Pileus Git - ~andy/linux/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
  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                           u_int8_t pf)
 491 {
 492         struct net *net = nf_ct_net(ct);
 493         struct ip_ct_tcp_state *sender = &state->seen[dir];
 494         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 495         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 496         __u32 seq, ack, sack, end, win, swin;
 497         bool res;
 498
 499         /*
 500          * Get the required data from the packet.
 501          */
 502         seq = ntohl(tcph->seq);
 503         ack = sack = ntohl(tcph->ack_seq);
 504         win = ntohs(tcph->window);
 505         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 506
 507         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 508                 tcp_sack(skb, dataoff, tcph, &sack);
 509
 510         pr_debug("tcp_in_window: START\n");
 511         pr_debug("tcp_in_window: ");
 512         nf_ct_dump_tuple(tuple);
 513         pr_debug("seq=%u ack=%u sack=%u win=%u end=%u\n",
 514                  seq, ack, sack, win, end);
 515         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 516                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 517                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 518                  sender->td_scale,
 519                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 520                  receiver->td_scale);
 521
 522         if (sender->td_end == 0) {
 523                 /*
 524                  * Initialize sender data.
 525                  */
 526                 if (tcph->syn && tcph->ack) {
 527                         /*
 528                          * Outgoing SYN-ACK in reply to a SYN.
 529                          */
 530                         sender->td_end =
 531                         sender->td_maxend = end;
 532                         sender->td_maxwin = (win == 0 ? 1 : win);
 533
 534                         tcp_options(skb, dataoff, tcph, sender);
 535                         /*
 536                          * RFC 1323:
 537                          * Both sides must send the Window Scale option
 538                          * to enable window scaling in either direction.
 539                          */
 540                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 541                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 542                                 sender->td_scale =
 543                                 receiver->td_scale = 0;
 544                 } else {
 545                         /*
 546                          * We are in the middle of a connection,
 547                          * its history is lost for us.
 548                          * Let's try to use the data from the packet.
 549                          */
 550                         sender->td_end = end;
 551                         sender->td_maxwin = (win == 0 ? 1 : win);
 552                         sender->td_maxend = end + sender->td_maxwin;
 553                 }
 554         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 555                      && dir == IP_CT_DIR_ORIGINAL)
 556                    || (state->state == TCP_CONNTRACK_SYN_RECV
 557                      && dir == IP_CT_DIR_REPLY))
 558                    && after(end, sender->td_end)) {
 559                 /*
 560                  * RFC 793: "if a TCP is reinitialized ... then it need
 561                  * not wait at all; it must only be sure to use sequence
 562                  * numbers larger than those recently used."
 563                  */
 564                 sender->td_end =
 565                 sender->td_maxend = end;
 566                 sender->td_maxwin = (win == 0 ? 1 : win);
 567
 568                 tcp_options(skb, dataoff, tcph, sender);
 569         }
 570
 571         if (!(tcph->ack)) {
 572                 /*
 573                  * If there is no ACK, just pretend it was set and OK.
 574                  */
 575                 ack = sack = receiver->td_end;
 576         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 577                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 578                    && (ack == 0)) {
 579                 /*
 580                  * Broken TCP stacks, that set ACK in RST packets as well
 581                  * with zero ack value.
 582                  */
 583                 ack = sack = receiver->td_end;
 584         }
 585
 586         if (seq == end
 587             && (!tcph->rst
 588                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
 589                 /*
 590                  * Packets contains no data: we assume it is valid
 591                  * and check the ack value only.
 592                  * However RST segments are always validated by their
 593                  * SEQ number, except when seq == 0 (reset sent answering
 594                  * SYN.
 595                  */
 596                 seq = end = sender->td_end;
 597
 598         pr_debug("tcp_in_window: ");
 599         nf_ct_dump_tuple(tuple);
 600         pr_debug("seq=%u ack=%u sack =%u win=%u end=%u\n",
 601                  seq, ack, sack, win, end);
 602         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 603                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 604                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 605                  sender->td_scale,
 606                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 607                  receiver->td_scale);
 608
 609         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 610                  before(seq, sender->td_maxend + 1),
 611                  after(end, sender->td_end - receiver->td_maxwin - 1),
 612                  before(sack, receiver->td_end + 1),
 613                  after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 614
 615         if (before(seq, sender->td_maxend + 1) &&
 616             after(end, sender->td_end - receiver->td_maxwin - 1) &&
 617             before(sack, receiver->td_end + 1) &&
 618             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
 619                 /*
 620                  * Take into account window scaling (RFC 1323).
