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_PUSH]                = 1,
 773         [TH_SYN|TH_URG]                 = 1,
 774         [TH_SYN|TH_PUSH|TH_URG]         = 1,
 775         [TH_SYN|TH_ACK]                 = 1,
 776         [TH_SYN|TH_ACK|TH_PUSH]         = 1,
 777         [TH_RST]                        = 1,
 778         [TH_RST|TH_ACK]                 = 1,
 779         [TH_RST|TH_ACK|TH_PUSH]         = 1,
 780         [TH_FIN|TH_ACK]                 = 1,
 781         [TH_ACK]                        = 1,
 782         [TH_ACK|TH_PUSH]                = 1,
 783         [TH_ACK|TH_URG]                 = 1,
 784         [TH_ACK|TH_URG|TH_PUSH]         = 1,
 785         [TH_FIN|TH_ACK|TH_PUSH]         = 1,
 786         [TH_FIN|TH_ACK|TH_URG]          = 1,
 787         [TH_FIN|TH_ACK|TH_URG|TH_PUSH]  = 1,
 788 };
 789
 790 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 791 static int tcp_error(struct sk_buff *skb,
 792                      unsigned int dataoff,
 793                      enum ip_conntrack_info *ctinfo,
 794                      int pf,
 795                      unsigned int hooknum)
 796 {
 797         struct tcphdr _tcph, *th;
 798         unsigned int tcplen = skb->len - dataoff;
 799         u_int8_t tcpflags;
 800
 801         /* Smaller that minimal TCP header? */
 802         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 803         if (th == NULL) {
 804                 if (LOG_INVALID(IPPROTO_TCP))
 805                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 806                                 "nf_ct_tcp: short packet ");
 807                 return -NF_ACCEPT;
 808         }
 809
 810         /* Not whole TCP header or malformed packet */
 811         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 812                 if (LOG_INVALID(IPPROTO_TCP))
 813                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 814                                 "nf_ct_tcp: truncated/malformed packet ");
 815                 return -NF_ACCEPT;
 816         }
 817
 818         /* Checksum invalid? Ignore.
 819          * We skip checking packets on the outgoing path
 820          * because the checksum is assumed to be correct.
 821          */
 822         /* FIXME: Source route IP option packets --RR */
 823         if (nf_conntrack_checksum &&
 824             ((pf == PF_INET && hooknum == NF_IP_PRE_ROUTING) ||
 825              (pf == PF_INET6 && hooknum == NF_IP6_PRE_ROUTING)) &&
 826             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
 827                 if (LOG_INVALID(IPPROTO_TCP))
 828                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 829                                   "nf_ct_tcp: bad TCP checksum ");
 830                 return -NF_ACCEPT;
 831         }
 832
 833         /* Check TCP flags. */
 834         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
 835         if (!tcp_valid_flags[tcpflags]) {
 836                 if (LOG_INVALID(IPPROTO_TCP))
 837                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 838                                   "nf_ct_tcp: invalid TCP flag combination ");
 839                 return -NF_ACCEPT;
 840         }
 841
 842         return NF_ACCEPT;
 843 }
 844
 845 /* Returns verdict for packet, or -1 for invalid. */
 846 static int tcp_packet(struct nf_conn *conntrack,
 847                       const struct sk_buff *skb,
 848                       unsigned int dataoff,
 849                       enum ip_conntrack_info ctinfo,
 850                       int pf,
 851                       unsigned int hooknum)
 852 {
 853         enum tcp_conntrack new_state, old_state;
 854         enum ip_conntrack_dir dir;
 855         struct tcphdr *th, _tcph;
 856         unsigned long timeout;
 857         unsigned int index;
 858
 859         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 860         BUG_ON(th == NULL);
 861
 862         write_lock_bh(&tcp_lock);
 863         old_state = conntrack->proto.tcp.state;
 864         dir = CTINFO2DIR(ctinfo);
 865         index = get_conntrack_index(th);
 866         new_state = tcp_conntracks[dir][index][old_state];
 867
 868         switch (new_state) {
 869         case TCP_CONNTRACK_IGNORE:
 870                 /* Ignored packets:
 871                  *
 872                  * a) SYN in ORIGINAL
 873                  * b) SYN/ACK in REPLY
 874                  * c) ACK in reply direction after initial SYN in original.
 875                  */
 876                 if (index == TCP_SYNACK_SET
 877                     && conntrack->proto.tcp.last_index == TCP_SYN_SET
 878                     && conntrack->proto.tcp.last_dir != dir
 879                     && ntohl(th->ack_seq) ==
 880                              conntrack->proto.tcp.last_end) {
 881                         /* This SYN/ACK acknowledges a SYN that we earlier
 882                          * ignored as invalid. This means that the client and
 883                          * the server are both in sync, while the firewall is
 884                          * not. We kill this session and block the SYN/ACK so
 885                          * that the client cannot but retransmit its SYN and
 886                          * thus initiate a clean new session.
