SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/netfilter/nf_conntrack_proto_tcp.c

   1 /* (C) 1999-2001 Paul `Rusty' Russell
   2  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>:
   9  *      - Real stateful connection tracking
  10  *      - Modified state transitions table
  11  *      - Window scaling support added
  12  *      - SACK support added
  13  *
  14  * Willy Tarreau:
  15  *      - State table bugfixes
  16  *      - More robust state changes
  17  *      - Tuning timer parameters
  18  *
  19  * 27 Oct 2004: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
  20  *      - genelized Layer 3 protocol part.
  21  *
  22  * Derived from net/ipv4/netfilter/ip_conntrack_proto_tcp.c
  23  *
  24  * version 2.2
  25  */
  26
  27 #include <linux/types.h>
  28 #include <linux/sched.h>
  29 #include <linux/timer.h>
  30 #include <linux/netfilter.h>
  31 #include <linux/module.h>
  32 #include <linux/in.h>
  33 #include <linux/tcp.h>
  34 #include <linux/spinlock.h>
  35 #include <linux/skbuff.h>
  36 #include <linux/ipv6.h>
  37 #include <net/ip6_checksum.h>
  38
  39 #include <net/tcp.h>
  40
  41 #include <linux/netfilter.h>
  42 #include <linux/netfilter_ipv4.h>
  43 #include <linux/netfilter_ipv6.h>
  44 #include <net/netfilter/nf_conntrack.h>
  45 #include <net/netfilter/nf_conntrack_l4proto.h>
  46 #include <net/netfilter/nf_conntrack_ecache.h>
  47
  48 #if 0
  49 #define DEBUGP printk
  50 #define DEBUGP_VARS
  51 #else
  52 #define DEBUGP(format, args...)
  53 #endif
  54
  55 /* Protects conntrack->proto.tcp */
  56 static DEFINE_RWLOCK(tcp_lock);
  57
  58 /* "Be conservative in what you do,
  59     be liberal in what you accept from others."
  60     If it's non-zero, we mark only out of window RST segments as INVALID. */
  61 int nf_ct_tcp_be_liberal __read_mostly = 0;
  62
  63 /* If it is set to zero, we disable picking up already established
  64    connections. */
  65 int nf_ct_tcp_loose __read_mostly = 1;
  66
  67 /* Max number of the retransmitted packets without receiving an (acceptable)
  68    ACK from the destination. If this number is reached, a shorter timer
  69    will be started. */
  70 int nf_ct_tcp_max_retrans __read_mostly = 3;
  71
  72   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  73      closely.  They're more complex. --RR */
  74
  75 static const char *tcp_conntrack_names[] = {
  76         "NONE",
  77         "SYN_SENT",
  78         "SYN_RECV",
  79         "ESTABLISHED",
  80         "FIN_WAIT",
  81         "CLOSE_WAIT",
  82         "LAST_ACK",
  83         "TIME_WAIT",
  84         "CLOSE",
  85         "LISTEN"
  86 };
  87
  88 #define SECS * HZ
  89 #define MINS * 60 SECS
  90 #define HOURS * 60 MINS
  91 #define DAYS * 24 HOURS
  92
  93 static unsigned int nf_ct_tcp_timeout_syn_sent __read_mostly =      2 MINS;
  94 static unsigned int nf_ct_tcp_timeout_syn_recv __read_mostly =     60 SECS;
  95 static unsigned int nf_ct_tcp_timeout_established __read_mostly =   5 DAYS;
  96 static unsigned int nf_ct_tcp_timeout_fin_wait __read_mostly =      2 MINS;
  97 static unsigned int nf_ct_tcp_timeout_close_wait __read_mostly =   60 SECS;
  98 static unsigned int nf_ct_tcp_timeout_last_ack __read_mostly =     30 SECS;
  99 static unsigned int nf_ct_tcp_timeout_time_wait __read_mostly =     2 MINS;
 100 static unsigned int nf_ct_tcp_timeout_close __read_mostly =        10 SECS;
 101
 102 /* RFC1122 says the R2 limit should be at least 100 seconds.
 103    Linux uses 15 packets as limit, which corresponds
 104    to ~13-30min depending on RTO. */
 105 static unsigned int nf_ct_tcp_timeout_max_retrans __read_mostly =   5 MINS;
 106
 107 static unsigned int * tcp_timeouts[] = {
 108     NULL,                              /* TCP_CONNTRACK_NONE */
 109     &nf_ct_tcp_timeout_syn_sent,       /* TCP_CONNTRACK_SYN_SENT, */
 110     &nf_ct_tcp_timeout_syn_recv,       /* TCP_CONNTRACK_SYN_RECV, */
 111     &nf_ct_tcp_timeout_established,    /* TCP_CONNTRACK_ESTABLISHED, */
 112     &nf_ct_tcp_timeout_fin_wait,       /* TCP_CONNTRACK_FIN_WAIT, */
 113     &nf_ct_tcp_timeout_close_wait,     /* TCP_CONNTRACK_CLOSE_WAIT, */
 114     &nf_ct_tcp_timeout_last_ack,       /* TCP_CONNTRACK_LAST_ACK, */
 115     &nf_ct_tcp_timeout_time_wait,      /* TCP_CONNTRACK_TIME_WAIT, */
 116     &nf_ct_tcp_timeout_close,          /* TCP_CONNTRACK_CLOSE, */
 117     NULL,                              /* TCP_CONNTRACK_LISTEN */
 118  };
 119
 120 #define sNO TCP_CONNTRACK_NONE
 121 #define sSS TCP_CONNTRACK_SYN_SENT
 122 #define sSR TCP_CONNTRACK_SYN_RECV
 123 #define sES TCP_CONNTRACK_ESTABLISHED
 124 #define sFW TCP_CONNTRACK_FIN_WAIT
 125 #define sCW TCP_CONNTRACK_CLOSE_WAIT
 126 #define sLA TCP_CONNTRACK_LAST_ACK
 127 #define sTW TCP_CONNTRACK_TIME_WAIT
 128 #define sCL TCP_CONNTRACK_CLOSE
 129 #define sLI TCP_CONNTRACK_LISTEN
 130 #define sIV TCP_CONNTRACK_MAX
 131 #define sIG TCP_CONNTRACK_IGNORE
 132
 133 /* What TCP flags are set from RST/SYN/FIN/ACK. */
 134 enum tcp_bit_set {
 135         TCP_SYN_SET,
 136         TCP_SYNACK_SET,
 137         TCP_FIN_SET,
 138         TCP_ACK_SET,
 139         TCP_RST_SET,
 140         TCP_NONE_SET,
 141 };
 142
 143 /*
 144  * The TCP state transition table needs a few words...
 145  *
 146  * We are the man in the middle. All the packets go through us
 147  * but might get lost in transit to the destination.
 148  * It is assumed that the destinations can't receive segments
 149  * we haven't seen.
