP_SYN_SET
 869                     && conntrack->proto.tcp.last_dir != dir
 870                     && ntohl(th->ack_seq) ==
 871                              conntrack->proto.tcp.last_end) {
 872                         /* This SYN/ACK acknowledges a SYN that we earlier
 873                          * ignored as invalid. This means that the client and
 874                          * the server are both in sync, while the firewall is
 875                          * not. We kill this session and block the SYN/ACK so
 876                          * that the client cannot but retransmit its SYN and
 877                          * thus initiate a clean new session.
 878                          */
 879                         write_unlock_bh(&tcp_lock);
 880                         if (LOG_INVALID(IPPROTO_TCP))
 881                                 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 882                                           "nf_ct_tcp: killing out of sync session ");
 883                         if (del_timer(&conntrack->timeout))
 884                                 conntrack->timeout.function((unsigned long)
 885                                                             conntrack);
 886                         return -NF_DROP;
 887                 }
 888                 conntrack->proto.tcp.last_index = index;
 889                 conntrack->proto.tcp.last_dir = dir;
 890                 conntrack->proto.tcp.last_seq = ntohl(th->seq);
 891                 conntrack->proto.tcp.last_end =
 892                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 893
 894                 write_unlock_bh(&tcp_lock);
 895                 if (LOG_INVALID(IPPROTO_TCP))
 896                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 897                                   "nf_ct_tcp: invalid packed ignored ");
 898                 return NF_ACCEPT;
 899         case TCP_CONNTRACK_MAX:
 900                 /* Invalid packet */
 901                 DEBUGP("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 902                        dir, get_conntrack_index(th),
 903                        old_state);
 904                 write_unlock_bh(&tcp_lock);
 905                 if (LOG_INVALID(IPPROTO_TCP))
 906                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 907                                   "nf_ct_tcp: invalid state ");
 908                 return -NF_ACCEPT;
 909         case TCP_CONNTRACK_SYN_SENT:
 910                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 911                         break;
 912                 if ((conntrack->proto.tcp.seen[dir].flags &
 913                         IP_CT_TCP_FLAG_CLOSE_INIT)
 914                     || after(ntohl(th->seq),
 915                              conntrack->proto.tcp.seen[dir].td_end)) {
 916                         /* Attempt to reopen a closed connection.
 917                         * Delete this connection and look up again. */
 918                         write_unlock_bh(&tcp_lock);
 919                         if (del_timer(&conntrack->timeout))
 920                                 conntrack->timeout.function((unsigned long)
 921                                                             conntrack);
 922                         return -NF_REPEAT;
 923                 } else {
 924                         write_unlock_bh(&tcp_lock);
 925                         if (LOG_INVALID(IPPROTO_TCP))
 926                                 nf_log_packet(pf, 0, skb, NULL, NULL,
 927                                               NULL, "nf_ct_tcp: invalid SYN");
 928                         return -NF_ACCEPT;
 929                 }
 930         case TCP_CONNTRACK_CLOSE:
 931                 if (index == TCP_RST_SET
 932                     && ((test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
 933                          && conntrack->proto.tcp.last_index == TCP_SYN_SET)
 934                         || (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
 935                             && conntrack->proto.tcp.last_index == TCP_ACK_SET))
 936                     && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
 937                         /* RST sent to invalid SYN or ACK we had let through
 938                          * at a) and c) above:
 939                          *
 940                          * a) SYN was in window then
 941                          * c) we hold a half-open connection.
 942                          *
 943                          * Delete our connection entry.
