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
   9 #include <linux/types.h>
  10 #include <linux/timer.h>
  11 #include <linux/module.h>
  12 #include <linux/in.h>
  13 #include <linux/tcp.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/ipv6.h>
  17 #include <net/ip6_checksum.h>
  18 #include <asm/unaligned.h>
  19
  20 #include <net/tcp.h>
  21
  22 #include <linux/netfilter.h>
  23 #include <linux/netfilter_ipv4.h>
  24 #include <linux/netfilter_ipv6.h>
  25 #include <net/netfilter/nf_conntrack.h>
  26 #include <net/netfilter/nf_conntrack_l4proto.h>
  27 #include <net/netfilter/nf_conntrack_ecache.h>
  28 #include <net/netfilter/nf_log.h>
  29
  30 /* Protects ct->proto.tcp */
  31 static DEFINE_RWLOCK(tcp_lock);
  32
  33 /* "Be conservative in what you do,
  34     be liberal in what you accept from others."
  35     If it's non-zero, we mark only out of window RST segments as INVALID. */
  36 static int nf_ct_tcp_be_liberal __read_mostly = 0;
  37
  38 /* If it is set to zero, we disable picking up already established
  39    connections. */
  40 static int nf_ct_tcp_loose __read_mostly = 1;
  41
  42 /* Max number of the retransmitted packets without receiving an (acceptable)
  43    ACK from the destination. If this number is reached, a shorter timer
  44    will be started. */
  45 static int nf_ct_tcp_max_retrans __read_mostly = 3;
  46
  47   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  48      closely.  They're more complex. --RR */
  49
  50 static const char *const tcp_conntrack_names[] = {
  51         "NONE",
  52         "SYN_SENT",
  53         "SYN_RECV",
  54         "ESTABLISHED",
  55         "FIN_WAIT",
  56         "CLOSE_WAIT",
  57         "LAST_ACK",
  58         "TIME_WAIT",
  59         "CLOSE",
  60         "LISTEN"
  61 };
  62
  63 #define SECS * HZ
  64 #define MINS * 60 SECS
  65 #define HOURS * 60 MINS
  66 #define DAYS * 24 HOURS
  67
  68 /* RFC1122 says the R2 limit should be at least 100 seconds.
  69    Linux uses 15 packets as limit, which corresponds
  70    to ~13-30min depending on RTO. */
  71 static unsigned int nf_ct_tcp_timeout_max_retrans __read_mostly    =   5 MINS;
  72 static unsigned int nf_ct_tcp_timeout_unacknowledged __read_mostly =   5 MINS;
  73
  74 static unsigned int tcp_timeouts[TCP_CONNTRACK_MAX] __read_mostly = {
  75         [TCP_CONNTRACK_SYN_SENT]        = 2 MINS,
  76         [TCP_CONNTRACK_SYN_RECV]        = 60 SECS,
  77         [TCP_CONNTRACK_ESTABLISHED]     = 5 DAYS,
  78         [TCP_CONNTRACK_FIN_WAIT]        = 2 MINS,
  79         [TCP_CONNTRACK_CLOSE_WAIT]      = 60 SECS,
  80         [TCP_CONNTRACK_LAST_ACK]        = 30 SECS,
  81         [TCP_CONNTRACK_TIME_WAIT]       = 2 MINS,
  82         [TCP_CONNTRACK_CLOSE]           = 10 SECS,
  83 };
  84
  85 #define sNO TCP_CONNTRACK_NONE
  86 #define sSS TCP_CONNTRACK_SYN_SENT
  87 #define sSR TCP_CONNTRACK_SYN_RECV
  88 #define sES TCP_CONNTRACK_ESTABLISHED
  89 #define sFW TCP_CONNTRACK_FIN_WAIT
  90 #define sCW TCP_CONNTRACK_CLOSE_WAIT
  91 #define sLA TCP_CONNTRACK_LAST_ACK
  92 #define sTW TCP_CONNTRACK_TIME_WAIT
  93 #define sCL TCP_CONNTRACK_CLOSE
  94 #define sLI TCP_CONNTRACK_LISTEN
  95 #define sIV TCP_CONNTRACK_MAX
  96 #define sIG TCP_CONNTRACK_IGNORE
  97
  98 /* What TCP flags are set from RST/SYN/FIN/ACK. */
  99 enum tcp_bit_set {
 100         TCP_SYN_SET,
 101         TCP_SYNACK_SET,
 102         TCP_FIN_SET,
 103         TCP_ACK_SET,
 104         TCP_RST_SET,
 105         TCP_NONE_SET,
 106 };
 107
 108 /*
 109  * The TCP state transition table needs a few words...
 110  *
 111  * We are the man in the middle. All the packets go through us
 112  * but might get lost in transit to the destination.
 113  * It is assumed that the destinations can't receive segments
 114  * we haven't seen.
 115  *
 116  * The checked segment is in window, but our windows are *not*
 117  * equivalent with the ones of the sender/receiver. We always
 118  * try to guess the state of the current sender.
 119  *
 120  * The meaning of the states are:
 121  *
 122  * NONE:        initial state
 123  * SYN_SENT:    SYN-only packet seen
 124  * SYN_RECV:    SYN-ACK packet seen
 125  * ESTABLISHED: ACK packet seen
 126  * FIN_WAIT:    FIN packet seen
 127  * CLOSE_WAIT:  ACK seen (after FIN)
 128  * LAST_ACK:    FIN seen (after FIN)
 129  * TIME_WAIT:   last ACK seen
 130  * CLOSE:       closed connection (RST)
 131  *
 132  * LISTEN state is not used.
 133  *
 134  * Packets marked as IGNORED (sIG):
 135  *      if they may be either invalid or valid
 136  *      and the receiver may send back a connection
 137  *      closing RST or a SYN/ACK.
 138  *
 139  * Packets marked as INVALID (sIV):
 140  *      if they are invalid
 141  *      or we do not support the request (simultaneous open)
 142  */
 143 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 144         {
 145 /* ORIGINAL */
 146 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 147 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
 148 /*
 149  *      sNO -> sSS      Initialize a new connection
 150  *      sSS -> sSS      Retransmitted SYN
 151  *      sSR -> sIG      Late retransmitted SYN?
