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

   1 /* (C) 1999-2001 Paul `Rusty' Russell
   2  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  */
   8
   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 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
  30 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
  31
  32 /* "Be conservative in what you do,
  33     be liberal in what you accept from others."
  34     If it's non-zero, we mark only out of window RST segments as INVALID. */
  35 static int nf_ct_tcp_be_liberal __read_mostly = 0;
  36
  37 /* If it is set to zero, we disable picking up already established
  38    connections. */
  39 static int nf_ct_tcp_loose __read_mostly = 1;
  40
  41 /* Max number of the retransmitted packets without receiving an (acceptable)
  42    ACK from the destination. If this number is reached, a shorter timer
  43    will be started. */
  44 static int nf_ct_tcp_max_retrans __read_mostly = 3;
  45
  46   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  47      closely.  They're more complex. --RR */
  48
  49 static const char *const tcp_conntrack_names[] = {
  50         "NONE",
  51         "SYN_SENT",
  52         "SYN_RECV",
  53         "ESTABLISHED",
  54         "FIN_WAIT",
  55         "CLOSE_WAIT",
  56         "LAST_ACK",
  57         "TIME_WAIT",
  58         "CLOSE",
  59         "SYN_SENT2",
  60 };
  61
  62 #define SECS * HZ
  63 #define MINS * 60 SECS
  64 #define HOURS * 60 MINS
  65 #define DAYS * 24 HOURS
  66
  67 /* RFC1122 says the R2 limit should be at least 100 seconds.
  68    Linux uses 15 packets as limit, which corresponds
  69    to ~13-30min depending on RTO. */
  70 static unsigned int nf_ct_tcp_timeout_max_retrans __read_mostly    =   5 MINS;
  71 static unsigned int nf_ct_tcp_timeout_unacknowledged __read_mostly =   5 MINS;
  72
  73 static unsigned int tcp_timeouts[TCP_CONNTRACK_MAX] __read_mostly = {
  74         [TCP_CONNTRACK_SYN_SENT]        = 2 MINS,
  75         [TCP_CONNTRACK_SYN_RECV]        = 60 SECS,
  76         [TCP_CONNTRACK_ESTABLISHED]     = 5 DAYS,
  77         [TCP_CONNTRACK_FIN_WAIT]        = 2 MINS,
  78         [TCP_CONNTRACK_CLOSE_WAIT]      = 60 SECS,
  79         [TCP_CONNTRACK_LAST_ACK]        = 30 SECS,
  80         [TCP_CONNTRACK_TIME_WAIT]       = 2 MINS,
  81         [TCP_CONNTRACK_CLOSE]           = 10 SECS,
  82         [TCP_CONNTRACK_SYN_SENT2]       = 2 MINS,
  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 sS2 TCP_CONNTRACK_SYN_SENT2
  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_SENT2:   SYN-only packet seen from reply dir, simultaneous open
 125  * SYN_RECV:    SYN-ACK packet seen
 126  * ESTABLISHED: ACK packet seen
 127  * FIN_WAIT:    FIN packet seen
 128  * CLOSE_WAIT:  ACK seen (after FIN)
 129  * LAST_ACK:    FIN seen (after FIN)
 130  * TIME_WAIT:   last ACK seen
 131  * CLOSE:       closed connection (RST)
 132  *
 133  * Packets marked as IGNORED (sIG):
 134  *      if they may be either invalid or valid
 135  *      and the receiver may send back a connection
 136  *      closing RST or a SYN/ACK.
 137  *
 138  * Packets marked as INVALID (sIV):
 139  *      if we regard them as truly invalid packets
 140  */
 141 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 142         {
 143 /* ORIGINAL */
 144 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 145 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sS2 },
 146 /*
 147  *      sNO -> sSS      Initialize a new connection
 148  *      sSS -> sSS      Retransmitted SYN
 149  *      sS2 -> sS2      Late retransmitted SYN
 150  *      sSR -> sIG
 151  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 152  *                      are errors. Receiver will reply with RST
 153  *                      and close the connection.
 154  *                      Or we are not in sync and hold a dead connection.
 155  *      sFW -> sIG
 156  *      sCW -> sIG
 157  *      sLA -> sIG
 158  *      sTW -> sSS      Reopened connection (RFC 1122).
 159  *      sCL -> sSS
 160  */
 161 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 162 /*synack*/ { sIV, sIV, sIG, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
 163 /*
 164  *      sNO -> sIV      Too late and no reason to do anything
 165  *      sSS -> sIV      Client can't send SYN and then SYN/ACK
 166  *      sS2 -> sSR      SYN/ACK sent to SYN2 in simultaneous open
 167  *      sSR -> sIG
 168  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 169  *                      are errors. Receiver will reply with RST
 170  *                      and close the connection.
 171  *                      Or we are not in sync and hold a dead connection.
 172  *      sFW -> sIG
 173  *      sCW -> sIG
 174  *      sLA -> sIG
 175  *      sTW -> sIG
 176  *      sCL -> sIG
 177  */
 178 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 179 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 180 /*
 181  *      sNO -> sIV      Too late and no reason to do anything...
 182  *      sSS -> sIV      Client migth not send FIN in this state:
 183  *                      we enforce waiting for a SYN/ACK reply first.
 184  *      sS2 -> sIV
 185  *      sSR -> sFW      Close started.
 186  *      sES -> sFW
 187  *      sFW -> sLA      FIN seen in both directions, waiting for
 188  *                      the last ACK.
 189  *                      Migth be a retransmitted FIN as well...
 190  *      sCW -> sLA
 191  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 192  *      sTW -> sTW
 193  *      sCL -> sCL
 194  */
 195 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 196 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 197 /*
 198  *      sNO -> sES      Assumed.
 199  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 200  *      sS2 -> sIV
 201  *      sSR -> sES      Established state is reached.
 202  *      sES -> sES      :-)
 203  *      sFW -> sCW      Normal close request answered by ACK.
 204  *      sCW -> sCW
 205  *      sLA -> sTW      Last ACK detected.
 206  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 207  *      sCL -> sCL
 208  */
 209 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 210 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
 211 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 212         },
 213         {
 214 /* REPLY */
 215 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 216 /*syn*/    { sIV, sS2, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sS2 },
 217 /*
 218  *      sNO -> sIV      Never reached.