 150  *
 151  * The checked segment is in window, but our windows are *not*
 152  * equivalent with the ones of the sender/receiver. We always
 153  * try to guess the state of the current sender.
 154  *
 155  * The meaning of the states are:
 156  *
 157  * NONE:        initial state
 158  * SYN_SENT:    SYN-only packet seen
 159  * SYN_RECV:    SYN-ACK packet seen
 160  * ESTABLISHED: ACK packet seen
 161  * FIN_WAIT:    FIN packet seen
 162  * CLOSE_WAIT:  ACK seen (after FIN)
 163  * LAST_ACK:    FIN seen (after FIN)
 164  * TIME_WAIT:   last ACK seen
 165  * CLOSE:       closed connection
 166  *
 167  * LISTEN state is not used.
 168  *
 169  * Packets marked as IGNORED (sIG):
 170  *      if they may be either invalid or valid
 171  *      and the receiver may send back a connection
 172  *      closing RST or a SYN/ACK.
 173  *
 174  * Packets marked as INVALID (sIV):
 175  *      if they are invalid
 176  *      or we do not support the request (simultaneous open)
 177  */
 178 static enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 179         {
 180 /* ORIGINAL */
 181 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 182 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
 183 /*
 184  *      sNO -> sSS      Initialize a new connection
 185  *      sSS -> sSS      Retransmitted SYN
 186  *      sSR -> sIG      Late retransmitted SYN?
 187  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 188  *                      are errors. Receiver will reply with RST
 189  *                      and close the connection.
 190  *                      Or we are not in sync and hold a dead connection.
 191  *      sFW -> sIG
 192  *      sCW -> sIG
 193  *      sLA -> sIG
 194  *      sTW -> sSS      Reopened connection (RFC 1122).
 195  *      sCL -> sSS
 196  */
 197 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 198 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 199 /*
 200  * A SYN/ACK from the client is always invalid:
 201  *      - either it tries to set up a simultaneous open, which is
 202  *        not supported;
 203  *      - or the firewall has just been inserted between the two hosts
 204  *        during the session set-up. The SYN will be retransmitted
 205  *        by the true client (or it'll time out).
 206  */
 207 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 208 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 209 /*
 210  *      sNO -> sIV      Too late and no reason to do anything...
 211  *      sSS -> sIV      Client migth not send FIN in this state:
 212  *                      we enforce waiting for a SYN/ACK reply first.
 213  *      sSR -> sFW      Close started.
 214  *      sES -> sFW
 215  *      sFW -> sLA      FIN seen in both directions, waiting for
 216  *                      the last ACK.
 217  *                      Migth be a retransmitted FIN as well...
 218  *      sCW -> sLA
 219  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 220  *      sTW -> sTW
 221  *      sCL -> sCL
 222  */
 223 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 224 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 225 /*
 226  *      sNO -> sES      Assumed.
 227  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 228  *      sSR -> sES      Established state is reached.
 229  *      sES -> sES      :-)
 230  *      sFW -> sCW      Normal close request answered by ACK.
 231  *      sCW -> sCW
 232  *      sLA -> sTW      Last ACK detected.
 233  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 234  *      sCL -> sCL
 235  */
 236 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 237 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 238 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 239         },
 240         {
 241 /* REPLY */
 242 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 243 /*syn*/    { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 244 /*
 245  *      sNO -> sIV      Never reached.
 246  *      sSS -> sIV      Simultaneous open, not supported
 247  *      sSR -> sIV      Simultaneous open, not supported.
 248  *      sES -> sIV      Server may not initiate a connection.
 249  *      sFW -> sIV
 250  *      sCW -> sIV
 251  *      sLA -> sIV
 252  *      sTW -> sIV      Reopened connection, but server may not do it.
 253  *      sCL -> sIV
 254  */
 255 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 256 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
 257 /*
 258  *      sSS -> sSR      Standard open.
 259  *      sSR -> sSR      Retransmitted SYN/ACK.
 260  *      sES -> sIG      Late retransmitted SYN/ACK?
 261  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 262  *      sCW -> sIG
 263  *      sLA -> sIG
 264  *      sTW -> sIG
 265  *      sCL -> sIG
 266  */
 267 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 268 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 269 /*
 270  *      sSS -> sIV      Server might not send FIN in this state.
 271  *      sSR -> sFW      Close started.
 272  *      sES -> sFW
 273  *      sFW -> sLA      FIN seen in both directions.
 274  *      sCW -> sLA
 275  *      sLA -> sLA      Retransmitted FIN.
 276  *      sTW -> sTW
 277  *      sCL -> sCL
 278  */
 279 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 280 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 281 /*
 282  *      sSS -> sIG      Might be a half-open connection.
 283  *      sSR -> sSR      Might answer late resent SYN.
 284  *      sES -> sES      :-)
 285  *      sFW -> sCW      Normal close request answered by ACK.
 286  *      sCW -> sCW
 287  *      sLA -> sTW      Last ACK detected.
 288  *      sTW -> sTW      Retransmitted last ACK.
 289  *      sCL -> sCL
 290  */
 291 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 292 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 293 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 294         }
 295 };
 296
 297 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
 298                             unsigned int dataoff,
 299                             struct nf_conntrack_tuple *tuple)
 300 {
 301         struct tcphdr _hdr, *hp;
 302
 303         /* Actually only need first 8 bytes. */
 304         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
 305         if (hp == NULL)
 306                 return 0;
 307
 308         tuple->src.u.tcp.port = hp->source;
 309         tuple->dst.u.tcp.port = hp->dest;
 310
 311         return 1;
 312 }
 313
 314 static int tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
 315                             const struct nf_conntrack_tuple *orig)
 316 {
 317         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
 318         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
 319         return 1;
 320 }
 321
 322 /* Print out the per-protocol part of the tuple. */
 323 static int tcp_print_tuple(struct seq_file *s,
 324                            const struct nf_conntrack_tuple *tuple)
 325 {
 326         return seq_printf(s, "sport=%hu dport=%hu ",
 327                           ntohs(tuple->src.u.tcp.port),
 328                           ntohs(tuple->dst.u.tcp.port));
 329 }
 330
 331 /* Print out the private part of the conntrack. */
 332 static int tcp_print_conntrack(struct seq_file *s,
 333                                const struct nf_conn *conntrack)
 334 {
 335         enum tcp_conntrack state;
 336
 337         read_lock_bh(&tcp_lock);
 338         state = conntrack->proto.tcp.state;
 339         read_unlock_bh(&tcp_lock);
 340
 341         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
 342 }
 343
 344 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 345 {
 346         if (tcph->rst) return TCP_RST_SET;
 347         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 348         else if (tcph->fin) return TCP_FIN_SET;
 349         else if (tcph->ack) return TCP_ACK_SET;
 350         else return TCP_NONE_SET;
 351 }
 352
 353 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 354    in IP Filter' by Guido van Rooij.