 152  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 153  *                      are errors. Receiver will reply with RST
 154  *                      and close the connection.
 155  *                      Or we are not in sync and hold a dead connection.
 156  *      sFW -> sIG
 157  *      sCW -> sIG
 158  *      sLA -> sIG
 159  *      sTW -> sSS      Reopened connection (RFC 1122).
 160  *      sCL -> sSS
 161  */
 162 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 163 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 164 /*
 165  * A SYN/ACK from the client is always invalid:
 166  *      - either it tries to set up a simultaneous open, which is
 167  *        not supported;
 168  *      - or the firewall has just been inserted between the two hosts
 169  *        during the session set-up. The SYN will be retransmitted
 170  *        by the true client (or it'll time out).
 171  */
 172 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 173 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 174 /*
 175  *      sNO -> sIV      Too late and no reason to do anything...
 176  *      sSS -> sIV      Client migth not send FIN in this state:
 177  *                      we enforce waiting for a SYN/ACK reply first.
 178  *      sSR -> sFW      Close started.
 179  *      sES -> sFW
 180  *      sFW -> sLA      FIN seen in both directions, waiting for
 181  *                      the last ACK.
 182  *                      Migth be a retransmitted FIN as well...
 183  *      sCW -> sLA
 184  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 185  *      sTW -> sTW
 186  *      sCL -> sCL
 187  */
 188 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 189 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 190 /*
 191  *      sNO -> sES      Assumed.
 192  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 193  *      sSR -> sES      Established state is reached.
 194  *      sES -> sES      :-)
 195  *      sFW -> sCW      Normal close request answered by ACK.
 196  *      sCW -> sCW
 197  *      sLA -> sTW      Last ACK detected.
 198  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 199  *      sCL -> sCL
 200  */
 201 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 202 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 203 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 204         },
 205         {
 206 /* REPLY */
 207 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 208 /*syn*/    { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 209 /*
 210  *      sNO -> sIV      Never reached.
 211  *      sSS -> sIV      Simultaneous open, not supported
 212  *      sSR -> sIV      Simultaneous open, not supported.
 213  *      sES -> sIV      Server may not initiate a connection.
 214  *      sFW -> sIV
 215  *      sCW -> sIV
 216  *      sLA -> sIV
 217  *      sTW -> sIV      Reopened connection, but server may not do it.
 218  *      sCL -> sIV
 219  */
 220 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 221 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
 222 /*
 223  *      sSS -> sSR      Standard open.
 224  *      sSR -> sSR      Retransmitted SYN/ACK.
 225  *      sES -> sIG      Late retransmitted SYN/ACK?
 226  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 227  *      sCW -> sIG
 228  *      sLA -> sIG
 229  *      sTW -> sIG
 230  *      sCL -> sIG
 231  */
 232 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 233 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 234 /*
 235  *      sSS -> sIV      Server might not send FIN in this state.
 236  *      sSR -> sFW      Close started.
 237  *      sES -> sFW
 238  *      sFW -> sLA      FIN seen in both directions.
 239  *      sCW -> sLA
 240  *      sLA -> sLA      Retransmitted FIN.
 241  *      sTW -> sTW
 242  *      sCL -> sCL
 243  */
 244 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 245 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 246 /*
 247  *      sSS -> sIG      Might be a half-open connection.
 248  *      sSR -> sSR      Might answer late resent SYN.
 249  *      sES -> sES      :-)
 250  *      sFW -> sCW      Normal close request answered by ACK.
 251  *      sCW -> sCW
 252  *      sLA -> sTW      Last ACK detected.
 253  *      sTW -> sTW      Retransmitted last ACK.
 254  *      sCL -> sCL
 255  */
 256 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 257 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 258 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 259         }
 260 };
 261
 262 static bool tcp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
 263                              struct nf_conntrack_tuple *tuple)
 264 {
 265         const struct tcphdr *hp;
 266         struct tcphdr _hdr;
 267
 268         /* Actually only need first 8 bytes. */
 269         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
 270         if (hp == NULL)
 271                 return false;
 272
 273         tuple->src.u.tcp.port = hp->source;
 274         tuple->dst.u.tcp.port = hp->dest;
 275
 276         return true;
 277 }
 278
 279 static bool tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
 280                              const struct nf_conntrack_tuple *orig)
 281 {
 282         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
 283         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
 284         return true;
 285 }
 286
 287 /* Print out the per-protocol part of the tuple. */
 288 static int tcp_print_tuple(struct seq_file *s,
 289                            const struct nf_conntrack_tuple *tuple)
 290 {
 291         return seq_printf(s, "sport=%hu dport=%hu ",
 292                           ntohs(tuple->src.u.tcp.port),
 293                           ntohs(tuple->dst.u.tcp.port));
 294 }
 295
 296 /* Print out the private part of the conntrack. */
 297 static int tcp_print_conntrack(struct seq_file *s, const struct nf_conn *ct)
 298 {
 299         enum tcp_conntrack state;
 300
 301         read_lock_bh(&tcp_lock);
 302         state = ct->proto.tcp.state;
 303         read_unlock_bh(&tcp_lock);
 304
 305         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
 306 }
 307
 308 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 309 {
 310         if (tcph->rst) return TCP_RST_SET;
 311         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 312         else if (tcph->fin) return TCP_FIN_SET;
 313         else if (tcph->ack) return TCP_ACK_SET;
 314         else return TCP_NONE_SET;
 315 }
 316
 317 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 318    in IP Filter' by Guido van Rooij.
 319
 320    http://www.nluug.nl/events/sane2000/papers.html
 321    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
 322
 323    The boundaries and the conditions are changed according to RFC793:
 324    the packet must intersect the window (i.e. segments may be
 325    after the right or before the left edge) and thus receivers may ACK
 326    segments after the right edge of the window.
 327
 328         td_maxend = max(sack + max(win,1)) seen in reply packets
 329         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 330         td_maxwin += seq + len - sender.td_maxend
 331                         if seq + len > sender.td_maxend
 332         td_end    = max(seq + len) seen in sent packets
 333
 334    I.   Upper bound for valid data:     seq <= sender.td_maxend
 335    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 336    III. Upper bound for valid (s)ack:   sack <= receiver.td_end
 337    IV.  Lower bound for valid (s)ack:   sack >= receiver.td_end - MAXACKWINDOW
 338
 339    where sack is the highest right edge of sack block found in the packet
 340    or ack in the case of packet without SACK option.