 219  *      sSS -> sS2      Simultaneous open
 220  *      sS2 -> sS2      Retransmitted simultaneous SYN
 221  *      sSR -> sIV      Invalid SYN packets sent by the server
 222  *      sES -> sIV
 223  *      sFW -> sIV
 224  *      sCW -> sIV
 225  *      sLA -> sIV
 226  *      sTW -> sIV      Reopened connection, but server may not do it.
 227  *      sCL -> sIV
 228  */
 229 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 230 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
 231 /*
 232  *      sSS -> sSR      Standard open.
 233  *      sS2 -> sSR      Simultaneous open
 234  *      sSR -> sSR      Retransmitted SYN/ACK.
 235  *      sES -> sIG      Late retransmitted SYN/ACK?
 236  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 237  *      sCW -> sIG
 238  *      sLA -> sIG
 239  *      sTW -> sIG
 240  *      sCL -> sIG
 241  */
 242 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 243 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 244 /*
 245  *      sSS -> sIV      Server might not send FIN in this state.
 246  *      sS2 -> sIV
 247  *      sSR -> sFW      Close started.
 248  *      sES -> sFW
 249  *      sFW -> sLA      FIN seen in both directions.
 250  *      sCW -> sLA
 251  *      sLA -> sLA      Retransmitted FIN.
 252  *      sTW -> sTW
 253  *      sCL -> sCL
 254  */
 255 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 256 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIG },
 257 /*
 258  *      sSS -> sIG      Might be a half-open connection.
 259  *      sS2 -> sIG
 260  *      sSR -> sSR      Might answer late resent SYN.
 261  *      sES -> sES      :-)
 262  *      sFW -> sCW      Normal close request answered by ACK.
 263  *      sCW -> sCW
 264  *      sLA -> sTW      Last ACK detected.
 265  *      sTW -> sTW      Retransmitted last ACK.
 266  *      sCL -> sCL
 267  */
 268 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 269 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
 270 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 271         }
 272 };
 273
 274 static bool tcp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
 275                              struct nf_conntrack_tuple *tuple)
 276 {
 277         const struct tcphdr *hp;
 278         struct tcphdr _hdr;
 279
 280         /* Actually only need first 8 bytes. */
 281         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
 282         if (hp == NULL)
 283                 return false;
 284
 285         tuple->src.u.tcp.port = hp->source;
 286         tuple->dst.u.tcp.port = hp->dest;
 287
 288         return true;
 289 }
 290
 291 static bool tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
 292                              const struct nf_conntrack_tuple *orig)
 293 {
 294         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
 295         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
 296         return true;
 297 }
 298
 299 /* Print out the per-protocol part of the tuple. */
 300 static int tcp_print_tuple(struct seq_file *s,
 301                            const struct nf_conntrack_tuple *tuple)
 302 {
 303         return seq_printf(s, "sport=%hu dport=%hu ",
 304                           ntohs(tuple->src.u.tcp.port),
 305                           ntohs(tuple->dst.u.tcp.port));
 306 }
 307
 308 /* Print out the private part of the conntrack. */
 309 static int tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
 310 {
 311         enum tcp_conntrack state;
 312
 313         spin_lock_bh(&ct->lock);
 314         state = ct->proto.tcp.state;
 315         spin_unlock_bh(&ct->lock);
 316
 317         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
 318 }
 319
 320 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 321 {
 322         if (tcph->rst) return TCP_RST_SET;
 323         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 324         else if (tcph->fin) return TCP_FIN_SET;
 325         else if (tcph->ack) return TCP_ACK_SET;
 326         else return TCP_NONE_SET;
 327 }
 328
 329 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 330    in IP Filter' by Guido van Rooij.
 331
 332    http://www.nluug.nl/events/sane2000/papers.html
 333    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
 334
 335    The boundaries and the conditions are changed according to RFC793:
 336    the packet must intersect the window (i.e. segments may be
 337    after the right or before the left edge) and thus receivers may ACK
 338    segments after the right edge of the window.
 339
 340         td_maxend = max(sack + max(win,1)) seen in reply packets
 341         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 342         td_maxwin += seq + len - sender.td_maxend
 343                         if seq + len > sender.td_maxend
 344         td_end    = max(seq + len) seen in sent packets
 345
 346    I.   Upper bound for valid data:     seq <= sender.td_maxend
 347    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 348    III. Upper bound for valid (s)ack:   sack <= receiver.td_end
 349    IV.  Lower bound for valid (s)ack:   sack >= receiver.td_end - MAXACKWINDOW
 350
 351    where sack is the highest right edge of sack block found in the packet
 352    or ack in the case of packet without SACK option.
 353
 354    The upper bound limit for a valid (s)ack is not ignored -
 355    we doesn't have to deal with fragments.
 356 */
 357
 358 static inline __u32 segment_seq_plus_len(__u32 seq,
 359                                          size_t len,
 360                                          unsigned int dataoff,
 361                                          const struct tcphdr *tcph)
 362 {
 363         /* XXX Should I use payload length field in IP/IPv6 header ?
