net/ipv4/netfilter/ip_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  * version 2.2
  20  */
  21
  22 #include <linux/config.h>
  23 #include <linux/types.h>
  24 #include <linux/sched.h>
  25 #include <linux/timer.h>
  26 #include <linux/netfilter.h>
  27 #include <linux/module.h>
  28 #include <linux/in.h>
  29 #include <linux/ip.h>
  30 #include <linux/tcp.h>
  31 #include <linux/spinlock.h>
  32
  33 #include <net/tcp.h>
  34
  35 #include <linux/netfilter.h>
  36 #include <linux/netfilter_ipv4.h>
  37 #include <linux/netfilter_ipv4/ip_conntrack.h>
  38 #include <linux/netfilter_ipv4/ip_conntrack_protocol.h>
  39
  40 #if 0
  41 #define DEBUGP printk
  42 #define DEBUGP_VARS
  43 #else
  44 #define DEBUGP(format, args...)
  45 #endif
  46
  47 /* Protects conntrack->proto.tcp */
  48 static DEFINE_RWLOCK(tcp_lock);
  49
  50 /* "Be conservative in what you do,
  51     be liberal in what you accept from others."
  52     If it's non-zero, we mark only out of window RST segments as INVALID. */
  53 int ip_ct_tcp_be_liberal = 0;
  54
  55 /* When connection is picked up from the middle, how many packets are required
  56    to pass in each direction when we assume we are in sync - if any side uses
  57    window scaling, we lost the game.
  58    If it is set to zero, we disable picking up already established
  59    connections. */
  60 int ip_ct_tcp_loose = 3;
  61
  62 /* Max number of the retransmitted packets without receiving an (acceptable)
  63    ACK from the destination. If this number is reached, a shorter timer
  64    will be started. */
  65 int ip_ct_tcp_max_retrans = 3;
  66
  67   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  68      closely.  They're more complex. --RR */
  69
  70 static const char *tcp_conntrack_names[] = {
  71         "NONE",
  72         "SYN_SENT",
  73         "SYN_RECV",
  74         "ESTABLISHED",
  75         "FIN_WAIT",
  76         "CLOSE_WAIT",
  77         "LAST_ACK",
  78         "TIME_WAIT",
  79         "CLOSE",
  80         "LISTEN"
  81 };
  82
  83 #define SECS * HZ
  84 #define MINS * 60 SECS
  85 #define HOURS * 60 MINS
  86 #define DAYS * 24 HOURS
  87
  88 unsigned long ip_ct_tcp_timeout_syn_sent =      2 MINS;
  89 unsigned long ip_ct_tcp_timeout_syn_recv =     60 SECS;
  90 unsigned long ip_ct_tcp_timeout_established =   5 DAYS;
  91 unsigned long ip_ct_tcp_timeout_fin_wait =      2 MINS;
  92 unsigned long ip_ct_tcp_timeout_close_wait =   60 SECS;
  93 unsigned long ip_ct_tcp_timeout_last_ack =     30 SECS;
  94 unsigned long ip_ct_tcp_timeout_time_wait =     2 MINS;
  95 unsigned long ip_ct_tcp_timeout_close =        10 SECS;
  96
  97 /* RFC1122 says the R2 limit should be at least 100 seconds.
  98    Linux uses 15 packets as limit, which corresponds
  99    to ~13-30min depending on RTO. */
 100 unsigned long ip_ct_tcp_timeout_max_retrans =     5 MINS;
 101
 102 static unsigned long * tcp_timeouts[]
 103 = { NULL,                              /*      TCP_CONNTRACK_NONE */
 104     &ip_ct_tcp_timeout_syn_sent,       /*      TCP_CONNTRACK_SYN_SENT, */
 105     &ip_ct_tcp_timeout_syn_recv,       /*      TCP_CONNTRACK_SYN_RECV, */
 106     &ip_ct_tcp_timeout_established,    /*      TCP_CONNTRACK_ESTABLISHED,      */
 107     &ip_ct_tcp_timeout_fin_wait,       /*      TCP_CONNTRACK_FIN_WAIT, */
 108     &ip_ct_tcp_timeout_close_wait,     /*      TCP_CONNTRACK_CLOSE_WAIT,       */
 109     &ip_ct_tcp_timeout_last_ack,       /*      TCP_CONNTRACK_LAST_ACK, */
 110     &ip_ct_tcp_timeout_time_wait,      /*      TCP_CONNTRACK_TIME_WAIT,        */
 111     &ip_ct_tcp_timeout_close,          /*      TCP_CONNTRACK_CLOSE,    */
 112     NULL,                              /*      TCP_CONNTRACK_LISTEN */
 113  };
 114
 115 #define sNO TCP_CONNTRACK_NONE
 116 #define sSS TCP_CONNTRACK_SYN_SENT
 117 #define sSR TCP_CONNTRACK_SYN_RECV
 118 #define sES TCP_CONNTRACK_ESTABLISHED
 119 #define sFW TCP_CONNTRACK_FIN_WAIT
 120 #define sCW TCP_CONNTRACK_CLOSE_WAIT
 121 #define sLA TCP_CONNTRACK_LAST_ACK
 122 #define sTW TCP_CONNTRACK_TIME_WAIT
 123 #define sCL TCP_CONNTRACK_CLOSE
 124 #define sLI TCP_CONNTRACK_LISTEN
 125 #define sIV TCP_CONNTRACK_MAX
 126 #define sIG TCP_CONNTRACK_IGNORE
 127
 128 /* What TCP flags are set from RST/SYN/FIN/ACK. */
 129 enum tcp_bit_set {
 130         TCP_SYN_SET,
 131         TCP_SYNACK_SET,
 132         TCP_FIN_SET,
 133         TCP_ACK_SET,
 134         TCP_RST_SET,
 135         TCP_NONE_SET,
 136 };
 137
 138 /*
 139  * The TCP state transition table needs a few words...