 341
 342    The upper bound limit for a valid (s)ack is not ignored -
 343    we doesn't have to deal with fragments.
 344 */
 345
 346 static inline __u32 segment_seq_plus_len(__u32 seq,
 347                                          size_t len,
 348                                          unsigned int dataoff,
 349                                          const struct tcphdr *tcph)
 350 {
 351         /* XXX Should I use payload length field in IP/IPv6 header ?
 352          * - YK */
 353         return (seq + len - dataoff - tcph->doff*4
 354                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 355 }
 356
 357 /* Fixme: what about big packets? */
 358 #define MAXACKWINCONST                  66000
 359 #define MAXACKWINDOW(sender)                                            \
 360         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 361                                               : MAXACKWINCONST)
 362
 363 /*
 364  * Simplified tcp_parse_options routine from tcp_input.c
 365  */
 366 static void tcp_options(const struct sk_buff *skb,
 367                         unsigned int dataoff,
 368                         const struct tcphdr *tcph,
 369                         struct ip_ct_tcp_state *state)
 370 {
 371         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 372         const unsigned char *ptr;
 373         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 374
 375         if (!length)
 376                 return;
 377
 378         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 379                                  length, buff);
 380         BUG_ON(ptr == NULL);
 381
 382         state->td_scale =
 383         state->flags = 0;
 384
 385         while (length > 0) {
 386                 int opcode=*ptr++;
 387                 int opsize;
 388
 389                 switch (opcode) {
 390                 case TCPOPT_EOL:
 391                         return;
 392                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 393                         length--;
 394                         continue;
 395                 default:
 396                         opsize=*ptr++;
 397                         if (opsize < 2) /* "silly options" */
 398                                 return;
 399                         if (opsize > length)
 400                                 break;  /* don't parse partial options */
 401
 402                         if (opcode == TCPOPT_SACK_PERM
 403                             && opsize == TCPOLEN_SACK_PERM)
 404                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 405                         else if (opcode == TCPOPT_WINDOW
 406                                  && opsize == TCPOLEN_WINDOW) {
 407                                 state->td_scale = *(u_int8_t *)ptr;
 408
 409                                 if (state->td_scale > 14) {
 410                                         /* See RFC1323 */
 411                                         state->td_scale = 14;
 412                                 }
 413                                 state->flags |=
 414                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 415                         }
 416                         ptr += opsize - 2;
 417                         length -= opsize;
 418                 }
 419         }
 420 }
 421
 422 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 423                      const struct tcphdr *tcph, __u32 *sack)
 424 {
 425         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 426         const unsigned char *ptr;
 427         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 428         __u32 tmp;
 429
 430         if (!length)
 431                 return;
 432
 433         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 434                                  length, buff);
 435         BUG_ON(ptr == NULL);
 436
 437         /* Fast path for timestamp-only option */
 438         if (length == TCPOLEN_TSTAMP_ALIGNED*4
 439             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
 440                                        | (TCPOPT_NOP << 16)
 441                                        | (TCPOPT_TIMESTAMP << 8)
 442                                        | TCPOLEN_TIMESTAMP))
 443                 return;
 444
 445         while (length > 0) {
 446                 int opcode = *ptr++;
 447                 int opsize, i;
 448
 449                 switch (opcode) {
 450                 case TCPOPT_EOL:
 451                         return;
 452                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 453                         length--;
 454                         continue;
 455                 default:
 456                         opsize = *ptr++;
 457                         if (opsize < 2) /* "silly options" */
 458                                 return;
 459                         if (opsize > length)
 460                                 break;  /* don't parse partial options */
 461
 462                         if (opcode == TCPOPT_SACK
 463                             && opsize >= (TCPOLEN_SACK_BASE
 464                                           + TCPOLEN_SACK_PERBLOCK)
 465                             && !((opsize - TCPOLEN_SACK_BASE)
 466                                  % TCPOLEN_SACK_PERBLOCK)) {
 467                                 for (i = 0;
 468                                      i < (opsize - TCPOLEN_SACK_BASE);
 469                                      i += TCPOLEN_SACK_PERBLOCK) {
 470                                         tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
 471
 472                                         if (after(tmp, *sack))
 473                                                 *sack = tmp;
 474                                 }
 475                                 return;
 476                         }
 477                         ptr += opsize - 2;
 478                         length -= opsize;
 479                 }
 480         }
 481 }
 482
 483 static bool tcp_in_window(const struct nf_conn *ct,
 484                           struct ip_ct_tcp *state,
 485                           enum ip_conntrack_dir dir,
 486                           unsigned int index,
 487                           const struct sk_buff *skb,
 488                           unsigned int dataoff,
 489                           const struct tcphdr *tcph,
 490                           int pf)
 491 {
 492         struct ip_ct_tcp_state *sender = &state->seen[dir];
 493         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 494         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 495         __u32 seq, ack, sack, end, win, swin;
 496         bool res;
 497
 498         /*
 499          * Get the required data from the packet.
 500          */
 501         seq = ntohl(tcph->seq);
 502         ack = sack = ntohl(tcph->ack_seq);
 503         win = ntohs(tcph->window);
 504         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 505
 506         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 507                 tcp_sack(skb, dataoff, tcph, &sack);
 508
 509         pr_debug("tcp_in_window: START\n");
 510         pr_debug("tcp_in_window: ");
 511         nf_ct_dump_tuple(tuple);
 512         pr_debug("seq=%u ack=%u sack=%u win=%u end=%u\n",
 513                  seq, ack, sack, win, end);
 514         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 515                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 516                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 517                  sender->td_scale,
 518                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 519                  receiver->td_scale);
 520
 521         if (sender->td_end == 0) {
 522                 /*
 523                  * Initialize sender data.
 524                  */
 525                 if (tcph->syn && tcph->ack) {
 526                         /*
 527                          * Outgoing SYN-ACK in reply to a SYN.
 528                          */
 529                         sender->td_end =
 530                         sender->td_maxend = end;
 531                         sender->td_maxwin = (win == 0 ? 1 : win);
 532
 533                         tcp_options(skb, dataoff, tcph, sender);
 534                         /*
 535                          * RFC 1323:
 536                          * Both sides must send the Window Scale option
 537                          * to enable window scaling in either direction.