 364          * - YK */
 365         return (seq + len - dataoff - tcph->doff*4
 366                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 367 }
 368
 369 /* Fixme: what about big packets? */
 370 #define MAXACKWINCONST                  66000
 371 #define MAXACKWINDOW(sender)                                            \
 372         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 373                                               : MAXACKWINCONST)
 374
 375 /*
 376  * Simplified tcp_parse_options routine from tcp_input.c
 377  */
 378 static void tcp_options(const struct sk_buff *skb,
 379                         unsigned int dataoff,
 380                         const struct tcphdr *tcph,
 381                         struct ip_ct_tcp_state *state)
 382 {
 383         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 384         const unsigned char *ptr;
 385         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 386
 387         if (!length)
 388                 return;
 389
 390         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 391                                  length, buff);
 392         BUG_ON(ptr == NULL);
 393
 394         state->td_scale =
 395         state->flags = 0;
 396
 397         while (length > 0) {
 398                 int opcode=*ptr++;
 399                 int opsize;
 400
 401                 switch (opcode) {
 402                 case TCPOPT_EOL:
 403                         return;
 404                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 405                         length--;
 406                         continue;
 407                 default:
 408                         opsize=*ptr++;
 409                         if (opsize < 2) /* "silly options" */
 410                                 return;
 411                         if (opsize > length)
 412                                 break;  /* don't parse partial options */
 413
 414                         if (opcode == TCPOPT_SACK_PERM
 415                             && opsize == TCPOLEN_SACK_PERM)
 416                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 417                         else if (opcode == TCPOPT_WINDOW
 418                                  && opsize == TCPOLEN_WINDOW) {
 419                                 state->td_scale = *(u_int8_t *)ptr;
 420
 421                                 if (state->td_scale > 14) {
 422                                         /* See RFC1323 */
 423                                         state->td_scale = 14;
 424                                 }
 425                                 state->flags |=
 426                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 427                         }
 428                         ptr += opsize - 2;
 429                         length -= opsize;
 430                 }
 431         }
 432 }
 433
 434 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 435                      const struct tcphdr *tcph, __u32 *sack)
 436 {
 437         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 438         const unsigned char *ptr;
 439         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 440         __u32 tmp;
 441
 442         if (!length)
 443                 return;
 444
 445         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 446                                  length, buff);
 447         BUG_ON(ptr == NULL);
 448
 449         /* Fast path for timestamp-only option */
 450         if (length == TCPOLEN_TSTAMP_ALIGNED*4
 451             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
 452                                        | (TCPOPT_NOP << 16)
 453                                        | (TCPOPT_TIMESTAMP << 8)
 454                                        | TCPOLEN_TIMESTAMP))
 455                 return;
 456
 457         while (length > 0) {
 458                 int opcode = *ptr++;
 459                 int opsize, i;
 460
 461                 switch (opcode) {
 462                 case TCPOPT_EOL:
 463                         return;
 464                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 465                         length--;
 466                         continue;
 467                 default:
 468                         opsize = *ptr++;
 469                         if (opsize < 2) /* "silly options" */
 470                                 return;
 471                         if (opsize > length)
 472                                 break;  /* don't parse partial options */
 473
 474                         if (opcode == TCPOPT_SACK
 475                             && opsize >= (TCPOLEN_SACK_BASE
 476                                           + TCPOLEN_SACK_PERBLOCK)
 477                             && !((opsize - TCPOLEN_SACK_BASE)
 478                                  % TCPOLEN_SACK_PERBLOCK)) {
 479                                 for (i = 0;
 480                                      i < (opsize - TCPOLEN_SACK_BASE);
 481                                      i += TCPOLEN_SACK_PERBLOCK) {
 482                                         tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
 483
 484                                         if (after(tmp, *sack))
 485                                                 *sack = tmp;
 486                                 }
 487                                 return;
 488                         }
 489                         ptr += opsize - 2;
 490                         length -= opsize;
 491                 }
 492         }
 493 }
 494
 495 static bool tcp_in_window(const struct nf_conn *ct,
 496                           struct ip_ct_tcp *state,
 497                           enum ip_conntrack_dir dir,
 498                           unsigned int index,
 499                           const struct sk_buff *skb,
 500                           unsigned int dataoff,
 501                           const struct tcphdr *tcph,
 502                           u_int8_t pf)
 503 {
 504         struct net *net = nf_ct_net(ct);
 505         struct ip_ct_tcp_state *sender = &state->seen[dir];
 506         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 507         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 508         __u32 seq, ack, sack, end, win, swin;
 509         bool res;
 510
 511         /*
 512          * Get the required data from the packet.
 513          */
 514         seq = ntohl(tcph->seq);
 515         ack = sack = ntohl(tcph->ack_seq);
 516         win = ntohs(tcph->window);
 517         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 518
 519         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 520                 tcp_sack(skb, dataoff, tcph, &sack);
 521
 522         pr_debug("tcp_in_window: START\n");
 523         pr_debug("tcp_in_window: ");
 524         nf_ct_dump_tuple(tuple);
 525         pr_debug("seq=%u ack=%u sack=%u win=%u end=%u\n",
 526                  seq, ack, sack, win, end);
 527         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 528                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 529                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 530                  sender->td_scale,
 531                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 532                  receiver->td_scale);
 533
 534         if (sender->td_maxwin == 0) {
 535                 /*
 536                  * Initialize sender data.
 537                  */
 538                 if (tcph->syn) {
 539                         /*
 540                          * SYN-ACK in reply to a SYN
 541                          * or SYN from reply direction in simultaneous open.
 542                          */
 543                         sender->td_end =
 544                         sender->td_maxend = end;
 545                         sender->td_maxwin = (win == 0 ? 1 : win);
 546
 547                         tcp_options(skb, dataoff, tcph, sender);
 548                         /*
 549                          * RFC 1323:
 550                          * Both sides must send the Window Scale option
 551                          * to enable window scaling in either direction.
 552                          */
 553                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 554                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 555                                 sender->td_scale =
 556                                 receiver->td_scale = 0;
 557                         if (!tcph->ack)
 558                                 /* Simultaneous open */
 559                                 return true;
 560                 } else {
 561                         /*
 562                          * We are in the middle of a connection,
 563                          * its history is lost for us.
 564                          * Let's try to use the data from the packet.
 565                          */
 566                         sender->td_end = end;
 567                         sender->td_maxwin = (win == 0 ? 1 : win);
 568                         sender->td_maxend = end + sender->td_maxwin;
 569                 }
 570         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 571                      && dir == IP_CT_DIR_ORIGINAL)
 572                    || (state->state == TCP_CONNTRACK_SYN_RECV
 573                      && dir == IP_CT_DIR_REPLY))
 574                    && after(end, sender->td_end)) {
 575                 /*
 576                  * RFC 793: "if a TCP is reinitialized ... then it need
 577                  * not wait at all; it must only be sure to use sequence
 578                  * numbers larger than those recently used."
 579                  */
 580                 sender->td_end =
 581                 sender->td_maxend = end;
 582                 sender->td_maxwin = (win == 0 ? 1 : win);
 583
 584                 tcp_options(skb, dataoff, tcph, sender);
 585         }
 586
 587         if (!(tcph->ack)) {
 588                 /*
 589                  * If there is no ACK, just pretend it was set and OK.
 590                  */
 591                 ack = sack = receiver->td_end;
 592         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 593                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 594                    && (ack == 0)) {
 595                 /*
 596                  * Broken TCP stacks, that set ACK in RST packets as well
 597                  * with zero ack value.