 621                  */
 622                 if (!tcph->syn)
 623                         win <<= sender->td_scale;
 624
 625                 /*
 626                  * Update sender data.
 627                  */
 628                 swin = win + (sack - ack);
 629                 if (sender->td_maxwin < swin)
 630                         sender->td_maxwin = swin;
 631                 if (after(end, sender->td_end)) {
 632                         sender->td_end = end;
 633                         sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 634                 }
 635                 /*
 636                  * Update receiver data.
 637                  */
 638                 if (after(end, sender->td_maxend))
 639                         receiver->td_maxwin += end - sender->td_maxend;
 640                 if (after(sack + win, receiver->td_maxend - 1)) {
 641                         receiver->td_maxend = sack + win;
 642                         if (win == 0)
 643                                 receiver->td_maxend++;
 644                 }
 645                 if (ack == receiver->td_end)
 646                         receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 647
 648                 /*
 649                  * Check retransmissions.
 650                  */
 651                 if (index == TCP_ACK_SET) {
 652                         if (state->last_dir == dir
 653                             && state->last_seq == seq
 654                             && state->last_ack == ack
 655                             && state->last_end == end
 656                             && state->last_win == win)
 657                                 state->retrans++;
 658                         else {
 659                                 state->last_dir = dir;
 660                                 state->last_seq = seq;
 661                                 state->last_ack = ack;
 662                                 state->last_end = end;
 663                                 state->last_win = win;
 664                                 state->retrans = 0;
 665                         }
 666                 }
 667                 res = true;
 668         } else {
 669                 res = false;
 670                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 671                     nf_ct_tcp_be_liberal)
 672                         res = true;
 673                 if (!res && LOG_INVALID(net, IPPROTO_TCP))
 674                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 675                         "nf_ct_tcp: %s ",
 676                         before(seq, sender->td_maxend + 1) ?
 677                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
 678                         before(sack, receiver->td_end + 1) ?
 679                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
 680                         : "ACK is under the lower bound (possible overly delayed ACK)"
 681                         : "ACK is over the upper bound (ACKed data not seen yet)"
 682                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 683                         : "SEQ is over the upper bound (over the window of the receiver)");
 684         }
 685
 686         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 687                  "receiver end=%u maxend=%u maxwin=%u\n",
 688                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 689                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 690
 691         return res;
 692 }
 693
 694 #ifdef CONFIG_NF_NAT_NEEDED
 695 /* Update sender->td_end after NAT successfully mangled the packet */
 696 /* Caller must linearize skb at tcp header. */
 697 void nf_conntrack_tcp_update(const struct sk_buff *skb,
 698                              unsigned int dataoff,
 699                              struct nf_conn *ct,
 700                              int dir)
 701 {
 702         const struct tcphdr *tcph = (const void *)skb->data + dataoff;
 703         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[dir];
 704         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[!dir];
 705         __u32 end;
 706
 707         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
 708
 709         write_lock_bh(&tcp_lock);
 710         /*
 711          * We have to worry for the ack in the reply packet only...
