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 /* Protects ct->proto.tcp */
  33 static DEFINE_RWLOCK(tcp_lock);
  34
  35 /* "Be conservative in what you do,
  36     be liberal in what you accept from others."
  37     If it's non-zero, we mark only out of window RST segments as INVALID. */
  38 static int nf_ct_tcp_be_liberal __read_mostly = 0;
  39
  40 /* If it is set to zero, we disable picking up already established
  41    connections. */
  42 static int nf_ct_tcp_loose __read_mostly = 1;
  43
  44 /* Max number of the retransmitted packets without receiving an (acceptable)
  45    ACK from the destination. If this number is reached, a shorter timer
  46    will be started. */
  47 static int nf_ct_tcp_max_retrans __read_mostly = 3;
  48
  49   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  50      closely.  They're more complex. --RR */
  51
  52 static const char *const tcp_conntrack_names[] = {
  53         "NONE",
  54         "SYN_SENT",
  55         "SYN_RECV",
  56         "ESTABLISHED",
  57         "FIN_WAIT",
  58         "CLOSE_WAIT",
  59         "LAST_ACK",
  60         "TIME_WAIT",
  61         "CLOSE",
  62         "LISTEN"
  63 };
  64
  65 #define SECS * HZ
  66 #define MINS * 60 SECS
  67 #define HOURS * 60 MINS
  68 #define DAYS * 24 HOURS
  69
  70 /* RFC1122 says the R2 limit should be at least 100 seconds.
  71    Linux uses 15 packets as limit, which corresponds
  72    to ~13-30min depending on RTO. */
  73 static unsigned int nf_ct_tcp_timeout_max_retrans __read_mostly    =   5 MINS;
  74 static unsigned int nf_ct_tcp_timeout_unacknowledged __read_mostly =   5 MINS;
  75
  76 static unsigned int tcp_timeouts[TCP_CONNTRACK_MAX] __read_mostly = {
  77         [TCP_CONNTRACK_SYN_SENT]        = 2 MINS,
  78         [TCP_CONNTRACK_SYN_RECV]        = 60 SECS,
  79         [TCP_CONNTRACK_ESTABLISHED]     = 5 DAYS,
  80         [TCP_CONNTRACK_FIN_WAIT]        = 2 MINS,
  81         [TCP_CONNTRACK_CLOSE_WAIT]      = 60 SECS,
  82         [TCP_CONNTRACK_LAST_ACK]        = 30 SECS,
  83         [TCP_CONNTRACK_TIME_WAIT]       = 2 MINS,
  84         [TCP_CONNTRACK_CLOSE]           = 10 SECS,
  85 };
  86
  87 #define sNO TCP_CONNTRACK_NONE
  88 #define sSS TCP_CONNTRACK_SYN_SENT
  89 #define sSR TCP_CONNTRACK_SYN_RECV
  90 #define sES TCP_CONNTRACK_ESTABLISHED
  91 #define sFW TCP_CONNTRACK_FIN_WAIT
  92 #define sCW TCP_CONNTRACK_CLOSE_WAIT
  93 #define sLA TCP_CONNTRACK_LAST_ACK
  94 #define sTW TCP_CONNTRACK_TIME_WAIT
  95 #define sCL TCP_CONNTRACK_CLOSE
  96 #define sLI TCP_CONNTRACK_LISTEN
  97 #define sIV TCP_CONNTRACK_MAX
  98 #define sIG TCP_CONNTRACK_IGNORE
  99
 100 /* What TCP flags are set from RST/SYN/FIN/ACK. */
 101 enum tcp_bit_set {
 102         TCP_SYN_SET,
 103         TCP_SYNACK_SET,
 104         TCP_FIN_SET,
 105         TCP_ACK_SET,
 106         TCP_RST_SET,
 107         TCP_NONE_SET,
 108 };
 109
 110 /*
 111  * The TCP state transition table needs a few words...
 112  *
 113  * We are the man in the middle. All the packets go through us
 114  * but might get lost in transit to the destination.
 115  * It is assumed that the destinations can't receive segments
 116  * we haven't seen.
 117  *
 118  * The checked segment is in window, but our windows are *not*
 119  * equivalent with the ones of the sender/receiver. We always
 120  * try to guess the state of the current sender.
 121  *
 122  * The meaning of the states are:
 123  *
 124  * NONE:        initial state
 125  * SYN_SENT:    SYN-only packet seen
 126  * SYN_RECV:    SYN-ACK packet seen
 127  * ESTABLISHED: ACK packet seen
 128  * FIN_WAIT:    FIN packet seen
 129  * CLOSE_WAIT:  ACK seen (after FIN)
 130  * LAST_ACK:    FIN seen (after FIN)
 131  * TIME_WAIT:   last ACK seen
 132  * CLOSE:       closed connection (RST)
 133  *
 134  * LISTEN state is not used.
 135  *
 136  * Packets marked as IGNORED (sIG):
 137  *      if they may be either invalid or valid
 138  *      and the receiver may send back a connection
 139  *      closing RST or a SYN/ACK.
 140  *
 141  * Packets marked as INVALID (sIV):
 142  *      if they are invalid
 143  *      or we do not support the request (simultaneous open)
 144  */
 145 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 146         {
 147 /* ORIGINAL */
 148 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 149 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
 150 /*
 151  *      sNO -> sSS      Initialize a new connection
 152  *      sSS -> sSS      Retransmitted SYN
 153  *      sSR -> sIG      Late retransmitted SYN?
 154  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 155  *                      are errors. Receiver will reply with RST
 156  *                      and close the connection.
 157  *                      Or we are not in sync and hold a dead connection.
 158  *      sFW -> sIG
 159  *      sCW -> sIG
 160  *      sLA -> sIG
 161  *      sTW -> sSS      Reopened connection (RFC 1122).
 162  *      sCL -> sSS
 163  */
 164 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 165 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 166 /*
 167  * A SYN/ACK from the client is always invalid:
 168  *      - either it tries to set up a simultaneous open, which is
 169  *        not supported;
 170  *      - or the firewall has just been inserted between the two hosts
 171  *        during the session set-up. The SYN will be retransmitted
 172  *        by the true client (or it'll time out).
 173  */
 174 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 175 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 176 /*
 177  *      sNO -> sIV      Too late and no reason to do anything...
 178  *      sSS -> sIV      Client migth not send FIN in this state:
 179  *                      we enforce waiting for a SYN/ACK reply first.