 538                          */
 539                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 540                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 541                                 sender->td_scale =
 542                                 receiver->td_scale = 0;
 543                 } else {
 544                         /*
 545                          * We are in the middle of a connection,
 546                          * its history is lost for us.
 547                          * Let's try to use the data from the packet.
 548                          */
 549                         sender->td_end = end;
 550                         sender->td_maxwin = (win == 0 ? 1 : win);
 551                         sender->td_maxend = end + sender->td_maxwin;
 552                 }
 553         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 554                      && dir == IP_CT_DIR_ORIGINAL)
 555                    || (state->state == TCP_CONNTRACK_SYN_RECV
 556                      && dir == IP_CT_DIR_REPLY))
 557                    && after(end, sender->td_end)) {
 558                 /*
 559                  * RFC 793: "if a TCP is reinitialized ... then it need
 560                  * not wait at all; it must only be sure to use sequence
 561                  * numbers larger than those recently used."
 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
 570         if (!(tcph->ack)) {
 571                 /*
 572                  * If there is no ACK, just pretend it was set and OK.
 573                  */
 574                 ack = sack = receiver->td_end;
 575         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 576                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 577                    && (ack == 0)) {
 578                 /*
 579                  * Broken TCP stacks, that set ACK in RST packets as well
 580                  * with zero ack value.
 581                  */
 582                 ack = sack = receiver->td_end;
 583         }
 584
 585         if (seq == end
 586             && (!tcph->rst
 587                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
 588                 /*
 589                  * Packets contains no data: we assume it is valid
 590                  * and check the ack value only.
 591                  * However RST segments are always validated by their
 592                  * SEQ number, except when seq == 0 (reset sent answering
 593                  * SYN.
 594                  */
 595                 seq = end = sender->td_end;
 596
 597         pr_debug("tcp_in_window: ");
 598         nf_ct_dump_tuple(tuple);
 599         pr_debug("seq=%u ack=%u sack =%u win=%u end=%u\n",
 600                  seq, ack, sack, win, end);
 601         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 602                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 603                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 604                  sender->td_scale,
 605                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 606                  receiver->td_scale);
 607
 608         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 609                  before(seq, sender->td_maxend + 1),
 610                  after(end, sender->td_end - receiver->td_maxwin - 1),
 611                  before(sack, receiver->td_end + 1),
 612                  after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 613
 614         if (before(seq, sender->td_maxend + 1) &&
 615             after(end, sender->td_end - receiver->td_maxwin - 1) &&
 616             before(sack, receiver->td_end + 1) &&
 617             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
 618                 /*
 619                  * Take into account window scaling (RFC 1323).
 620                  */
 621                 if (!tcph->syn)
 622                         win <<= sender->td_scale;
 623
 624                 /*
 625                  * Update sender data.
 626                  */
 627                 swin = win + (sack - ack);
 628                 if (sender->td_maxwin < swin)
 629                         sender->td_maxwin = swin;
 630                 if (after(end, sender->td_end)) {
 631                         sender->td_end = end;
 632                         sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 633                 }
 634                 /*
 635                  * Update receiver data.
 636                  */
 637                 if (after(end, sender->td_maxend))
 638                         receiver->td_maxwin += end - sender->td_maxend;
 639                 if (after(sack + win, receiver->td_maxend - 1)) {
 640                         receiver->td_maxend = sack + win;
 641                         if (win == 0)
 642                                 receiver->td_maxend++;
 643                 }
 644                 if (ack == receiver->td_end)
 645                         receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 646
 647                 /*
 648                  * Check retransmissions.
 649                  */
 650                 if (index == TCP_ACK_SET) {
 651                         if (state->last_dir == dir
 652                             && state->last_seq == seq
 653                             && state->last_ack == ack
 654                             && state->last_end == end
 655                             && state->last_win == win)
 656                                 state->retrans++;
 657                         else {
 658                                 state->last_dir = dir;
 659                                 state->last_seq = seq;
 660                                 state->last_ack = ack;
 661                                 state->last_end = end;
 662                                 state->last_win = win;
 663                                 state->retrans = 0;
 664                         }
 665                 }
 666                 res = true;
 667         } else {
 668                 res = false;
 669                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 670                     nf_ct_tcp_be_liberal)
 671                         res = true;
 672                 if (!res && LOG_INVALID(IPPROTO_TCP))
 673                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 674                         "nf_ct_tcp: %s ",
 675                         before(seq, sender->td_maxend + 1) ?
 676                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
 677                         before(sack, receiver->td_end + 1) ?
 678                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
 679                         : "ACK is under the lower bound (possible overly delayed ACK)"
 680                         : "ACK is over the upper bound (ACKed data not seen yet)"
 681                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 682                         : "SEQ is over the upper bound (over the window of the receiver)");
 683         }
 684
 685         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 686                  "receiver end=%u maxend=%u maxwin=%u\n",
 687                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 688                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 689
 690         return res;
 691 }
 692
 693 #ifdef CONFIG_NF_NAT_NEEDED
 694 /* Update sender->td_end after NAT successfully mangled the packet */
 695 /* Caller must linearize skb at tcp header. */
 696 void nf_conntrack_tcp_update(const struct sk_buff *skb,
 697                              unsigned int dataoff,
 698                              struct nf_conn *ct,
 699                              int dir)
 700 {
 701         const struct tcphdr *tcph = (const void *)skb->data + dataoff;
 702         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[dir];
 703         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[!dir];
 704         __u32 end;
 705
 706         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
 707
 708         write_lock_bh(&tcp_lock);
 709         /*
 710          * We have to worry for the ack in the reply packet only...