 712          */
 713         if (after(end, ct->proto.tcp.seen[dir].td_end))
 714                 ct->proto.tcp.seen[dir].td_end = end;
 715         ct->proto.tcp.last_end = end;
 716         write_unlock_bh(&tcp_lock);
 717         pr_debug("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
 718                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 719                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 720                  sender->td_scale,
 721                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 722                  receiver->td_scale);
 723 }
 724 EXPORT_SYMBOL_GPL(nf_conntrack_tcp_update);
 725 #endif
 726
 727 #define TH_FIN  0x01
 728 #define TH_SYN  0x02
 729 #define TH_RST  0x04
 730 #define TH_PUSH 0x08
 731 #define TH_ACK  0x10
 732 #define TH_URG  0x20
 733 #define TH_ECE  0x40
 734 #define TH_CWR  0x80
 735
 736 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 737 static const u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_ACK|TH_URG) + 1] =
 738 {
 739         [TH_SYN]                        = 1,
 740         [TH_SYN|TH_URG]                 = 1,
 741         [TH_SYN|TH_ACK]                 = 1,
 742         [TH_RST]                        = 1,
 743         [TH_RST|TH_ACK]                 = 1,
 744         [TH_FIN|TH_ACK]                 = 1,
 745         [TH_FIN|TH_ACK|TH_URG]          = 1,
 746         [TH_ACK]                        = 1,
 747         [TH_ACK|TH_URG]                 = 1,
 748 };
 749
 750 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 751 static int tcp_error(struct net *net,
 752                      struct sk_buff *skb,
 753                      unsigned int dataoff,
 754                      enum ip_conntrack_info *ctinfo,
 755                      u_int8_t pf,
 756                      unsigned int hooknum)
 757 {
 758         const struct tcphdr *th;
 759         struct tcphdr _tcph;
 760         unsigned int tcplen = skb->len - dataoff;
 761         u_int8_t tcpflags;
 762
 763         /* Smaller that minimal TCP header? */
 764         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 765         if (th == NULL) {
 766                 if (LOG_INVALID(net, IPPROTO_TCP))
 767                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 768                                 "nf_ct_tcp: short packet ");
 769                 return -NF_ACCEPT;
 770         }
 771
 772         /* Not whole TCP header or malformed packet */
 773         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 774                 if (LOG_INVALID(net, IPPROTO_TCP))
 775                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 776                                 "nf_ct_tcp: truncated/malformed packet ");
 777                 return -NF_ACCEPT;
 778         }
 779
 780         /* Checksum invalid? Ignore.
 781          * We skip checking packets on the outgoing path
 782          * because the checksum is assumed to be correct.
 783          */
 784         /* FIXME: Source route IP option packets --RR */
 785         if (net->ct.sysctl_checksum && hooknum == NF_INET_PRE_ROUTING &&
 786             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
 787                 if (LOG_INVALID(net, IPPROTO_TCP))
 788                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 789                                   "nf_ct_tcp: bad TCP checksum ");
 790                 return -NF_ACCEPT;
 791         }
 792
 793         /* Check TCP flags. */
 794         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR|TH_PUSH));
 795         if (!tcp_valid_flags[tcpflags]) {
 796                 if (LOG_INVALID(net, IPPROTO_TCP))
 797                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 798                                   "nf_ct_tcp: invalid TCP flag combination ");
 799                 return -NF_ACCEPT;
 800         }
 801
 802         return NF_ACCEPT;
 803 }
 804
 805 /* Returns verdict for packet, or -1 for invalid. */
 806 static int tcp_packet(struct nf_conn *ct,
 807                       const struct sk_buff *skb,
 808                       unsigned int dataoff,
 809                       enum ip_conntrack_info ctinfo,
 810                       u_int8_t pf,
 811                       unsigned int hooknum)
 812 {
 813         struct net *net = nf_ct_net(ct);
 814         struct nf_conntrack_tuple *tuple;
 815         enum tcp_conntrack new_state, old_state;
 816         enum ip_conntrack_dir dir;
 817         const struct tcphdr *th;
 818         struct tcphdr _tcph;
 819         unsigned long timeout;
 820         unsigned int index;
 821
 822         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 823         BUG_ON(th == NULL);
 824
 825         write_lock_bh(&tcp_lock);
 826         old_state = ct->proto.tcp.state;
 827         dir = CTINFO2DIR(ctinfo);
 828         index = get_conntrack_index(th);
 829         new_state = tcp_conntracks[dir][index][old_state];
 830         tuple = &ct->tuplehash[dir].tuple;
 831
 832         switch (new_state) {
 833         case TCP_CONNTRACK_SYN_SENT:
 834                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 835                         break;
 836                 /* RFC 1122: "When a connection is closed actively,
 837                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 838                  * (Maximum Segment Lifetime). However, it MAY accept
 839                  * a new SYN from the remote TCP to reopen the connection
 840                  * directly from TIME-WAIT state, if..."
 841                  * We ignore the conditions because we are in the
 842                  * TIME-WAIT state anyway.
 843                  *
 844                  * Handle aborted connections: we and the server
 845                  * think there is an existing connection but the client
 846                  * aborts it and starts a new one.