 180  *      sSR -> sFW      Close started.
 181  *      sES -> sFW
 182  *      sFW -> sLA      FIN seen in both directions, waiting for
 183  *                      the last ACK.
 184  *                      Migth be a retransmitted FIN as well...
 185  *      sCW -> sLA
 186  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 187  *      sTW -> sTW
 188  *      sCL -> sCL
 189  */
 190 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 191 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 192 /*
 193  *      sNO -> sES      Assumed.
 194  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 195  *      sSR -> sES      Established state is reached.
 196  *      sES -> sES      :-)
 197  *      sFW -> sCW      Normal close request answered by ACK.
 198  *      sCW -> sCW
 199  *      sLA -> sTW      Last ACK detected.
 200  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 201  *      sCL -> sCL
 202  */
 203 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 204 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 205 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 206         },
 207         {
 208 /* REPLY */
 209 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 210 /*syn*/    { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 211 /*
 212  *      sNO -> sIV      Never reached.
 213  *      sSS -> sIV      Simultaneous open, not supported
 214  *      sSR -> sIV      Simultaneous open, not supported.
 215  *      sES -> sIV      Server may not initiate a connection.
 216  *      sFW -> sIV
 217  *      sCW -> sIV
 218  *      sLA -> sIV
 219  *      sTW -> sIV      Reopened connection, but server may not do it.
 220  *      sCL -> sIV
 221  */
 222 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 223 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
 224 /*
 225  *      sSS -> sSR      Standard open.
 226  *      sSR -> sSR      Retransmitted SYN/ACK.
 227  *      sES -> sIG      Late retransmitted SYN/ACK?
 228  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 229  *      sCW -> sIG
 230  *      sLA -> sIG
 231  *      sTW -> sIG
 232  *      sCL -> sIG
 233  */
 234 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 235 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 236 /*
 237  *      sSS -> sIV      Server might not send FIN in this state.
 238  *      sSR -> sFW      Close started.
 239  *      sES -> sFW
 240  *      sFW -> sLA      FIN seen in both directions.
 241  *      sCW -> sLA
 242  *      sLA -> sLA      Retransmitted FIN.
 243  *      sTW -> sTW
 244  *      sCL -> sCL
 245  */
 246 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 247 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 248 /*
 249  *      sSS -> sIG      Might be a half-open connection.
 250  *      sSR -> sSR      Might answer late resent SYN.
 251  *      sES -> sES      :-)
 252  *      sFW -> sCW      Normal close request answered by ACK.
 253  *      sCW -> sCW
 254  *      sLA -> sTW      Last ACK detected.
 255  *      sTW -> sTW      Retransmitted last ACK.
 256  *      sCL -> sCL
 257  */
 258 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 259 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 260 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 261         }
 262 };
 263
 264 static bool tcp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
 265                              struct nf_conntrack_tuple *tuple)
 266 {
 267         const struct tcphdr *hp;
 268         struct tcphdr _hdr;
 269
 270         /* Actually only need first 8 bytes. */
 271         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
 272         if (hp == NULL)
 273                 return false;
 274
 275         tuple->src.u.tcp.port = hp->source;
 276         tuple->dst.u.tcp.port = hp->dest;
 277
 278         return true;
 279 }
 280
 281 static bool tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
 282                              const struct nf_conntrack_tuple *orig)
 283 {
 284         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
 285         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
 286         return true;
 287 }
 288
 289 /* Print out the per-protocol part of the tuple. */
 290 static int tcp_print_tuple(struct seq_file *s,
 291                            const struct nf_conntrack_tuple *tuple)
 292 {
 293         return seq_printf(s, "sport=%hu dport=%hu ",
 294                           ntohs(tuple->src.u.tcp.port),
 295                           ntohs(tuple->dst.u.tcp.port));
 296 }
 297
 298 /* Print out the private part of the conntrack. */
 299 static int tcp_print_conntrack(struct seq_file *s, const struct nf_conn *ct)
 300 {
 301         enum tcp_conntrack state;
 302
 303         read_lock_bh(&tcp_lock);
 304         state = ct->proto.tcp.state;
 305         read_unlock_bh(&tcp_lock);
 306
 307         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
 308 }
 309
 310 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 311 {
 312         if (tcph->rst) return TCP_RST_SET;
 313         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 314         else if (tcph->fin) return TCP_FIN_SET;
 315         else if (tcph->ack) return TCP_ACK_SET;
 316         else return TCP_NONE_SET;
 317 }
 318
 319 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 320    in IP Filter' by Guido van Rooij.
 321
 322    http://www.nluug.nl/events/sane2000/papers.html
 323    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
 324
 325    The boundaries and the conditions are changed according to RFC793:
 326    the packet must intersect the window (i.e. segments may be
 327    after the right or before the left edge) and thus receivers may ACK
 328    segments after the right edge of the window.
 329
 330         td_maxend = max(sack + max(win,1)) seen in reply packets
 331         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 332         td_maxwin += seq + len - sender.td_maxend
 333                         if seq + len > sender.td_maxend
 334         td_end    = max(seq + len) seen in sent packets
 335
 336    I.   Upper bound for valid data:     seq <= sender.td_maxend
 337    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 338    III. Upper bound for valid (s)ack:   sack <= receiver.td_end
 339    IV.  Lower bound for valid (s)ack:   sack >= receiver.td_end - MAXACKWINDOW
 340
 341    where sack is the highest right edge of sack block found in the packet
 342    or ack in the case of packet without SACK option.
 343
 344    The upper bound limit for a valid (s)ack is not ignored -
 345    we doesn't have to deal with fragments.
 346 */
 347
 348 static inline __u32 segment_seq_plus_len(__u32 seq,
 349                                          size_t len,
 350                                          unsigned int dataoff,
 351                                          const struct tcphdr *tcph)
 352 {
 353         /* XXX Should I use payload length field in IP/IPv6 header ?