 711          */
 712         if (after(end, ct->proto.tcp.seen[dir].td_end))
 713                 ct->proto.tcp.seen[dir].td_end = end;
 714         ct->proto.tcp.last_end = end;
 715         write_unlock_bh(&tcp_lock);
 716         pr_debug("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
 717                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 718                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 719                  sender->td_scale,
 720                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 721                  receiver->td_scale);
 722 }
 723 EXPORT_SYMBOL_GPL(nf_conntrack_tcp_update);
 724 #endif
 725
 726 #define TH_FIN  0x01
 727 #define TH_SYN  0x02
 728 #define TH_RST  0x04
 729 #define TH_PUSH 0x08
 730 #define TH_ACK  0x10
 731 #define TH_URG  0x20
 732 #define TH_ECE  0x40
 733 #define TH_CWR  0x80
 734
 735 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 736 static const u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_ACK|TH_URG) + 1] =
 737 {
 738         [TH_SYN]                        = 1,
 739         [TH_SYN|TH_URG]                 = 1,
 740         [TH_SYN|TH_ACK]                 = 1,
 741         [TH_RST]                        = 1,
 742         [TH_RST|TH_ACK]                 = 1,
 743         [TH_FIN|TH_ACK]                 = 1,
 744         [TH_FIN|TH_ACK|TH_URG]          = 1,
 745         [TH_ACK]                        = 1,
 746         [TH_ACK|TH_URG]                 = 1,
 747 };
 748
 749 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 750 static int tcp_error(struct sk_buff *skb,
 751                      unsigned int dataoff,
 752                      enum ip_conntrack_info *ctinfo,
 753                      int pf,
 754                      unsigned int hooknum)
 755 {
 756         const struct tcphdr *th;
 757         struct tcphdr _tcph;
 758         unsigned int tcplen = skb->len - dataoff;
 759         u_int8_t tcpflags;
 760
 761         /* Smaller that minimal TCP header? */
 762         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 763         if (th == NULL) {
 764                 if (LOG_INVALID(IPPROTO_TCP))
 765                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 766                                 "nf_ct_tcp: short packet ");
 767                 return -NF_ACCEPT;
 768         }
 769
 770         /* Not whole TCP header or malformed packet */
 771         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 772                 if (LOG_INVALID(IPPROTO_TCP))
 773                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 774                                 "nf_ct_tcp: truncated/malformed packet ");
 775                 return -NF_ACCEPT;
 776         }
 777
 778         /* Checksum invalid? Ignore.
 779          * We skip checking packets on the outgoing path
 780          * because the checksum is assumed to be correct.
 781          */
 782         /* FIXME: Source route IP option packets --RR */
 783         if (nf_conntrack_checksum && hooknum == NF_INET_PRE_ROUTING &&
 784             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
 785                 if (LOG_INVALID(IPPROTO_TCP))
 786                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 787                                   "nf_ct_tcp: bad TCP checksum ");
 788                 return -NF_ACCEPT;
 789         }
 790
 791         /* Check TCP flags. */
 792         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR|TH_PUSH));
 793         if (!tcp_valid_flags[tcpflags]) {
 794                 if (LOG_INVALID(IPPROTO_TCP))
 795                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 796                                   "nf_ct_tcp: invalid TCP flag combination ");
 797                 return -NF_ACCEPT;
 798         }
 799
 800         return NF_ACCEPT;
 801 }
 802
 803 /* Returns verdict for packet, or -1 for invalid. */
 804 static int tcp_packet(struct nf_conn *ct,
 805                       const struct sk_buff *skb,
 806                       unsigned int dataoff,
 807                       enum ip_conntrack_info ctinfo,
 808                       int pf,
 809                       unsigned int hooknum)
 810 {
 811         struct nf_conntrack_tuple *tuple;
 812         enum tcp_conntrack new_state, old_state;
 813         enum ip_conntrack_dir dir;
 814         const struct tcphdr *th;
 815         struct tcphdr _tcph;
 816         unsigned long timeout;
 817         unsigned int index;
 818
 819         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 820         BUG_ON(th == NULL);
 821
 822         write_lock_bh(&tcp_lock);
 823         old_state = ct->proto.tcp.state;
 824         dir = CTINFO2DIR(ctinfo);
 825         index = get_conntrack_index(th);
 826         new_state = tcp_conntracks[dir][index][old_state];
 827         tuple = &ct->tuplehash[dir].tuple;
 828
 829         switch (new_state) {
 830         case TCP_CONNTRACK_SYN_SENT:
 831                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 832                         break;
 833                 /* RFC 1122: "When a connection is closed actively,
 834                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 835                  * (Maximum Segment Lifetime). However, it MAY accept
 836                  * a new SYN from the remote TCP to reopen the connection
 837                  * directly from TIME-WAIT state, if..."
 838                  * We ignore the conditions because we are in the
 839                  * TIME-WAIT state anyway.
 840                  *
 841                  * Handle aborted connections: we and the server
 842                  * think there is an existing connection but the client
 843                  * aborts it and starts a new one.
 844                  */
 845                 if (((ct->proto.tcp.seen[dir].flags
 846                       | ct->proto.tcp.seen[!dir].flags)
 847                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 848                     || (ct->proto.tcp.last_dir == dir
 849                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 850                         /* Attempt to reopen a closed/aborted connection.
 851                          * Delete this connection and look up again. */
 852                         write_unlock_bh(&tcp_lock);
 853
 854                         /* Only repeat if we can actually remove the timer.
 855                          * Destruction may already be in progress in process
 856                          * context and we must give it a chance to terminate.
 857                          */
 858                         if (nf_ct_kill(ct))
 859                                 return -NF_REPEAT;
 860                         return -NF_DROP;
 861                 }
 862                 /* Fall through */
 863         case TCP_CONNTRACK_IGNORE:
 864                 /* Ignored packets:
 865                  *
 866                  * Our connection entry may be out of sync, so ignore
 867                  * packets which may signal the real connection between
 868                  * the client and the server.
 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 the ignored packet is invalid, the receiver will send
 875                  * a RST we'll catch below.