 355
 356    http://www.nluug.nl/events/sane2000/papers.html
 357    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
 358
 359    The boundaries and the conditions are changed according to RFC793:
 360    the packet must intersect the window (i.e. segments may be
 361    after the right or before the left edge) and thus receivers may ACK
 362    segments after the right edge of the window.
 363
 364         td_maxend = max(sack + max(win,1)) seen in reply packets
 365         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 366         td_maxwin += seq + len - sender.td_maxend
 367                         if seq + len > sender.td_maxend
 368         td_end    = max(seq + len) seen in sent packets
 369
 370    I.   Upper bound for valid data:     seq <= sender.td_maxend
 371    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 372    III. Upper bound for valid ack:      sack <= receiver.td_end
 373    IV.  Lower bound for valid ack:      ack >= receiver.td_end - MAXACKWINDOW
 374
 375    where sack is the highest right edge of sack block found in the packet.
 376
 377    The upper bound limit for a valid ack is not ignored -
 378    we doesn't have to deal with fragments.
 379 */
 380
 381 static inline __u32 segment_seq_plus_len(__u32 seq,
 382                                          size_t len,
 383                                          unsigned int dataoff,
 384                                          struct tcphdr *tcph)
 385 {
 386         /* XXX Should I use payload length field in IP/IPv6 header ?
 387          * - YK */
 388         return (seq + len - dataoff - tcph->doff*4
 389                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 390 }
 391
 392 /* Fixme: what about big packets? */
 393 #define MAXACKWINCONST                  66000
 394 #define MAXACKWINDOW(sender)                                            \
 395         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 396                                               : MAXACKWINCONST)
 397
 398 /*
 399  * Simplified tcp_parse_options routine from tcp_input.c
 400  */
 401 static void tcp_options(const struct sk_buff *skb,
 402                         unsigned int dataoff,
 403                         struct tcphdr *tcph,
 404                         struct ip_ct_tcp_state *state)
 405 {
 406         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 407         unsigned char *ptr;
 408         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 409
 410         if (!length)
 411                 return;
 412
 413         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 414                                  length, buff);
 415         BUG_ON(ptr == NULL);
 416
 417         state->td_scale =
 418         state->flags = 0;
 419
 420         while (length > 0) {
 421                 int opcode=*ptr++;
 422                 int opsize;
 423
 424                 switch (opcode) {
 425                 case TCPOPT_EOL:
 426                         return;
 427                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 428                         length--;
 429                         continue;
 430                 default:
 431                         opsize=*ptr++;
 432                         if (opsize < 2) /* "silly options" */
 433                                 return;
 434                         if (opsize > length)
 435                                 break;  /* don't parse partial options */
 436
 437                         if (opcode == TCPOPT_SACK_PERM
 438                             && opsize == TCPOLEN_SACK_PERM)
 439                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 440                         else if (opcode == TCPOPT_WINDOW
 441                                  && opsize == TCPOLEN_WINDOW) {
 442                                 state->td_scale = *(u_int8_t *)ptr;
 443
 444                                 if (state->td_scale > 14) {
 445                                         /* See RFC1323 */
 446                                         state->td_scale = 14;
 447                                 }
 448                                 state->flags |=
 449                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 450                         }
 451                         ptr += opsize - 2;
 452                         length -= opsize;
 453                 }
 454         }
 455 }
 456
 457 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 458                      struct tcphdr *tcph, __u32 *sack)
 459 {
 460         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 461         unsigned char *ptr;
 462         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 463         __u32 tmp;
 464
 465         if (!length)
 466                 return;
 467
 468         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 469                                  length, buff);
 470         BUG_ON(ptr == NULL);
 471
 472         /* Fast path for timestamp-only option */
 473         if (length == TCPOLEN_TSTAMP_ALIGNED*4
 474             && *(__be32 *)ptr ==
 475                 __constant_htonl((TCPOPT_NOP << 24)
 476                                  | (TCPOPT_NOP << 16)
 477                                  | (TCPOPT_TIMESTAMP << 8)
 478                                  | TCPOLEN_TIMESTAMP))
 479                 return;
 480
 481         while (length > 0) {
 482                 int opcode = *ptr++;
 483                 int opsize, i;
 484
 485                 switch (opcode) {
 486                 case TCPOPT_EOL:
 487                         return;
 488                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 489                         length--;
 490                         continue;
 491                 default:
 492                         opsize = *ptr++;
 493                         if (opsize < 2) /* "silly options" */
 494                                 return;
 495                         if (opsize > length)
 496                                 break;  /* don't parse partial options */
 497
 498                         if (opcode == TCPOPT_SACK
 499                             && opsize >= (TCPOLEN_SACK_BASE
 500                                           + TCPOLEN_SACK_PERBLOCK)
 501                             && !((opsize - TCPOLEN_SACK_BASE)
 502                                  % TCPOLEN_SACK_PERBLOCK)) {
 503                                 for (i = 0;
 504                                      i < (opsize - TCPOLEN_SACK_BASE);
 505                                      i += TCPOLEN_SACK_PERBLOCK) {
 506                                         tmp = ntohl(*((__be32 *)(ptr+i)+1));
 507
 508                                         if (after(tmp, *sack))
 509                                                 *sack = tmp;
 510                                 }
 511                                 return;
 512                         }
 513                         ptr += opsize - 2;
 514                         length -= opsize;
 515                 }
 516         }
 517 }
 518
 519 static int tcp_in_window(struct ip_ct_tcp *state,
 520                          enum ip_conntrack_dir dir,
 521                          unsigned int index,
 522                          const struct sk_buff *skb,
 523                          unsigned int dataoff,
 524                          struct tcphdr *tcph,
 525                          int pf)
 526 {
 527         struct ip_ct_tcp_state *sender = &state->seen[dir];
 528         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 529         __u32 seq, ack, sack, end, win, swin;
 530         int res;
 531
 532         /*
 533          * Get the required data from the packet.
 534          */
 535         seq = ntohl(tcph->seq);
 536         ack = sack = ntohl(tcph->ack_seq);
 537         win = ntohs(tcph->window);
 538         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 539
 540         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 541                 tcp_sack(skb, dataoff, tcph, &sack);
 542
 543         DEBUGP("tcp_in_window: START\n");
 544         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
 545                "seq=%u ack=%u sack=%u win=%u end=%u\n",
 546                 NIPQUAD(iph->saddr), ntohs(tcph->source),
 547                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
 548                 seq, ack, sack, win, end);
 549         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 550                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 551                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 552                 sender->td_scale,
 553                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 554                 receiver->td_scale);
 555
 556         if (sender->td_end == 0) {
 557                 /*
 558                  * Initialize sender data.
 559                  */
 560                 if (tcph->syn && tcph->ack) {
 561                         /*
 562                          * Outgoing SYN-ACK in reply to a SYN.
