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/types.h>
  23 #include <linux/sched.h>
  24 #include <linux/timer.h>
  25 #include <linux/netfilter.h>
  26 #include <linux/module.h>
  27 #include <linux/in.h>
  28 #include <linux/ip.h>
  29 #include <linux/tcp.h>
  30 #include <linux/spinlock.h>
  31
  32 #include <net/tcp.h>
  33
  34 #include <linux/netfilter_ipv4.h>
  35 #include <linux/netfilter_ipv4/ip_conntrack.h>
  36 #include <linux/netfilter_ipv4/ip_conntrack_protocol.h>
  37
  38 #if 0
  39 #define DEBUGP printk
  40 #define DEBUGP_VARS
  41 #else
  42 #define DEBUGP(format, args...)
  43 #endif
  44
  45 /* Protects conntrack->proto.tcp */
  46 static DEFINE_RWLOCK(tcp_lock);
  47
  48 /* "Be conservative in what you do,
  49     be liberal in what you accept from others."
  50     If it's non-zero, we mark only out of window RST segments as INVALID. */
  51 int ip_ct_tcp_be_liberal __read_mostly = 0;
  52
  53 /* When connection is picked up from the middle, how many packets are required
  54    to pass in each direction when we assume we are in sync - if any side uses
  55    window scaling, we lost the game.
  56    If it is set to zero, we disable picking up already established
  57    connections. */
  58 int ip_ct_tcp_loose __read_mostly = 3;
  59
  60 /* Max number of the retransmitted packets without receiving an (acceptable)
  61    ACK from the destination. If this number is reached, a shorter timer
  62    will be started. */
  63 int ip_ct_tcp_max_retrans __read_mostly = 3;
  64
  65   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  66      closely.  They're more complex. --RR */
  67
  68 static const char *tcp_conntrack_names[] = {
  69         "NONE",
  70         "SYN_SENT",
  71         "SYN_RECV",
  72         "ESTABLISHED",
  73         "FIN_WAIT",
  74         "CLOSE_WAIT",
  75         "LAST_ACK",
  76         "TIME_WAIT",
  77         "CLOSE",
  78         "LISTEN"
  79 };
  80
  81 #define SECS * HZ
  82 #define MINS * 60 SECS
  83 #define HOURS * 60 MINS
  84 #define DAYS * 24 HOURS
  85
  86 unsigned int ip_ct_tcp_timeout_syn_sent __read_mostly =      2 MINS;
  87 unsigned int ip_ct_tcp_timeout_syn_recv __read_mostly =     60 SECS;
  88 unsigned int ip_ct_tcp_timeout_established __read_mostly =   5 DAYS;
  89 unsigned int ip_ct_tcp_timeout_fin_wait __read_mostly =      2 MINS;
  90 unsigned int ip_ct_tcp_timeout_close_wait __read_mostly =   60 SECS;
  91 unsigned int ip_ct_tcp_timeout_last_ack __read_mostly =     30 SECS;
  92 unsigned int ip_ct_tcp_timeout_time_wait __read_mostly =     2 MINS;
  93 unsigned int ip_ct_tcp_timeout_close __read_mostly =        10 SECS;
  94
  95 /* RFC1122 says the R2 limit should be at least 100 seconds.
  96    Linux uses 15 packets as limit, which corresponds
  97    to ~13-30min depending on RTO. */
  98 unsigned int ip_ct_tcp_timeout_max_retrans __read_mostly =   5 MINS;
  99
 100 static const unsigned int * tcp_timeouts[]
 101 = { NULL,                              /*      TCP_CONNTRACK_NONE */
 102     &ip_ct_tcp_timeout_syn_sent,       /*      TCP_CONNTRACK_SYN_SENT, */
 103     &ip_ct_tcp_timeout_syn_recv,       /*      TCP_CONNTRACK_SYN_RECV, */
 104     &ip_ct_tcp_timeout_established,    /*      TCP_CONNTRACK_ESTABLISHED,      */
 105     &ip_ct_tcp_timeout_fin_wait,       /*      TCP_CONNTRACK_FIN_WAIT, */
 106     &ip_ct_tcp_timeout_close_wait,     /*      TCP_CONNTRACK_CLOSE_WAIT,       */
 107     &ip_ct_tcp_timeout_last_ack,       /*      TCP_CONNTRACK_LAST_ACK, */
 108     &ip_ct_tcp_timeout_time_wait,      /*      TCP_CONNTRACK_TIME_WAIT,        */
 109     &ip_ct_tcp_timeout_close,          /*      TCP_CONNTRACK_CLOSE,    */
 110     NULL,                              /*      TCP_CONNTRACK_LISTEN */
 111  };
 112
 113 #define sNO TCP_CONNTRACK_NONE
 114 #define sSS TCP_CONNTRACK_SYN_SENT
 115 #define sSR TCP_CONNTRACK_SYN_RECV
 116 #define sES TCP_CONNTRACK_ESTABLISHED
 117 #define sFW TCP_CONNTRACK_FIN_WAIT
 118 #define sCW TCP_CONNTRACK_CLOSE_WAIT
 119 #define sLA TCP_CONNTRACK_LAST_ACK
 120 #define sTW TCP_CONNTRACK_TIME_WAIT
 121 #define sCL TCP_CONNTRACK_CLOSE
 122 #define sLI TCP_CONNTRACK_LISTEN
 123 #define sIV TCP_CONNTRACK_MAX
 124 #define sIG TCP_CONNTRACK_IGNORE
 125
 126 /* What TCP flags are set from RST/SYN/FIN/ACK. */
 127 enum tcp_bit_set {
 128         TCP_SYN_SET,
 129         TCP_SYNACK_SET,
 130         TCP_FIN_SET,
 131         TCP_ACK_SET,
 132         TCP_RST_SET,
 133         TCP_NONE_SET,
 134 };
 135
 136 /*
 137  * The TCP state transition table needs a few words...