 876                  */
 877                 if (index == TCP_SYNACK_SET
 878                     && ct->proto.tcp.last_index == TCP_SYN_SET
 879                     && ct->proto.tcp.last_dir != dir
 880                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 881                         /* b) 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                         nf_ct_kill(ct);
 893                         return -NF_DROP;
 894                 }
 895                 ct->proto.tcp.last_index = index;
 896                 ct->proto.tcp.last_dir = dir;
 897                 ct->proto.tcp.last_seq = ntohl(th->seq);
 898                 ct->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 packet ignored ");
 905                 return NF_ACCEPT;
 906         case TCP_CONNTRACK_MAX:
 907                 /* Invalid packet */
 908                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 909                          dir, get_conntrack_index(th), old_state);
 910                 write_unlock_bh(&tcp_lock);
 911                 if (LOG_INVALID(IPPROTO_TCP))
 912                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 913                                   "nf_ct_tcp: invalid state ");
 914                 return -NF_ACCEPT;
 915         case TCP_CONNTRACK_CLOSE:
 916                 if (index == TCP_RST_SET
 917                     && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
 918                          && ct->proto.tcp.last_index == TCP_SYN_SET)
 919                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
 920                             && ct->proto.tcp.last_index == TCP_ACK_SET))
 921                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 922                         /* RST sent to invalid SYN or ACK we had let through
 923                          * at a) and c) above:
 924                          *
 925                          * a) SYN was in window then
 926                          * c) we hold a half-open connection.
 927                          *
 928                          * Delete our connection entry.
 929                          * We skip window checking, because packet might ACK
 930                          * segments we ignored. */
 931                         goto in_window;
 932                 }
 933                 /* Just fall through */
 934         default:
 935                 /* Keep compilers happy. */
 936                 break;
 937         }
 938
 939         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
 940                            skb, dataoff, th, pf)) {
 941                 write_unlock_bh(&tcp_lock);
 942                 return -NF_ACCEPT;
 943         }
 944      in_window:
 945         /* From now on we have got in-window packets */
 946         ct->proto.tcp.last_index = index;
 947         ct->proto.tcp.last_dir = dir;
 948
 949         pr_debug("tcp_conntracks: ");
 950         nf_ct_dump_tuple(tuple);
 951         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
 952                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
 953                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
 954                  old_state, new_state);
 955
 956         ct->proto.tcp.state = new_state;
 957         if (old_state != new_state
 958             && new_state == TCP_CONNTRACK_FIN_WAIT)
 959                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
 960
 961         if (ct->proto.tcp.retrans >= nf_ct_tcp_max_retrans &&
 962             tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans)
 963                 timeout = nf_ct_tcp_timeout_max_retrans;
 964         else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
 965                  IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
 966                  tcp_timeouts[new_state] > nf_ct_tcp_timeout_unacknowledged)
 967                 timeout = nf_ct_tcp_timeout_unacknowledged;
 968         else
 969                 timeout = tcp_timeouts[new_state];
 970         write_unlock_bh(&tcp_lock);
 971
 972         nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
 973         if (new_state != old_state)
 974                 nf_conntrack_event_cache(IPCT_PROTOINFO, skb);
 975
 976         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 977                 /* If only reply is a RST, we can consider ourselves not to
 978                    have an established connection: this is a fairly common
 979                    problem case, so we can delete the conntrack
 980                    immediately.  --RR */
 981                 if (th->rst) {
 982                         nf_ct_kill_acct(ct, ctinfo, skb);
 983                         return NF_ACCEPT;
 984                 }
 985         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
 986                    && (old_state == TCP_CONNTRACK_SYN_RECV
 987                        || old_state == TCP_CONNTRACK_ESTABLISHED)
 988                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
 989                 /* Set ASSURED if we see see valid ack in ESTABLISHED
 990                    after SYN_RECV or a valid answer for a picked up
 991                    connection. */
 992                 set_bit(IPS_ASSURED_BIT, &ct->status);
 993                 nf_conntrack_event_cache(IPCT_STATUS, skb);
 994         }
 995         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
 996
 997         return NF_ACCEPT;
 998 }
 999
1000 /* Called when a new connection for this protocol found. */
1001 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
1002                     unsigned int dataoff)
1003 {
1004         enum tcp_conntrack new_state;
1005         const struct tcphdr *th;
1006         struct tcphdr _tcph;
1007         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
1008         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
1009
1010         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1011         BUG_ON(th == NULL);
1012
1013         /* Don't need lock here: this conntrack not in circulation yet */
1014         new_state
1015                 = tcp_conntracks[0][get_conntrack_index(th)]
1016                 [TCP_CONNTRACK_NONE];
1017
1018         /* Invalid: delete conntrack */
1019         if (new_state >= TCP_CONNTRACK_MAX) {
1020                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1021                 return false;
1022         }
1023
1024         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1025                 /* SYN packet */
1026                 ct->proto.tcp.seen[0].td_end =
1027                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1028                                              dataoff, th);
1029                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1030                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1031                         ct->proto.tcp.seen[0].td_maxwin = 1;
1032                 ct->proto.tcp.seen[0].td_maxend =
1033                         ct->proto.tcp.seen[0].td_end;
1034
1035                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1036                 ct->proto.tcp.seen[1].flags = 0;
1037         } else if (nf_ct_tcp_loose == 0) {
1038                 /* Don't try to pick up connections. */
1039                 return false;
1040         } else {
1041                 /*
1042                  * We are in the middle of a connection,
1043                  * its history is lost for us.
1044                  * Let's try to use the data from the packet.