 354          * - YK */
 355         return (seq + len - dataoff - tcph->doff*4
 356                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 357 }
 358
 359 /* Fixme: what about big packets? */
 360 #define MAXACKWINCONST                  66000
 361 #define MAXACKWINDOW(sender)                                            \
 362         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 363                                               : MAXACKWINCONST)
 364
 365 /*
 366  * Simplified tcp_parse_options routine from tcp_input.c
 367  */
 368 static void tcp_options(const struct sk_buff *skb,
 369                         unsigned int dataoff,
 370                         const struct tcphdr *tcph,
 371                         struct ip_ct_tcp_state *state)
 372 {
 373         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 374         const unsigned char *ptr;
 375         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 376
 377         if (!length)
 378                 return;
 379
 380         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 381                                  length, buff);
 382         BUG_ON(ptr == NULL);
 383
 384         state->td_scale =
 385         state->flags = 0;
 386
 387         while (length > 0) {
 388                 int opcode=*ptr++;
 389                 int opsize;
 390
 391                 switch (opcode) {
 392                 case TCPOPT_EOL:
 393                         return;
 394                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 395                         length--;
 396                         continue;
 397                 default:
 398                         opsize=*ptr++;
 399                         if (opsize < 2) /* "silly options" */
 400                                 return;
 401                         if (opsize > length)
 402                                 break;  /* don't parse partial options */
 403
 404                         if (opcode == TCPOPT_SACK_PERM
 405                             && opsize == TCPOLEN_SACK_PERM)
 406                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 407                         else if (opcode == TCPOPT_WINDOW
 408                                  && opsize == TCPOLEN_WINDOW) {
 409                                 state->td_scale = *(u_int8_t *)ptr;
 410
 411                                 if (state->td_scale > 14) {
 412                                         /* See RFC1323 */
 413                                         state->td_scale = 14;
 414                                 }
 415                                 state->flags |=
 416                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 417                         }
 418                         ptr += opsize - 2;
 419                         length -= opsize;
 420                 }
 421         }
 422 }
 423
 424 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 425                      const struct tcphdr *tcph, __u32 *sack)
 426 {
 427         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 428         const unsigned char *ptr;
 429         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 430         __u32 tmp;
 431
 432         if (!length)
 433                 return;
 434
 435         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 436                                  length, buff);
 437         BUG_ON(ptr == NULL);
 438
 439         /* Fast path for timestamp-only option */
 440         if (length == TCPOLEN_TSTAMP_ALIGNED*4
 441             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
 442                                        | (TCPOPT_NOP << 16)
 443                                        | (TCPOPT_TIMESTAMP << 8)
 444                                        | TCPOLEN_TIMESTAMP))
 445                 return;
 446
 447         while (length > 0) {
 448                 int opcode = *ptr++;
 449                 int opsize, i;
 450
 451                 switch (opcode) {
 452                 case TCPOPT_EOL:
 453                         return;
 454                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 455                         length--;
 456                         continue;
 457                 default:
 458                         opsize = *ptr++;
 459                         if (opsize < 2) /* "silly options" */
 460                                 return;
 461                         if (opsize > length)
 462                                 break;  /* don't parse partial options */
 463
 464                         if (opcode == TCPOPT_SACK
 465                             && opsize >= (TCPOLEN_SACK_BASE
 466                                           + TCPOLEN_SACK_PERBLOCK)
 467                             && !((opsize - TCPOLEN_SACK_BASE)
 468                                  % TCPOLEN_SACK_PERBLOCK)) {
 469                                 for (i = 0;
 470                                      i < (opsize - TCPOLEN_SACK_BASE);
 471                                      i += TCPOLEN_SACK_PERBLOCK) {
 472                                         tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
 473
 474                                         if (after(tmp, *sack))
 475                                                 *sack = tmp;
 476                                 }
 477                                 return;
 478                         }
 479                         ptr += opsize - 2;
 480                         length -= opsize;
 481                 }
 482         }
 483 }
 484
 485 static bool tcp_in_window(const struct nf_conn *ct,
 486                           struct ip_ct_tcp *state,
 487                           enum ip_conntrack_dir dir,
 488                           unsigned int index,
 489                           const struct sk_buff *skb,
 490                           unsigned int dataoff,
 491                           const struct tcphdr *tcph,
 492                           u_int8_t pf)
 493 {
 494         struct net *net = nf_ct_net(ct);
 495         struct ip_ct_tcp_state *sender = &state->seen[dir];
 496         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 497         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 498         __u32 seq, ack, sack, end, win, swin;
 499         bool res;
 500
 501         /*
 502          * Get the required data from the packet.
 503          */
 504         seq = ntohl(tcph->seq);
 505         ack = sack = ntohl(tcph->ack_seq);
 506         win = ntohs(tcph->window);
 507         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 508
 509         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 510                 tcp_sack(skb, dataoff, tcph, &sack);
 511
 512         pr_debug("tcp_in_window: START\n");
 513         pr_debug("tcp_in_window: ");
 514         nf_ct_dump_tuple(tuple);
 515         pr_debug("seq=%u ack=%u sack=%u win=%u end=%u\n",
 516                  seq, ack, sack, win, end);
 517         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 518                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 519                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 520                  sender->td_scale,
 521                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 522                  receiver->td_scale);
 523
 524         if (sender->td_end == 0) {
 525                 /*
 526                  * Initialize sender data.
 527                  */
 528                 if (tcph->syn && tcph->ack) {
 529                         /*
 530                          * Outgoing SYN-ACK in reply to a SYN.
 531                          */
 532                         sender->td_end =
 533                         sender->td_maxend = end;
 534                         sender->td_maxwin = (win == 0 ? 1 : win);
 535
 536                         tcp_options(skb, dataoff, tcph, sender);
 537                         /*
 538                          * RFC 1323:
 539                          * Both sides must send the Window Scale option
 540                          * to enable window scaling in either direction.
 541                          */
 542                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 543                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 544                                 sender->td_scale =
 545                                 receiver->td_scale = 0;
 546                 } else {
 547                         /*
 548                          * We are in the middle of a connection,
 549                          * its history is lost for us.
 550                          * Let's try to use the data from the packet.
 551                          */
 552                         sender->td_end = end;
 553                         sender->td_maxwin = (win == 0 ? 1 : win);
 554                         sender->td_maxend = end + sender->td_maxwin;
 555                 }
 556         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 557                      && dir == IP_CT_DIR_ORIGINAL)
 558                    || (state->state == TCP_CONNTRACK_SYN_RECV
 559                      && dir == IP_CT_DIR_REPLY))
 560                    && after(end, sender->td_end)) {
 561                 /*
 562                  * RFC 793: "if a TCP is reinitialized ... then it need
 563                  * not wait at all; it must only be sure to use sequence
 564                  * numbers larger than those recently used."