 138  *
 139  * We are the man in the middle. All the packets go through us
 140  * but might get lost in transit to the destination.
 141  * It is assumed that the destinations can't receive segments
 142  * we haven't seen.
 143  *
 144  * The checked segment is in window, but our windows are *not*
 145  * equivalent with the ones of the sender/receiver. We always
 146  * try to guess the state of the current sender.
 147  *
 148  * The meaning of the states are:
 149  *
 150  * NONE:        initial state
 151  * SYN_SENT:    SYN-only packet seen
 152  * SYN_RECV:    SYN-ACK packet seen
 153  * ESTABLISHED: ACK packet seen
 154  * FIN_WAIT:    FIN packet seen
 155  * CLOSE_WAIT:  ACK seen (after FIN)
 156  * LAST_ACK:    FIN seen (after FIN)
 157  * TIME_WAIT:   last ACK seen
 158  * CLOSE:       closed connection
 159  *
 160  * LISTEN state is not used.
 161  *
 162  * Packets marked as IGNORED (sIG):
 163  *      if they may be either invalid or valid
 164  *      and the receiver may send back a connection
 165  *      closing RST or a SYN/ACK.
 166  *
 167  * Packets marked as INVALID (sIV):
 168  *      if they are invalid
 169  *      or we do not support the request (simultaneous open)
 170  */
 171 static const enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 172         {
 173 /* ORIGINAL */
 174 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 175 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
 176 /*
 177  *      sNO -> sSS      Initialize a new connection
 178  *      sSS -> sSS      Retransmitted SYN
 179  *      sSR -> sIG      Late retransmitted SYN?
 180  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 181  *                      are errors. Receiver will reply with RST
 182  *                      and close the connection.
 183  *                      Or we are not in sync and hold a dead connection.
 184  *      sFW -> sIG
 185  *      sCW -> sIG
 186  *      sLA -> sIG
 187  *      sTW -> sSS      Reopened connection (RFC 1122).
 188  *      sCL -> sSS
 189  */
 190 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 191 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 192 /*
 193  * A SYN/ACK from the client is always invalid:
 194  *      - either it tries to set up a simultaneous open, which is
 195  *        not supported;
 196  *      - or the firewall has just been inserted between the two hosts
 197  *        during the session set-up. The SYN will be retransmitted
 198  *        by the true client (or it'll time out).
 199  */
 200 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 201 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 202 /*
 203  *      sNO -> sIV      Too late and no reason to do anything...
 204  *      sSS -> sIV      Client migth not send FIN in this state:
 205  *                      we enforce waiting for a SYN/ACK reply first.
 206  *      sSR -> sFW      Close started.
 207  *      sES -> sFW
 208  *      sFW -> sLA      FIN seen in both directions, waiting for
 209  *                      the last ACK.
 210  *                      Migth be a retransmitted FIN as well...
 211  *      sCW -> sLA
 212  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 213  *      sTW -> sTW
 214  *      sCL -> sCL
 215  */
 216 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 217 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 218 /*
 219  *      sNO -> sES      Assumed.
 220  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 221  *      sSR -> sES      Established state is reached.
 222  *      sES -> sES      :-)
 223  *      sFW -> sCW      Normal close request answered by ACK.
 224  *      sCW -> sCW
 225  *      sLA -> sTW      Last ACK detected.
 226  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 227  *      sCL -> sCL
 228  */
 229 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 230 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 231 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 232         },
 233         {
 234 /* REPLY */
 235 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 236 /*syn*/    { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 237 /*
 238  *      sNO -> sIV      Never reached.
 239  *      sSS -> sIV      Simultaneous open, not supported
 240  *      sSR -> sIV      Simultaneous open, not supported.
 241  *      sES -> sIV      Server may not initiate a connection.
 242  *      sFW -> sIV
 243  *      sCW -> sIV
 244  *      sLA -> sIV
 245  *      sTW -> sIV      Reopened connection, but server may not do it.
 246  *      sCL -> sIV
 247  */
 248 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 249 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
 250 /*
 251  *      sSS -> sSR      Standard open.
 252  *      sSR -> sSR      Retransmitted SYN/ACK.
 253  *      sES -> sIG      Late retransmitted SYN/ACK?
 254  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 255  *      sCW -> sIG
 256  *      sLA -> sIG
 257  *      sTW -> sIG
 258  *      sCL -> sIG
 259  */
 260 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 261 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 262 /*
 263  *      sSS -> sIV      Server might not send FIN in this state.
 264  *      sSR -> sFW      Close started.
 265  *      sES -> sFW
 266  *      sFW -> sLA      FIN seen in both directions.
 267  *      sCW -> sLA
 268  *      sLA -> sLA      Retransmitted FIN.
 269  *      sTW -> sTW
 270  *      sCL -> sCL
 271  */
 272 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 273 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 274 /*
 275  *      sSS -> sIG      Might be a half-open connection.
 276  *      sSR -> sSR      Might answer late resent SYN.
 277  *      sES -> sES      :-)
 278  *      sFW -> sCW      Normal close request answered by ACK.