 140  *
 141  * We are the man in the middle. All the packets go through us
 142  * but might get lost in transit to the destination.
 143  * It is assumed that the destinations can't receive segments
 144  * we haven't seen.
 145  *
 146  * The checked segment is in window, but our windows are *not*
 147  * equivalent with the ones of the sender/receiver. We always
 148  * try to guess the state of the current sender.
 149  *
 150  * The meaning of the states are:
 151  *
 152  * NONE:        initial state
 153  * SYN_SENT:    SYN-only packet seen
 154  * SYN_RECV:    SYN-ACK packet seen
 155  * ESTABLISHED: ACK packet seen
 156  * FIN_WAIT:    FIN packet seen
 157  * CLOSE_WAIT:  ACK seen (after FIN)
 158  * LAST_ACK:    FIN seen (after FIN)
 159  * TIME_WAIT:   last ACK seen
 160  * CLOSE:       closed connection
 161  *
 162  * LISTEN state is not used.
 163  *
 164  * Packets marked as IGNORED (sIG):
 165  *      if they may be either invalid or valid
 166  *      and the receiver may send back a connection
 167  *      closing RST or a SYN/ACK.
 168  *
 169  * Packets marked as INVALID (sIV):
 170  *      if they are invalid
 171  *      or we do not support the request (simultaneous open)
 172  */
 173 static enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 174         {
 175 /* ORIGINAL */
 176 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 177 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
 178 /*
 179  *      sNO -> sSS      Initialize a new connection
 180  *      sSS -> sSS      Retransmitted SYN
 181  *      sSR -> sIG      Late retransmitted SYN?
 182  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 183  *                      are errors. Receiver will reply with RST
 184  *                      and close the connection.
 185  *                      Or we are not in sync and hold a dead connection.
 186  *      sFW -> sIG
 187  *      sCW -> sIG
 188  *      sLA -> sIG
 189  *      sTW -> sSS      Reopened connection (RFC 1122).
 190  *      sCL -> sSS
 191  */
 192 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 193 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 194 /*
 195  * A SYN/ACK from the client is always invalid:
 196  *      - either it tries to set up a simultaneous open, which is
 197  *        not supported;
 198  *      - or the firewall has just been inserted between the two hosts
 199  *        during the session set-up. The SYN will be retransmitted
 200  *        by the true client (or it'll time out).
 201  */
 202 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 203 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 204 /*
 205  *      sNO -> sIV      Too late and no reason to do anything...
 206  *      sSS -> sIV      Client migth not send FIN in this state:
 207  *                      we enforce waiting for a SYN/ACK reply first.
 208  *      sSR -> sFW      Close started.
 209  *      sES -> sFW
 210  *      sFW -> sLA      FIN seen in both directions, waiting for
 211  *                      the last ACK.
 212  *                      Migth be a retransmitted FIN as well...
 213  *      sCW -> sLA
 214  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 215  *      sTW -> sTW
 216  *      sCL -> sCL
 217  */
 218 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 219 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 220 /*
 221  *      sNO -> sES      Assumed.
 222  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 223  *      sSR -> sES      Established state is reached.
 224  *      sES -> sES      :-)
 225  *      sFW -> sCW      Normal close request answered by ACK.
 226  *      sCW -> sCW
 227  *      sLA -> sTW      Last ACK detected.
 228  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 229  *      sCL -> sCL
 230  */
 231 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 232 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 233 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 234         },
 235         {
 236 /* REPLY */
 237 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 238 /*syn*/    { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 239 /*
 240  *      sNO -> sIV      Never reached.
 241  *      sSS -> sIV      Simultaneous open, not supported
 242  *      sSR -> sIV      Simultaneous open, not supported.
 243  *      sES -> sIV      Server may not initiate a connection.
 244  *      sFW -> sIV
 245  *      sCW -> sIV
 246  *      sLA -> sIV
 247  *      sTW -> sIV      Reopened connection, but server may not do it.
 248  *      sCL -> sIV
 249  */
 250 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 251 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
 252 /*
 253  *      sSS -> sSR      Standard open.
 254  *      sSR -> sSR      Retransmitted SYN/ACK.
 255  *      sES -> sIG      Late retransmitted SYN/ACK?
 256  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 257  *      sCW -> sIG
 258  *      sLA -> sIG
 259  *      sTW -> sIG
 260  *      sCL -> sIG
 261  */
 262 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 263 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 264 /*
 265  *      sSS -> sIV      Server might not send FIN in this state.
 266  *      sSR -> sFW      Close started.
 267  *      sES -> sFW
 268  *      sFW -> sLA      FIN seen in both directions.
 269  *      sCW -> sLA
 270  *      sLA -> sLA      Retransmitted FIN.
 271  *      sTW -> sTW
 272  *      sCL -> sCL
 273  */
 274 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 275 /*ack*/    { sIV, sIV, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 276 /*
 277  *      sSS -> sIV      Might be a half-open connection.
 278  *      sSR -> sSR      Might answer late resent SYN.