 887                          */
 888                         write_unlock_bh(&tcp_lock);
 889                         if (LOG_INVALID(IPPROTO_TCP))
 890                                 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 891                                           "nf_ct_tcp: killing out of sync session ");
 892                         if (del_timer(&conntrack->timeout))
 893                                 conntrack->timeout.function((unsigned long)
 894                                                             conntrack);
 895                         return -NF_DROP;
 896                 }
 897                 conntrack->proto.tcp.last_index = index;
 898                 conntrack->proto.tcp.last_dir = dir;
 899                 conntrack->proto.tcp.last_seq = ntohl(th->seq);
 900                 conntrack->proto.tcp.last_end =
 901                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 902
 903                 write_unlock_bh(&tcp_lock);
 904                 if (LOG_INVALID(IPPROTO_TCP))
 905                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 906                                   "nf_ct_tcp: invalid packed ignored ");
 907                 return NF_ACCEPT;
 908         case TCP_CONNTRACK_MAX:
 909                 /* Invalid packet */
 910                 DEBUGP("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 911                        dir, get_conntrack_index(th),
 912                        old_state);
 913                 write_unlock_bh(&tcp_lock);
 914                 if (LOG_INVALID(IPPROTO_TCP))
 915                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 916                                   "nf_ct_tcp: invalid state ");
 917                 return -NF_ACCEPT;
 918         case TCP_CONNTRACK_SYN_SENT:
 919                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 920                         break;
 921                 if ((conntrack->proto.tcp.seen[dir].flags &
 922                         IP_CT_TCP_FLAG_CLOSE_INIT)
 923                     || after(ntohl(th->seq),
 924                              conntrack->proto.tcp.seen[dir].td_end)) {
 925                         /* Attempt to reopen a closed connection.
 926                         * Delete this connection and look up again. */
 927                         write_unlock_bh(&tcp_lock);
 928                         if (del_timer(&conntrack->timeout))
 929                                 conntrack->timeout.function((unsigned long)
 930                                                             conntrack);
 931                         return -NF_REPEAT;
 932                 } else {
 933                         write_unlock_bh(&tcp_lock);
 934                         if (LOG_INVALID(IPPROTO_TCP))
 935                                 nf_log_packet(pf, 0, skb, NULL, NULL,
 936                                               NULL, "nf_ct_tcp: invalid SYN");
 937                         return -NF_ACCEPT;
 938                 }
 939         case TCP_CONNTRACK_CLOSE:
 940                 if (index == TCP_RST_SET
 941                     && ((test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
 942                          && conntrack->proto.tcp.last_index == TCP_SYN_SET)
 943                         || (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
 944                             && conntrack->proto.tcp.last_index == TCP_ACK_SET))
 945                     && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
 946                         /* RST sent to invalid SYN or ACK we had let through
 947                          * at a) and c) above:
 948                          *
 949                          * a) SYN was in window then
 950                          * c) we hold a half-open connection.
 951                          *
 952                          * Delete our connection entry.
 953                          * We skip window checking, because packet might ACK
 954                          * segments we ignored. */
 955                         goto in_window;
 956                 }
 957                 /* Just fall through */
 958         default:
 959                 /* Keep compilers happy. */
 960                 break;
 961         }
 962
 963         if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
 964                            skb, dataoff, th, pf)) {
 965                 write_unlock_bh(&tcp_lock);
 966                 return -NF_ACCEPT;
 967         }
 968      in_window:
 969         /* From now on we have got in-window packets */
 970         conntrack->proto.tcp.last_index = index;
 971
 972         DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
 973                "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
 974                 NIPQUAD(iph->saddr), ntohs(th->source),
 975                 NIPQUAD(iph->daddr), ntohs(th->dest),
 976                 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
 977                 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
 978                 old_state, new_state);
 979
 980         conntrack->proto.tcp.state = new_state;
 981         if (old_state != new_state
 982             && (new_state == TCP_CONNTRACK_FIN_WAIT
 983                 || new_state == TCP_CONNTRACK_CLOSE))
 984                 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
 985         timeout = conntrack->proto.tcp.retrans >= nf_ct_tcp_max_retrans
 986                   && *tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans
 987                   ? nf_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
 988         write_unlock_bh(&tcp_lock);
 989
 990         nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
 991         if (new_state != old_state)
 992                 nf_conntrack_event_cache(IPCT_PROTOINFO, skb);
 993
 994         if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
 995                 /* If only reply is a RST, we can consider ourselves not to
 996                    have an established connection: this is a fairly common
 997                    problem case, so we can delete the conntrack
 998                    immediately.  --RR */
 999                 if (th->rst) {
1000                         if (del_timer(&conntrack->timeout))
1001                                 conntrack->timeout.function((unsigned long)
1002                                                             conntrack);
1003                         return NF_ACCEPT;
1004                 }
1005         } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
1006                    && (old_state == TCP_CONNTRACK_SYN_RECV