 565                  */
 566                 sender->td_end =
 567                 sender->td_maxend = end;
 568                 sender->td_maxwin = (win == 0 ? 1 : win);
 569
 570                 tcp_options(skb, dataoff, tcph, sender);
 571         }
 572
 573         if (!(tcph->ack)) {
 574                 /*
 575                  * If there is no ACK, just pretend it was set and OK.
 576                  */
 577                 ack = sack = receiver->td_end;
 578         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 579                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 580                    && (ack == 0)) {
 581                 /*
 582                  * Broken TCP stacks, that set ACK in RST packets as well
 583                  * with zero ack value.
 584                  */
 585                 ack = sack = receiver->td_end;
 586         }
 587
 588         if (seq == end
 589             && (!tcph->rst
 590                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
 591                 /*
 592                  * Packets contains no data: we assume it is valid
 593                  * and check the ack value only.
 594                  * However RST segments are always validated by their
 595                  * SEQ number, except when seq == 0 (reset sent answering
 596                  * SYN.
 597                  */
 598                 seq = end = sender->td_end;
 599
 600         pr_debug("tcp_in_window: ");
 601         nf_ct_dump_tuple(tuple);
 602         pr_debug("seq=%u ack=%u sack =%u win=%u end=%u\n",
 603                  seq, ack, sack, win, end);
 604         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 605                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 606                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 607                  sender->td_scale,
 608                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 609                  receiver->td_scale);
 610
 611         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 612                  before(seq, sender->td_maxend + 1),
 613                  after(end, sender->td_end - receiver->td_maxwin - 1),
 614                  before(sack, receiver->td_end + 1),
 615                  after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 616
 617         if (before(seq, sender->td_maxend + 1) &&
 618             after(end, sender->td_end - receiver->td_maxwin - 1) &&
 619             before(sack, receiver->td_end + 1) &&
 620             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
 621                 /*
 622                  * Take into account window scaling (RFC 1323).
 623                  */
 624                 if (!tcph->syn)
 625                         win <<= sender->td_scale;
 626
 627                 /*
 628                  * Update sender data.
 629                  */
 630                 swin = win + (sack - ack);
 631                 if (sender->td_maxwin < swin)
 632                         sender->td_maxwin = swin;
 633                 if (after(end, sender->td_end)) {
 634                         sender->td_end = end;
 635                         sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 636                 }
 637                 /*
 638                  * Update receiver data.
 639                  */
 640                 if (after(end, sender->td_maxend))
 641                         receiver->td_maxwin += end - sender->td_maxend;
 642                 if (after(sack + win, receiver->td_maxend - 1)) {
 643                         receiver->td_maxend = sack + win;
 644                         if (win == 0)
 645                                 receiver->td_maxend++;
 646                 }
 647                 if (ack == receiver->td_end)
 648                         receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 649
 650                 /*
 651                  * Check retransmissions.
 652                  */
 653                 if (index == TCP_ACK_SET) {
 654                         if (state->last_dir == dir
 655                             && state->last_seq == seq
 656                             && state->last_ack == ack
 657                             && state->last_end == end
 658                             && state->last_win == win)
 659                                 state->retrans++;
 660                         else {
 661                                 state->last_dir = dir;
 662                                 state->last_seq = seq;
 663                                 state->last_ack = ack;
 664                                 state->last_end = end;
 665                                 state->last_win = win;
 666                                 state->retrans = 0;
 667                         }
 668                 }
 669                 res = true;
 670         } else {
 671                 res = false;
 672                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 673                     nf_ct_tcp_be_liberal)
 674                         res = true;
 675                 if (!res && LOG_INVALID(net, IPPROTO_TCP))
 676                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 677                         "nf_ct_tcp: %s ",
 678                         before(seq, sender->td_maxend + 1) ?
 679                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
 680                         before(sack, receiver->td_end + 1) ?
 681                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
 682                         : "ACK is under the lower bound (possible overly delayed ACK)"
 683                         : "ACK is over the upper bound (ACKed data not seen yet)"
 684                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 685                         : "SEQ is over the upper bound (over the window of the receiver)");
 686         }
 687
 688         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 689                  "receiver end=%u maxend=%u maxwin=%u\n",
 690                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 691                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 692
 693         return res;
 694 }
 695
 696 #ifdef CONFIG_NF_NAT_NEEDED
 697 /* Update sender->td_end after NAT successfully mangled the packet */
 698 /* Caller must linearize skb at tcp header. */
 699 void nf_conntrack_tcp_update(const struct sk_buff *skb,
 700                              unsigned int dataoff,
 701                              struct nf_conn *ct,
 702                              int dir)
 703 {
 704         const struct tcphdr *tcph = (const void *)skb->data + dataoff;
 705         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[dir];
 706         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[!dir];
 707         __u32 end;
 708
 709         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
 710
 711         write_lock_bh(&tcp_lock);
 712         /*
 713          * We have to worry for the ack in the reply packet only...
 714          */
 715         if (after(end, ct->proto.tcp.seen[dir].td_end))
 716                 ct->proto.tcp.seen[dir].td_end = end;
 717         ct->proto.tcp.last_end = end;
 718         write_unlock_bh(&tcp_lock);
 719         pr_debug("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
 720                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 721                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 722                  sender->td_scale,
 723                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 724                  receiver->td_scale);
 725 }
 726 EXPORT_SYMBOL_GPL(nf_conntrack_tcp_update);
 727 #endif
 728
 729 #define TH_FIN  0x01
 730 #define TH_SYN  0x02
 731 #define TH_RST  0x04
 732 #define TH_PUSH 0x08
 733 #define TH_ACK  0x10
 734 #define TH_URG  0x20
 735 #define TH_ECE  0x40
 736 #define TH_CWR  0x80
 737
 738 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 739 static const u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_ACK|TH_URG) + 1] =
 740 {
 741         [TH_SYN]                        = 1,
 742         [TH_SYN|TH_URG]                 = 1,
 743         [TH_SYN|TH_ACK]                 = 1,
 744         [TH_RST]                        = 1,
 745         [TH_RST|TH_ACK]                 = 1,
 746         [TH_FIN|TH_ACK]                 = 1,
 747         [TH_FIN|TH_ACK|TH_URG]          = 1,
 748         [TH_ACK]                        = 1,
 749         [TH_ACK|TH_URG]                 = 1,
 750 };
 751
 752 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 753 static int tcp_error(struct net *net,
 754                      struct sk_buff *skb,
 755                      unsigned int dataoff,
 756                      enum ip_conntrack_info *ctinfo,
 757                      u_int8_t pf,
 758                      unsigned int hooknum)
 759 {
 760         const struct tcphdr *th;
 761         struct tcphdr _tcph;
 762         unsigned int tcplen = skb->len - dataoff;
 763         u_int8_t tcpflags;
 764
 765         /* Smaller that minimal TCP header? */
 766         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 767         if (th == NULL) {
 768                 if (LOG_INVALID(net, IPPROTO_TCP))
 769                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 770                                 "nf_ct_tcp: short packet ");
 771                 return -NF_ACCEPT;
 772         }
 773
 774         /* Not whole TCP header or malformed packet */
 775         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 776                 if (LOG_INVALID(net, IPPROTO_TCP))
 777                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 778                                 "nf_ct_tcp: truncated/malformed packet ");
 779                 return -NF_ACCEPT;
 780         }
 781
 782         /* Checksum invalid? Ignore.