 279  *      sES -> sES      :-)
 280  *      sFW -> sCW      Normal close request answered by ACK.
 281  *      sCW -> sCW
 282  *      sLA -> sTW      Last ACK detected.
 283  *      sTW -> sTW      Retransmitted last ACK.
 284  *      sCL -> sCL
 285  */
 286 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 287 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 288 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 289         }
 290 };
 291
 292 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
 293                             unsigned int dataoff,
 294                             struct ip_conntrack_tuple *tuple)
 295 {
 296         struct tcphdr _hdr, *hp;
 297
 298         /* Actually only need first 8 bytes. */
 299         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
 300         if (hp == NULL)
 301                 return 0;
 302
 303         tuple->src.u.tcp.port = hp->source;
 304         tuple->dst.u.tcp.port = hp->dest;
 305
 306         return 1;
 307 }
 308
 309 static int tcp_invert_tuple(struct ip_conntrack_tuple *tuple,
 310                             const struct ip_conntrack_tuple *orig)
 311 {
 312         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
 313         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
 314         return 1;
 315 }
 316
 317 /* Print out the per-protocol part of the tuple. */
 318 static int tcp_print_tuple(struct seq_file *s,
 319                            const struct ip_conntrack_tuple *tuple)
 320 {
 321         return seq_printf(s, "sport=%hu dport=%hu ",
 322                           ntohs(tuple->src.u.tcp.port),
 323                           ntohs(tuple->dst.u.tcp.port));
 324 }
 325
 326 /* Print out the private part of the conntrack. */
 327 static int tcp_print_conntrack(struct seq_file *s,
 328                                const struct ip_conntrack *conntrack)
 329 {
 330         enum tcp_conntrack state;
 331
 332         read_lock_bh(&tcp_lock);
 333         state = conntrack->proto.tcp.state;
 334         read_unlock_bh(&tcp_lock);
 335
 336         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
 337 }
 338
 339 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
 340     defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
 341 static int tcp_to_nfattr(struct sk_buff *skb, struct nfattr *nfa,
 342                          const struct ip_conntrack *ct)
 343 {
 344         struct nfattr *nest_parms = NFA_NEST(skb, CTA_PROTOINFO_TCP);
 345
 346         read_lock_bh(&tcp_lock);
 347         NFA_PUT(skb, CTA_PROTOINFO_TCP_STATE, sizeof(u_int8_t),
 348                 &ct->proto.tcp.state);
 349         read_unlock_bh(&tcp_lock);
 350
 351         NFA_NEST_END(skb, nest_parms);
 352
 353         return 0;
 354
 355 nfattr_failure:
 356         read_unlock_bh(&tcp_lock);
 357         return -1;
 358 }
 359
 360 static int nfattr_to_tcp(struct nfattr *cda[], struct ip_conntrack *ct)
 361 {
 362         struct nfattr *attr = cda[CTA_PROTOINFO_TCP-1];
 363         struct nfattr *tb[CTA_PROTOINFO_TCP_MAX];
 364
 365         /* updates could not contain anything about the private
 366          * protocol info, in that case skip the parsing */
 367         if (!attr)
 368                 return 0;
 369
 370         nfattr_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, attr);
 371
 372         if (!tb[CTA_PROTOINFO_TCP_STATE-1])
 373                 return -EINVAL;
 374
 375         write_lock_bh(&tcp_lock);
 376         ct->proto.tcp.state =
 377                 *(u_int8_t *)NFA_DATA(tb[CTA_PROTOINFO_TCP_STATE-1]);
 378         write_unlock_bh(&tcp_lock);
 379
 380         return 0;
 381 }
 382 #endif
 383
 384 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 385 {
 386         if (tcph->rst) return TCP_RST_SET;
 387         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 388         else if (tcph->fin) return TCP_FIN_SET;
 389         else if (tcph->ack) return TCP_ACK_SET;
 390         else return TCP_NONE_SET;
 391 }
 392
 393 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 394    in IP Filter' by Guido van Rooij.
 395
 396    http://www.nluug.nl/events/sane2000/papers.html
 397    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
 398
 399    The boundaries and the conditions are changed according to RFC793:
 400    the packet must intersect the window (i.e. segments may be
 401    after the right or before the left edge) and thus receivers may ACK
 402    segments after the right edge of the window.
 403
 404         td_maxend = max(sack + max(win,1)) seen in reply packets
 405         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 406         td_maxwin += seq + len - sender.td_maxend
 407                         if seq + len > sender.td_maxend
 408         td_end    = max(seq + len) seen in sent packets
 409
 410    I.   Upper bound for valid data:     seq <= sender.td_maxend
 411    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 412    III. Upper bound for valid ack:      sack <= receiver.td_end
 413    IV.  Lower bound for valid ack:      ack >= receiver.td_end - MAXACKWINDOW
 414
 415    where sack is the highest right edge of sack block found in the packet.
 416
 417    The upper bound limit for a valid ack is not ignored -
 418    we doesn't have to deal with fragments.