 847                  */
 848                 if (((ct->proto.tcp.seen[dir].flags
 849                       | ct->proto.tcp.seen[!dir].flags)
 850                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 851                     || (ct->proto.tcp.last_dir == dir
 852                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 853                         /* Attempt to reopen a closed/aborted connection.
 854                          * Delete this connection and look up again. */
 855                         write_unlock_bh(&tcp_lock);
 856
 857                         /* Only repeat if we can actually remove the timer.
 858                          * Destruction may already be in progress in process
 859                          * context and we must give it a chance to terminate.
 860                          */
 861                         if (nf_ct_kill(ct))
 862                                 return -NF_REPEAT;
 863                         return NF_DROP;
 864                 }
 865                 /* Fall through */
 866         case TCP_CONNTRACK_IGNORE:
 867                 /* Ignored packets:
 868                  *
 869                  * Our connection entry may be out of sync, so ignore
 870                  * packets which may signal the real connection between
 871                  * the client and the server.
 872                  *
 873                  * a) SYN in ORIGINAL
 874                  * b) SYN/ACK in REPLY
 875                  * c) ACK in reply direction after initial SYN in original.
 876                  *
 877                  * If the ignored packet is invalid, the receiver will send
 878                  * a RST we'll catch below.
 879                  */
 880                 if (index == TCP_SYNACK_SET
 881                     && ct->proto.tcp.last_index == TCP_SYN_SET
 882                     && ct->proto.tcp.last_dir != dir
 883                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 884                         /* b) This SYN/ACK acknowledges a SYN that we earlier
 885                          * ignored as invalid. This means that the client and
 886                          * the server are both in sync, while the firewall is
 887                          * not. We kill this session and block the SYN/ACK so
 888                          * that the client cannot but retransmit its SYN and
 889                          * thus initiate a clean new session.
 890                          */
 891                         write_unlock_bh(&tcp_lock);
 892                         if (LOG_INVALID(net, IPPROTO_TCP))
 893                                 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 894                                           "nf_ct_tcp: killing out of sync session ");
 895                         nf_ct_kill(ct);
 896                         return NF_DROP;
 897                 }
 898                 ct->proto.tcp.last_index = index;
 899                 ct->proto.tcp.last_dir = dir;
 900                 ct->proto.tcp.last_seq = ntohl(th->seq);
 901                 ct->proto.tcp.last_end =
 902                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 903
 904                 write_unlock_bh(&tcp_lock);
 905                 if (LOG_INVALID(net, IPPROTO_TCP))
 906                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 907                                   "nf_ct_tcp: invalid packet ignored ");
 908                 return NF_ACCEPT;
 909         case TCP_CONNTRACK_MAX:
 910                 /* Invalid packet */
 911                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 912                          dir, get_conntrack_index(th), old_state);
 913                 write_unlock_bh(&tcp_lock);
 914                 if (LOG_INVALID(net, IPPROTO_TCP))
 915                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 916                                   "nf_ct_tcp: invalid state ");
 917                 return -NF_ACCEPT;
 918         case TCP_CONNTRACK_CLOSE:
 919                 if (index == TCP_RST_SET
 920                     && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
 921                          && ct->proto.tcp.last_index == TCP_SYN_SET)
 922                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
 923                             && ct->proto.tcp.last_index == TCP_ACK_SET))
 924                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 925                         /* RST sent to invalid SYN or ACK we had let through
 926                          * at a) and c) above:
 927                          *
 928                          * a) SYN was in window then
 929                          * c) we hold a half-open connection.
 930                          *
 931                          * Delete our connection entry.