 563                          */
 564                         sender->td_end =
 565                         sender->td_maxend = end;
 566                         sender->td_maxwin = (win == 0 ? 1 : win);
 567
 568                         tcp_options(skb, dataoff, tcph, sender);
 569                         /*
 570                          * RFC 1323:
 571                          * Both sides must send the Window Scale option
 572                          * to enable window scaling in either direction.
 573                          */
 574                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 575                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 576                                 sender->td_scale =
 577                                 receiver->td_scale = 0;
 578                 } else {
 579                         /*
 580                          * We are in the middle of a connection,
 581                          * its history is lost for us.
 582                          * Let's try to use the data from the packet.
 583                          */
 584                         sender->td_end = end;
 585                         sender->td_maxwin = (win == 0 ? 1 : win);
 586                         sender->td_maxend = end + sender->td_maxwin;
 587                 }
 588         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 589                      && dir == IP_CT_DIR_ORIGINAL)
 590                    || (state->state == TCP_CONNTRACK_SYN_RECV
 591                      && dir == IP_CT_DIR_REPLY))
 592                    && after(end, sender->td_end)) {
 593                 /*
 594                  * RFC 793: "if a TCP is reinitialized ... then it need
 595                  * not wait at all; it must only be sure to use sequence
 596                  * numbers larger than those recently used."
 597                  */
 598                 sender->td_end =
 599                 sender->td_maxend = end;
 600                 sender->td_maxwin = (win == 0 ? 1 : win);
 601
 602                 tcp_options(skb, dataoff, tcph, sender);
 603         }
 604
 605         if (!(tcph->ack)) {
 606                 /*
 607                  * If there is no ACK, just pretend it was set and OK.
 608                  */
 609                 ack = sack = receiver->td_end;
 610         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 611                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 612                    && (ack == 0)) {
 613                 /*
 614                  * Broken TCP stacks, that set ACK in RST packets as well
 615                  * with zero ack value.
 616                  */
 617                 ack = sack = receiver->td_end;
 618         }
 619
 620         if (seq == end
 621             && (!tcph->rst
 622                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
 623                 /*
 624                  * Packets contains no data: we assume it is valid
 625                  * and check the ack value only.
 626                  * However RST segments are always validated by their
 627                  * SEQ number, except when seq == 0 (reset sent answering
 628                  * SYN.
 629                  */
 630                 seq = end = sender->td_end;
 631
 632         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
 633                "seq=%u ack=%u sack =%u win=%u end=%u\n",
 634                 NIPQUAD(iph->saddr), ntohs(tcph->source),
 635                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
 636                 seq, ack, sack, win, end);
 637         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 638                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 639                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 640                 sender->td_scale,
 641                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 642                 receiver->td_scale);
 643
 644         DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 645                 before(seq, sender->td_maxend + 1),
 646                 after(end, sender->td_end - receiver->td_maxwin - 1),
 647                 before(sack, receiver->td_end + 1),
 648                 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
 649
 650         if (before(seq, sender->td_maxend + 1) &&
 651             after(end, sender->td_end - receiver->td_maxwin - 1) &&
 652             before(sack, receiver->td_end + 1) &&
 653             after(ack, receiver->td_end - MAXACKWINDOW(sender))) {
 654                 /*
 655                  * Take into account window scaling (RFC 1323).
 656                  */
 657                 if (!tcph->syn)
 658                         win <<= sender->td_scale;
 659
 660                 /*
 661                  * Update sender data.
 662                  */
 663                 swin = win + (sack - ack);
 664                 if (sender->td_maxwin < swin)
 665                         sender->td_maxwin = swin;
 666                 if (after(end, sender->td_end))
 667                         sender->td_end = end;
 668                 /*
 669                  * Update receiver data.
 670                  */
 671                 if (after(end, sender->td_maxend))
 672                         receiver->td_maxwin += end - sender->td_maxend;
 673                 if (after(sack + win, receiver->td_maxend - 1)) {
 674                         receiver->td_maxend = sack + win;
 675                         if (win == 0)
 676                                 receiver->td_maxend++;
 677                 }
 678
 679                 /*
 680                  * Check retransmissions.
 681                  */
 682                 if (index == TCP_ACK_SET) {
 683                         if (state->last_dir == dir
 684                             && state->last_seq == seq
 685                             && state->last_ack == ack
 686                             && state->last_end == end
 687                             && state->last_win == win)
 688                                 state->retrans++;
 689                         else {
 690                                 state->last_dir = dir;
 691                                 state->last_seq = seq;
 692                                 state->last_ack = ack;
 693                                 state->last_end = end;
 694                                 state->last_win = win;
 695                                 state->retrans = 0;
 696                         }
 697                 }
 698                 res = 1;
 699         } else {
 700                 res = 0;
 701                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 702                     nf_ct_tcp_be_liberal)
 703                         res = 1;
 704                 if (!res && LOG_INVALID(IPPROTO_TCP))
 705                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 706                         "nf_ct_tcp: %s ",
 707                         before(seq, sender->td_maxend + 1) ?
 708                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
 709                         before(sack, receiver->td_end + 1) ?
 710                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
 711                         : "ACK is under the lower bound (possible overly delayed ACK)"
 712                         : "ACK is over the upper bound (ACKed data not seen yet)"
 713                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 714                         : "SEQ is over the upper bound (over the window of the receiver)");
 715         }
 716
 717         DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
 718                "receiver end=%u maxend=%u maxwin=%u\n",
 719                 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 720                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 721
 722         return res;
 723 }
 724
 725 #ifdef CONFIG_NF_NAT_NEEDED
 726 /* Update sender->td_end after NAT successfully mangled the packet */
 727 /* Caller must linearize skb at tcp header. */
 728 void nf_conntrack_tcp_update(struct sk_buff *skb,
 729                              unsigned int dataoff,
 730                              struct nf_conn *conntrack,
 731                              int dir)
 732 {
 733         struct tcphdr *tcph = (void *)skb->data + dataoff;
 734         __u32 end;
 735 #ifdef DEBUGP_VARS
 736         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
 737         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
 738 #endif
 739
 740         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
 741
 742         write_lock_bh(&tcp_lock);
 743         /*
 744          * We have to worry for the ack in the reply packet only...