 419 */
 420
 421 static inline __u32 segment_seq_plus_len(__u32 seq,
 422                                          size_t len,
 423                                          struct iphdr *iph,
 424                                          struct tcphdr *tcph)
 425 {
 426         return (seq + len - (iph->ihl + tcph->doff)*4
 427                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 428 }
 429
 430 /* Fixme: what about big packets? */
 431 #define MAXACKWINCONST                  66000
 432 #define MAXACKWINDOW(sender)                                            \
 433         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 434                                               : MAXACKWINCONST)
 435
 436 /*
 437  * Simplified tcp_parse_options routine from tcp_input.c
 438  */
 439 static void tcp_options(const struct sk_buff *skb,
 440                         struct iphdr *iph,
 441                         struct tcphdr *tcph,
 442                         struct ip_ct_tcp_state *state)
 443 {
 444         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 445         unsigned char *ptr;
 446         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 447
 448         if (!length)
 449                 return;
 450
 451         ptr = skb_header_pointer(skb,
 452                                  (iph->ihl * 4) + sizeof(struct tcphdr),
 453                                  length, buff);
 454         BUG_ON(ptr == NULL);
 455
 456         state->td_scale =
 457         state->flags = 0;
 458
 459         while (length > 0) {
 460                 int opcode=*ptr++;
 461                 int opsize;
 462
 463                 switch (opcode) {
 464                 case TCPOPT_EOL:
 465                         return;
 466                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 467                         length--;
 468                         continue;
 469                 default:
 470                         opsize=*ptr++;
 471                         if (opsize < 2) /* "silly options" */
 472                                 return;
 473                         if (opsize > length)
 474                                 break;  /* don't parse partial options */
 475
 476                         if (opcode == TCPOPT_SACK_PERM
 477                             && opsize == TCPOLEN_SACK_PERM)
 478                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 479                         else if (opcode == TCPOPT_WINDOW
 480                                  && opsize == TCPOLEN_WINDOW) {
 481                                 state->td_scale = *(u_int8_t *)ptr;
 482
 483                                 if (state->td_scale > 14) {
 484                                         /* See RFC1323 */
 485                                         state->td_scale = 14;
 486                                 }
 487                                 state->flags |=
 488                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 489                         }
 490                         ptr += opsize - 2;
 491                         length -= opsize;
 492                 }
 493         }
 494 }
 495
 496 static void tcp_sack(const struct sk_buff *skb,
 497                      struct iphdr *iph,
 498                      struct tcphdr *tcph,
 499                      __u32 *sack)
 500 {
 501         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 502         unsigned char *ptr;
 503         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 504         __u32 tmp;
 505
 506         if (!length)
 507                 return;
 508
 509         ptr = skb_header_pointer(skb,
 510                                  (iph->ihl * 4) + sizeof(struct tcphdr),
 511                                  length, buff);
 512         BUG_ON(ptr == NULL);
 513
 514         /* Fast path for timestamp-only option */
 515         if (length == TCPOLEN_TSTAMP_ALIGNED*4
 516             && *(__u32 *)ptr ==
 517                 __constant_ntohl((TCPOPT_NOP << 24)
 518                                  | (TCPOPT_NOP << 16)
 519                                  | (TCPOPT_TIMESTAMP << 8)
 520                                  | TCPOLEN_TIMESTAMP))
 521                 return;
 522
 523         while (length > 0) {
 524                 int opcode=*ptr++;
 525                 int opsize, i;
 526
 527                 switch (opcode) {
 528                 case TCPOPT_EOL:
 529                         return;
 530                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 531                         length--;
 532                         continue;
 533                 default:
 534                         opsize=*ptr++;
 535                         if (opsize < 2) /* "silly options" */
 536                                 return;
 537                         if (opsize > length)
 538                                 break;  /* don't parse partial options */
 539
 540                         if (opcode == TCPOPT_SACK
 541                             && opsize >= (TCPOLEN_SACK_BASE
 542                                           + TCPOLEN_SACK_PERBLOCK)
 543                             && !((opsize - TCPOLEN_SACK_BASE)
 544                                  % TCPOLEN_SACK_PERBLOCK)) {
 545                                 for (i = 0;
 546                                      i < (opsize - TCPOLEN_SACK_BASE);
 547                                      i += TCPOLEN_SACK_PERBLOCK) {
 548                                         tmp = ntohl(*((u_int32_t *)(ptr+i)+1));
 549
 550                                         if (after(tmp, *sack))
 551                                                 *sack = tmp;
 552                                 }
 553                                 return;
 554                         }
 555                         ptr += opsize - 2;
 556                         length -= opsize;
 557                 }
 558         }
 559 }
 560
 561 static int tcp_in_window(struct ip_ct_tcp *state,
 562                          enum ip_conntrack_dir dir,
 563                          unsigned int index,
 564                          const struct sk_buff *skb,
 565                          struct iphdr *iph,
 566                          struct tcphdr *tcph)
 567 {
 568         struct ip_ct_tcp_state *sender = &state->seen[dir];
 569         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 570         __u32 seq, ack, sack, end, win, swin;
 571         int res;
 572
 573         /*
 574          * Get the required data from the packet.
 575          */
 576         seq = ntohl(tcph->seq);
 577         ack = sack = ntohl(tcph->ack_seq);
 578         win = ntohs(tcph->window);
 579         end = segment_seq_plus_len(seq, skb->len, iph, tcph);
 580
 581         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 582                 tcp_sack(skb, iph, tcph, &sack);
 583
 584         DEBUGP("tcp_in_window: START\n");
 585         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
 586                "seq=%u ack=%u sack=%u win=%u end=%u\n",
 587                 NIPQUAD(iph->saddr), ntohs(tcph->source),
 588                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
 589                 seq, ack, sack, win, end);
 590         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 591                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 592                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 593                 sender->td_scale,
 594                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 595                 receiver->td_scale);
 596
 597         if (sender->td_end == 0) {
 598                 /*
 599                  * Initialize sender data.