 783          * We skip checking packets on the outgoing path
 784          * because the checksum is assumed to be correct.
 785          */
 786         /* FIXME: Source route IP option packets --RR */
 787         if (net->ct.sysctl_checksum && hooknum == NF_INET_PRE_ROUTING &&
 788             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
 789                 if (LOG_INVALID(net, IPPROTO_TCP))
 790                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 791                                   "nf_ct_tcp: bad TCP checksum ");
 792                 return -NF_ACCEPT;
 793         }
 794
 795         /* Check TCP flags. */
 796         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR|TH_PUSH));
 797         if (!tcp_valid_flags[tcpflags]) {
 798                 if (LOG_INVALID(net, IPPROTO_TCP))
 799                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 800                                   "nf_ct_tcp: invalid TCP flag combination ");
 801                 return -NF_ACCEPT;
 802         }
 803
 804         return NF_ACCEPT;
 805 }
 806
 807 /* Returns verdict for packet, or -1 for invalid. */
 808 static int tcp_packet(struct nf_conn *ct,
 809                       const struct sk_buff *skb,
 810                       unsigned int dataoff,
 811                       enum ip_conntrack_info ctinfo,
 812                       u_int8_t pf,
 813                       unsigned int hooknum)
 814 {
 815         struct net *net = nf_ct_net(ct);
 816         struct nf_conntrack_tuple *tuple;
 817         enum tcp_conntrack new_state, old_state;
 818         enum ip_conntrack_dir dir;
 819         const struct tcphdr *th;
 820         struct tcphdr _tcph;
 821         unsigned long timeout;
 822         unsigned int index;
 823
 824         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 825         BUG_ON(th == NULL);
 826
 827         write_lock_bh(&tcp_lock);
 828         old_state = ct->proto.tcp.state;
 829         dir = CTINFO2DIR(ctinfo);
 830         index = get_conntrack_index(th);
 831         new_state = tcp_conntracks[dir][index][old_state];
 832         tuple = &ct->tuplehash[dir].tuple;
 833
 834         switch (new_state) {
 835         case TCP_CONNTRACK_SYN_SENT:
 836                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 837                         break;
 838                 /* RFC 1122: "When a connection is closed actively,
 839                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 840                  * (Maximum Segment Lifetime). However, it MAY accept
 841                  * a new SYN from the remote TCP to reopen the connection
 842                  * directly from TIME-WAIT state, if..."
 843                  * We ignore the conditions because we are in the
 844                  * TIME-WAIT state anyway.
 845                  *
 846                  * Handle aborted connections: we and the server
 847                  * think there is an existing connection but the client
 848                  * aborts it and starts a new one.
 849                  */
 850                 if (((ct->proto.tcp.seen[dir].flags
 851                       | ct->proto.tcp.seen[!dir].flags)
 852                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 853                     || (ct->proto.tcp.last_dir == dir
 854                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 855                         /* Attempt to reopen a closed/aborted connection.
 856                          * Delete this connection and look up again. */
 857                         write_unlock_bh(&tcp_lock);
 858
 859                         /* Only repeat if we can actually remove the timer.
 860                          * Destruction may already be in progress in process
 861                          * context and we must give it a chance to terminate.
 862                          */
 863                         if (nf_ct_kill(ct))
 864                                 return -NF_REPEAT;
 865                         return NF_DROP;
 866                 }
 867                 /* Fall through */
 868         case TCP_CONNTRACK_IGNORE:
 869                 /* Ignored packets:
 870                  *
 871                  * Our connection entry may be out of sync, so ignore
 872                  * packets which may signal the real connection between
 873                  * the client and the server.
 874                  *
 875                  * a) SYN in ORIGINAL
 876                  * b) SYN/ACK in REPLY
 877                  * c) ACK in reply direction after initial SYN in original.
 878                  *
 879                  * If the ignored packet is invalid, the receiver will send
 880                  * a RST we'll catch below.
 881                  */
 882                 if (index == TCP_SYNACK_SET
 883                     && ct->proto.tcp.last_index == TCP_SYN_SET
 884                     && ct->proto.tcp.last_dir != dir
 885                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 886                         /* b) This SYN/ACK acknowledges a SYN that we earlier
 887                          * ignored as invalid. This means that the client and
 888                          * the server are both in sync, while the firewall is
 889                          * not. We kill this session and block the SYN/ACK so
 890                          * that the client cannot but retransmit its SYN and
 891                          * thus initiate a clean new session.