 745          */
 746         if (after(end, conntrack->proto.tcp.seen[dir].td_end))
 747                 conntrack->proto.tcp.seen[dir].td_end = end;
 748         conntrack->proto.tcp.last_end = end;
 749         write_unlock_bh(&tcp_lock);
 750         DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
 751                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 752                 sender->td_end, sender->td_maxend, sender->td_maxwin,
 753                 sender->td_scale,
 754                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 755                 receiver->td_scale);
 756 }
 757 EXPORT_SYMBOL_GPL(nf_conntrack_tcp_update);
 758 #endif
 759
 760 #define TH_FIN  0x01
 761 #define TH_SYN  0x02
 762 #define TH_RST  0x04
 763 #define TH_PUSH 0x08
 764 #define TH_ACK  0x10
 765 #define TH_URG  0x20
 766 #define TH_ECE  0x40
 767 #define TH_CWR  0x80
 768
 769 /* table of valid flag combinations - ECE and CWR are always valid */
 770 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
 771 {
 772         [TH_SYN]                        = 1,
 773         [TH_SYN|TH_ACK]                 = 1,
 774         [TH_SYN|TH_PUSH]                = 1,
 775         [TH_SYN|TH_ACK|TH_PUSH]         = 1,
 776         [TH_RST]                        = 1,
 777         [TH_RST|TH_ACK]                 = 1,
 778         [TH_RST|TH_ACK|TH_PUSH]         = 1,
 779         [TH_FIN|TH_ACK]                 = 1,
 780         [TH_ACK]                        = 1,
 781         [TH_ACK|TH_PUSH]                = 1,
 782         [TH_ACK|TH_URG]                 = 1,
 783         [TH_ACK|TH_URG|TH_PUSH]         = 1,
 784         [TH_FIN|TH_ACK|TH_PUSH]         = 1,
 785         [TH_FIN|TH_ACK|TH_URG]          = 1,
 786         [TH_FIN|TH_ACK|TH_URG|TH_PUSH]  = 1,
 787 };
 788
 789 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 790 static int tcp_error(struct sk_buff *skb,
 791                      unsigned int dataoff,
 792                      enum ip_conntrack_info *ctinfo,
 793                      int pf,
 794                      unsigned int hooknum)
 795 {
 796         struct tcphdr _tcph, *th;
 797         unsigned int tcplen = skb->len - dataoff;
 798         u_int8_t tcpflags;
 799
 800         /* Smaller that minimal TCP header? */
 801         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 802         if (th == NULL) {
 803                 if (LOG_INVALID(IPPROTO_TCP))
 804                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 805                                 "nf_ct_tcp: short packet ");
 806                 return -NF_ACCEPT;
 807         }
 808
 809         /* Not whole TCP header or malformed packet */
 810         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 811                 if (LOG_INVALID(IPPROTO_TCP))
 812                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 813                                 "nf_ct_tcp: truncated/malformed packet ");
 814                 return -NF_ACCEPT;
 815         }
 816
 817         /* Checksum invalid? Ignore.
 818          * We skip checking packets on the outgoing path
 819          * because the checksum is assumed to be correct.
 820          */
 821         /* FIXME: Source route IP option packets --RR */
 822         if (nf_conntrack_checksum &&
 823             ((pf == PF_INET && hooknum == NF_IP_PRE_ROUTING) ||
 824              (pf == PF_INET6 && hooknum == NF_IP6_PRE_ROUTING)) &&
 825             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
 826                 if (LOG_INVALID(IPPROTO_TCP))
 827                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 828                                   "nf_ct_tcp: bad TCP checksum ");
 829                 return -NF_ACCEPT;
 830         }
 831
 832         /* Check TCP flags. */
 833         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
 834         if (!tcp_valid_flags[tcpflags]) {
 835                 if (LOG_INVALID(IPPROTO_TCP))
 836                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 837                                   "nf_ct_tcp: invalid TCP flag combination ");
 838                 return -NF_ACCEPT;
 839         }
 840
 841         return NF_ACCEPT;
 842 }
 843
 844 /* Returns verdict for packet, or -1 for invalid. */
 845 static int tcp_packet(struct nf_conn *conntrack,
 846                       const struct sk_buff *skb,
 847                       unsigned int dataoff,
 848                       enum ip_conntrack_info ctinfo,
 849                       int pf,
 850                       unsigned int hooknum)
 851 {
 852         enum tcp_conntrack new_state, old_state;
 853         enum ip_conntrack_dir dir;
 854         struct tcphdr *th, _tcph;
 855         unsigned long timeout;
 856         unsigned int index;
 857
 858         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 859         BUG_ON(th == NULL);
 860
 861         write_lock_bh(&tcp_lock);
 862         old_state = conntrack->proto.tcp.state;
 863         dir = CTINFO2DIR(ctinfo);
 864         index = get_conntrack_index(th);
 865         new_state = tcp_conntracks[dir][index][old_state];
 866
 867         switch (new_state) {
 868         case TCP_CONNTRACK_IGNORE:
 869                 /* Ignored packets:
 870                  *
 871                  * a) SYN in ORIGINAL
 872                  * b) SYN/ACK in REPLY
 873                  * c) ACK in reply direction after initial SYN in original.
 874                  */
 875                 if (index == TCP_SYNACK_SET
 876                     && conntrack->proto.tcp.last_index == TCP_SYN_SET
 877                     && conntrack->proto.tcp.last_dir != dir
 878                     && ntohl(th->ack_seq) ==
 879                              conntrack->proto.tcp.last_end) {
 880                         /* This SYN/ACK acknowledges a SYN that we earlier
 881                          * ignored as invalid. This means that the client and
 882                          * the server are both in sync, while the firewall is
 883                          * not. We kill this session and block the SYN/ACK so
 884                          * that the client cannot but retransmit its SYN and
 885                          * thus initiate a clean new session.
 886                          */
 887                         write_unlock_bh(&tcp_lock);
 888                         if (LOG_INVALID(IPPROTO_TCP))
 889                                 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 890                                           "nf_ct_tcp: killing out of sync session ");
 891                         if (del_timer(&conntrack->timeout))
 892                                 conntrack->timeout.function((unsigned long)
 893                                                             conntrack);
 894                         return -NF_DROP;
 895                 }
 896                 conntrack->proto.tcp.last_index = index;
 897                 conntrack->proto.tcp.last_dir = dir;
 898                 conntrack->proto.tcp.last_seq = ntohl(th->seq);
 899                 conntrack->proto.tcp.last_end =
 900                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 901
 902                 write_unlock_bh(&tcp_lock);
 903                 if (LOG_INVALID(IPPROTO_TCP))
 904                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 905                                   "nf_ct_tcp: invalid packed ignored ");
 906                 return NF_ACCEPT;
 907         case TCP_CONNTRACK_MAX:
 908                 /* Invalid packet */
 909                 DEBUGP("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 910                        dir, get_conntrack_index(th),
 911                        old_state);
 912                 write_unlock_bh(&tcp_lock);
 913                 if (LOG_INVALID(IPPROTO_TCP))
 914                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 915                                   "nf_ct_tcp: invalid state ");
 916                 return -NF_ACCEPT;
 917         case TCP_CONNTRACK_SYN_SENT:
 918                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 919                         break;
 920                 if ((conntrack->proto.tcp.seen[dir].flags &
 921                         IP_CT_TCP_FLAG_CLOSE_INIT)
 922                     || after(ntohl(th->seq),
 923                              conntrack->proto.tcp.seen[dir].td_end)) {
 924                         /* Attempt to reopen a closed connection.