 600                  */
 601                 if (tcph->syn && tcph->ack) {
 602                         /*
 603                          * Outgoing SYN-ACK in reply to a SYN.
 604                          */
 605                         sender->td_end =
 606                         sender->td_maxend = end;
 607                         sender->td_maxwin = (win == 0 ? 1 : win);
 608
 609                         tcp_options(skb, iph, tcph, sender);
 610                         /*
 611                          * RFC 1323:
 612                          * Both sides must send the Window Scale option
 613                          * to enable window scaling in either direction.
 614                          */
 615                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 616                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 617                                 sender->td_scale =
 618                                 receiver->td_scale = 0;
 619                 } else {
 620                         /*
 621                          * We are in the middle of a connection,
 622                          * its history is lost for us.
 623                          * Let's try to use the data from the packet.
 624                          */
 625                         sender->td_end = end;
 626                         sender->td_maxwin = (win == 0 ? 1 : win);
 627                         sender->td_maxend = end + sender->td_maxwin;
 628                 }
 629         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 630                      && dir == IP_CT_DIR_ORIGINAL)
 631                     || (state->state == TCP_CONNTRACK_SYN_RECV
 632                         && dir == IP_CT_DIR_REPLY))
 633                     && after(end, sender->td_end)) {
 634                 /*
 635                  * RFC 793: "if a TCP is reinitialized ... then it need
 636                  * not wait at all; it must only be sure to use sequence
 637                  * numbers larger than those recently used."
 638                  */
 639                 sender->td_end =
 640                 sender->td_maxend = end;
 641                 sender->td_maxwin = (win == 0 ? 1 : win);
 642
 643                 tcp_options(skb, iph, tcph, sender);
 644         }
 645
 646         if (!(tcph->ack)) {
 647                 /*
 648                  * If there is no ACK, just pretend it was set and OK.
 649                  */
 650                 ack = sack = receiver->td_end;
 651         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 652                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 653                    && (ack == 0)) {
 654                 /*
 655                  * Broken TCP stacks, that set ACK in RST packets as well
 656                  * with zero ack value.
 657                  */
 658                 ack = sack = receiver->td_end;
 659         }
 660
 661         if (seq == end
 662             && (!tcph->rst
 663                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
 664                 /*
 665                  * Packets contains no data: we assume it is valid
 666                  * and check the ack value only.
 667                  * However RST segments are always validated by their
 668                  * SEQ number, except when seq == 0 (reset sent answering
 669                  * SYN.
 670                  */
 671                 seq = end = sender->td_end;
 672
 673         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
 674                "seq=%u ack=%u sack =%u win=%u end=%u\n",
 675                 NIPQUAD(iph->saddr), ntohs(tcph->source),
 676                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
 677                 seq, ack, sack, win, end);
 678         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 679                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 680                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 681                 sender->td_scale,
 682                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 683                 receiver->td_scale);
 684
 685         DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 686                 before(seq, sender->td_maxend + 1),
 687                 after(end, sender->td_end - receiver->td_maxwin - 1),
 688                 before(sack, receiver->td_end + 1),
 689                 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
 690
 691         if (sender->loose || receiver->loose ||
 692             (before(seq, sender->td_maxend + 1) &&
 693              after(end, sender->td_end - receiver->td_maxwin - 1) &&
 694              before(sack, receiver->td_end + 1) &&
 695              after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
 696                 /*
 697                  * Take into account window scaling (RFC 1323).
 698                  */
 699                 if (!tcph->syn)
 700                         win <<= sender->td_scale;
 701
 702                 /*
 703                  * Update sender data.
 704                  */
 705                 swin = win + (sack - ack);
 706                 if (sender->td_maxwin < swin)
 707                         sender->td_maxwin = swin;
 708                 if (after(end, sender->td_end))
 709                         sender->td_end = end;
 710                 /*
 711                  * Update receiver data.
 712                  */
 713                 if (after(end, sender->td_maxend))
 714                         receiver->td_maxwin += end - sender->td_maxend;
 715                 if (after(sack + win, receiver->td_maxend - 1)) {
 716                         receiver->td_maxend = sack + win;
 717                         if (win == 0)
 718                                 receiver->td_maxend++;
 719                 }
 720
 721                 /*
 722                  * Check retransmissions.
 723                  */
 724                 if (index == TCP_ACK_SET) {
 725                         if (state->last_dir == dir
 726                             && state->last_seq == seq
 727                             && state->last_ack == ack
 728                             && state->last_end == end)
 729                                 state->retrans++;
 730                         else {
 731                                 state->last_dir = dir;
 732                                 state->last_seq = seq;
 733                                 state->last_ack = ack;
 734                                 state->last_end = end;
 735                                 state->retrans = 0;
 736                         }
 737                 }
 738                 /*
 739                  * Close the window of disabled window tracking :-)
 740                  */
 741                 if (sender->loose)
 742                         sender->loose--;
 743
 744                 res = 1;
 745         } else {
 746                 if (LOG_INVALID(IPPROTO_TCP))
 747                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 748                         "ip_ct_tcp: %s ",
 749                         before(seq, sender->td_maxend + 1) ?
 750                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
 751                         before(sack, receiver->td_end + 1) ?