 892                          */
 893                         write_unlock_bh(&tcp_lock);
 894                         if (LOG_INVALID(net, IPPROTO_TCP))
 895                                 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 896                                           "nf_ct_tcp: killing out of sync session ");
 897                         nf_ct_kill(ct);
 898                         return NF_DROP;
 899                 }
 900                 ct->proto.tcp.last_index = index;
 901                 ct->proto.tcp.last_dir = dir;
 902                 ct->proto.tcp.last_seq = ntohl(th->seq);
 903                 ct->proto.tcp.last_end =
 904                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 905
 906                 write_unlock_bh(&tcp_lock);
 907                 if (LOG_INVALID(net, IPPROTO_TCP))
 908                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 909                                   "nf_ct_tcp: invalid packet ignored ");
 910                 return NF_ACCEPT;
 911         case TCP_CONNTRACK_MAX:
 912                 /* Invalid packet */
 913                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 914                          dir, get_conntrack_index(th), old_state);
 915                 write_unlock_bh(&tcp_lock);
 916                 if (LOG_INVALID(net, IPPROTO_TCP))
 917                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 918                                   "nf_ct_tcp: invalid state ");
 919                 return -NF_ACCEPT;
 920         case TCP_CONNTRACK_CLOSE:
 921                 if (index == TCP_RST_SET
 922                     && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
 923                          && ct->proto.tcp.last_index == TCP_SYN_SET)
 924                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
 925                             && ct->proto.tcp.last_index == TCP_ACK_SET))
 926                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 927                         /* RST sent to invalid SYN or ACK we had let through
 928                          * at a) and c) above:
 929                          *
 930                          * a) SYN was in window then
 931                          * c) we hold a half-open connection.
 932                          *
 933                          * Delete our connection entry.
 934                          * We skip window checking, because packet might ACK
 935                          * segments we ignored. */
 936                         goto in_window;
 937                 }
 938                 /* Just fall through */
 939         default:
 940                 /* Keep compilers happy. */
 941                 break;
 942         }
 943
 944         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
 945                            skb, dataoff, th, pf)) {
 946                 write_unlock_bh(&tcp_lock);
 947                 return -NF_ACCEPT;
 948         }
 949      in_window:
 950         /* From now on we have got in-window packets */
 951         ct->proto.tcp.last_index = index;
 952         ct->proto.tcp.last_dir = dir;
 953
 954         pr_debug("tcp_conntracks: ");
 955         nf_ct_dump_tuple(tuple);
 956         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
 957                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
 958                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
 959                  old_state, new_state);
 960
 961         ct->proto.tcp.state = new_state;
 962         if (old_state != new_state
 963             && new_state == TCP_CONNTRACK_FIN_WAIT)
 964                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
 965
 966         if (ct->proto.tcp.retrans >= nf_ct_tcp_max_retrans &&
 967             tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans)
 968                 timeout = nf_ct_tcp_timeout_max_retrans;
 969         else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
 970                  IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
 971                  tcp_timeouts[new_state] > nf_ct_tcp_timeout_unacknowledged)
 972                 timeout = nf_ct_tcp_timeout_unacknowledged;
 973         else
 974                 timeout = tcp_timeouts[new_state];
 975         write_unlock_bh(&tcp_lock);
 976
 977         nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, ct);
 978         if (new_state != old_state)
 979                 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
 980
 981         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 982                 /* If only reply is a RST, we can consider ourselves not to
 983                    have an established connection: this is a fairly common
 984                    problem case, so we can delete the conntrack
 985                    immediately.  --RR */
 986                 if (th->rst) {
 987                         nf_ct_kill_acct(ct, ctinfo, skb);
 988                         return NF_ACCEPT;
 989                 }
 990         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
 991                    && (old_state == TCP_CONNTRACK_SYN_RECV
 992                        || old_state == TCP_CONNTRACK_ESTABLISHED)
 993                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
 994                 /* Set ASSURED if we see see valid ack in ESTABLISHED
 995                    after SYN_RECV or a valid answer for a picked up
 996                    connection. */
 997                 set_bit(IPS_ASSURED_BIT, &ct->status);
 998                 nf_conntrack_event_cache(IPCT_STATUS, ct);
 999         }
1000         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1001
1002         return NF_ACCEPT;
1003 }
1004
1005 /* Called when a new connection for this protocol found. */
1006 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
1007                     unsigned int dataoff)
1008 {
1009         enum tcp_conntrack new_state;
1010         const struct tcphdr *th;
1011         struct tcphdr _tcph;
1012         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
1013         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
1014
1015         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1016         BUG_ON(th == NULL);
1017
1018         /* Don't need lock here: this conntrack not in circulation yet */
1019         new_state
1020                 = tcp_conntracks[0][get_conntrack_index(th)]
1021                 [TCP_CONNTRACK_NONE];
1022
1023         /* Invalid: delete conntrack */
1024         if (new_state >= TCP_CONNTRACK_MAX) {
1025                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1026                 return false;
1027         }
1028
1029         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1030                 /* SYN packet */
1031                 ct->proto.tcp.seen[0].td_end =
1032                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1033                                              dataoff, th);
1034                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1035                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1036                         ct->proto.tcp.seen[0].td_maxwin = 1;
1037                 ct->proto.tcp.seen[0].td_maxend =
1038                         ct->proto.tcp.seen[0].td_end;
1039
1040                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1041                 ct->proto.tcp.seen[1].flags = 0;
1042         } else if (nf_ct_tcp_loose == 0) {
1043                 /* Don't try to pick up connections. */
1044                 return false;
1045         } else {
1046                 /*
1047                  * We are in the middle of a connection,
1048                  * its history is lost for us.
1049                  * Let's try to use the data from the packet.