 925                         * Delete this connection and look up again. */
 926                         write_unlock_bh(&tcp_lock);
 927                         if (del_timer(&conntrack->timeout))
 928                                 conntrack->timeout.function((unsigned long)
 929                                                             conntrack);
 930                         return -NF_REPEAT;
 931                 } else {
 932                         write_unlock_bh(&tcp_lock);
 933                         if (LOG_INVALID(IPPROTO_TCP))
 934                                 nf_log_packet(pf, 0, skb, NULL, NULL,
 935                                               NULL, "nf_ct_tcp: invalid SYN");
 936                         return -NF_ACCEPT;
 937                 }
 938         case TCP_CONNTRACK_CLOSE:
 939                 if (index == TCP_RST_SET
 940                     && ((test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
 941                          && conntrack->proto.tcp.last_index == TCP_SYN_SET)
 942                         || (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
 943                             && conntrack->proto.tcp.last_index == TCP_ACK_SET))
 944                     && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
 945                         /* RST sent to invalid SYN or ACK we had let through
 946                          * at a) and c) above:
 947                          *
 948                          * a) SYN was in window then
 949                          * c) we hold a half-open connection.
 950                          *
 951                          * Delete our connection entry.
 952                          * We skip window checking, because packet might ACK
 953                          * segments we ignored. */
 954                         goto in_window;
 955                 }
 956                 /* Just fall through */
 957         default:
 958                 /* Keep compilers happy. */
 959                 break;
 960         }
 961
 962         if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
 963                            skb, dataoff, th, pf)) {
 964                 write_unlock_bh(&tcp_lock);
 965                 return -NF_ACCEPT;
 966         }
 967      in_window:
 968         /* From now on we have got in-window packets */
 969         conntrack->proto.tcp.last_index = index;
 970
 971         DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
 972                "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
 973                 NIPQUAD(iph->saddr), ntohs(th->source),
 974                 NIPQUAD(iph->daddr), ntohs(th->dest),
 975                 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
 976                 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
 977                 old_state, new_state);
 978
 979         conntrack->proto.tcp.state = new_state;
 980         if (old_state != new_state
 981             && (new_state == TCP_CONNTRACK_FIN_WAIT
 982                 || new_state == TCP_CONNTRACK_CLOSE))
 983                 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
 984         timeout = conntrack->proto.tcp.retrans >= nf_ct_tcp_max_retrans
 985                   && *tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans
 986                   ? nf_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
 987         write_unlock_bh(&tcp_lock);
 988
 989         nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
 990         if (new_state != old_state)
 991                 nf_conntrack_event_cache(IPCT_PROTOINFO, skb);
 992
 993         if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
 994                 /* If only reply is a RST, we can consider ourselves not to
 995                    have an established connection: this is a fairly common
 996                    problem case, so we can delete the conntrack
 997                    immediately.  --RR */
 998                 if (th->rst) {
 999                         if (del_timer(&conntrack->timeout))
1000                                 conntrack->timeout.function((unsigned long)
1001                                                             conntrack);
1002                         return NF_ACCEPT;
1003                 }
1004         } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
1005                    && (old_state == TCP_CONNTRACK_SYN_RECV
1006                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1007                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1008                 /* Set ASSURED if we see see valid ack in ESTABLISHED
1009                    after SYN_RECV or a valid answer for a picked up
1010                    connection. */
1011                 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1012                 nf_conntrack_event_cache(IPCT_STATUS, skb);
1013         }
1014         nf_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1015
1016         return NF_ACCEPT;
1017 }
1018
1019 /* Called when a new connection for this protocol found. */
1020 static int tcp_new(struct nf_conn *conntrack,
1021                    const struct sk_buff *skb,
1022                    unsigned int dataoff)
1023 {
1024         enum tcp_conntrack new_state;
1025         struct tcphdr *th, _tcph;
1026 #ifdef DEBUGP_VARS
1027         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1028         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1029 #endif
1030
1031         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1032         BUG_ON(th == NULL);
1033
1034         /* Don't need lock here: this conntrack not in circulation yet */
1035         new_state
1036                 = tcp_conntracks[0][get_conntrack_index(th)]
1037                 [TCP_CONNTRACK_NONE];
1038
1039         /* Invalid: delete conntrack */
1040         if (new_state >= TCP_CONNTRACK_MAX) {
1041                 DEBUGP("nf_ct_tcp: invalid new deleting.\n");
1042                 return 0;
1043         }
1044
1045         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1046                 /* SYN packet */
1047                 conntrack->proto.tcp.seen[0].td_end =
1048                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1049                                              dataoff, th);
1050                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1051                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1052                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1053                 conntrack->proto.tcp.seen[0].td_maxend =
1054                         conntrack->proto.tcp.seen[0].td_end;
1055
1056                 tcp_options(skb, dataoff, th, &conntrack->proto.tcp.seen[0]);
1057                 conntrack->proto.tcp.seen[1].flags = 0;
1058         } else if (nf_ct_tcp_loose == 0) {
1059                 /* Don't try to pick up connections. */
1060                 return 0;
1061         } else {
1062                 /*
1063                  * We are in the middle of a connection,
1064                  * its history is lost for us.
1065                  * Let's try to use the data from the packet.