1045                  */
1046                 ct->proto.tcp.seen[0].td_end =
1047                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1048                                              dataoff, th);
1049                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1050                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1051                         ct->proto.tcp.seen[0].td_maxwin = 1;
1052                 ct->proto.tcp.seen[0].td_maxend =
1053                         ct->proto.tcp.seen[0].td_end +
1054                         ct->proto.tcp.seen[0].td_maxwin;
1055                 ct->proto.tcp.seen[0].td_scale = 0;
1056
1057                 /* We assume SACK and liberal window checking to handle
1058                  * window scaling */
1059                 ct->proto.tcp.seen[0].flags =
1060                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1061                                               IP_CT_TCP_FLAG_BE_LIBERAL;
1062         }
1063
1064         ct->proto.tcp.seen[1].td_end = 0;
1065         ct->proto.tcp.seen[1].td_maxend = 0;
1066         ct->proto.tcp.seen[1].td_maxwin = 1;
1067         ct->proto.tcp.seen[1].td_scale = 0;
1068
1069         /* tcp_packet will set them */
1070         ct->proto.tcp.state = TCP_CONNTRACK_NONE;
1071         ct->proto.tcp.last_index = TCP_NONE_SET;
1072
1073         pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1074                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1075                  sender->td_end, sender->td_maxend, sender->td_maxwin,
1076                  sender->td_scale,
1077                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1078                  receiver->td_scale);
1079         return true;
1080 }
1081
1082 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1083
1084 #include <linux/netfilter/nfnetlink.h>
1085 #include <linux/netfilter/nfnetlink_conntrack.h>
1086
1087 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1088                          const struct nf_conn *ct)
1089 {
1090         struct nlattr *nest_parms;
1091         struct nf_ct_tcp_flags tmp = {};
1092
1093         read_lock_bh(&tcp_lock);
1094         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1095         if (!nest_parms)
1096                 goto nla_put_failure;
1097
1098         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state);
1099
1100         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1101                    ct->proto.tcp.seen[0].td_scale);
1102
1103         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1104                    ct->proto.tcp.seen[1].td_scale);
1105
1106         tmp.flags = ct->proto.tcp.seen[0].flags;
1107         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1108                 sizeof(struct nf_ct_tcp_flags), &tmp);
1109
1110         tmp.flags = ct->proto.tcp.seen[1].flags;
1111         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1112                 sizeof(struct nf_ct_tcp_flags), &tmp);
1113         read_unlock_bh(&tcp_lock);
1114
1115         nla_nest_end(skb, nest_parms);
1116
1117         return 0;
1118
1119 nla_put_failure:
1120         read_unlock_bh(&tcp_lock);
1121         return -1;
1122 }
1123
1124 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1125         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1126         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1127         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1128         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1129         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len =  sizeof(struct nf_ct_tcp_flags) },
1130 };
1131
1132 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1133 {
1134         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1135         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1136         int err;
1137
1138         /* updates could not contain anything about the private
1139          * protocol info, in that case skip the parsing */
1140         if (!pattr)
1141                 return 0;
1142
1143         err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr, tcp_nla_policy);
1144         if (err < 0)
1145                 return err;
1146
1147         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1148             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1149                 return -EINVAL;
1150
1151         write_lock_bh(&tcp_lock);
1152         if (tb[CTA_PROTOINFO_TCP_STATE])
1153                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1154
1155         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1156                 struct nf_ct_tcp_flags *attr =
1157                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1158                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1159                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1160         }
1161
1162         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1163                 struct nf_ct_tcp_flags *attr =
1164                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1165                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1166                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1167         }
1168
1169         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1170             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1171             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1172             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1173                 ct->proto.tcp.seen[0].td_scale =
1174                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1175                 ct->proto.tcp.seen[1].td_scale =
1176                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1177         }
1178         write_unlock_bh(&tcp_lock);
1179
1180         return 0;
1181 }
1182 #endif
1183
1184 #ifdef CONFIG_SYSCTL
1185 static unsigned int tcp_sysctl_table_users;
1186 static struct ctl_table_header *tcp_sysctl_header;
1187 static struct ctl_table tcp_sysctl_table[] = {
1188         {
1189                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1190                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1191                 .maxlen         = sizeof(unsigned int),
1192                 .mode           = 0644,
1193                 .proc_handler   = &proc_dointvec_jiffies,
1194         },
1195         {
1196                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1197                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1198                 .maxlen         = sizeof(unsigned int),
1199                 .mode           = 0644,
1200                 .proc_handler   = &proc_dointvec_jiffies,
1201         },
1202         {
1203                 .procname       = "nf_conntrack_tcp_timeout_established",
1204                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1205                 .maxlen         = sizeof(unsigned int),
1206                 .mode           = 0644,
1207                 .proc_handler   = &proc_dointvec_jiffies,
1208         },
1209         {
1210                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1211                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1212                 .maxlen         = sizeof(unsigned int),
1213                 .mode           = 0644,
1214                 .proc_handler   = &proc_dointvec_jiffies,
1215         },
1216         {
1217                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1218                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1219                 .maxlen         = sizeof(unsigned int),
1220                 .mode           = 0644,
1221                 .proc_handler   = &proc_dointvec_jiffies,
1222         },
1223         {
1224                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1225                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1226                 .maxlen         = sizeof(unsigned int),
1227                 .mode           = 0644,
1228                 .proc_handler   = &proc_dointvec_jiffies,
1229         },
1230         {
1231                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1232                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1233                 .maxlen         = sizeof(unsigned int),
1234                 .mode           = 0644,
1235                 .proc_handler   = &proc_dointvec_jiffies,
1236         },
1237         {
1238                 .procname       = "nf_conntrack_tcp_timeout_close",
1239                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1240                 .maxlen         = sizeof(unsigned int),
1241                 .mode           = 0644,
1242                 .proc_handler   = &proc_dointvec_jiffies,
1243         },
1244         {
1245                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1246                 .data           = &nf_ct_tcp_timeout_max_retrans,