1050                  */
1051                 ct->proto.tcp.seen[0].td_end =
1052                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1053                                              dataoff, th);
1054                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1055                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1056                         ct->proto.tcp.seen[0].td_maxwin = 1;
1057                 ct->proto.tcp.seen[0].td_maxend =
1058                         ct->proto.tcp.seen[0].td_end +
1059                         ct->proto.tcp.seen[0].td_maxwin;
1060                 ct->proto.tcp.seen[0].td_scale = 0;
1061
1062                 /* We assume SACK and liberal window checking to handle
1063                  * window scaling */
1064                 ct->proto.tcp.seen[0].flags =
1065                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1066                                               IP_CT_TCP_FLAG_BE_LIBERAL;
1067         }
1068
1069         ct->proto.tcp.seen[1].td_end = 0;
1070         ct->proto.tcp.seen[1].td_maxend = 0;
1071         ct->proto.tcp.seen[1].td_maxwin = 1;
1072         ct->proto.tcp.seen[1].td_scale = 0;
1073
1074         /* tcp_packet will set them */
1075         ct->proto.tcp.state = TCP_CONNTRACK_NONE;
1076         ct->proto.tcp.last_index = TCP_NONE_SET;
1077
1078         pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1079                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1080                  sender->td_end, sender->td_maxend, sender->td_maxwin,
1081                  sender->td_scale,
1082                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1083                  receiver->td_scale);
1084         return true;
1085 }
1086
1087 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1088
1089 #include <linux/netfilter/nfnetlink.h>
1090 #include <linux/netfilter/nfnetlink_conntrack.h>
1091
1092 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1093                          const struct nf_conn *ct)
1094 {
1095         struct nlattr *nest_parms;
1096         struct nf_ct_tcp_flags tmp = {};
1097
1098         read_lock_bh(&tcp_lock);
1099         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1100         if (!nest_parms)
1101                 goto nla_put_failure;
1102
1103         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state);
1104
1105         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1106                    ct->proto.tcp.seen[0].td_scale);
1107
1108         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1109                    ct->proto.tcp.seen[1].td_scale);
1110
1111         tmp.flags = ct->proto.tcp.seen[0].flags;
1112         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1113                 sizeof(struct nf_ct_tcp_flags), &tmp);
1114
1115         tmp.flags = ct->proto.tcp.seen[1].flags;
1116         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1117                 sizeof(struct nf_ct_tcp_flags), &tmp);
1118         read_unlock_bh(&tcp_lock);
1119
1120         nla_nest_end(skb, nest_parms);
1121
1122         return 0;
1123
1124 nla_put_failure:
1125         read_unlock_bh(&tcp_lock);
1126         return -1;
1127 }
1128
1129 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1130         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1131         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1132         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1133         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1134         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len =  sizeof(struct nf_ct_tcp_flags) },
1135 };
1136
1137 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1138 {
1139         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1140         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1141         int err;
1142
1143         /* updates could not contain anything about the private
1144          * protocol info, in that case skip the parsing */
1145         if (!pattr)
1146                 return 0;
1147
1148         err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr, tcp_nla_policy);
1149         if (err < 0)
1150                 return err;
1151
1152         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1153             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1154                 return -EINVAL;
1155
1156         write_lock_bh(&tcp_lock);
1157         if (tb[CTA_PROTOINFO_TCP_STATE])
1158                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1159
1160         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1161                 struct nf_ct_tcp_flags *attr =
1162                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1163                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1164                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1165         }
1166
1167         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1168                 struct nf_ct_tcp_flags *attr =
1169                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1170                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1171                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1172         }
1173
1174         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1175             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1176             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1177             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1178                 ct->proto.tcp.seen[0].td_scale =
1179                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1180                 ct->proto.tcp.seen[1].td_scale =
1181                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1182         }
1183         write_unlock_bh(&tcp_lock);
1184
1185         return 0;
1186 }
1187
1188 static int tcp_nlattr_size(void)
1189 {
1190         return nla_total_size(0)           /* CTA_PROTOINFO_TCP */
1191                 + nla_policy_len(tcp_nla_policy, CTA_PROTOINFO_TCP_MAX + 1);
1192 }
1193
1194 static int tcp_nlattr_tuple_size(void)
1195 {
1196         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1197 }
1198 #endif
1199
1200 #ifdef CONFIG_SYSCTL
1201 static unsigned int tcp_sysctl_table_users;
1202 static struct ctl_table_header *tcp_sysctl_header;
1203 static struct ctl_table tcp_sysctl_table[] = {
1204         {
1205                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1206                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1207                 .maxlen         = sizeof(unsigned int),
1208                 .mode           = 0644,
1209                 .proc_handler   = proc_dointvec_jiffies,
1210         },
1211         {
1212                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1213                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1214                 .maxlen         = sizeof(unsigned int),
1215                 .mode           = 0644,
1216                 .proc_handler   = proc_dointvec_jiffies,
1217         },
1218         {
1219                 .procname       = "nf_conntrack_tcp_timeout_established",
1220                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1221                 .maxlen         = sizeof(unsigned int),
1222                 .mode           = 0644,
1223                 .proc_handler   = proc_dointvec_jiffies,
1224         },
1225         {
1226                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1227                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1228                 .maxlen         = sizeof(unsigned int),
1229                 .mode           = 0644,
1230                 .proc_handler   = proc_dointvec_jiffies,
1231         },
1232         {
1233                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1234                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1235                 .maxlen         = sizeof(unsigned int),
1236                 .mode           = 0644,
1237                 .proc_handler   = proc_dointvec_jiffies,
1238         },
1239         {
1240                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1241                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1242                 .maxlen         = sizeof(unsigned int),
1243                 .mode           = 0644,
1244                 .proc_handler   = proc_dointvec_jiffies,
1245         },
1246         {
1247                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1248                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1249                 .maxlen         = sizeof(unsigned int),
1250                 .mode           = 0644,
1251                 .proc_handler   = proc_dointvec_jiffies,
1252         },
1253         {
1254                 .procname       = "nf_conntrack_tcp_timeout_close",
1255                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1256                 .maxlen         = sizeof(unsigned int),
1257                 .mode           = 0644,
1258                 .proc_handler   = proc_dointvec_jiffies,
1259         },
1260         {
1261                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1262                 .data           = &nf_ct_tcp_timeout_max_retrans,