 752                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
 753                         : "ACK is under the lower bound (possible overly delayed ACK)"
 754                         : "ACK is over the upper bound (ACKed data not seen yet)"
 755                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 756                         : "SEQ is over the upper bound (over the window of the receiver)");
 757
 758                 res = ip_ct_tcp_be_liberal;
 759         }
 760
 761         DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
 762                "receiver end=%u maxend=%u maxwin=%u\n",
 763                 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 764                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 765
 766         return res;
 767 }
 768
 769 #ifdef CONFIG_IP_NF_NAT_NEEDED
 770 /* Update sender->td_end after NAT successfully mangled the packet */
 771 void ip_conntrack_tcp_update(struct sk_buff *skb,
 772                              struct ip_conntrack *conntrack,
 773                              enum ip_conntrack_dir dir)
 774 {
 775         struct iphdr *iph = skb->nh.iph;
 776         struct tcphdr *tcph = (void *)skb->nh.iph + skb->nh.iph->ihl*4;
 777         __u32 end;
 778 #ifdef DEBUGP_VARS
 779         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
 780         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
 781 #endif
 782
 783         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, iph, tcph);
 784
 785         write_lock_bh(&tcp_lock);
 786         /*
 787          * We have to worry for the ack in the reply packet only...
 788          */
 789         if (after(end, conntrack->proto.tcp.seen[dir].td_end))
 790                 conntrack->proto.tcp.seen[dir].td_end = end;
 791         conntrack->proto.tcp.last_end = end;
 792         write_unlock_bh(&tcp_lock);
 793         DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
 794                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 795                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 796                 sender->td_scale,
 797                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 798                 receiver->td_scale);
 799 }
 800
 801 #endif
 802
 803 #define TH_FIN  0x01
 804 #define TH_SYN  0x02
 805 #define TH_RST  0x04
 806 #define TH_PUSH 0x08
 807 #define TH_ACK  0x10
 808 #define TH_URG  0x20
 809 #define TH_ECE  0x40
 810 #define TH_CWR  0x80
 811
 812 /* table of valid flag combinations - ECE and CWR are always valid */
 813 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
 814 {
 815         [TH_SYN]                        = 1,
 816         [TH_SYN|TH_ACK]                 = 1,
 817         [TH_SYN|TH_PUSH]                = 1,
 818         [TH_SYN|TH_ACK|TH_PUSH]         = 1,
 819         [TH_RST]                        = 1,
 820         [TH_RST|TH_ACK]                 = 1,
 821         [TH_RST|TH_ACK|TH_PUSH]         = 1,
 822         [TH_FIN|TH_ACK]                 = 1,
 823         [TH_ACK]                        = 1,
 824         [TH_ACK|TH_PUSH]                = 1,
 825         [TH_ACK|TH_URG]                 = 1,
 826         [TH_ACK|TH_URG|TH_PUSH]         = 1,
 827         [TH_FIN|TH_ACK|TH_PUSH]         = 1,
 828         [TH_FIN|TH_ACK|TH_URG]          = 1,
 829         [TH_FIN|TH_ACK|TH_URG|TH_PUSH]  = 1,
 830 };
 831
 832 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 833 static int tcp_error(struct sk_buff *skb,
 834                      enum ip_conntrack_info *ctinfo,
 835                      unsigned int hooknum)
 836 {
 837         struct iphdr *iph = skb->nh.iph;
 838         struct tcphdr _tcph, *th;
 839         unsigned int tcplen = skb->len - iph->ihl * 4;
 840         u_int8_t tcpflags;
 841
 842         /* Smaller that minimal TCP header? */
 843         th = skb_header_pointer(skb, iph->ihl * 4,
 844                                 sizeof(_tcph), &_tcph);
 845         if (th == NULL) {
 846                 if (LOG_INVALID(IPPROTO_TCP))
 847                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 848                                 "ip_ct_tcp: short packet ");
 849                 return -NF_ACCEPT;
 850         }
 851
 852         /* Not whole TCP header or malformed packet */
 853         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 854                 if (LOG_INVALID(IPPROTO_TCP))
 855                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 856                                 "ip_ct_tcp: truncated/malformed packet ");
 857                 return -NF_ACCEPT;
 858         }
 859
 860         /* Checksum invalid? Ignore.
 861          * We skip checking packets on the outgoing path
 862          * because the semantic of CHECKSUM_HW is different there
 863          * and moreover root might send raw packets.