 944                          * We skip window checking, because packet might ACK
 945                          * segments we ignored. */
 946                         goto in_window;
 947                 }
 948                 /* Just fall through */
 949         default:
 950                 /* Keep compilers happy. */
 951                 break;
 952         }
 953
 954         if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
 955                            skb, dataoff, th, pf)) {
 956                 write_unlock_bh(&tcp_lock);
 957                 return -NF_ACCEPT;
 958         }
 959      in_window:
 960         /* From now on we have got in-window packets */
 961         conntrack->proto.tcp.last_index = index;
 962
 963         DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
 964                "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
 965                 NIPQUAD(iph->saddr), ntohs(th->source),
 966                 NIPQUAD(iph->daddr), ntohs(th->dest),
 967                 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
 968                 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
 969                 old_state, new_state);
 970
 971         conntrack->proto.tcp.state = new_state;
 972         if (old_state != new_state
 973             && (new_state == TCP_CONNTRACK_FIN_WAIT
 974                 || new_state == TCP_CONNTRACK_CLOSE))
 975                 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
 976         timeout = conntrack->proto.tcp.retrans >= nf_ct_tcp_max_retrans
 977                   && *tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans
 978                   ? nf_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
 979         write_unlock_bh(&tcp_lock);
 980
 981         nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
 982         if (new_state != old_state)
 983                 nf_conntrack_event_cache(IPCT_PROTOINFO, skb);
 984
 985         if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
 986                 /* If only reply is a RST, we can consider ourselves not to
 987                    have an established connection: this is a fairly common
 988                    problem case, so we can delete the conntrack
 989                    immediately.  --RR */
 990                 if (th->rst) {
 991                         if (del_timer(&conntrack->timeout))
 992                                 conntrack->timeout.function((unsigned long)
 993                                                             conntrack);
 994                         return NF_ACCEPT;
 995                 }
 996         } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
 997                    && (old_state == TCP_CONNTRACK_SYN_RECV
 998                        || old_state == TCP_CONNTRACK_ESTABLISHED)
 999                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1000                 /* Set ASSURED if we see see valid ack in ESTABLISHED
1001                    after SYN_RECV or a valid answer for a picked up
1002                    connection. */
1003                 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1004                 nf_conntrack_event_cache(IPCT_STATUS, skb);
1005         }
1006         nf_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1007
1008         return NF_ACCEPT;
1009 }
1010
1011 /* Called when a new connection for this protocol found. */
1012 static int tcp_new(struct nf_conn *conntrack,
1013                    const struct sk_buff *skb,
1014                    unsigned int dataoff)
1015 {
1016         enum tcp_conntrack new_state;
1017         struct tcphdr *th, _tcph;
1018 #ifdef DEBUGP_VARS
1019         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1020         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1021 #endif
1022
1023         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1024         BUG_ON(th == NULL);
1025
1026         /* Don't need lock here: this conntrack not in circulation yet */
1027         new_state
1028                 = tcp_conntracks[0][get_conntrack_index(th)]
1029                 [TCP_CONNTRACK_NONE];
1030
1031         /* Invalid: delete conntrack */
1032         if (new_state >= TCP_CONNTRACK_MAX) {
1033                 DEBUGP("nf_ct_tcp: invalid new deleting.\n");
1034                 return 0;
1035         }
1036
1037         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1038                 /* SYN packet */
1039                 conntrack->proto.tcp.seen[0].td_end =
1040                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1041                                              dataoff, th);
1042                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1043                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1044                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1045                 conntrack->proto.tcp.seen[0].td_maxend =
1046                         conntrack->proto.tcp.seen[0].td_end;
1047
1048                 tcp_options(skb, dataoff, th, &conntrack->proto.tcp.seen[0]);
1049                 conntrack->proto.tcp.seen[1].flags = 0;
1050         } else if (nf_ct_tcp_loose == 0) {
1051                 /* Don't try to pick up connections. */
1052                 return 0;
1053         } else {
1054                 /*
1055                  * We are in the middle of a connection,
1056                  * its history is lost for us.
1057                  * Let's try to use the data from the packet.
1058                  */
1059                 conntrack->proto.tcp.seen[0].td_end =