 279  *      sCW -> sCW
 280  *      sLA -> sTW      Last ACK detected.
 281  *      sTW -> sTW      Retransmitted last ACK.
 282  *      sCL -> sCL
 283  */
 284 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 285 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 286 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 287         }
 288 };
 289
 290 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
 291                             unsigned int dataoff,
 292                             struct ip_conntrack_tuple *tuple)
 293 {
 294         struct tcphdr _hdr, *hp;
 295
 296         /* Actually only need first 8 bytes. */
 297         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
 298         if (hp == NULL)
 299                 return 0;
 300
 301         tuple->src.u.tcp.port = hp->source;
 302         tuple->dst.u.tcp.port = hp->dest;
 303
 304         return 1;
 305 }
 306
 307 static int tcp_invert_tuple(struct ip_conntrack_tuple *tuple,
 308                             const struct ip_conntrack_tuple *orig)
 309 {
 310         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
 311         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
 312         return 1;
 313 }
 314
 315 /* Print out the per-protocol part of the tuple. */
 316 static int tcp_print_tuple(struct seq_file *s,
 317                            const struct ip_conntrack_tuple *tuple)
 318 {
 319         return seq_printf(s, "sport=%hu dport=%hu ",
 320                           ntohs(tuple->src.u.tcp.port),
 321                           ntohs(tuple->dst.u.tcp.port));
 322 }
 323
 324 /* Print out the private part of the conntrack. */
 325 static int tcp_print_conntrack(struct seq_file *s,
 326                                const struct ip_conntrack *conntrack)
 327 {
 328         enum tcp_conntrack state;
 329
 330         read_lock_bh(&tcp_lock);
 331         state = conntrack->proto.tcp.state;
 332         read_unlock_bh(&tcp_lock);
 333
 334         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
 335 }
 336
 337 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
 338     defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
 339 static int tcp_to_nfattr(struct sk_buff *skb, struct nfattr *nfa,
 340                          const struct ip_conntrack *ct)
 341 {
 342         struct nfattr *nest_parms;
 343
 344         read_lock_bh(&tcp_lock);
 345         nest_parms = NFA_NEST(skb, CTA_PROTOINFO_TCP);
 346         NFA_PUT(skb, CTA_PROTOINFO_TCP_STATE, sizeof(u_int8_t),
 347                 &ct->proto.tcp.state);
 348         read_unlock_bh(&tcp_lock);
 349
 350         NFA_NEST_END(skb, nest_parms);
 351
 352         return 0;
 353
 354 nfattr_failure:
 355         read_unlock_bh(&tcp_lock);
 356         return -1;
 357 }
 358
 359 static const size_t cta_min_tcp[CTA_PROTOINFO_TCP_MAX] = {
 360         [CTA_PROTOINFO_TCP_STATE-1]     = sizeof(u_int8_t),
 361 };
 362
 363 static int nfattr_to_tcp(struct nfattr *cda[], struct ip_conntrack *ct)
 364 {
 365         struct nfattr *attr = cda[CTA_PROTOINFO_TCP-1];
 366         struct nfattr *tb[CTA_PROTOINFO_TCP_MAX];
 367
 368         /* updates could not contain anything about the private
 369          * protocol info, in that case skip the parsing */
 370         if (!attr)
 371                 return 0;
 372
 373         nfattr_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, attr);
 374
 375         if (nfattr_bad_size(tb, CTA_PROTOINFO_TCP_MAX, cta_min_tcp))
 376                 return -EINVAL;
 377
 378         if (!tb[CTA_PROTOINFO_TCP_STATE-1])
 379                 return -EINVAL;
 380
 381         write_lock_bh(&tcp_lock);
 382         ct->proto.tcp.state =
 383                 *(u_int8_t *)NFA_DATA(tb[CTA_PROTOINFO_TCP_STATE-1]);
 384         write_unlock_bh(&tcp_lock);
 385
 386         return 0;
 387 }
 388 #endif
 389
 390 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 391 {
 392         if (tcph->rst) return TCP_RST_SET;
 393         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 394         else if (tcph->fin) return TCP_FIN_SET;
 395         else if (tcph->ack) return TCP_ACK_SET;
 396         else return TCP_NONE_SET;
 397 }
 398
 399 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 400    in IP Filter' by Guido van Rooij.
 401
 402    http://www.nluug.nl/events/sane2000/papers.html
 403    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
 404
 405    The boundaries and the conditions are changed according to RFC793:
 406    the packet must intersect the window (i.e. segments may be
 407    after the right or before the left edge) and thus receivers may ACK
 408    segments after the right edge of the window.
 409
 410         td_maxend = max(sack + max(win,1)) seen in reply packets
 411         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 412         td_maxwin += seq + len - sender.td_maxend
 413                         if seq + len > sender.td_maxend
 414         td_end    = max(seq + len) seen in sent packets
 415
 416    I.   Upper bound for valid data:     seq <= sender.td_maxend
 417    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 418    III. Upper bound for valid ack:      sack <= receiver.td_end
 419    IV.  Lower bound for valid ack:      ack >= receiver.td_end - MAXACKWINDOW
 420
 421    where sack is the highest right edge of sack block found in the packet.
 422
 423    The upper bound limit for a valid ack is not ignored -
 424    we doesn't have to deal with fragments.