 864          */
 865         /* FIXME: Source route IP option packets --RR */
 866         if (hooknum == NF_IP_PRE_ROUTING
 867             && skb->ip_summed != CHECKSUM_UNNECESSARY
 868             && csum_tcpudp_magic(iph->saddr, iph->daddr, tcplen, IPPROTO_TCP,
 869                                  skb->ip_summed == CHECKSUM_HW ? skb->csum
 870                                  : skb_checksum(skb, iph->ihl*4, tcplen, 0))) {
 871                 if (LOG_INVALID(IPPROTO_TCP))
 872                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 873                                   "ip_ct_tcp: bad TCP checksum ");
 874                 return -NF_ACCEPT;
 875         }
 876
 877         /* Check TCP flags. */
 878         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
 879         if (!tcp_valid_flags[tcpflags]) {
 880                 if (LOG_INVALID(IPPROTO_TCP))
 881                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 882                                   "ip_ct_tcp: invalid TCP flag combination ");
 883                 return -NF_ACCEPT;
 884         }
 885
 886         return NF_ACCEPT;
 887 }
 888
 889 /* Returns verdict for packet, or -1 for invalid. */
 890 static int tcp_packet(struct ip_conntrack *conntrack,
 891                       const struct sk_buff *skb,
 892                       enum ip_conntrack_info ctinfo)
 893 {
 894         enum tcp_conntrack new_state, old_state;
 895         enum ip_conntrack_dir dir;
 896         struct iphdr *iph = skb->nh.iph;
 897         struct tcphdr *th, _tcph;
 898         unsigned long timeout;
 899         unsigned int index;
 900
 901         th = skb_header_pointer(skb, iph->ihl * 4,
 902                                 sizeof(_tcph), &_tcph);
 903         BUG_ON(th == NULL);
 904
 905         write_lock_bh(&tcp_lock);
 906         old_state = conntrack->proto.tcp.state;
 907         dir = CTINFO2DIR(ctinfo);
 908         index = get_conntrack_index(th);
 909         new_state = tcp_conntracks[dir][index][old_state];
 910
 911         switch (new_state) {
 912         case TCP_CONNTRACK_IGNORE:
 913                 /* Either SYN in ORIGINAL
 914                  * or SYN/ACK in REPLY. */
 915                 if (index == TCP_SYNACK_SET
 916                     && conntrack->proto.tcp.last_index == TCP_SYN_SET
 917                     && conntrack->proto.tcp.last_dir != dir
 918                     && ntohl(th->ack_seq) ==
 919                              conntrack->proto.tcp.last_end) {
 920                         /* This SYN/ACK acknowledges a SYN that we earlier
 921                          * ignored as invalid. This means that the client and
 922                          * the server are both in sync, while the firewall is
 923                          * not. We kill this session and block the SYN/ACK so
 924                          * that the client cannot but retransmit its SYN and
 925                          * thus initiate a clean new session.
 926                          */
 927                         write_unlock_bh(&tcp_lock);
 928                         if (LOG_INVALID(IPPROTO_TCP))
 929                                 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
 930                                               NULL, "ip_ct_tcp: "
 931                                               "killing out of sync session ");
 932                         if (del_timer(&conntrack->timeout))
 933                                 conntrack->timeout.function((unsigned long)
 934                                                             conntrack);
 935                         return -NF_DROP;
 936                 }
 937                 conntrack->proto.tcp.last_index = index;
 938                 conntrack->proto.tcp.last_dir = dir;
 939                 conntrack->proto.tcp.last_seq = ntohl(th->seq);
 940                 conntrack->proto.tcp.last_end =
 941                     segment_seq_plus_len(ntohl(th->seq), skb->len, iph, th);
 942
 943                 write_unlock_bh(&tcp_lock);
 944                 if (LOG_INVALID(IPPROTO_TCP))
 945                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 946                                   "ip_ct_tcp: invalid packet ignored ");
 947                 return NF_ACCEPT;
 948         case TCP_CONNTRACK_MAX:
 949                 /* Invalid packet */
 950                 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 951                        dir, get_conntrack_index(th),
 952                        old_state);
 953                 write_unlock_bh(&tcp_lock);
 954                 if (LOG_INVALID(IPPROTO_TCP))
 955                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 956                                   "ip_ct_tcp: invalid state ");
 957                 return -NF_ACCEPT;
 958         case TCP_CONNTRACK_SYN_SENT:
 959                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 960                         break;
 961                 if ((conntrack->proto.tcp.seen[dir].flags &
 962                          IP_CT_TCP_FLAG_CLOSE_INIT)
 963                     || after(ntohl(th->seq),
 964                              conntrack->proto.tcp.seen[dir].td_end)) {
 965                         /* Attempt to reopen a closed connection.
 966                         * Delete this connection and look up again. */
 967                         write_unlock_bh(&tcp_lock);
 968                         if (del_timer(&conntrack->timeout))
 969                                 conntrack->timeout.function((unsigned long)
 970                                                             conntrack);
 971                         return -NF_REPEAT;
 972                 } else {
 973                         write_unlock_bh(&tcp_lock);
 974                         if (LOG_INVALID(IPPROTO_TCP))
 975                                 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
 976                                               NULL, "ip_ct_tcp: invalid SYN");
 977                         return -NF_ACCEPT;
 978                 }
 979         case TCP_CONNTRACK_CLOSE:
 980                 if (index == TCP_RST_SET
 981                     && test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
 982                     && conntrack->proto.tcp.last_index == TCP_SYN_SET
 983                     && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
 984                         /* RST sent to invalid SYN we had let trough
 985                          * SYN was in window then, tear down connection.