1060                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1061                                              dataoff, th);
1062                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1063                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1064                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1065                 conntrack->proto.tcp.seen[0].td_maxend =
1066                         conntrack->proto.tcp.seen[0].td_end +
1067                         conntrack->proto.tcp.seen[0].td_maxwin;
1068                 conntrack->proto.tcp.seen[0].td_scale = 0;
1069
1070                 /* We assume SACK and liberal window checking to handle
1071                  * window scaling */
1072                 conntrack->proto.tcp.seen[0].flags =
1073                 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1074                                                      IP_CT_TCP_FLAG_BE_LIBERAL;
1075         }
1076
1077         conntrack->proto.tcp.seen[1].td_end = 0;
1078         conntrack->proto.tcp.seen[1].td_maxend = 0;
1079         conntrack->proto.tcp.seen[1].td_maxwin = 1;
1080         conntrack->proto.tcp.seen[1].td_scale = 0;
1081
1082         /* tcp_packet will set them */
1083         conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1084         conntrack->proto.tcp.last_index = TCP_NONE_SET;
1085
1086         DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1087                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1088                 sender->td_end, sender->td_maxend, sender->td_maxwin,
1089                 sender->td_scale,
1090                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1091                 receiver->td_scale);
1092         return 1;
1093 }
1094
1095 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1096
1097 #include <linux/netfilter/nfnetlink.h>
1098 #include <linux/netfilter/nfnetlink_conntrack.h>
1099
1100 static int tcp_to_nfattr(struct sk_buff *skb, struct nfattr *nfa,
1101                          const struct nf_conn *ct)
1102 {
1103         struct nfattr *nest_parms;
1104
1105         read_lock_bh(&tcp_lock);
1106         nest_parms = NFA_NEST(skb, CTA_PROTOINFO_TCP);
1107         NFA_PUT(skb, CTA_PROTOINFO_TCP_STATE, sizeof(u_int8_t),
1108                 &ct->proto.tcp.state);
1109         read_unlock_bh(&tcp_lock);
1110
1111         NFA_NEST_END(skb, nest_parms);
1112
1113         return 0;
1114
1115 nfattr_failure:
1116         read_unlock_bh(&tcp_lock);
1117         return -1;
1118 }
1119
1120 static const size_t cta_min_tcp[CTA_PROTOINFO_TCP_MAX] = {
1121         [CTA_PROTOINFO_TCP_STATE-1]     = sizeof(u_int8_t),
1122 };
1123
1124 static int nfattr_to_tcp(struct nfattr *cda[], struct nf_conn *ct)
1125 {
1126         struct nfattr *attr = cda[CTA_PROTOINFO_TCP-1];
1127         struct nfattr *tb[CTA_PROTOINFO_TCP_MAX];
1128
1129         /* updates could not contain anything about the private
1130          * protocol info, in that case skip the parsing */
1131         if (!attr)
1132                 return 0;
1133
1134         nfattr_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, attr);
1135
1136         if (nfattr_bad_size(tb, CTA_PROTOINFO_TCP_MAX, cta_min_tcp))
1137                 return -EINVAL;
1138
1139         if (!tb[CTA_PROTOINFO_TCP_STATE-1])
1140                 return -EINVAL;
1141
1142         write_lock_bh(&tcp_lock);
1143         ct->proto.tcp.state =
1144                 *(u_int8_t *)NFA_DATA(tb[CTA_PROTOINFO_TCP_STATE-1]);
1145         write_unlock_bh(&tcp_lock);
1146
1147         return 0;
1148 }
1149 #endif
1150
1151 #ifdef CONFIG_SYSCTL
1152 static unsigned int tcp_sysctl_table_users;
1153 static struct ctl_table_header *tcp_sysctl_header;
1154 static struct ctl_table tcp_sysctl_table[] = {
1155         {
1156                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_SYN_SENT,
1157                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1158                 .data           = &nf_ct_tcp_timeout_syn_sent,
1159                 .maxlen         = sizeof(unsigned int),
1160                 .mode           = 0644,
1161                 .proc_handler   = &proc_dointvec_jiffies,
1162         },
1163         {
1164                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_SYN_RECV,
1165                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1166                 .data           = &nf_ct_tcp_timeout_syn_recv,
1167                 .maxlen         = sizeof(unsigned int),
1168                 .mode           = 0644,
1169                 .proc_handler   = &proc_dointvec_jiffies,
1170         },
1171         {
1172                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_ESTABLISHED,
1173                 .procname       = "nf_conntrack_tcp_timeout_established",
1174                 .data           = &nf_ct_tcp_timeout_established,
1175                 .maxlen         = sizeof(unsigned int),
1176                 .mode           = 0644,
1177                 .proc_handler   = &proc_dointvec_jiffies,
1178         },
1179         {
1180                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_FIN_WAIT,
1181                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1182                 .data           = &nf_ct_tcp_timeout_fin_wait,
1183                 .maxlen         = sizeof(unsigned int),
1184                 .mode           = 0644,
1185                 .proc_handler   = &proc_dointvec_jiffies,
1186         },
1187         {
1188                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_CLOSE_WAIT,