 425 */
 426
 427 static inline __u32 segment_seq_plus_len(__u32 seq,
 428                                          size_t len,
 429                                          struct iphdr *iph,
 430                                          struct tcphdr *tcph)
 431 {
 432         return (seq + len - (iph->ihl + tcph->doff)*4
 433                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 434 }
 435
 436 /* Fixme: what about big packets? */
 437 #define MAXACKWINCONST                  66000
 438 #define MAXACKWINDOW(sender)                                            \
 439         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 440                                               : MAXACKWINCONST)
 441
 442 /*
 443  * Simplified tcp_parse_options routine from tcp_input.c
 444  */
 445 static void tcp_options(const struct sk_buff *skb,
 446                         struct iphdr *iph,
 447                         struct tcphdr *tcph,
 448                         struct ip_ct_tcp_state *state)
 449 {
 450         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 451         unsigned char *ptr;
 452         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 453
 454         if (!length)
 455                 return;
 456
 457         ptr = skb_header_pointer(skb,
 458                                  (iph->ihl * 4) + sizeof(struct tcphdr),
 459                                  length, buff);
 460         BUG_ON(ptr == NULL);
 461
 462         state->td_scale =
 463         state->flags = 0;
 464
 465         while (length > 0) {
 466                 int opcode=*ptr++;
 467                 int opsize;
 468
 469                 switch (opcode) {
 470                 case TCPOPT_EOL:
 471                         return;
 472                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 473                         length--;
 474                         continue;
 475                 default:
 476                         opsize=*ptr++;
 477                         if (opsize < 2) /* "silly options" */
 478                                 return;
 479                         if (opsize > length)
 480                                 break;  /* don't parse partial options */
 481
 482                         if (opcode == TCPOPT_SACK_PERM
 483                             && opsize == TCPOLEN_SACK_PERM)
 484                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 485                         else if (opcode == TCPOPT_WINDOW
 486                                  && opsize == TCPOLEN_WINDOW) {
 487                                 state->td_scale = *(u_int8_t *)ptr;
 488
 489                                 if (state->td_scale > 14) {
 490                                         /* See RFC1323 */
 491                                         state->td_scale = 14;
 492                                 }
 493                                 state->flags |=
 494                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 495                         }
 496                         ptr += opsize - 2;
 497                         length -= opsize;
 498                 }
 499         }
 500 }
 501
 502 static void tcp_sack(const struct sk_buff *skb,
 503                      struct iphdr *iph,
 504                      struct tcphdr *tcph,
 505                      __u32 *sack)
 506 {
 507         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 508         unsigned char *ptr;
 509         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 510         __u32 tmp;
 511
 512         if (!length)
 513                 return;
 514
 515         ptr = skb_header_pointer(skb,
 516                                  (iph->ihl * 4) + sizeof(struct tcphdr),
 517                                  length, buff);
 518         BUG_ON(ptr == NULL);
 519
 520         /* Fast path for timestamp-only option */
 521         if (length == TCPOLEN_TSTAMP_ALIGNED*4
 522             && *(__u32 *)ptr ==
 523                 __constant_ntohl((TCPOPT_NOP << 24)
 524                                  | (TCPOPT_NOP << 16)
 525                                  | (TCPOPT_TIMESTAMP << 8)
 526                                  | TCPOLEN_TIMESTAMP))
 527                 return;
 528
 529         while (length > 0) {
 530                 int opcode=*ptr++;
 531                 int opsize, i;
 532
 533                 switch (opcode) {
 534                 case TCPOPT_EOL:
 535                         return;
 536                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 537                         length--;
 538                         continue;
 539                 default:
 540                         opsize=*ptr++;
 541                         if (opsize < 2) /* "silly options" */
 542                                 return;
 543                         if (opsize > length)
 544                                 break;  /* don't parse partial options */
 545
 546                         if (opcode == TCPOPT_SACK
 547                             && opsize >= (TCPOLEN_SACK_BASE
 548                                           + TCPOLEN_SACK_PERBLOCK)
 549                             && !((opsize - TCPOLEN_SACK_BASE)
 550                                  % TCPOLEN_SACK_PERBLOCK)) {
 551                                 for (i = 0;
 552                                      i < (opsize - TCPOLEN_SACK_BASE);
 553                                      i += TCPOLEN_SACK_PERBLOCK) {
 554                                         tmp = ntohl(*((u_int32_t *)(ptr+i)+1));
 555
 556                                         if (after(tmp, *sack))
 557                                                 *sack = tmp;
 558                                 }
 559                                 return;
 560                         }
 561                         ptr += opsize - 2;
 562                         length -= opsize;
 563                 }
 564         }
 565 }
 566
 567 static int tcp_in_window(struct ip_ct_tcp *state,
 568                          enum ip_conntrack_dir dir,
 569                          unsigned int index,
 570                          const struct sk_buff *skb,
 571                          struct iphdr *iph,
 572                          struct tcphdr *tcph)
 573 {
 574         struct ip_ct_tcp_state *sender = &state->seen[dir];
 575         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 576         __u32 seq, ack, sack, end, win, swin;
 577         int res;
 578
 579         /*
 580          * Get the required data from the packet.
 581          */
 582         seq = ntohl(tcph->seq);
 583         ack = sack = ntohl(tcph->ack_seq);
 584         win = ntohs(tcph->window);
 585         end = segment_seq_plus_len(seq, skb->len, iph, tcph);
 586
 587         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 588                 tcp_sack(skb, iph, tcph, &sack);
 589
 590         DEBUGP("tcp_in_window: START\n");
 591         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
 592                "seq=%u ack=%u sack=%u win=%u end=%u\n",
 593                 NIPQUAD(iph->saddr), ntohs(tcph->source),
 594                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
 595                 seq, ack, sack, win, end);
 596         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 597                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 598                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 599                 sender->td_scale,
 600                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 601                 receiver->td_scale);
 602
 603         if (sender->td_end == 0) {
 604                 /*
 605                  * Initialize sender data.