 598                  */
 599                 ack = sack = receiver->td_end;
 600         }
 601
 602         if (seq == end
 603             && (!tcph->rst
 604                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
 605                 /*
 606                  * Packets contains no data: we assume it is valid
 607                  * and check the ack value only.
 608                  * However RST segments are always validated by their
 609                  * SEQ number, except when seq == 0 (reset sent answering
 610                  * SYN.
 611                  */
 612                 seq = end = sender->td_end;
 613
 614         pr_debug("tcp_in_window: ");
 615         nf_ct_dump_tuple(tuple);
 616         pr_debug("seq=%u ack=%u sack =%u win=%u end=%u\n",
 617                  seq, ack, sack, win, end);
 618         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 619                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 620                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 621                  sender->td_scale,
 622                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 623                  receiver->td_scale);
 624
 625         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 626                  before(seq, sender->td_maxend + 1),
 627                  after(end, sender->td_end - receiver->td_maxwin - 1),
 628                  before(sack, receiver->td_end + 1),
 629                  after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 630
 631         if (before(seq, sender->td_maxend + 1) &&
 632             after(end, sender->td_end - receiver->td_maxwin - 1) &&
 633             before(sack, receiver->td_end + 1) &&
 634             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
 635                 /*
 636                  * Take into account window scaling (RFC 1323).
 637                  */
 638                 if (!tcph->syn)
 639                         win <<= sender->td_scale;
 640
 641                 /*
 642                  * Update sender data.
 643                  */
 644                 swin = win + (sack - ack);
 645                 if (sender->td_maxwin < swin)
 646                         sender->td_maxwin = swin;
 647                 if (after(end, sender->td_end)) {
 648                         sender->td_end = end;
 649                         sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 650                 }
 651                 /*
 652                  * Update receiver data.
 653                  */
 654                 if (after(end, sender->td_maxend))
 655                         receiver->td_maxwin += end - sender->td_maxend;
 656                 if (after(sack + win, receiver->td_maxend - 1)) {
 657                         receiver->td_maxend = sack + win;
 658                         if (win == 0)
 659                                 receiver->td_maxend++;
 660                 }
 661                 if (ack == receiver->td_end)
 662                         receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 663
 664                 /*
 665                  * Check retransmissions.
 666                  */
 667                 if (index == TCP_ACK_SET) {
 668                         if (state->last_dir == dir
 669                             && state->last_seq == seq
 670                             && state->last_ack == ack
 671                             && state->last_end == end
 672                             && state->last_win == win)
 673                                 state->retrans++;
 674                         else {
 675                                 state->last_dir = dir;
 676                                 state->last_seq = seq;
 677                                 state->last_ack = ack;
 678                                 state->last_end = end;
 679                                 state->last_win = win;
 680                                 state->retrans = 0;
 681                         }
 682                 }
 683                 res = true;
 684         } else {
 685                 res = false;
 686                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 687                     nf_ct_tcp_be_liberal)
 688                         res = true;
 689                 if (!res && LOG_INVALID(net, IPPROTO_TCP))
 690                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 691                         "nf_ct_tcp: %s ",
 692                         before(seq, sender->td_maxend + 1) ?
 693                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
 694                         before(sack, receiver->td_end + 1) ?
 695                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
 696                         : "ACK is under the lower bound (possible overly delayed ACK)"
 697                         : "ACK is over the upper bound (ACKed data not seen yet)"
 698                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 699                         : "SEQ is over the upper bound (over the window of the receiver)");
 700         }
 701
 702         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 703                  "receiver end=%u maxend=%u maxwin=%u\n",
 704                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 705                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 706
 707         return res;
 708 }
 709
 710 #ifdef CONFIG_NF_NAT_NEEDED
 711 /* Update sender->td_end after NAT successfully mangled the packet */
 712 /* Caller must linearize skb at tcp header. */
 713 void nf_conntrack_tcp_update(const struct sk_buff *skb,
 714                              unsigned int dataoff,
 715                              struct nf_conn *ct,
 716                              int dir)
 717 {
 718         const struct tcphdr *tcph = (const void *)skb->data + dataoff;
 719         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[dir];
 720         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[!dir];
 721         __u32 end;
 722
 723         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
 724
 725         spin_lock_bh(&ct->lock);
 726         /*
 727          * We have to worry for the ack in the reply packet only...
 728          */
 729         if (after(end, ct->proto.tcp.seen[dir].td_end))
 730                 ct->proto.tcp.seen[dir].td_end = end;
 731         ct->proto.tcp.last_end = end;
 732         spin_unlock_bh(&ct->lock);
 733         pr_debug("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
 734                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 735                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 736                  sender->td_scale,
 737                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 738                  receiver->td_scale);
 739 }
 740 EXPORT_SYMBOL_GPL(nf_conntrack_tcp_update);
 741 #endif
 742
 743 #define TH_FIN  0x01
 744 #define TH_SYN  0x02
 745 #define TH_RST  0x04
 746 #define TH_PUSH 0x08
 747 #define TH_ACK  0x10
 748 #define TH_URG  0x20
 749 #define TH_ECE  0x40
 750 #define TH_CWR  0x80
 751
 752 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 753 static const u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_ACK|TH_URG) + 1] =
 754 {
 755         [TH_SYN]                        = 1,
 756         [TH_SYN|TH_URG]                 = 1,
 757         [TH_SYN|TH_ACK]                 = 1,
 758         [TH_RST]                        = 1,
 759         [TH_RST|TH_ACK]                 = 1,
 760         [TH_FIN|TH_ACK]                 = 1,
 761         [TH_FIN|TH_ACK|TH_URG]          = 1,
 762         [TH_ACK]                        = 1,
 763         [TH_ACK|TH_URG]                 = 1,
 764 };
 765
 766 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 767 static int tcp_error(struct net *net,
 768                      struct sk_buff *skb,
 769                      unsigned int dataoff,
 770                      enum ip_conntrack_info *ctinfo,
 771                      u_int8_t pf,
 772                      unsigned int hooknum)
 773 {
 774         const struct tcphdr *th;
 775         struct tcphdr _tcph;
 776         unsigned int tcplen = skb->len - dataoff;
 777         u_int8_t tcpflags;
 778
 779         /* Smaller that minimal TCP header? */
 780         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 781         if (th == NULL) {
 782                 if (LOG_INVALID(net, IPPROTO_TCP))
 783                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 784                                 "nf_ct_tcp: short packet ");
 785                 return -NF_ACCEPT;
 786         }
 787
 788         /* Not whole TCP header or malformed packet */
 789         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 790                 if (LOG_INVALID(net, IPPROTO_TCP))
 791                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 792                                 "nf_ct_tcp: truncated/malformed packet ");
 793                 return -NF_ACCEPT;
 794         }
 795
 796         /* Checksum invalid? Ignore.