 932                          * We skip window checking, because packet might ACK
 933                          * segments we ignored. */
 934                         goto in_window;
 935                 }
 936                 /* Just fall through */
 937         default:
 938                 /* Keep compilers happy. */
 939                 break;
 940         }
 941
 942         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
 943                            skb, dataoff, th, pf)) {
 944                 write_unlock_bh(&tcp_lock);
 945                 return -NF_ACCEPT;
 946         }
 947      in_window:
 948         /* From now on we have got in-window packets */
 949         ct->proto.tcp.last_index = index;
 950         ct->proto.tcp.last_dir = dir;
 951
 952         pr_debug("tcp_conntracks: ");
 953         nf_ct_dump_tuple(tuple);
 954         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
 955                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
 956                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
 957                  old_state, new_state);
 958
 959         ct->proto.tcp.state = new_state;
 960         if (old_state != new_state
 961             && new_state == TCP_CONNTRACK_FIN_WAIT)
 962                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
 963
 964         if (ct->proto.tcp.retrans >= nf_ct_tcp_max_retrans &&
 965             tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans)
 966                 timeout = nf_ct_tcp_timeout_max_retrans;
 967         else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
 968                  IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
 969                  tcp_timeouts[new_state] > nf_ct_tcp_timeout_unacknowledged)
 970                 timeout = nf_ct_tcp_timeout_unacknowledged;
 971         else
 972                 timeout = tcp_timeouts[new_state];
 973         write_unlock_bh(&tcp_lock);
 974
 975         nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, ct);
 976         if (new_state != old_state)
 977                 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
 978
 979         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 980                 /* If only reply is a RST, we can consider ourselves not to
 981                    have an established connection: this is a fairly common
 982                    problem case, so we can delete the conntrack
 983                    immediately.  --RR */
 984                 if (th->rst) {
 985                         nf_ct_kill_acct(ct, ctinfo, skb);
 986                         return NF_ACCEPT;
 987                 }
 988         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
 989                    && (old_state == TCP_CONNTRACK_SYN_RECV
 990                        || old_state == TCP_CONNTRACK_ESTABLISHED)
 991                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
 992                 /* Set ASSURED if we see see valid ack in ESTABLISHED
 993                    after SYN_RECV or a valid answer for a picked up
 994                    connection. */
 995                 set_bit(IPS_ASSURED_BIT, &ct->status);
 996                 nf_conntrack_event_cache(IPCT_STATUS, ct);
 997         }
 998         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
 999
1000         return NF_ACCEPT;
1001 }
1002
1003 /* Called when a new connection for this protocol found. */
1004 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
1005                     unsigned int dataoff)
1006 {
1007         enum tcp_conntrack new_state;
1008         const struct tcphdr *th;
1009         struct tcphdr _tcph;
1010         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
1011         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
1012
1013         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1014         BUG_ON(th == NULL);
1015
1016         /* Don't need lock here: this conntrack not in circulation yet */
1017         new_state
1018                 = tcp_conntracks[0][get_conntrack_index(th)]
1019                 [TCP_CONNTRACK_NONE];
1020
1021         /* Invalid: delete conntrack */
1022         if (new_state >= TCP_CONNTRACK_MAX) {
1023                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1024                 return false;
1025         }
1026
1027         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1028                 /* SYN packet */
1029                 ct->proto.tcp.seen[0].td_end =
1030                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1031                                              dataoff, th);
1032                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1033                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1034                         ct->proto.tcp.seen[0].td_maxwin = 1;
1035                 ct->proto.tcp.seen[0].td_maxend =
1036                         ct->proto.tcp.seen[0].td_end;
1037
1038                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1039                 ct->proto.tcp.seen[1].flags = 0;
1040         } else if (nf_ct_tcp_loose == 0) {
1041                 /* Don't try to pick up connections. */
1042                 return false;
1043         } else {
1044                 /*
1045                  * We are in the middle of a connection,
1046                  * its history is lost for us.
1047                  * Let's try to use the data from the packet.