 606                  */
 607                 if (tcph->syn && tcph->ack) {
 608                         /*
 609                          * Outgoing SYN-ACK in reply to a SYN.
 610                          */
 611                         sender->td_end =
 612                         sender->td_maxend = end;
 613                         sender->td_maxwin = (win == 0 ? 1 : win);
 614
 615                         tcp_options(skb, iph, tcph, sender);
 616                         /*
 617                          * RFC 1323:
 618                          * Both sides must send the Window Scale option
 619                          * to enable window scaling in either direction.
 620                          */
 621                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 622                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 623                                 sender->td_scale =
 624                                 receiver->td_scale = 0;
 625                 } else {
 626                         /*
 627                          * We are in the middle of a connection,
 628                          * its history is lost for us.
 629                          * Let's try to use the data from the packet.
 630                          */
 631                         sender->td_end = end;
 632                         sender->td_maxwin = (win == 0 ? 1 : win);
 633                         sender->td_maxend = end + sender->td_maxwin;
 634                 }
 635         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 636                      && dir == IP_CT_DIR_ORIGINAL)
 637                     || (state->state == TCP_CONNTRACK_SYN_RECV
 638                         && dir == IP_CT_DIR_REPLY))
 639                     && after(end, sender->td_end)) {
 640                 /*
 641                  * RFC 793: "if a TCP is reinitialized ... then it need
 642                  * not wait at all; it must only be sure to use sequence
 643                  * numbers larger than those recently used."
 644                  */
 645                 sender->td_end =
 646                 sender->td_maxend = end;
 647                 sender->td_maxwin = (win == 0 ? 1 : win);
 648
 649                 tcp_options(skb, iph, tcph, sender);
 650         }
 651
 652         if (!(tcph->ack)) {
 653                 /*
 654                  * If there is no ACK, just pretend it was set and OK.
 655                  */
 656                 ack = sack = receiver->td_end;
 657         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 658                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 659                    && (ack == 0)) {
 660                 /*
 661                  * Broken TCP stacks, that set ACK in RST packets as well
 662                  * with zero ack value.
 663                  */
 664                 ack = sack = receiver->td_end;
 665         }
 666
 667         if (seq == end
 668             && (!tcph->rst
 669                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
 670                 /*
 671                  * Packets contains no data: we assume it is valid
 672                  * and check the ack value only.
 673                  * However RST segments are always validated by their
 674                  * SEQ number, except when seq == 0 (reset sent answering
 675                  * SYN.
 676                  */
 677                 seq = end = sender->td_end;
 678
 679         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
 680                "seq=%u ack=%u sack =%u win=%u end=%u\n",
 681                 NIPQUAD(iph->saddr), ntohs(tcph->source),
 682                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
 683                 seq, ack, sack, win, end);
 684         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 685                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 686                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 687                 sender->td_scale,
 688                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 689                 receiver->td_scale);
 690
 691         DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 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         if (sender->loose || receiver->loose ||
 698             (before(seq, sender->td_maxend + 1) &&
 699              after(end, sender->td_end - receiver->td_maxwin - 1) &&
 700              before(sack, receiver->td_end + 1) &&
 701              after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
 702                 /*
 703                  * Take into account window scaling (RFC 1323).
 704                  */
 705                 if (!tcph->syn)
 706                         win <<= sender->td_scale;
 707
 708                 /*
 709                  * Update sender data.
 710                  */
 711                 swin = win + (sack - ack);
 712                 if (sender->td_maxwin < swin)
 713                         sender->td_maxwin = swin;
 714                 if (after(end, sender->td_end))
 715                         sender->td_end = end;
 716                 /*
 717                  * Update receiver data.
 718                  */
 719                 if (after(end, sender->td_maxend))
 720                         receiver->td_maxwin += end - sender->td_maxend;
 721                 if (after(sack + win, receiver->td_maxend - 1)) {
 722                         receiver->td_maxend = sack + win;
 723                         if (win == 0)
 724                                 receiver->td_maxend++;
 725                 }
 726
 727                 /*
 728                  * Check retransmissions.
 729                  */
 730                 if (index == TCP_ACK_SET) {
 731                         if (state->last_dir == dir
 732                             && state->last_seq == seq
 733                             && state->last_ack == ack
 734                             && state->last_end == end
 735                             && state->last_win == win)
 736                                 state->retrans++;
 737                         else {
 738                                 state->last_dir = dir;
 739                                 state->last_seq = seq;
 740                                 state->last_ack = ack;
 741                                 state->last_end = end;
 742                                 state->last_win = win;
 743                                 state->retrans = 0;
 744                         }
 745                 }
 746                 /*
 747                  * Close the window of disabled window tracking :-)
 748                  */
 749                 if (sender->loose)
 750                         sender->loose--;
 751
 752                 res = 1;
 753         } else {
 754                 if (LOG_INVALID(IPPROTO_TCP))
 755                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 756                         "ip_ct_tcp: %s ",
 757                         before(seq, sender->td_maxend + 1) ?
 758                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
 759                         before(sack, receiver->td_end + 1) ?