 986                          * We skip window checking, because packet might ACK
 987                          * segments we ignored in the SYN. */
 988                         goto in_window;
 989                 }
 990                 /* Just fall trough */
 991         default:
 992                 /* Keep compilers happy. */
 993                 break;
 994         }
 995
 996         if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
 997                            skb, iph, th)) {
 998                 write_unlock_bh(&tcp_lock);
 999                 return -NF_ACCEPT;
1000         }
1001     in_window:
1002         /* From now on we have got in-window packets */
1003         conntrack->proto.tcp.last_index = index;
1004
1005         DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
1006                "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1007                 NIPQUAD(iph->saddr), ntohs(th->source),
1008                 NIPQUAD(iph->daddr), ntohs(th->dest),
1009                 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1010                 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1011                 old_state, new_state);
1012
1013         conntrack->proto.tcp.state = new_state;
1014         if (old_state != new_state
1015             && (new_state == TCP_CONNTRACK_FIN_WAIT
1016                 || new_state == TCP_CONNTRACK_CLOSE))
1017                 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1018         timeout = conntrack->proto.tcp.retrans >= ip_ct_tcp_max_retrans
1019                   && *tcp_timeouts[new_state] > ip_ct_tcp_timeout_max_retrans
1020                   ? ip_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
1021         write_unlock_bh(&tcp_lock);
1022
1023         ip_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
1024         if (new_state != old_state)
1025                 ip_conntrack_event_cache(IPCT_PROTOINFO, skb);
1026
1027         if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
1028                 /* If only reply is a RST, we can consider ourselves not to
1029                    have an established connection: this is a fairly common
1030                    problem case, so we can delete the conntrack
1031                    immediately.  --RR */
1032                 if (th->rst) {
1033                         if (del_timer(&conntrack->timeout))
1034                                 conntrack->timeout.function((unsigned long)
1035                                                             conntrack);
1036                         return NF_ACCEPT;
1037                 }
1038         } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
1039                    && (old_state == TCP_CONNTRACK_SYN_RECV
1040                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1041                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1042                 /* Set ASSURED if we see see valid ack in ESTABLISHED
1043                    after SYN_RECV or a valid answer for a picked up
1044                    connection. */
1045                 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1046                 ip_conntrack_event_cache(IPCT_STATUS, skb);
1047         }
1048         ip_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1049
1050         return NF_ACCEPT;
1051 }
1052
1053 /* Called when a new connection for this protocol found. */
1054 static int tcp_new(struct ip_conntrack *conntrack,
1055                    const struct sk_buff *skb)
1056 {
1057         enum tcp_conntrack new_state;
1058         struct iphdr *iph = skb->nh.iph;
1059         struct tcphdr *th, _tcph;
1060 #ifdef DEBUGP_VARS
1061         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1062         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1063 #endif
1064
1065         th = skb_header_pointer(skb, iph->ihl * 4,
1066                                 sizeof(_tcph), &_tcph);
1067         BUG_ON(th == NULL);
1068
1069         /* Don't need lock here: this conntrack not in circulation yet */
1070         new_state
1071                 = tcp_conntracks[0][get_conntrack_index(th)]
1072                 [TCP_CONNTRACK_NONE];
1073
1074         /* Invalid: delete conntrack */
1075         if (new_state >= TCP_CONNTRACK_MAX) {
1076                 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1077                 return 0;
1078         }
1079
1080         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1081                 /* SYN packet */
1082                 conntrack->proto.tcp.seen[0].td_end =
1083                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1084                                              iph, th);
1085                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1086                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1087                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1088                 conntrack->proto.tcp.seen[0].td_maxend =
1089                         conntrack->proto.tcp.seen[0].td_end;
1090
1091                 tcp_options(skb, iph, th, &conntrack->proto.tcp.seen[0]);
1092                 conntrack->proto.tcp.seen[1].flags = 0;
1093                 conntrack->proto.tcp.seen[0].loose =
1094                 conntrack->proto.tcp.seen[1].loose = 0;
1095         } else if (ip_ct_tcp_loose == 0) {
1096                 /* Don't try to pick up connections. */
1097                 return 0;
1098         } else {
1099                 /*
1100                  * We are in the middle of a connection,
1101                  * its history is lost for us.
1102                  * Let's try to use the data from the packet.
1103                  */
1104                 conntrack->proto.tcp.seen[0].td_end =
1105                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1106                                              iph, th);
1107                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1108                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1109                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1110                 conntrack->proto.tcp.seen[0].td_maxend =
1111                         conntrack->proto.tcp.seen[0].td_end +
1112                         conntrack->proto.tcp.seen[0].td_maxwin;
1113                 conntrack->proto.tcp.seen[0].td_scale = 0;
1114
1115                 /* We assume SACK. Should we assume window scaling too? */
1116                 conntrack->proto.tcp.seen[0].flags =
1117                 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1118                 conntrack->proto.tcp.seen[0].loose =
1119                 conntrack->proto.tcp.seen[1].loose = ip_ct_tcp_loose;
1120         }
1121
1122         conntrack->proto.tcp.seen[1].td_end = 0;
1123         conntrack->proto.tcp.seen[1].td_maxend = 0;
1124         conntrack->proto.tcp.seen[1].td_maxwin = 1;
1125         conntrack->proto.tcp.seen[1].td_scale = 0;
1126
1127         /* tcp_packet will set them */
1128         conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1129         conntrack->proto.tcp.last_index = TCP_NONE_SET;
1130
1131         DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1132                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1133                 sender->td_end, sender->td_maxend, sender->td_maxwin,
1134                 sender->td_scale,
1135                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1136                 receiver->td_scale);
1137         return 1;
1138 }
1139
1140 struct ip_conntrack_protocol ip_conntrack_protocol_tcp =
1141 {
1142         .proto                  = IPPROTO_TCP,
1143         .name                   = "tcp",
1144         .pkt_to_tuple           = tcp_pkt_to_tuple,
1145         .invert_tuple           = tcp_invert_tuple,
1146         .print_tuple            = tcp_print_tuple,
1147         .print_conntrack        = tcp_print_conntrack,
1148         .packet                 = tcp_packet,
1149         .new                    = tcp_new,
1150         .error                  = tcp_error,
1151 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
1152     defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
1153         .to_nfattr              = tcp_to_nfattr,
1154         .from_nfattr            = nfattr_to_tcp,
1155         .tuple_to_nfattr        = ip_ct_port_tuple_to_nfattr,
1156         .nfattr_to_tuple        = ip_ct_port_nfattr_to_tuple,
1157 #endif
1158 };