 797          * We skip checking packets on the outgoing path
 798          * because the checksum is assumed to be correct.
 799          */
 800         /* FIXME: Source route IP option packets --RR */
 801         if (net->ct.sysctl_checksum && hooknum == NF_INET_PRE_ROUTING &&
 802             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
 803                 if (LOG_INVALID(net, IPPROTO_TCP))
 804                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 805                                   "nf_ct_tcp: bad TCP checksum ");
 806                 return -NF_ACCEPT;
 807         }
 808
 809         /* Check TCP flags. */
 810         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR|TH_PUSH));
 811         if (!tcp_valid_flags[tcpflags]) {
 812                 if (LOG_INVALID(net, IPPROTO_TCP))
 813                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 814                                   "nf_ct_tcp: invalid TCP flag combination ");
 815                 return -NF_ACCEPT;
 816         }
 817
 818         return NF_ACCEPT;
 819 }
 820
 821 /* Returns verdict for packet, or -1 for invalid. */
 822 static int tcp_packet(struct nf_conn *ct,
 823                       const struct sk_buff *skb,
 824                       unsigned int dataoff,
 825                       enum ip_conntrack_info ctinfo,
 826                       u_int8_t pf,
 827                       unsigned int hooknum)
 828 {
 829         struct net *net = nf_ct_net(ct);
 830         struct nf_conntrack_tuple *tuple;
 831         enum tcp_conntrack new_state, old_state;
 832         enum ip_conntrack_dir dir;
 833         const struct tcphdr *th;
 834         struct tcphdr _tcph;
 835         unsigned long timeout;
 836         unsigned int index;
 837
 838         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 839         BUG_ON(th == NULL);
 840
 841         spin_lock_bh(&ct->lock);
 842         old_state = ct->proto.tcp.state;
 843         dir = CTINFO2DIR(ctinfo);
 844         index = get_conntrack_index(th);
 845         new_state = tcp_conntracks[dir][index][old_state];
 846         tuple = &ct->tuplehash[dir].tuple;
 847
 848         switch (new_state) {
 849         case TCP_CONNTRACK_SYN_SENT:
 850                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 851                         break;
 852                 /* RFC 1122: "When a connection is closed actively,
 853                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 854                  * (Maximum Segment Lifetime). However, it MAY accept
 855                  * a new SYN from the remote TCP to reopen the connection
 856                  * directly from TIME-WAIT state, if..."
 857                  * We ignore the conditions because we are in the
 858                  * TIME-WAIT state anyway.
 859                  *
 860                  * Handle aborted connections: we and the server
 861                  * think there is an existing connection but the client
 862                  * aborts it and starts a new one.
 863                  */
 864                 if (((ct->proto.tcp.seen[dir].flags
 865                       | ct->proto.tcp.seen[!dir].flags)
 866                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 867                     || (ct->proto.tcp.last_dir == dir
 868                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 869                         /* Attempt to reopen a closed/aborted connection.
 870                          * Delete this connection and look up again. */
 871                         spin_unlock_bh(&ct->lock);
 872
 873                         /* Only repeat if we can actually remove the timer.
 874                          * Destruction may already be in progress in process
 875                          * context and we must give it a chance to terminate.
 876                          */
 877                         if (nf_ct_kill(ct))
 878                                 return -NF_REPEAT;
 879                         return NF_DROP;
 880                 }
 881                 /* Fall through */
 882         case TCP_CONNTRACK_IGNORE:
 883                 /* Ignored packets:
 884                  *
 885                  * Our connection entry may be out of sync, so ignore
 886                  * packets which may signal the real connection between
 887                  * the client and the server.
 888                  *
 889                  * a) SYN in ORIGINAL
 890                  * b) SYN/ACK in REPLY
 891                  * c) ACK in reply direction after initial SYN in original.
 892                  *
 893                  * If the ignored packet is invalid, the receiver will send
 894                  * a RST we'll catch below.
 895                  */
 896                 if (index == TCP_SYNACK_SET
 897                     && ct->proto.tcp.last_index == TCP_SYN_SET
 898                     && ct->proto.tcp.last_dir != dir
 899                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 900                         /* b) This SYN/ACK acknowledges a SYN that we earlier
 901                          * ignored as invalid. This means that the client and
 902                          * the server are both in sync, while the firewall is
 903                          * not. We kill this session and block the SYN/ACK so
 904                          * that the client cannot but retransmit its SYN and
 905                          * thus initiate a clean new session.
 906                          */
 907                         spin_unlock_bh(&ct->lock);
 908                         if (LOG_INVALID(net, IPPROTO_TCP))
 909                                 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 910                                           "nf_ct_tcp: killing out of sync session ");
 911                         nf_ct_kill(ct);
 912                         return NF_DROP;
 913                 }
 914                 ct->proto.tcp.last_index = index;
 915                 ct->proto.tcp.last_dir = dir;
 916                 ct->proto.tcp.last_seq = ntohl(th->seq);
 917                 ct->proto.tcp.last_end =
 918                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 919
 920                 spin_unlock_bh(&ct->lock);
 921                 if (LOG_INVALID(net, IPPROTO_TCP))
 922                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 923                                   "nf_ct_tcp: invalid packet ignored ");
 924                 return NF_ACCEPT;
 925         case TCP_CONNTRACK_MAX:
 926                 /* Invalid packet */
 927                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 928                          dir, get_conntrack_index(th), old_state);
 929                 spin_unlock_bh(&ct->lock);
 930                 if (LOG_INVALID(net, IPPROTO_TCP))
 931                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 932                                   "nf_ct_tcp: invalid state ");
 933                 return -NF_ACCEPT;
 934         case TCP_CONNTRACK_CLOSE:
 935                 if (index == TCP_RST_SET
 936                     && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
 937                          && ct->proto.tcp.last_index == TCP_SYN_SET)
 938                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
 939                             && ct->proto.tcp.last_index == TCP_ACK_SET))
 940                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 941                         /* RST sent to invalid SYN or ACK we had let through
 942                          * at a) and c) above:
 943                          *
 944                          * a) SYN was in window then
 945                          * c) we hold a half-open connection.