 760                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
 761                         : "ACK is under the lower bound (possible overly delayed ACK)"
 762                         : "ACK is over the upper bound (ACKed data not seen yet)"
 763                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 764                         : "SEQ is over the upper bound (over the window of the receiver)");
 765
 766                 res = ip_ct_tcp_be_liberal;
 767         }
 768
 769         DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
 770                "receiver end=%u maxend=%u maxwin=%u\n",
 771                 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 772                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 773
 774         return res;
 775 }
 776
 777 #ifdef CONFIG_IP_NF_NAT_NEEDED
 778 /* Update sender->td_end after NAT successfully mangled the packet */
 779 void ip_conntrack_tcp_update(struct sk_buff *skb,
 780                              struct ip_conntrack *conntrack,
 781                              enum ip_conntrack_dir dir)
 782 {
 783         struct iphdr *iph = skb->nh.iph;
 784         struct tcphdr *tcph = (void *)skb->nh.iph + skb->nh.iph->ihl*4;
 785         __u32 end;
 786 #ifdef DEBUGP_VARS
 787         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
 788         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
 789 #endif
 790
 791         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, iph, tcph);
 792
 793         write_lock_bh(&tcp_lock);
 794         /*
 795          * We have to worry for the ack in the reply packet only...
 796          */
 797         if (after(end, conntrack->proto.tcp.seen[dir].td_end))
 798                 conntrack->proto.tcp.seen[dir].td_end = end;
 799         conntrack->proto.tcp.last_end = end;
 800         write_unlock_bh(&tcp_lock);
 801         DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
 802                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 803                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 804                 sender->td_scale,
 805                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 806                 receiver->td_scale);
 807 }
 808
 809 #endif
 810
 811 #define TH_FIN  0x01
 812 #define TH_SYN  0x02
 813 #define TH_RST  0x04
 814 #define TH_PUSH 0x08
 815 #define TH_ACK  0x10
 816 #define TH_URG  0x20
 817 #define TH_ECE  0x40
 818 #define TH_CWR  0x80
 819
 820 /* table of valid flag combinations - ECE and CWR are always valid */
 821 static const u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
 822 {
 823         [TH_SYN]                        = 1,
 824         [TH_SYN|TH_ACK]                 = 1,
 825         [TH_SYN|TH_PUSH]                = 1,
 826         [TH_SYN|TH_ACK|TH_PUSH]         = 1,
 827         [TH_RST]                        = 1,
 828         [TH_RST|TH_ACK]                 = 1,
 829         [TH_RST|TH_ACK|TH_PUSH]         = 1,
 830         [TH_FIN|TH_ACK]                 = 1,
 831         [TH_ACK]                        = 1,
 832         [TH_ACK|TH_PUSH]                = 1,
 833         [TH_ACK|TH_URG]                 = 1,
 834         [TH_ACK|TH_URG|TH_PUSH]         = 1,
 835         [TH_FIN|TH_ACK|TH_PUSH]         = 1,
 836         [TH_FIN|TH_ACK|TH_URG]          = 1,
 837         [TH_FIN|TH_ACK|TH_URG|TH_PUSH]  = 1,
 838 };
 839
 840 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 841 static int tcp_error(struct sk_buff *skb,
 842                      enum ip_conntrack_info *ctinfo,
 843                      unsigned int hooknum)
 844 {
 845         struct iphdr *iph = skb->nh.iph;
 846         struct tcphdr _tcph, *th;
 847         unsigned int tcplen = skb->len - iph->ihl * 4;
 848         u_int8_t tcpflags;
 849
 850         /* Smaller that minimal TCP header? */
 851         th = skb_header_pointer(skb, iph->ihl * 4,
 852                                 sizeof(_tcph), &_tcph);
 853         if (th == NULL) {
 854                 if (LOG_INVALID(IPPROTO_TCP))
 855                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 856                                 "ip_ct_tcp: short packet ");
 857                 return -NF_ACCEPT;
 858         }
 859
 860         /* Not whole TCP header or malformed packet */
 861         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 862                 if (LOG_INVALID(IPPROTO_TCP))
 863                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 864                                 "ip_ct_tcp: truncated/malformed packet ");
 865                 return -NF_ACCEPT;
 866         }
 867
 868         /* Checksum invalid? Ignore.
 869          * We skip checking packets on the outgoing path
 870          * because it is assumed to be correct.
 871          */
 872         /* FIXME: Source route IP option packets --RR */
 873         if (ip_conntrack_checksum && hooknum == NF_IP_PRE_ROUTING &&
 874             nf_ip_checksum(skb, hooknum, iph->ihl * 4, IPPROTO_TCP)) {
 875                 if (LOG_INVALID(IPPROTO_TCP))
 876                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 877                                   "ip_ct_tcp: bad TCP checksum ");
 878                 return -NF_ACCEPT;
 879         }
 880
 881         /* Check TCP flags. */
 882         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
 883         if (!tcp_valid_flags[tcpflags]) {
 884                 if (LOG_INVALID(IPPROTO_TCP))
 885                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 886                                   "ip_ct_tcp: invalid TCP flag combination ");
 887                 return -NF_ACCEPT;
 888         }
 889
 890         return NF_ACCEPT;
 891 }
 892
 893 /* Returns verdict for packet, or -1 for invalid. */
 894 static int tcp_packet(struct ip_conntrack *conntrack,
 895                       const struct sk_buff *skb,
 896                       enum ip_conntrack_info ctinfo)
 897 {
 898         enum tcp_conntrack new_state, old_state;
 899         enum ip_conntrack_dir dir;
 900         struct iphdr *iph = skb->nh.iph;
 901         struct tcphdr *th, _tcph;
 902         unsigned long timeout;
 903         unsigned int index;
 904
 905         th = skb_header_pointer(skb, iph->ihl * 4,
 906                                 sizeof(_tcph), &_tcph);
 907         BUG_ON(th == NULL);
 908
 909         write_lock_bh(&tcp_lock);
 910         old_state = conntrack->proto.tcp.state;
 911         dir = CTINFO2DIR(ctinfo);
 912         index = get_conntrack_index(th);
 913         new_state = tcp_conntracks[dir][index][old_state];
 914
 915         switch (new_state) {
 916         case TCP_CONNTRACK_IGNORE:
 917                 /* Ignored packets:
 918                  *
 919                  * a) SYN in ORIGINAL
 920                  * b) SYN/ACK in REPLY
 921                  * c) ACK in reply direction after initial SYN in original.