1066                  */
1067                 conntrack->proto.tcp.seen[0].td_end =
1068                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1069                                              dataoff, th);
1070                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1071                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1072                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1073                 conntrack->proto.tcp.seen[0].td_maxend =
1074                         conntrack->proto.tcp.seen[0].td_end +
1075                         conntrack->proto.tcp.seen[0].td_maxwin;
1076                 conntrack->proto.tcp.seen[0].td_scale = 0;
1077
1078                 /* We assume SACK and liberal window checking to handle
1079                  * window scaling */
1080                 conntrack->proto.tcp.seen[0].flags =
1081                 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1082                                                      IP_CT_TCP_FLAG_BE_LIBERAL;
1083         }
1084
1085         conntrack->proto.tcp.seen[1].td_end = 0;
1086         conntrack->proto.tcp.seen[1].td_maxend = 0;
1087         conntrack->proto.tcp.seen[1].td_maxwin = 1;
1088         conntrack->proto.tcp.seen[1].td_scale = 0;
1089
1090         /* tcp_packet will set them */
1091         conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1092         conntrack->proto.tcp.last_index = TCP_NONE_SET;
1093
1094         DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1095                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1096                 sender->td_end, sender->td_maxend, sender->td_maxwin,
1097                 sender->td_scale,
1098                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1099                 receiver->td_scale);
1100         return 1;
1101 }
1102
1103 #if defined(CONFIG_NF_CT_NETLINK) || \
1104     defined(CONFIG_NF_CT_NETLINK_MODULE)
1105
1106 #include <linux/netfilter/nfnetlink.h>
1107 #include <linux/netfilter/nfnetlink_conntrack.h>
1108
1109 static int tcp_to_nfattr(struct sk_buff *skb, struct nfattr *nfa,
1110                          const struct nf_conn *ct)
1111 {
1112         struct nfattr *nest_parms;
1113
1114         read_lock_bh(&tcp_lock);
1115         nest_parms = NFA_NEST(skb, CTA_PROTOINFO_TCP);
1116         NFA_PUT(skb, CTA_PROTOINFO_TCP_STATE, sizeof(u_int8_t),
1117                 &ct->proto.tcp.state);
1118         read_unlock_bh(&tcp_lock);
1119
1120         NFA_NEST_END(skb, nest_parms);
1121
1122         return 0;
1123
1124 nfattr_failure:
1125         read_unlock_bh(&tcp_lock);
1126         return -1;
1127 }
1128
1129 static const size_t cta_min_tcp[CTA_PROTOINFO_TCP_MAX] = {
1130         [CTA_PROTOINFO_TCP_STATE-1]     = sizeof(u_int8_t),
1131 };
1132
1133 static int nfattr_to_tcp(struct nfattr *cda[], struct nf_conn *ct)
1134 {
1135         struct nfattr *attr = cda[CTA_PROTOINFO_TCP-1];
1136         struct nfattr *tb[CTA_PROTOINFO_TCP_MAX];
1137
1138         /* updates could not contain anything about the private
1139          * protocol info, in that case skip the parsing */
1140         if (!attr)
1141                 return 0;
1142
1143         nfattr_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, attr);
1144
1145         if (nfattr_bad_size(tb, CTA_PROTOINFO_TCP_MAX, cta_min_tcp))
1146                 return -EINVAL;
1147
1148         if (!tb[CTA_PROTOINFO_TCP_STATE-1])
1149                 return -EINVAL;
1150
1151         write_lock_bh(&tcp_lock);
1152         ct->proto.tcp.state =
1153                 *(u_int8_t *)NFA_DATA(tb[CTA_PROTOINFO_TCP_STATE-1]);
1154         write_unlock_bh(&tcp_lock);
1155
1156         return 0;
1157 }
1158 #endif
1159
1160 #ifdef CONFIG_SYSCTL
1161 static unsigned int tcp_sysctl_table_users;
1162 static struct ctl_table_header *tcp_sysctl_header;
1163 static struct ctl_table tcp_sysctl_table[] = {
1164         {
1165                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_SYN_SENT,
1166                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1167                 .data           = &nf_ct_tcp_timeout_syn_sent,
1168                 .maxlen         = sizeof(unsigned int),
1169                 .mode           = 0644,
1170                 .proc_handler   = &proc_dointvec_jiffies,
1171         },
1172         {
1173                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_SYN_RECV,
1174                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1175                 .data           = &nf_ct_tcp_timeout_syn_recv,
1176                 .maxlen         = sizeof(unsigned int),
1177                 .mode           = 0644,
1178                 .proc_handler   = &proc_dointvec_jiffies,
1179         },
1180         {
1181                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_ESTABLISHED,
1182                 .procname       = "nf_conntrack_tcp_timeout_established",
1183                 .data           = &nf_ct_tcp_timeout_established,
1184                 .maxlen         = sizeof(unsigned int),
1185                 .mode           = 0644,
1186                 .proc_handler   = &proc_dointvec_jiffies,
1187         },
1188         {
1189                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_FIN_WAIT,
1190                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1191                 .data           = &nf_ct_tcp_timeout_fin_wait,
1192                 .maxlen         = sizeof(unsigned int),
1193                 .mode           = 0644,
1194                 .proc_handler   = &proc_dointvec_jiffies,
1195         },
1196         {
1197                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_CLOSE_WAIT,
1198                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1199                 .data           = &nf_ct_tcp_timeout_close_wait,
1200                 .maxlen         = sizeof(unsigned int),
1201                 .mode           = 0644,
1202                 .proc_handler   = &proc_dointvec_jiffies,
1203         },
1204         {
1205                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_LAST_ACK,
1206                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1207                 .data           = &nf_ct_tcp_timeout_last_ack,
1208                 .maxlen         = sizeof(unsigned int),
1209                 .mode           = 0644,
1210                 .proc_handler   = &proc_dointvec_jiffies,
1211         },
1212         {
1213                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_TIME_WAIT,
1214                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1215                 .data           = &nf_ct_tcp_timeout_time_wait,
1216                 .maxlen         = sizeof(unsigned int),
1217                 .mode           = 0644,
1218                 .proc_handler   = &proc_dointvec_jiffies,
1219         },
1220         {
1221                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_CLOSE,
1222                 .procname       = "nf_conntrack_tcp_timeout_close",
1223                 .data           = &nf_ct_tcp_timeout_close,
1224                 .maxlen         = sizeof(unsigned int),
1225                 .mode           = 0644,
1226                 .proc_handler   = &proc_dointvec_jiffies,
1227         },
1228         {
1229                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_MAX_RETRANS,
1230                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1231                 .data           = &nf_ct_tcp_timeout_max_retrans,
1232                 .maxlen         = sizeof(unsigned int),
1233                 .mode           = 0644,
1234                 .proc_handler   = &proc_dointvec_jiffies,
1235         },
1236         {
1237                 .ctl_name       = NET_NF_CONNTRACK_TCP_LOOSE,
1238                 .procname       = "nf_conntrack_tcp_loose",
1239                 .data           = &nf_ct_tcp_loose,
1240                 .maxlen         = sizeof(unsigned int),
1241                 .mode           = 0644,
1242                 .proc_handler   = &proc_dointvec,
1243         },
1244         {
1245                 .ctl_name       = NET_NF_CONNTRACK_TCP_BE_LIBERAL,
1246                 .procname       = "nf_conntrack_tcp_be_liberal",
1247                 .data           = &nf_ct_tcp_be_liberal,
1248                 .maxlen         = sizeof(unsigned int),
1249                 .mode           = 0644,