 946                          *
 947                          * Delete our connection entry.
 948                          * We skip window checking, because packet might ACK
 949                          * segments we ignored. */
 950                         goto in_window;
 951                 }
 952                 /* Just fall through */
 953         default:
 954                 /* Keep compilers happy. */
 955                 break;
 956         }
 957
 958         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
 959                            skb, dataoff, th, pf)) {
 960                 spin_unlock_bh(&ct->lock);
 961                 return -NF_ACCEPT;
 962         }
 963      in_window:
 964         /* From now on we have got in-window packets */
 965         ct->proto.tcp.last_index = index;
 966         ct->proto.tcp.last_dir = dir;
 967
 968         pr_debug("tcp_conntracks: ");
 969         nf_ct_dump_tuple(tuple);
 970         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
 971                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
 972                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
 973                  old_state, new_state);
 974
 975         ct->proto.tcp.state = new_state;
 976         if (old_state != new_state
 977             && new_state == TCP_CONNTRACK_FIN_WAIT)
 978                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
 979
 980         if (ct->proto.tcp.retrans >= nf_ct_tcp_max_retrans &&
 981             tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans)
 982                 timeout = nf_ct_tcp_timeout_max_retrans;
 983         else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
 984                  IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
 985                  tcp_timeouts[new_state] > nf_ct_tcp_timeout_unacknowledged)
 986                 timeout = nf_ct_tcp_timeout_unacknowledged;
 987         else
 988                 timeout = tcp_timeouts[new_state];
 989         spin_unlock_bh(&ct->lock);
 990
 991         if (new_state != old_state)
 992                 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
 993
 994         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->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                         nf_ct_kill_acct(ct, ctinfo, skb);
1001                         return NF_ACCEPT;
1002                 }
1003         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1004                    && (old_state == TCP_CONNTRACK_SYN_RECV
1005                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1006                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1007                 /* Set ASSURED if we see see valid ack in ESTABLISHED
1008                    after SYN_RECV or a valid answer for a picked up
1009                    connection. */
1010                 set_bit(IPS_ASSURED_BIT, &ct->status);
1011                 nf_conntrack_event_cache(IPCT_STATUS, ct);
1012         }
1013         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1014
1015         return NF_ACCEPT;
1016 }
1017
1018 /* Called when a new connection for this protocol found. */
1019 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
1020                     unsigned int dataoff)
1021 {
1022         enum tcp_conntrack new_state;
1023         const struct tcphdr *th;
1024         struct tcphdr _tcph;
1025         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
1026         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
1027
1028         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1029         BUG_ON(th == NULL);
1030
1031         /* Don't need lock here: this conntrack not in circulation yet */
1032         new_state
1033                 = tcp_conntracks[0][get_conntrack_index(th)]
1034                 [TCP_CONNTRACK_NONE];
1035
1036         /* Invalid: delete conntrack */
1037         if (new_state >= TCP_CONNTRACK_MAX) {
1038                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1039                 return false;
1040         }
1041
1042         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1043                 /* SYN packet */
1044                 ct->proto.tcp.seen[0].td_end =
1045                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1046                                              dataoff, th);
1047                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1048                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1049                         ct->proto.tcp.seen[0].td_maxwin = 1;
1050                 ct->proto.tcp.seen[0].td_maxend =
1051                         ct->proto.tcp.seen[0].td_end;
1052
1053                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1054                 ct->proto.tcp.seen[1].flags = 0;
1055         } else if (nf_ct_tcp_loose == 0) {
1056                 /* Don't try to pick up connections. */
1057                 return false;
1058         } else {
1059                 /*
1060                  * We are in the middle of a connection,
1061                  * its history is lost for us.
1062                  * Let's try to use the data from the packet.
1063                  */
1064                 ct->proto.tcp.seen[0].td_end =
1065                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1066                                              dataoff, th);
1067                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1068                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1069                         ct->proto.tcp.seen[0].td_maxwin = 1;
1070                 ct->proto.tcp.seen[0].td_maxend =
1071                         ct->proto.tcp.seen[0].td_end +
1072                         ct->proto.tcp.seen[0].td_maxwin;
1073                 ct->proto.tcp.seen[0].td_scale = 0;
1074
1075                 /* We assume SACK and liberal window checking to handle
1076                  * window scaling */
1077                 ct->proto.tcp.seen[0].flags =
1078                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1079                                               IP_CT_TCP_FLAG_BE_LIBERAL;
1080         }
1081
1082         ct->proto.tcp.seen[1].td_end = 0;
1083         ct->proto.tcp.seen[1].td_maxend = 0;
1084         ct->proto.tcp.seen[1].td_maxwin = 0;
1085         ct->proto.tcp.seen[1].td_scale = 0;
1086
1087         /* tcp_packet will set them */
1088         ct->proto.tcp.state = TCP_CONNTRACK_NONE;
1089         ct->proto.tcp.last_index = TCP_NONE_SET;
1090
1091         pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1092                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1093                  sender->td_end, sender->td_maxend, sender->td_maxwin,
1094                  sender->td_scale,
1095                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1096                  receiver->td_scale);
1097         return true;
1098 }
1099
1100 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1101
1102 #include <linux/netfilter/nfnetlink.h>
1103 #include <linux/netfilter/nfnetlink_conntrack.h>
1104
1105 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1106                          struct nf_conn *ct)
1107 {
1108         struct nlattr *nest_parms;
1109         struct nf_ct_tcp_flags tmp = {};
1110
1111         spin_lock_bh(&ct->lock);
1112         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1113         if (!nest_parms)
1114                 goto nla_put_failure;
1115
1116         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state);
1117
1118         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1119                    ct->proto.tcp.seen[0].td_scale);
1120
1121         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1122                    ct->proto.tcp.seen[1].td_scale);
1123
1124         tmp.flags = ct->proto.tcp.seen[0].flags;
1125         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1126                 sizeof(struct nf_ct_tcp_flags), &tmp);
1127