1189                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1190                 .data           = &nf_ct_tcp_timeout_close_wait,
1191                 .maxlen         = sizeof(unsigned int),
1192                 .mode           = 0644,
1193                 .proc_handler   = &proc_dointvec_jiffies,
1194         },
1195         {
1196                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_LAST_ACK,
1197                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1198                 .data           = &nf_ct_tcp_timeout_last_ack,
1199                 .maxlen         = sizeof(unsigned int),
1200                 .mode           = 0644,
1201                 .proc_handler   = &proc_dointvec_jiffies,
1202         },
1203         {
1204                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_TIME_WAIT,
1205                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1206                 .data           = &nf_ct_tcp_timeout_time_wait,
1207                 .maxlen         = sizeof(unsigned int),
1208                 .mode           = 0644,
1209                 .proc_handler   = &proc_dointvec_jiffies,
1210         },
1211         {
1212                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_CLOSE,
1213                 .procname       = "nf_conntrack_tcp_timeout_close",
1214                 .data           = &nf_ct_tcp_timeout_close,
1215                 .maxlen         = sizeof(unsigned int),
1216                 .mode           = 0644,
1217                 .proc_handler   = &proc_dointvec_jiffies,
1218         },
1219         {
1220                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_MAX_RETRANS,
1221                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1222                 .data           = &nf_ct_tcp_timeout_max_retrans,
1223                 .maxlen         = sizeof(unsigned int),
1224                 .mode           = 0644,
1225                 .proc_handler   = &proc_dointvec_jiffies,
1226         },
1227         {
1228                 .ctl_name       = NET_NF_CONNTRACK_TCP_LOOSE,
1229                 .procname       = "nf_conntrack_tcp_loose",
1230                 .data           = &nf_ct_tcp_loose,
1231                 .maxlen         = sizeof(unsigned int),
1232                 .mode           = 0644,
1233                 .proc_handler   = &proc_dointvec,
1234         },
1235         {
1236                 .ctl_name       = NET_NF_CONNTRACK_TCP_BE_LIBERAL,
1237                 .procname       = "nf_conntrack_tcp_be_liberal",
1238                 .data           = &nf_ct_tcp_be_liberal,
1239                 .maxlen         = sizeof(unsigned int),
1240                 .mode           = 0644,
1241                 .proc_handler   = &proc_dointvec,
1242         },
1243         {
1244                 .ctl_name       = NET_NF_CONNTRACK_TCP_MAX_RETRANS,
1245                 .procname       = "nf_conntrack_tcp_max_retrans",
1246                 .data           = &nf_ct_tcp_max_retrans,
1247                 .maxlen         = sizeof(unsigned int),
1248                 .mode           = 0644,
1249                 .proc_handler   = &proc_dointvec,
1250         },
1251         {
1252                 .ctl_name       = 0
1253         }
1254 };
1255
1256 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1257 static struct ctl_table tcp_compat_sysctl_table[] = {
1258         {
1259                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_SYN_SENT,
1260                 .procname       = "ip_conntrack_tcp_timeout_syn_sent",
1261                 .data           = &nf_ct_tcp_timeout_syn_sent,
1262                 .maxlen         = sizeof(unsigned int),
1263                 .mode           = 0644,
1264                 .proc_handler   = &proc_dointvec_jiffies,
1265         },
1266         {
1267                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_SYN_RECV,
1268                 .procname       = "ip_conntrack_tcp_timeout_syn_recv",
1269                 .data           = &nf_ct_tcp_timeout_syn_recv,
1270                 .maxlen         = sizeof(unsigned int),
1271                 .mode           = 0644,
1272                 .proc_handler   = &proc_dointvec_jiffies,
1273         },
1274         {
1275                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_ESTABLISHED,
1276                 .procname       = "ip_conntrack_tcp_timeout_established",
1277                 .data           = &nf_ct_tcp_timeout_established,
1278                 .maxlen         = sizeof(unsigned int),
1279                 .mode           = 0644,
1280                 .proc_handler   = &proc_dointvec_jiffies,
1281         },
1282         {
1283                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_FIN_WAIT,
1284                 .procname       = "ip_conntrack_tcp_timeout_fin_wait",
1285                 .data           = &nf_ct_tcp_timeout_fin_wait,
1286                 .maxlen         = sizeof(unsigned int),
1287                 .mode           = 0644,
1288                 .proc_handler   = &proc_dointvec_jiffies,
1289         },
1290         {
1291                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_CLOSE_WAIT,
1292                 .procname       = "ip_conntrack_tcp_timeout_close_wait",
1293                 .data           = &nf_ct_tcp_timeout_close_wait,
1294                 .maxlen         = sizeof(unsigned int),
1295                 .mode           = 0644,
1296                 .proc_handler   = &proc_dointvec_jiffies,
1297         },
1298         {