 922                  */
 923                 if (index == TCP_SYNACK_SET
 924                     && conntrack->proto.tcp.last_index == TCP_SYN_SET
 925                     && conntrack->proto.tcp.last_dir != dir
 926                     && ntohl(th->ack_seq) ==
 927                              conntrack->proto.tcp.last_end) {
 928                         /* This SYN/ACK acknowledges a SYN that we earlier
 929                          * ignored as invalid. This means that the client and
 930                          * the server are both in sync, while the firewall is
 931                          * not. We kill this session and block the SYN/ACK so
 932                          * that the client cannot but retransmit its SYN and
 933                          * thus initiate a clean new session.
 934                          */
 935                         write_unlock_bh(&tcp_lock);
 936                         if (LOG_INVALID(IPPROTO_TCP))
 937                                 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
 938                                               NULL, "ip_ct_tcp: "
 939                                               "killing out of sync session ");
 940                         if (del_timer(&conntrack->timeout))
 941                                 conntrack->timeout.function((unsigned long)
 942                                                             conntrack);
 943                         return -NF_DROP;
 944                 }
 945                 conntrack->proto.tcp.last_index = index;
 946                 conntrack->proto.tcp.last_dir = dir;
 947                 conntrack->proto.tcp.last_seq = ntohl(th->seq);
 948                 conntrack->proto.tcp.last_end =
 949                     segment_seq_plus_len(ntohl(th->seq), skb->len, iph, th);
 950
 951                 write_unlock_bh(&tcp_lock);
 952                 if (LOG_INVALID(IPPROTO_TCP))
 953                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 954                                   "ip_ct_tcp: invalid packet ignored ");
 955                 return NF_ACCEPT;
 956         case TCP_CONNTRACK_MAX:
 957                 /* Invalid packet */
 958                 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 959                        dir, get_conntrack_index(th),
 960                        old_state);
 961                 write_unlock_bh(&tcp_lock);
 962                 if (LOG_INVALID(IPPROTO_TCP))
 963                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
 964                                   "ip_ct_tcp: invalid state ");
 965                 return -NF_ACCEPT;
 966         case TCP_CONNTRACK_SYN_SENT:
 967                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 968                         break;
 969                 if ((conntrack->proto.tcp.seen[dir].flags &
 970                          IP_CT_TCP_FLAG_CLOSE_INIT)
 971                     || after(ntohl(th->seq),
 972                              conntrack->proto.tcp.seen[dir].td_end)) {
 973                         /* Attempt to reopen a closed connection.
 974                         * Delete this connection and look up again. */
 975                         write_unlock_bh(&tcp_lock);
 976                         if (del_timer(&conntrack->timeout))
 977                                 conntrack->timeout.function((unsigned long)
 978                                                             conntrack);
 979                         return -NF_REPEAT;
 980                 } else {
 981                         write_unlock_bh(&tcp_lock);
 982                         if (LOG_INVALID(IPPROTO_TCP))
 983                                 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
 984                                               NULL, "ip_ct_tcp: invalid SYN");
 985                         return -NF_ACCEPT;
 986                 }
 987         case TCP_CONNTRACK_CLOSE:
 988                 if (index == TCP_RST_SET
 989                     && ((test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
 990                          && conntrack->proto.tcp.last_index == TCP_SYN_SET)
 991                         || (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
 992                             && conntrack->proto.tcp.last_index == TCP_ACK_SET))
 993                     && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
 994                         /* RST sent to invalid SYN or ACK we had let through
 995                          * at a) and c) above:
 996                          *
 997                          * a) SYN was in window then
 998                          * c) we hold a half-open connection.
 999                          *
1000                          * Delete our connection entry.