1247                 .maxlen         = sizeof(unsigned int),
1248                 .mode           = 0644,
1249                 .proc_handler   = &proc_dointvec_jiffies,
1250         },
1251         {
1252                 .procname       = "nf_conntrack_tcp_timeout_unacknowledged",
1253                 .data           = &nf_ct_tcp_timeout_unacknowledged,
1254                 .maxlen         = sizeof(unsigned int),
1255                 .mode           = 0644,
1256                 .proc_handler   = &proc_dointvec_jiffies,
1257         },
1258         {
1259                 .ctl_name       = NET_NF_CONNTRACK_TCP_LOOSE,
1260                 .procname       = "nf_conntrack_tcp_loose",
1261                 .data           = &nf_ct_tcp_loose,
1262                 .maxlen         = sizeof(unsigned int),
1263                 .mode           = 0644,
1264                 .proc_handler   = &proc_dointvec,
1265         },
1266         {
1267                 .ctl_name       = NET_NF_CONNTRACK_TCP_BE_LIBERAL,
1268                 .procname       = "nf_conntrack_tcp_be_liberal",
1269                 .data           = &nf_ct_tcp_be_liberal,
1270                 .maxlen         = sizeof(unsigned int),
1271                 .mode           = 0644,
1272                 .proc_handler   = &proc_dointvec,
1273         },
1274         {
1275                 .ctl_name       = NET_NF_CONNTRACK_TCP_MAX_RETRANS,
1276                 .procname       = "nf_conntrack_tcp_max_retrans",
1277                 .data           = &nf_ct_tcp_max_retrans,
1278                 .maxlen         = sizeof(unsigned int),
1279                 .mode           = 0644,
1280                 .proc_handler   = &proc_dointvec,
1281         },
1282         {
1283                 .ctl_name       = 0
1284         }
1285 };
1286
1287 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1288 static struct ctl_table tcp_compat_sysctl_table[] = {
1289         {
1290                 .procname       = "ip_conntrack_tcp_timeout_syn_sent",
1291                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1292                 .maxlen         = sizeof(unsigned int),
1293                 .mode           = 0644,
1294                 .proc_handler   = &proc_dointvec_jiffies,
1295         },
1296         {
1297                 .procname       = "ip_conntrack_tcp_timeout_syn_recv",
1298                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1299                 .maxlen         = sizeof(unsigned int),
1300                 .mode           = 0644,
1301                 .proc_handler   = &proc_dointvec_jiffies,
1302         },
1303         {
1304                 .procname       = "ip_conntrack_tcp_timeout_established",
1305                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1306                 .maxlen         = sizeof(unsigned int),
1307                 .mode           = 0644,
1308                 .proc_handler   = &proc_dointvec_jiffies,
1309         },
1310         {
1311                 .procname       = "ip_conntrack_tcp_timeout_fin_wait",
1312                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1313                 .maxlen         = sizeof(unsigned int),
1314                 .mode           = 0644,
1315                 .proc_handler   = &proc_dointvec_jiffies,
1316         },
1317         {
1318                 .procname       = "ip_conntrack_tcp_timeout_close_wait",
1319                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1320                 .maxlen         = sizeof(unsigned int),
1321                 .mode           = 0644,
1322                 .proc_handler   = &proc_dointvec_jiffies,
1323         },
1324         {
1325                 .procname       = "ip_conntrack_tcp_timeout_last_ack",
1326                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1327                 .maxlen         = sizeof(unsigned int),
1328                 .mode           = 0644,
1329                 .proc_handler   = &proc_dointvec_jiffies,
1330         },
1331         {
1332                 .procname       = "ip_conntrack_tcp_timeout_time_wait",
1333                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1334                 .maxlen         = sizeof(unsigned int),
1335                 .mode           = 0644,
1336                 .proc_handler   = &proc_dointvec_jiffies,
1337         },
1338         {
1339                 .procname       = "ip_conntrack_tcp_timeout_close",
1340                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1341                 .maxlen         = sizeof(unsigned int),
1342                 .mode           = 0644,
1343                 .proc_handler   = &proc_dointvec_jiffies,
1344         },
1345         {
1346                 .procname       = "ip_conntrack_tcp_timeout_max_retrans",
1347                 .data           = &nf_ct_tcp_timeout_max_retrans,
1348                 .maxlen         = sizeof(unsigned int),
1349                 .mode           = 0644,
1350                 .proc_handler   = &proc_dointvec_jiffies,
1351         },
1352         {
1353                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_LOOSE,
1354                 .procname       = "ip_conntrack_tcp_loose",
1355                 .data           = &nf_ct_tcp_loose,
1356                 .maxlen         = sizeof(unsigned int),
1357                 .mode           = 0644,
1358                 .proc_handler   = &proc_dointvec,
1359         },
1360         {
1361                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_BE_LIBERAL,
1362                 .procname       = "ip_conntrack_tcp_be_liberal",
1363                 .data           = &nf_ct_tcp_be_liberal,
1364                 .maxlen         = sizeof(unsigned int),
1365                 .mode           = 0644,
1366                 .proc_handler   = &proc_dointvec,
1367         },
1368         {
1369                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_MAX_RETRANS,
1370                 .procname       = "ip_conntrack_tcp_max_retrans",
1371                 .data           = &nf_ct_tcp_max_retrans,
1372                 .maxlen         = sizeof(unsigned int),
1373                 .mode           = 0644,
1374                 .proc_handler   = &proc_dointvec,
1375         },
1376         {
1377                 .ctl_name       = 0
1378         }
1379 };
1380 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
1381 #endif /* CONFIG_SYSCTL */
1382
1383 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1384 {
1385         .l3proto                = PF_INET,
1386         .l4proto                = IPPROTO_TCP,
1387         .name                   = "tcp",
1388         .pkt_to_tuple           = tcp_pkt_to_tuple,
1389         .invert_tuple           = tcp_invert_tuple,
1390         .print_tuple            = tcp_print_tuple,
1391         .print_conntrack        = tcp_print_conntrack,
1392         .packet                 = tcp_packet,
1393         .new                    = tcp_new,
1394         .error                  = tcp_error,
1395 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1396         .to_nlattr              = tcp_to_nlattr,
1397         .from_nlattr            = nlattr_to_tcp,
1398         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1399         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1400         .nla_policy             = nf_ct_port_nla_policy,
1401 #endif
1402 #ifdef CONFIG_SYSCTL
1403         .ctl_table_users        = &tcp_sysctl_table_users,
1404         .ctl_table_header       = &tcp_sysctl_header,
1405         .ctl_table              = tcp_sysctl_table,
1406 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1407         .ctl_compat_table       = tcp_compat_sysctl_table,
1408 #endif
1409 #endif
1410 };
1411 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1412
1413 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1414 {
1415         .l3proto                = PF_INET6,
1416         .l4proto                = IPPROTO_TCP,
1417         .name                   = "tcp",
1418         .pkt_to_tuple           = tcp_pkt_to_tuple,
1419         .invert_tuple           = tcp_invert_tuple,
1420         .print_tuple            = tcp_print_tuple,
1421         .print_conntrack        = tcp_print_conntrack,
1422         .packet                 = tcp_packet,
1423         .new                    = tcp_new,
1424         .error                  = tcp_error,
1425 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1426         .to_nlattr              = tcp_to_nlattr,
1427         .from_nlattr            = nlattr_to_tcp,
1428         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1429         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1430         .nla_policy             = nf_ct_port_nla_policy,
1431 #endif
1432 #ifdef CONFIG_SYSCTL
1433         .ctl_table_users        = &tcp_sysctl_table_users,
1434         .ctl_table_header       = &tcp_sysctl_header,
1435         .ctl_table              = tcp_sysctl_table,
1436 #endif
1437 };
1438 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);