1263                 .maxlen         = sizeof(unsigned int),
1264                 .mode           = 0644,
1265                 .proc_handler   = proc_dointvec_jiffies,
1266         },
1267         {
1268                 .procname       = "nf_conntrack_tcp_timeout_unacknowledged",
1269                 .data           = &nf_ct_tcp_timeout_unacknowledged,
1270                 .maxlen         = sizeof(unsigned int),
1271                 .mode           = 0644,
1272                 .proc_handler   = proc_dointvec_jiffies,
1273         },
1274         {
1275                 .ctl_name       = NET_NF_CONNTRACK_TCP_LOOSE,
1276                 .procname       = "nf_conntrack_tcp_loose",
1277                 .data           = &nf_ct_tcp_loose,
1278                 .maxlen         = sizeof(unsigned int),
1279                 .mode           = 0644,
1280                 .proc_handler   = proc_dointvec,
1281         },
1282         {
1283                 .ctl_name       = NET_NF_CONNTRACK_TCP_BE_LIBERAL,
1284                 .procname       = "nf_conntrack_tcp_be_liberal",
1285                 .data           = &nf_ct_tcp_be_liberal,
1286                 .maxlen         = sizeof(unsigned int),
1287                 .mode           = 0644,
1288                 .proc_handler   = proc_dointvec,
1289         },
1290         {
1291                 .ctl_name       = NET_NF_CONNTRACK_TCP_MAX_RETRANS,
1292                 .procname       = "nf_conntrack_tcp_max_retrans",
1293                 .data           = &nf_ct_tcp_max_retrans,
1294                 .maxlen         = sizeof(unsigned int),
1295                 .mode           = 0644,
1296                 .proc_handler   = proc_dointvec,
1297         },
1298         {
1299                 .ctl_name       = 0
1300         }
1301 };
1302
1303 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1304 static struct ctl_table tcp_compat_sysctl_table[] = {
1305         {
1306                 .procname       = "ip_conntrack_tcp_timeout_syn_sent",
1307                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1308                 .maxlen         = sizeof(unsigned int),
1309                 .mode           = 0644,
1310                 .proc_handler   = proc_dointvec_jiffies,
1311         },
1312         {
1313                 .procname       = "ip_conntrack_tcp_timeout_syn_recv",
1314                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1315                 .maxlen         = sizeof(unsigned int),
1316                 .mode           = 0644,
1317                 .proc_handler   = proc_dointvec_jiffies,
1318         },
1319         {
1320                 .procname       = "ip_conntrack_tcp_timeout_established",
1321                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1322                 .maxlen         = sizeof(unsigned int),
1323                 .mode           = 0644,
1324                 .proc_handler   = proc_dointvec_jiffies,
1325         },
1326         {
1327                 .procname       = "ip_conntrack_tcp_timeout_fin_wait",
1328                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1329                 .maxlen         = sizeof(unsigned int),
1330                 .mode           = 0644,
1331                 .proc_handler   = proc_dointvec_jiffies,
1332         },
1333         {
1334                 .procname       = "ip_conntrack_tcp_timeout_close_wait",
1335                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1336                 .maxlen         = sizeof(unsigned int),
1337                 .mode           = 0644,
1338                 .proc_handler   = proc_dointvec_jiffies,
1339         },
1340         {
1341                 .procname       = "ip_conntrack_tcp_timeout_last_ack",
1342                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1343                 .maxlen         = sizeof(unsigned int),
1344                 .mode           = 0644,
1345                 .proc_handler   = proc_dointvec_jiffies,
1346         },
1347         {
1348                 .procname       = "ip_conntrack_tcp_timeout_time_wait",
1349                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1350                 .maxlen         = sizeof(unsigned int),
1351                 .mode           = 0644,
1352                 .proc_handler   = proc_dointvec_jiffies,
1353         },
1354         {
1355                 .procname       = "ip_conntrack_tcp_timeout_close",
1356                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1357                 .maxlen         = sizeof(unsigned int),
1358                 .mode           = 0644,
1359                 .proc_handler   = proc_dointvec_jiffies,
1360         },
1361         {
1362                 .procname       = "ip_conntrack_tcp_timeout_max_retrans",
1363                 .data           = &nf_ct_tcp_timeout_max_retrans,
1364                 .maxlen         = sizeof(unsigned int),
1365                 .mode           = 0644,
1366                 .proc_handler   = proc_dointvec_jiffies,
1367         },
1368         {
1369                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_LOOSE,
1370                 .procname       = "ip_conntrack_tcp_loose",
1371                 .data           = &nf_ct_tcp_loose,
1372                 .maxlen         = sizeof(unsigned int),
1373                 .mode           = 0644,
1374                 .proc_handler   = proc_dointvec,
1375         },
1376         {
1377                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_BE_LIBERAL,
1378                 .procname       = "ip_conntrack_tcp_be_liberal",
1379                 .data           = &nf_ct_tcp_be_liberal,
1380                 .maxlen         = sizeof(unsigned int),
1381                 .mode           = 0644,
1382                 .proc_handler   = proc_dointvec,
1383         },
1384         {
1385                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_MAX_RETRANS,
1386                 .procname       = "ip_conntrack_tcp_max_retrans",
1387                 .data           = &nf_ct_tcp_max_retrans,
1388                 .maxlen         = sizeof(unsigned int),
1389                 .mode           = 0644,
1390                 .proc_handler   = proc_dointvec,
1391         },
1392         {
1393                 .ctl_name       = 0
1394         }
1395 };
1396 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
1397 #endif /* CONFIG_SYSCTL */
1398
1399 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1400 {
1401         .l3proto                = PF_INET,
1402         .l4proto                = IPPROTO_TCP,
1403         .name                   = "tcp",
1404         .pkt_to_tuple           = tcp_pkt_to_tuple,
1405         .invert_tuple           = tcp_invert_tuple,
1406         .print_tuple            = tcp_print_tuple,
1407         .print_conntrack        = tcp_print_conntrack,
1408         .packet                 = tcp_packet,
1409         .new                    = tcp_new,
1410         .error                  = tcp_error,
1411 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1412         .to_nlattr              = tcp_to_nlattr,
1413         .nlattr_size            = tcp_nlattr_size,
1414         .from_nlattr            = nlattr_to_tcp,
1415         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1416         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1417         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1418         .nla_policy             = nf_ct_port_nla_policy,
1419 #endif
1420 #ifdef CONFIG_SYSCTL
1421         .ctl_table_users        = &tcp_sysctl_table_users,
1422         .ctl_table_header       = &tcp_sysctl_header,
1423         .ctl_table              = tcp_sysctl_table,
1424 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1425         .ctl_compat_table       = tcp_compat_sysctl_table,
1426 #endif
1427 #endif
1428 };
1429 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1430
1431 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1432 {
1433         .l3proto                = PF_INET6,
1434         .l4proto                = IPPROTO_TCP,
1435         .name                   = "tcp",
1436         .pkt_to_tuple           = tcp_pkt_to_tuple,
1437         .invert_tuple           = tcp_invert_tuple,
1438         .print_tuple            = tcp_print_tuple,
1439         .print_conntrack        = tcp_print_conntrack,
1440         .packet                 = tcp_packet,
1441         .new                    = tcp_new,
1442         .error                  = tcp_error,
1443 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1444         .to_nlattr              = tcp_to_nlattr,
1445         .nlattr_size            = tcp_nlattr_size,
1446         .from_nlattr            = nlattr_to_tcp,
1447         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1448         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1449         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1450         .nla_policy             = nf_ct_port_nla_policy,
1451 #endif
1452 #ifdef CONFIG_SYSCTL
1453         .ctl_table_users        = &tcp_sysctl_table_users,
1454         .ctl_table_header       = &tcp_sysctl_header,
1455         .ctl_table              = tcp_sysctl_table,
1456 #endif
1457 };
1458 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);