1001                          * We skip window checking, because packet might ACK
1002                          * segments we ignored. */
1003                         goto in_window;
1004                 }
1005                 /* Just fall through */
1006         default:
1007                 /* Keep compilers happy. */
1008                 break;
1009         }
1010
1011         if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
1012                            skb, iph, th)) {
1013                 write_unlock_bh(&tcp_lock);
1014                 return -NF_ACCEPT;
1015         }
1016     in_window:
1017         /* From now on we have got in-window packets */
1018         conntrack->proto.tcp.last_index = index;
1019
1020         DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
1021                "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1022                 NIPQUAD(iph->saddr), ntohs(th->source),
1023                 NIPQUAD(iph->daddr), ntohs(th->dest),
1024                 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1025                 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1026                 old_state, new_state);
1027
1028         conntrack->proto.tcp.state = new_state;
1029         if (old_state != new_state
1030             && (new_state == TCP_CONNTRACK_FIN_WAIT
1031                 || new_state == TCP_CONNTRACK_CLOSE))
1032                 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1033         timeout = conntrack->proto.tcp.retrans >= ip_ct_tcp_max_retrans
1034                   && *tcp_timeouts[new_state] > ip_ct_tcp_timeout_max_retrans
1035                   ? ip_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
1036         write_unlock_bh(&tcp_lock);
1037
1038         ip_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
1039         if (new_state != old_state)
1040                 ip_conntrack_event_cache(IPCT_PROTOINFO, skb);
1041
1042         if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
1043                 /* If only reply is a RST, we can consider ourselves not to
1044                    have an established connection: this is a fairly common
1045                    problem case, so we can delete the conntrack
1046                    immediately.  --RR */
1047                 if (th->rst) {
1048                         if (del_timer(&conntrack->timeout))
1049                                 conntrack->timeout.function((unsigned long)
1050                                                             conntrack);
1051                         return NF_ACCEPT;
1052                 }
1053         } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
1054                    && (old_state == TCP_CONNTRACK_SYN_RECV
1055                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1056                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1057                 /* Set ASSURED if we see see valid ack in ESTABLISHED
1058                    after SYN_RECV or a valid answer for a picked up
1059                    connection. */
1060                 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1061                 ip_conntrack_event_cache(IPCT_STATUS, skb);
1062         }
1063         ip_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1064
1065         return NF_ACCEPT;
1066 }
1067
1068 /* Called when a new connection for this protocol found. */
1069 static int tcp_new(struct ip_conntrack *conntrack,
1070                    const struct sk_buff *skb)
1071 {
1072         enum tcp_conntrack new_state;
1073         struct iphdr *iph = skb->nh.iph;
1074         struct tcphdr *th, _tcph;
1075 #ifdef DEBUGP_VARS
1076         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1077         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1078 #endif
1079
1080         th = skb_header_pointer(skb, iph->ihl * 4,
1081                                 sizeof(_tcph), &_tcph);
1082         BUG_ON(th == NULL);
1083
1084         /* Don't need lock here: this conntrack not in circulation yet */
1085         new_state
1086                 = tcp_conntracks[0][get_conntrack_index(th)]
1087                 [TCP_CONNTRACK_NONE];
1088
1089         /* Invalid: delete conntrack */
1090         if (new_state >= TCP_CONNTRACK_MAX) {
1091                 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1092                 return 0;
1093         }
1094
1095         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1096                 /* SYN packet */
1097                 conntrack->proto.tcp.seen[0].td_end =
1098                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1099                                              iph, th);
1100                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1101                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1102                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1103                 conntrack->proto.tcp.seen[0].td_maxend =
1104                         conntrack->proto.tcp.seen[0].td_end;
1105
1106                 tcp_options(skb, iph, th, &conntrack->proto.tcp.seen[0]);
1107                 conntrack->proto.tcp.seen[1].flags = 0;
1108                 conntrack->proto.tcp.seen[0].loose =
1109                 conntrack->proto.tcp.seen[1].loose = 0;
1110         } else if (ip_ct_tcp_loose == 0) {
1111                 /* Don't try to pick up connections. */
1112                 return 0;
1113         } else {
1114                 /*
1115                  * We are in the middle of a connection,
1116                  * its history is lost for us.
1117                  * Let's try to use the data from the packet.
1118                  */
1119                 conntrack->proto.tcp.seen[0].td_end =
1120                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1121                                              iph, th);
1122                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1123                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1124                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1125                 conntrack->proto.tcp.seen[0].td_maxend =
1126                         conntrack->proto.tcp.seen[0].td_end +
1127                         conntrack->proto.tcp.seen[0].td_maxwin;
1128                 conntrack->proto.tcp.seen[0].td_scale = 0;
1129
1130                 /* We assume SACK. Should we assume window scaling too? */
1131                 conntrack->proto.tcp.seen[0].flags =
1132                 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1133                 conntrack->proto.tcp.seen[0].loose =
1134                 conntrack->proto.tcp.seen[1].loose = ip_ct_tcp_loose;
1135         }
1136
1137         conntrack->proto.tcp.seen[1].td_end = 0;
1138         conntrack->proto.tcp.seen[1].td_maxend = 0;
1139         conntrack->proto.tcp.seen[1].td_maxwin = 1;
1140         conntrack->proto.tcp.seen[1].td_scale = 0;
1141
1142         /* tcp_packet will set them */
1143         conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1144         conntrack->proto.tcp.last_index = TCP_NONE_SET;
1145
1146         DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1147                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1148                 sender->td_end, sender->td_maxend, sender->td_maxwin,
1149                 sender->td_scale,
1150                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1151                 receiver->td_scale);
1152         return 1;
1153 }
1154
1155 struct ip_conntrack_protocol ip_conntrack_protocol_tcp =
1156 {
1157         .proto                  = IPPROTO_TCP,
1158         .name                   = "tcp",
1159         .pkt_to_tuple           = tcp_pkt_to_tuple,
1160         .invert_tuple           = tcp_invert_tuple,
1161         .print_tuple            = tcp_print_tuple,
1162         .print_conntrack        = tcp_print_conntrack,
1163         .packet                 = tcp_packet,
1164         .new                    = tcp_new,
1165         .error                  = tcp_error,
1166 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
1167     defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
1168         .to_nfattr              = tcp_to_nfattr,
1169         .from_nfattr            = nfattr_to_tcp,
1170         .tuple_to_nfattr        = ip_ct_port_tuple_to_nfattr,
1171         .nfattr_to_tuple        = ip_ct_port_nfattr_to_tuple,
1172 #endif
1173 };