1048                  */
1049                 ct->proto.tcp.seen[0].td_end =
1050                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1051                                              dataoff, th);
1052                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1053                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1054                         ct->proto.tcp.seen[0].td_maxwin = 1;
1055                 ct->proto.tcp.seen[0].td_maxend =
1056                         ct->proto.tcp.seen[0].td_end +
1057                         ct->proto.tcp.seen[0].td_maxwin;
1058                 ct->proto.tcp.seen[0].td_scale = 0;
1059
1060                 /* We assume SACK and liberal window checking to handle
1061                  * window scaling */
1062                 ct->proto.tcp.seen[0].flags =
1063                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1064                                               IP_CT_TCP_FLAG_BE_LIBERAL;
1065         }
1066
1067         ct->proto.tcp.seen[1].td_end = 0;
1068         ct->proto.tcp.seen[1].td_maxend = 0;
1069         ct->proto.tcp.seen[1].td_maxwin = 1;
1070         ct->proto.tcp.seen[1].td_scale = 0;
1071
1072         /* tcp_packet will set them */
1073         ct->proto.tcp.state = TCP_CONNTRACK_NONE;
1074         ct->proto.tcp.last_index = TCP_NONE_SET;
1075
1076         pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1077                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1078                  sender->td_end, sender->td_maxend, sender->td_maxwin,
1079                  sender->td_scale,
1080                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1081                  receiver->td_scale);
1082         return true;
1083 }
1084
1085 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1086
1087 #include <linux/netfilter/nfnetlink.h>
1088 #include <linux/netfilter/nfnetlink_conntrack.h>
1089
1090 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1091                          const struct nf_conn *ct)
1092 {
1093         struct nlattr *nest_parms;
1094         struct nf_ct_tcp_flags tmp = {};
1095
1096         read_lock_bh(&tcp_lock);
1097         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1098         if (!nest_parms)
1099                 goto nla_put_failure;
1100
1101         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state);
1102
1103         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1104                    ct->proto.tcp.seen[0].td_scale);
1105
1106         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1107                    ct->proto.tcp.seen[1].td_scale);
1108
1109         tmp.flags = ct->proto.tcp.seen[0].flags;
1110         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1111                 sizeof(struct nf_ct_tcp_flags), &tmp);
1112
1113         tmp.flags = ct->proto.tcp.seen[1].flags;
1114         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1115                 sizeof(struct nf_ct_tcp_flags), &tmp);
1116         read_unlock_bh(&tcp_lock);
1117
1118         nla_nest_end(skb, nest_parms);
1119
1120         return 0;
1121
1122 nla_put_failure:
1123         read_unlock_bh(&tcp_lock);
1124         return -1;
1125 }
1126
1127 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1128         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1129         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1130         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1131         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1132         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len =  sizeof(struct nf_ct_tcp_flags) },
1133 };
1134
1135 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1136 {
1137         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1138         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1139         int err;
1140
1141         /* updates could not contain anything about the private
1142          * protocol info, in that case skip the parsing */
1143         if (!pattr)
1144                 return 0;
1145
1146         err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr, tcp_nla_policy);
1147         if (err < 0)
1148                 return err;
1149
1150         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1151             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1152                 return -EINVAL;
1153
1154         write_lock_bh(&tcp_lock);
1155         if (tb[CTA_PROTOINFO_TCP_STATE])
1156                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1157
1158         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1159                 struct nf_ct_tcp_flags *attr =
1160                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1161                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1162                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1163         }
1164
1165         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1166                 struct nf_ct_tcp_flags *attr =
1167                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1168                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1169                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1170         }
1171
1172         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1173             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1174             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1175             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1176                 ct->proto.tcp.seen[0].td_scale =
1177                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1178                 ct->proto.tcp.seen[1].td_scale =
1179                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1180         }
1181         write_unlock_bh(&tcp_lock);
1182
1183         return 0;
1184 }
1185 #endif
1186
1187 #ifdef CONFIG_SYSCTL
1188 static unsigned int tcp_sysctl_table_users;