1128         tmp.flags = ct->proto.tcp.seen[1].flags;
1129         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1130                 sizeof(struct nf_ct_tcp_flags), &tmp);
1131         spin_unlock_bh(&ct->lock);
1132
1133         nla_nest_end(skb, nest_parms);
1134
1135         return 0;
1136
1137 nla_put_failure:
1138         spin_unlock_bh(&ct->lock);
1139         return -1;
1140 }
1141
1142 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1143         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1144         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1145         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1146         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1147         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len =  sizeof(struct nf_ct_tcp_flags) },
1148 };
1149
1150 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1151 {
1152         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1153         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1154         int err;
1155
1156         /* updates could not contain anything about the private
1157          * protocol info, in that case skip the parsing */
1158         if (!pattr)
1159                 return 0;
1160
1161         err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr, tcp_nla_policy);
1162         if (err < 0)
1163                 return err;
1164
1165         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1166             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1167                 return -EINVAL;
1168
1169         spin_lock_bh(&ct->lock);
1170         if (tb[CTA_PROTOINFO_TCP_STATE])
1171                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1172
1173         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1174                 struct nf_ct_tcp_flags *attr =
1175                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1176                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1177                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1178         }
1179
1180         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1181                 struct nf_ct_tcp_flags *attr =
1182                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1183                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1184                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1185         }
1186
1187         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1188             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1189             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1190             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1191                 ct->proto.tcp.seen[0].td_scale =
1192                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1193                 ct->proto.tcp.seen[1].td_scale =
1194                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1195         }
1196         spin_unlock_bh(&ct->lock);
1197
1198         return 0;
1199 }
1200
1201 static int tcp_nlattr_size(void)
1202 {
1203         return nla_total_size(0)           /* CTA_PROTOINFO_TCP */
1204                 + nla_policy_len(tcp_nla_policy, CTA_PROTOINFO_TCP_MAX + 1);
1205 }
1206
1207 static int tcp_nlattr_tuple_size(void)
1208 {
1209         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1210 }
1211 #endif
1212
1213 #ifdef CONFIG_SYSCTL
1214 static unsigned int tcp_sysctl_table_users;
1215 static struct ctl_table_header *tcp_sysctl_header;
1216 static struct ctl_table tcp_sysctl_table[] = {
1217         {
1218                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1219                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1220                 .maxlen         = sizeof(unsigned int),
1221                 .mode           = 0644,
1222                 .proc_handler   = proc_dointvec_jiffies,
1223         },
1224         {
1225                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1226                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1227                 .maxlen         = sizeof(unsigned int),
1228                 .mode           = 0644,
1229                 .proc_handler   = proc_dointvec_jiffies,
1230         },
1231         {
1232                 .procname       = "nf_conntrack_tcp_timeout_established",
1233                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1234                 .maxlen         = sizeof(unsigned int),
1235                 .mode           = 0644,
1236                 .proc_handler   = proc_dointvec_jiffies,
1237         },
1238         {
1239                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1240                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1241                 .maxlen         = sizeof(unsigned int),
1242                 .mode           = 0644,
1243                 .proc_handler   = proc_dointvec_jiffies,
1244         },
1245         {
1246                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1247                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1248                 .maxlen         = sizeof(unsigned int),
1249                 .mode           = 0644,
1250                 .proc_handler   = proc_dointvec_jiffies,
1251         },
1252         {
1253                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1254                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1255                 .maxlen         = sizeof(unsigned int),
1256                 .mode           = 0644,
1257                 .proc_handler   = proc_dointvec_jiffies,
1258         },
1259         {
1260                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1261                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1262                 .maxlen         = sizeof(unsigned int),
1263                 .mode           = 0644,
1264                 .proc_handler   = proc_dointvec_jiffies,
1265         },
1266         {
1267                 .procname       = "nf_conntrack_tcp_timeout_close",
1268                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1269                 .maxlen         = sizeof(unsigned int),
1270                 .mode           = 0644,
1271                 .proc_handler   = proc_dointvec_jiffies,
1272         },
1273         {
1274                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1275                 .data           = &nf_ct_tcp_timeout_max_retrans,
1276                 .maxlen         = sizeof(unsigned int),
1277                 .mode           = 0644,
1278                 .proc_handler   = proc_dointvec_jiffies,
1279         },
1280         {
1281                 .procname       = "nf_conntrack_tcp_timeout_unacknowledged",
1282                 .data           = &nf_ct_tcp_timeout_unacknowledged,
1283                 .maxlen         = sizeof(unsigned int),
1284                 .mode           = 0644,
1285                 .proc_handler   = proc_dointvec_jiffies,
1286         },
1287         {
1288                 .ctl_name       = NET_NF_CONNTRACK_TCP_LOOSE,
1289                 .procname       = "nf_conntrack_tcp_loose",
1290                 .data           = &nf_ct_tcp_loose,
1291                 .maxlen         = sizeof(unsigned int),
1292                 .mode           = 0644,
1293                 .proc_handler   = proc_dointvec,
1294         },
1295         {
1296                 .ctl_name       = NET_NF_CONNTRACK_TCP_BE_LIBERAL,
1297                 .procname       = "nf_conntrack_tcp_be_liberal",
1298                 .data           = &nf_ct_tcp_be_liberal,
1299                 .maxlen         = sizeof(unsigned int),
1300                 .mode           = 0644,
1301                 .proc_handler   = proc_dointvec,
1302         },
1303         {
1304                 .ctl_name       = NET_NF_CONNTRACK_TCP_MAX_RETRANS,
1305                 .procname       = "nf_conntrack_tcp_max_retrans",
1306                 .data           = &nf_ct_tcp_max_retrans,
1307                 .maxlen         = sizeof(unsigned int),