1007                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1008                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1009                 /* Set ASSURED if we see see valid ack in ESTABLISHED
1010                    after SYN_RECV or a valid answer for a picked up
1011                    connection. */
1012                 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1013                 nf_conntrack_event_cache(IPCT_STATUS, skb);
1014         }
1015         nf_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1016
1017         return NF_ACCEPT;
1018 }
1019
1020 /* Called when a new connection for this protocol found. */
1021 static int tcp_new(struct nf_conn *conntrack,
1022                    const struct sk_buff *skb,
1023                    unsigned int dataoff)
1024 {
1025         enum tcp_conntrack new_state;
1026         struct tcphdr *th, _tcph;
1027 #ifdef DEBUGP_VARS
1028         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1029         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1030 #endif
1031
1032         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1033         BUG_ON(th == NULL);
1034
1035         /* Don't need lock here: this conntrack not in circulation yet */
1036         new_state
1037                 = tcp_conntracks[0][get_conntrack_index(th)]
1038                 [TCP_CONNTRACK_NONE];
1039
1040         /* Invalid: delete conntrack */
1041         if (new_state >= TCP_CONNTRACK_MAX) {
1042                 DEBUGP("nf_ct_tcp: invalid new deleting.\n");
1043                 return 0;
1044         }
1045
1046         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1047                 /* SYN packet */
1048                 conntrack->proto.tcp.seen[0].td_end =
1049                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1050                                              dataoff, th);
1051                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1052                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1053                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1054                 conntrack->proto.tcp.seen[0].td_maxend =
1055                         conntrack->proto.tcp.seen[0].td_end;
1056
1057                 tcp_options(skb, dataoff, th, &conntrack->proto.tcp.seen[0]);
1058                 conntrack->proto.tcp.seen[1].flags = 0;
1059         } else if (nf_ct_tcp_loose == 0) {
1060                 /* Don't try to pick up connections. */
1061                 return 0;
1062         } else {
1063                 /*
1064                  * We are in the middle of a connection,
1065                  * its history is lost for us.
1066                  * Let's try to use the data from the packet.
1067                  */
1068                 conntrack->proto.tcp.seen[0].td_end =
1069                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1070                                              dataoff, th);
1071                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1072                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1073                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1074                 conntrack->proto.tcp.seen[0].td_maxend =
1075                         conntrack->proto.tcp.seen[0].td_end +
1076                         conntrack->proto.tcp.seen[0].td_maxwin;
1077                 conntrack->proto.tcp.seen[0].td_scale = 0;
1078
1079                 /* We assume SACK and liberal window checking to handle
1080                  * window scaling */
1081                 conntrack->proto.tcp.seen[0].flags =
1082                 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1083                                                      IP_CT_TCP_FLAG_BE_LIBERAL;
1084         }
1085
1086         conntrack->proto.tcp.seen[1].td_end = 0;
1087         conntrack->proto.tcp.seen[1].td_maxend = 0;
1088         conntrack->proto.tcp.seen[1].td_maxwin = 1;
1089         conntrack->proto.tcp.seen[1].td_scale = 0;
1090
1091         /* tcp_packet will set them */
1092         conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1093         conntrack->proto.tcp.last_index = TCP_NONE_SET;
1094
1095         DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1096                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1097                 sender->td_end, sender->td_maxend, sender->td_maxwin,
1098                 sender->td_scale,
1099                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1100                 receiver->td_scale);
1101         return 1;
1102 }
1103
1104 #if defined(CONFIG_NF_CT_NETLINK) || 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) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1383         .to_nfattr              = tcp_to_nfattr,
1384         .from_nfattr            = nfattr_to_tcp,
1385         .tuple_to_nfattr        = nf_ct_port_tuple_to_nfattr,
1386         .nfattr_to_tuple        = nf_ct_port_nfattr_to_tuple,
1387 #endif
1388 #ifdef CONFIG_SYSCTL
1389         .ctl_table_users        = &tcp_sysctl_table_users,
1390         .ctl_table_header       = &tcp_sysctl_header,
1391         .ctl_table              = tcp_sysctl_table,
1392 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1393         .ctl_compat_table       = tcp_compat_sysctl_table,
1394 #endif
1395 #endif
1396 };
1397 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1398
1399 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 =
1400 {
1401         .l3proto                = PF_INET6,
1402         .l4proto                = IPPROTO_TCP,
1403         .name                   = "tcp",
1404         .pkt_to_tuple           = tcp_pkt_to_tuple,
1405         .invert_tuple           = tcp_invert_tuple,
1406         .print_tuple            = tcp_print_tuple,
1407         .print_conntrack        = tcp_print_conntrack,
1408         .packet                 = tcp_packet,
1409         .new                    = tcp_new,
1410         .error                  = tcp_error,
1411 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1412         .to_nfattr              = tcp_to_nfattr,
1413         .from_nfattr            = nfattr_to_tcp,
1414         .tuple_to_nfattr        = nf_ct_port_tuple_to_nfattr,
1415         .nfattr_to_tuple        = nf_ct_port_nfattr_to_tuple,
1416 #endif
1417 #ifdef CONFIG_SYSCTL
1418         .ctl_table_users        = &tcp_sysctl_table_users,
1419         .ctl_table_header       = &tcp_sysctl_header,
1420         .ctl_table              = tcp_sysctl_table,
1421 #endif
1422 };
1423 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);