1250                 .proc_handler   = &proc_dointvec,
1251         },
1252         {
1253                 .ctl_name       = NET_NF_CONNTRACK_TCP_MAX_RETRANS,
1254                 .procname       = "nf_conntrack_tcp_max_retrans",
1255                 .data           = &nf_ct_tcp_max_retrans,
1256                 .maxlen         = sizeof(unsigned int),
1257                 .mode           = 0644,
1258                 .proc_handler   = &proc_dointvec,
1259         },
1260         {
1261                 .ctl_name       = 0
1262         }
1263 };
1264
1265 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1266 static struct ctl_table tcp_compat_sysctl_table[] = {
1267         {
1268                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_SYN_SENT,
1269                 .procname       = "ip_conntrack_tcp_timeout_syn_sent",
1270                 .data           = &nf_ct_tcp_timeout_syn_sent,
1271                 .maxlen         = sizeof(unsigned int),
1272                 .mode           = 0644,
1273                 .proc_handler   = &proc_dointvec_jiffies,
1274         },
1275         {
1276                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_SYN_RECV,
1277                 .procname       = "ip_conntrack_tcp_timeout_syn_recv",
1278                 .data           = &nf_ct_tcp_timeout_syn_recv,
1279                 .maxlen         = sizeof(unsigned int),
1280                 .mode           = 0644,
1281                 .proc_handler   = &proc_dointvec_jiffies,
1282         },
1283         {
1284                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_ESTABLISHED,
1285                 .procname       = "ip_conntrack_tcp_timeout_established",
1286                 .data           = &nf_ct_tcp_timeout_established,
1287                 .maxlen         = sizeof(unsigned int),
1288                 .mode           = 0644,
1289                 .proc_handler   = &proc_dointvec_jiffies,
1290         },
1291         {
1292                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_FIN_WAIT,
1293                 .procname       = "ip_conntrack_tcp_timeout_fin_wait",
1294                 .data           = &nf_ct_tcp_timeout_fin_wait,
1295                 .maxlen         = sizeof(unsigned int),
1296                 .mode           = 0644,
1297                 .proc_handler   = &proc_dointvec_jiffies,
1298         },
1299         {
1300                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_CLOSE_WAIT,
1301                 .procname       = "ip_conntrack_tcp_timeout_close_wait",
1302                 .data           = &nf_ct_tcp_timeout_close_wait,
1303                 .maxlen         = sizeof(unsigned int),
1304                 .mode           = 0644,
1305                 .proc_handler   = &proc_dointvec_jiffies,
1306         },
1307         {
1308                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_LAST_ACK,
1309                 .procname       = "ip_conntrack_tcp_timeout_last_ack",
1310                 .data           = &nf_ct_tcp_timeout_last_ack,
1311                 .maxlen         = sizeof(unsigned int),
1312                 .mode           = 0644,
1313                 .proc_handler   = &proc_dointvec_jiffies,
1314         },
1315         {
1316                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_TIME_WAIT,
1317                 .procname       = "ip_conntrack_tcp_timeout_time_wait",
1318                 .data           = &nf_ct_tcp_timeout_time_wait,
1319                 .maxlen         = sizeof(unsigned int),
1320                 .mode           = 0644,
1321                 .proc_handler   = &proc_dointvec_jiffies,
1322         },
1323         {
1324                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_CLOSE,
1325                 .procname       = "ip_conntrack_tcp_timeout_close",
1326                 .data           = &nf_ct_tcp_timeout_close,
1327                 .maxlen         = sizeof(unsigned int),
1328                 .mode           = 0644,
1329                 .proc_handler   = &proc_dointvec_jiffies,
1330         },
1331         {
1332                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_MAX_RETRANS,
1333                 .procname       = "ip_conntrack_tcp_timeout_max_retrans",
1334                 .data           = &nf_ct_tcp_timeout_max_retrans,
1335                 .maxlen         = sizeof(unsigned int),
1336                 .mode           = 0644,
1337                 .proc_handler   = &proc_dointvec_jiffies,
1338         },
1339         {
1340                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_LOOSE,
1341                 .procname       = "ip_conntrack_tcp_loose",
1342                 .data           = &nf_ct_tcp_loose,
1343                 .maxlen         = sizeof(unsigned int),
1344                 .mode           = 0644,
1345                 .proc_handler   = &proc_dointvec,
1346         },
1347         {
1348                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_BE_LIBERAL,
1349                 .procname       = "ip_conntrack_tcp_be_liberal",
1350                 .data           = &nf_ct_tcp_be_liberal,
1351                 .maxlen         = sizeof(unsigned int),
1352                 .mode           = 0644,
1353                 .proc_handler   = &proc_dointvec,
1354         },
1355         {
1356                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_MAX_RETRANS,
1357                 .procname       = "ip_conntrack_tcp_max_retrans",
1358                 .data           = &nf_ct_tcp_max_retrans,
1359                 .maxlen         = sizeof(unsigned int),
1360                 .mode           = 0644,
1361                 .proc_handler   = &proc_dointvec,
1362         },
1363         {
1364                 .ctl_name       = 0
1365         }
1366 };
1367 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
1368 #endif /* CONFIG_SYSCTL */
1369
1370 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 =
1371 {
1372         .l3proto                = PF_INET,
1373         .l4proto                = IPPROTO_TCP,
1374         .name                   = "tcp",
1375         .pkt_to_tuple           = tcp_pkt_to_tuple,
1376         .invert_tuple           = tcp_invert_tuple,
1377         .print_tuple            = tcp_print_tuple,
1378         .print_conntrack        = tcp_print_conntrack,
1379         .packet                 = tcp_packet,
1380         .new                    = tcp_new,
1381         .error                  = tcp_error,
1382 #if defined(CONFIG_NF_CT_NETLINK) || \
1383     defined(CONFIG_NF_CT_NETLINK_MODULE)
1384         .to_nfattr              = tcp_to_nfattr,
1385         .from_nfattr            = nfattr_to_tcp,
1386         .tuple_to_nfattr        = nf_ct_port_tuple_to_nfattr,
1387         .nfattr_to_tuple        = nf_ct_port_nfattr_to_tuple,
1388 #endif
1389 #ifdef CONFIG_SYSCTL
1390         .ctl_table_users        = &tcp_sysctl_table_users,
1391         .ctl_table_header       = &tcp_sysctl_header,
1392         .ctl_table              = tcp_sysctl_table,
1393 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1394         .ctl_compat_table       = tcp_compat_sysctl_table,
1395 #endif
1396 #endif
1397 };
1398 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1399
1400 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 =
1401 {
1402         .l3proto                = PF_INET6,
1403         .l4proto                = IPPROTO_TCP,
1404         .name                   = "tcp",
1405         .pkt_to_tuple           = tcp_pkt_to_tuple,
1406         .invert_tuple           = tcp_invert_tuple,
1407         .print_tuple            = tcp_print_tuple,
1408         .print_conntrack        = tcp_print_conntrack,
1409         .packet                 = tcp_packet,
1410         .new                    = tcp_new,
1411         .error                  = tcp_error,
1412 #if defined(CONFIG_NF_CT_NETLINK) || \
1413     defined(CONFIG_NF_CT_NETLINK_MODULE)
1414         .to_nfattr              = tcp_to_nfattr,
1415         .from_nfattr            = nfattr_to_tcp,
1416         .tuple_to_nfattr        = nf_ct_port_tuple_to_nfattr,
1417         .nfattr_to_tuple        = nf_ct_port_nfattr_to_tuple,
1418 #endif
1419 #ifdef CONFIG_SYSCTL
1420         .ctl_table_users        = &tcp_sysctl_table_users,
1421         .ctl_table_header       = &tcp_sysctl_header,
1422         .ctl_table              = tcp_sysctl_table,
1423 #endif
1424 };
1425 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);