net/netfilter/nf_conntrack_proto_tcp.c

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