1189 static struct ctl_table_header *tcp_sysctl_header;
1190 static struct ctl_table tcp_sysctl_table[] = {
1191         {
1192                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1193                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1194                 .maxlen         = sizeof(unsigned int),
1195                 .mode           = 0644,
1196                 .proc_handler   = proc_dointvec_jiffies,
1197         },
1198         {
1199                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1200                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1201                 .maxlen         = sizeof(unsigned int),
1202                 .mode           = 0644,
1203                 .proc_handler   = proc_dointvec_jiffies,
1204         },
1205         {
1206                 .procname       = "nf_conntrack_tcp_timeout_established",
1207                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1208                 .maxlen         = sizeof(unsigned int),
1209                 .mode           = 0644,
1210                 .proc_handler   = proc_dointvec_jiffies,
1211         },
1212         {
1213                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1214                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1215                 .maxlen         = sizeof(unsigned int),
1216                 .mode           = 0644,
1217                 .proc_handler   = proc_dointvec_jiffies,
1218         },
1219         {
1220                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1221                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1222                 .maxlen         = sizeof(unsigned int),
1223                 .mode           = 0644,
1224                 .proc_handler   = proc_dointvec_jiffies,
1225         },
1226         {
1227                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1228                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1229                 .maxlen         = sizeof(unsigned int),
1230                 .mode           = 0644,
1231                 .proc_handler   = proc_dointvec_jiffies,
1232         },
1233         {
1234                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1235                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1236                 .maxlen         = sizeof(unsigned int),
1237                 .mode           = 0644,
1238                 .proc_handler   = proc_dointvec_jiffies,
1239         },
1240         {
1241                 .procname       = "nf_conntrack_tcp_timeout_close",
1242                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1243                 .maxlen         = sizeof(unsigned int),
1244                 .mode           = 0644,
1245                 .proc_handler   = proc_dointvec_jiffies,
1246         },
1247         {
1248                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1249                 .data           = &nf_ct_tcp_timeout_max_retrans,
1250                 .maxlen         = sizeof(unsigned int),
1251                 .mode           = 0644,
1252                 .proc_handler   = proc_dointvec_jiffies,
1253         },
1254         {
1255                 .procname       = "nf_conntrack_tcp_timeout_unacknowledged",
1256                 .data           = &nf_ct_tcp_timeout_unacknowledged,
1257                 .maxlen         = sizeof(unsigned int),
1258                 .mode           = 0644,
1259                 .proc_handler   = proc_dointvec_jiffies,
1260         },
1261         {
1262                 .ctl_name       = NET_NF_CONNTRACK_TCP_LOOSE,
1263                 .procname       = "nf_conntrack_tcp_loose",
1264                 .data           = &nf_ct_tcp_loose,
1265                 .maxlen         = sizeof(unsigned int),
1266                 .mode           = 0644,
1267                 .proc_handler   = proc_dointvec,
1268         },
1269         {
1270                 .ctl_name       = NET_NF_CONNTRACK_TCP_BE_LIBERAL,
1271                 .procname       = "nf_conntrack_tcp_be_liberal",
1272                 .data           = &nf_ct_tcp_be_liberal,
1273                 .maxlen         = sizeof(unsigned int),
1274                 .mode           = 0644,
1275                 .proc_handler   = proc_dointvec,
1276         },
1277         {
1278                 .ctl_name       = NET_NF_CONNTRACK_TCP_MAX_RETRANS,
1279                 .procname       = "nf_conntrack_tcp_max_retrans",
1280                 .data           = &nf_ct_tcp_max_retrans,
1281                 .maxlen         = sizeof(unsigned int),
1282                 .mode           = 0644,
1283                 .proc_handler   = proc_dointvec,
1284         },
1285         {
1286                 .ctl_name       = 0
1287         }
1288 };
1289
1290 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1291 static struct ctl_table tcp_compat_sysctl_table[] = {
1292         {
1293                 .procname       = "ip_conntrack_tcp_timeout_syn_sent",
1294                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1295                 .maxlen         = sizeof(unsigned int),
1296                 .mode           = 0644,
1297                 .proc_handler   = proc_dointvec_jiffies,
1298         },
1299         {
1300                 .procname       = "ip_conntrack_tcp_timeout_syn_recv",
1301                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1302                 .maxlen         = sizeof(unsigned int),
1303                 .mode           = 0644,
1304                 .proc_handler   = proc_dointvec_jiffies,
1305         },
1306         {
1307                 .procname       = "ip_conntrack_tcp_timeout_established",
1308                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1309                 .maxlen         = sizeof(unsigned int),
1310                 .mode           = 0644,
1311                 .proc_handler   = proc_dointvec_jiffies,
1312         },
1313         {
1314                 .procname       = "ip_conntrack_tcp_timeout_fin_wait",