1308                 .mode           = 0644,
1309                 .proc_handler   = proc_dointvec,
1310         },
1311         {
1312                 .ctl_name       = 0
1313         }
1314 };
1315
1316 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1317 static struct ctl_table tcp_compat_sysctl_table[] = {
1318         {
1319                 .procname       = "ip_conntrack_tcp_timeout_syn_sent",
1320                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1321                 .maxlen         = sizeof(unsigned int),
1322                 .mode           = 0644,
1323                 .proc_handler   = proc_dointvec_jiffies,
1324         },
1325         {
1326                 .procname       = "ip_conntrack_tcp_timeout_syn_sent2",
1327                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT2],
1328                 .maxlen         = sizeof(unsigned int),
1329                 .mode           = 0644,
1330                 .proc_handler   = proc_dointvec_jiffies,
1331         },
1332         {
1333                 .procname       = "ip_conntrack_tcp_timeout_syn_recv",
1334                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1335                 .maxlen         = sizeof(unsigned int),
1336                 .mode           = 0644,
1337                 .proc_handler   = proc_dointvec_jiffies,
1338         },
1339         {
1340                 .procname       = "ip_conntrack_tcp_timeout_established",
1341                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1342                 .maxlen         = sizeof(unsigned int),
1343                 .mode           = 0644,
1344                 .proc_handler   = proc_dointvec_jiffies,
1345         },
1346         {
1347                 .procname       = "ip_conntrack_tcp_timeout_fin_wait",
1348                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1349                 .maxlen         = sizeof(unsigned int),
1350                 .mode           = 0644,
1351                 .proc_handler   = proc_dointvec_jiffies,
1352         },
1353         {
1354                 .procname       = "ip_conntrack_tcp_timeout_close_wait",
1355                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1356                 .maxlen         = sizeof(unsigned int),
1357                 .mode           = 0644,
1358                 .proc_handler   = proc_dointvec_jiffies,
1359         },
1360         {
1361                 .procname       = "ip_conntrack_tcp_timeout_last_ack",
1362                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1363                 .maxlen         = sizeof(unsigned int),
1364                 .mode           = 0644,
1365                 .proc_handler   = proc_dointvec_jiffies,
1366         },
1367         {
1368                 .procname       = "ip_conntrack_tcp_timeout_time_wait",
1369                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1370                 .maxlen         = sizeof(unsigned int),
1371                 .mode           = 0644,
1372                 .proc_handler   = proc_dointvec_jiffies,
1373         },
1374         {
1375                 .procname       = "ip_conntrack_tcp_timeout_close",
1376                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1377                 .maxlen         = sizeof(unsigned int),
1378                 .mode           = 0644,
1379                 .proc_handler   = proc_dointvec_jiffies,
1380         },
1381         {
1382                 .procname       = "ip_conntrack_tcp_timeout_max_retrans",
1383                 .data           = &nf_ct_tcp_timeout_max_retrans,
1384                 .maxlen         = sizeof(unsigned int),
1385                 .mode           = 0644,
1386                 .proc_handler   = proc_dointvec_jiffies,
1387         },
1388         {
1389                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_LOOSE,
1390                 .procname       = "ip_conntrack_tcp_loose",
1391                 .data           = &nf_ct_tcp_loose,
1392                 .maxlen         = sizeof(unsigned int),
1393                 .mode           = 0644,
1394                 .proc_handler   = proc_dointvec,
1395         },
1396         {
1397                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_BE_LIBERAL,
1398                 .procname       = "ip_conntrack_tcp_be_liberal",
1399                 .data           = &nf_ct_tcp_be_liberal,
1400                 .maxlen         = sizeof(unsigned int),
1401                 .mode           = 0644,
1402                 .proc_handler   = proc_dointvec,
1403         },
1404         {
1405                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_MAX_RETRANS,
1406                 .procname       = "ip_conntrack_tcp_max_retrans",
1407                 .data           = &nf_ct_tcp_max_retrans,
1408                 .maxlen         = sizeof(unsigned int),
1409                 .mode           = 0644,
1410                 .proc_handler   = proc_dointvec,
1411         },
1412         {
1413                 .ctl_name       = 0
1414         }
1415 };
1416 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
1417 #endif /* CONFIG_SYSCTL */
1418
1419 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1420 {
1421         .l3proto                = PF_INET,
1422         .l4proto                = IPPROTO_TCP,
1423         .name                   = "tcp",
1424         .pkt_to_tuple           = tcp_pkt_to_tuple,
1425         .invert_tuple           = tcp_invert_tuple,
1426         .print_tuple            = tcp_print_tuple,
1427         .print_conntrack        = tcp_print_conntrack,
1428         .packet                 = tcp_packet,
1429         .new                    = tcp_new,
1430         .error                  = tcp_error,
1431 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1432         .to_nlattr              = tcp_to_nlattr,
1433         .nlattr_size            = tcp_nlattr_size,
1434         .from_nlattr            = nlattr_to_tcp,
1435         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1436         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1437         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1438         .nla_policy             = nf_ct_port_nla_policy,
1439 #endif
1440 #ifdef CONFIG_SYSCTL
1441         .ctl_table_users        = &tcp_sysctl_table_users,
1442         .ctl_table_header       = &tcp_sysctl_header,
1443         .ctl_table              = tcp_sysctl_table,
1444 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1445         .ctl_compat_table       = tcp_compat_sysctl_table,
1446 #endif
1447 #endif
1448 };
1449 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1450
1451 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1452 {
1453         .l3proto                = PF_INET6,
1454         .l4proto                = IPPROTO_TCP,
1455         .name                   = "tcp",
1456         .pkt_to_tuple           = tcp_pkt_to_tuple,
1457         .invert_tuple           = tcp_invert_tuple,
1458         .print_tuple            = tcp_print_tuple,
1459         .print_conntrack        = tcp_print_conntrack,
1460         .packet                 = tcp_packet,
1461         .new                    = tcp_new,
1462         .error                  = tcp_error,
1463 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1464         .to_nlattr              = tcp_to_nlattr,
1465         .nlattr_size            = tcp_nlattr_size,
1466         .from_nlattr            = nlattr_to_tcp,
1467         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1468         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1469         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1470         .nla_policy             = nf_ct_port_nla_policy,
1471 #endif
1472 #ifdef CONFIG_SYSCTL
1473         .ctl_table_users        = &tcp_sysctl_table_users,
1474         .ctl_table_header       = &tcp_sysctl_header,
1475         .ctl_table              = tcp_sysctl_table,
1476 #endif
1477 };
1478 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);