1299                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_LAST_ACK,
1300                 .procname       = "ip_conntrack_tcp_timeout_last_ack",
1301                 .data           = &nf_ct_tcp_timeout_last_ack,
1302                 .maxlen         = sizeof(unsigned int),
1303                 .mode           = 0644,
1304                 .proc_handler   = &proc_dointvec_jiffies,
1305         },
1306         {
1307                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_TIME_WAIT,
1308                 .procname       = "ip_conntrack_tcp_timeout_time_wait",
1309                 .data           = &nf_ct_tcp_timeout_time_wait,
1310                 .maxlen         = sizeof(unsigned int),
1311                 .mode           = 0644,
1312                 .proc_handler   = &proc_dointvec_jiffies,
1313         },
1314         {
1315                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_CLOSE,
1316                 .procname       = "ip_conntrack_tcp_timeout_close",
1317                 .data           = &nf_ct_tcp_timeout_close,
1318                 .maxlen         = sizeof(unsigned int),
1319                 .mode           = 0644,
1320                 .proc_handler   = &proc_dointvec_jiffies,
1321         },
1322         {
1323                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_MAX_RETRANS,
1324                 .procname       = "ip_conntrack_tcp_timeout_max_retrans",
1325                 .data           = &nf_ct_tcp_timeout_max_retrans,
1326                 .maxlen         = sizeof(unsigned int),
1327                 .mode           = 0644,
1328                 .proc_handler   = &proc_dointvec_jiffies,
1329         },
1330         {
1331                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_LOOSE,
1332                 .procname       = "ip_conntrack_tcp_loose",
1333                 .data           = &nf_ct_tcp_loose,
1334                 .maxlen         = sizeof(unsigned int),
1335                 .mode           = 0644,
1336                 .proc_handler   = &proc_dointvec,
1337         },
1338         {
1339                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_BE_LIBERAL,
1340                 .procname       = "ip_conntrack_tcp_be_liberal",
1341                 .data           = &nf_ct_tcp_be_liberal,
1342                 .maxlen         = sizeof(unsigned int),
1343                 .mode           = 0644,
1344                 .proc_handler   = &proc_dointvec,
1345         },
1346         {
1347                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_MAX_RETRANS,
1348                 .procname       = "ip_conntrack_tcp_max_retrans",
1349                 .data           = &nf_ct_tcp_max_retrans,
1350                 .maxlen         = sizeof(unsigned int),
1351                 .mode           = 0644,
1352                 .proc_handler   = &proc_dointvec,
1353         },
1354         {
1355                 .ctl_name       = 0
1356         }
1357 };
1358 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
1359 #endif /* CONFIG_SYSCTL */
1360
1361 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 =
1362 {
1363         .l3proto                = PF_INET,
1364         .l4proto                = IPPROTO_TCP,
1365         .name                   = "tcp",
1366         .pkt_to_tuple           = tcp_pkt_to_tuple,
1367         .invert_tuple           = tcp_invert_tuple,
1368         .print_tuple            = tcp_print_tuple,
1369         .print_conntrack        = tcp_print_conntrack,
1370         .packet                 = tcp_packet,
1371         .new                    = tcp_new,
1372         .error                  = tcp_error,
1373 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1374         .to_nfattr              = tcp_to_nfattr,
1375         .from_nfattr            = nfattr_to_tcp,
1376         .tuple_to_nfattr        = nf_ct_port_tuple_to_nfattr,
1377         .nfattr_to_tuple        = nf_ct_port_nfattr_to_tuple,
1378 #endif
1379 #ifdef CONFIG_SYSCTL
1380         .ctl_table_users        = &tcp_sysctl_table_users,
1381         .ctl_table_header       = &tcp_sysctl_header,
1382         .ctl_table              = tcp_sysctl_table,
1383 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1384         .ctl_compat_table       = tcp_compat_sysctl_table,
1385 #endif
1386 #endif
1387 };
1388 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1389
1390 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 =
1391 {
1392         .l3proto                = PF_INET6,
1393         .l4proto                = IPPROTO_TCP,
1394         .name                   = "tcp",
1395         .pkt_to_tuple           = tcp_pkt_to_tuple,
1396         .invert_tuple           = tcp_invert_tuple,
1397         .print_tuple            = tcp_print_tuple,
1398         .print_conntrack        = tcp_print_conntrack,
1399         .packet                 = tcp_packet,
1400         .new                    = tcp_new,
1401         .error                  = tcp_error,
1402 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1403         .to_nfattr              = tcp_to_nfattr,
1404         .from_nfattr            = nfattr_to_tcp,
1405         .tuple_to_nfattr        = nf_ct_port_tuple_to_nfattr,
1406         .nfattr_to_tuple        = nf_ct_port_nfattr_to_tuple,
1407 #endif
1408 #ifdef CONFIG_SYSCTL
1409         .ctl_table_users        = &tcp_sysctl_table_users,
1410         .ctl_table_header       = &tcp_sysctl_header,
1411         .ctl_table              = tcp_sysctl_table,
1412 #endif
1413 };
1414 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);