1315                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1316                 .maxlen         = sizeof(unsigned int),
1317                 .mode           = 0644,
1318                 .proc_handler   = proc_dointvec_jiffies,
1319         },
1320         {
1321                 .procname       = "ip_conntrack_tcp_timeout_close_wait",
1322                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1323                 .maxlen         = sizeof(unsigned int),
1324                 .mode           = 0644,
1325                 .proc_handler   = proc_dointvec_jiffies,
1326         },
1327         {
1328                 .procname       = "ip_conntrack_tcp_timeout_last_ack",
1329                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1330                 .maxlen         = sizeof(unsigned int),
1331                 .mode           = 0644,
1332                 .proc_handler   = proc_dointvec_jiffies,
1333         },
1334         {
1335                 .procname       = "ip_conntrack_tcp_timeout_time_wait",
1336                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1337                 .maxlen         = sizeof(unsigned int),
1338                 .mode           = 0644,
1339                 .proc_handler   = proc_dointvec_jiffies,
1340         },
1341         {
1342                 .procname       = "ip_conntrack_tcp_timeout_close",
1343                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1344                 .maxlen         = sizeof(unsigned int),
1345                 .mode           = 0644,
1346                 .proc_handler   = proc_dointvec_jiffies,
1347         },
1348         {
1349                 .procname       = "ip_conntrack_tcp_timeout_max_retrans",
1350                 .data           = &nf_ct_tcp_timeout_max_retrans,
1351                 .maxlen         = sizeof(unsigned int),
1352                 .mode           = 0644,
1353                 .proc_handler   = proc_dointvec_jiffies,
1354         },
1355         {
1356                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_LOOSE,
1357                 .procname       = "ip_conntrack_tcp_loose",
1358                 .data           = &nf_ct_tcp_loose,
1359                 .maxlen         = sizeof(unsigned int),
1360                 .mode           = 0644,
1361                 .proc_handler   = proc_dointvec,
1362         },
1363         {
1364                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_BE_LIBERAL,
1365                 .procname       = "ip_conntrack_tcp_be_liberal",
1366                 .data           = &nf_ct_tcp_be_liberal,
1367                 .maxlen         = sizeof(unsigned int),
1368                 .mode           = 0644,
1369                 .proc_handler   = proc_dointvec,
1370         },
1371         {
1372                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_MAX_RETRANS,
1373                 .procname       = "ip_conntrack_tcp_max_retrans",
1374                 .data           = &nf_ct_tcp_max_retrans,
1375                 .maxlen         = sizeof(unsigned int),
1376                 .mode           = 0644,
1377                 .proc_handler   = proc_dointvec,
1378         },
1379         {
1380                 .ctl_name       = 0
1381         }
1382 };
1383 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
1384 #endif /* CONFIG_SYSCTL */
1385
1386 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1387 {
1388         .l3proto                = PF_INET,
1389         .l4proto                = IPPROTO_TCP,
1390         .name                   = "tcp",
1391         .pkt_to_tuple           = tcp_pkt_to_tuple,
1392         .invert_tuple           = tcp_invert_tuple,
1393         .print_tuple            = tcp_print_tuple,
1394         .print_conntrack        = tcp_print_conntrack,
1395         .packet                 = tcp_packet,
1396         .new                    = tcp_new,
1397         .error                  = tcp_error,
1398 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1399         .to_nlattr              = tcp_to_nlattr,
1400         .from_nlattr            = nlattr_to_tcp,
1401         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1402         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1403         .nla_policy             = nf_ct_port_nla_policy,
1404 #endif
1405 #ifdef CONFIG_SYSCTL
1406         .ctl_table_users        = &tcp_sysctl_table_users,
1407         .ctl_table_header       = &tcp_sysctl_header,
1408         .ctl_table              = tcp_sysctl_table,
1409 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1410         .ctl_compat_table       = tcp_compat_sysctl_table,
1411 #endif
1412 #endif
1413 };
1414 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1415
1416 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1417 {
1418         .l3proto                = PF_INET6,
1419         .l4proto                = IPPROTO_TCP,
1420         .name                   = "tcp",
1421         .pkt_to_tuple           = tcp_pkt_to_tuple,
1422         .invert_tuple           = tcp_invert_tuple,
1423         .print_tuple            = tcp_print_tuple,
1424         .print_conntrack        = tcp_print_conntrack,
1425         .packet                 = tcp_packet,
1426         .new                    = tcp_new,
1427         .error                  = tcp_error,
1428 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1429         .to_nlattr              = tcp_to_nlattr,
1430         .from_nlattr            = nlattr_to_tcp,
1431         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1432         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1433         .nla_policy             = nf_ct_port_nla_policy,
1434 #endif
1435 #ifdef CONFIG_SYSCTL
1436         .ctl_table_users        = &tcp_sysctl_table_users,
1437         .ctl_table_header       = &tcp_sysctl_header,
1438         .ctl_table              = tcp_sysctl_table,
1439 #endif
1440 };
1441 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);