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
  19 #include <net/tcp.h>
  20
  21 #include <linux/netfilter.h>
  22 #include <linux/netfilter_ipv4.h>
  23 #include <linux/netfilter_ipv6.h>
  24 #include <net/netfilter/nf_conntrack.h>
  25 #include <net/netfilter/nf_conntrack_l4proto.h>
  26 #include <net/netfilter/nf_conntrack_ecache.h>
  27 #include <net/netfilter/nf_log.h>
  28
  29 /* Protects ct->proto.tcp */
  30 static DEFINE_RWLOCK(tcp_lock);
  31
  32 /* "Be conservative in what you do,
  33     be liberal in what you accept from others."
  34     If it's non-zero, we mark only out of window RST segments as INVALID. */
  35 static int nf_ct_tcp_be_liberal __read_mostly = 0;
  36
  37 /* If it is set to zero, we disable picking up already established
  38    connections. */
  39 static int nf_ct_tcp_loose __read_mostly = 1;
  40
  41 /* Max number of the retransmitted packets without receiving an (acceptable)
  42    ACK from the destination. If this number is reached, a shorter timer
  43    will be started. */
  44 static int nf_ct_tcp_max_retrans __read_mostly = 3;
  45
  46   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  47      closely.  They're more complex. --RR */
  48
  49 static const char *const tcp_conntrack_names[] = {
  50         "NONE",
  51         "SYN_SENT",
  52         "SYN_RECV",
  53         "ESTABLISHED",
  54         "FIN_WAIT",
  55         "CLOSE_WAIT",
  56         "LAST_ACK",
  57         "TIME_WAIT",
  58         "CLOSE",
  59         "LISTEN"
  60 };
  61
  62 #define SECS * HZ
  63 #define MINS * 60 SECS
  64 #define HOURS * 60 MINS
  65 #define DAYS * 24 HOURS
  66
  67 /* RFC1122 says the R2 limit should be at least 100 seconds.
  68    Linux uses 15 packets as limit, which corresponds
  69    to ~13-30min depending on RTO. */
  70 static unsigned int nf_ct_tcp_timeout_max_retrans __read_mostly =   5 MINS;
  71
  72 static unsigned int tcp_timeouts[TCP_CONNTRACK_MAX] __read_mostly = {
  73         [TCP_CONNTRACK_SYN_SENT]        = 2 MINS,
  74         [TCP_CONNTRACK_SYN_RECV]        = 60 SECS,
  75         [TCP_CONNTRACK_ESTABLISHED]     = 5 DAYS,
  76         [TCP_CONNTRACK_FIN_WAIT]        = 2 MINS,
  77         [TCP_CONNTRACK_CLOSE_WAIT]      = 60 SECS,
  78         [TCP_CONNTRACK_LAST_ACK]        = 30 SECS,
  79         [TCP_CONNTRACK_TIME_WAIT]       = 2 MINS,
  80         [TCP_CONNTRACK_CLOSE]           = 10 SECS,
  81 };
  82
  83 #define sNO TCP_CONNTRACK_NONE
  84 #define sSS TCP_CONNTRACK_SYN_SENT
  85 #define sSR TCP_CONNTRACK_SYN_RECV
  86 #define sES TCP_CONNTRACK_ESTABLISHED
  87 #define sFW TCP_CONNTRACK_FIN_WAIT
  88 #define sCW TCP_CONNTRACK_CLOSE_WAIT
  89 #define sLA TCP_CONNTRACK_LAST_ACK
  90 #define sTW TCP_CONNTRACK_TIME_WAIT
  91 #define sCL TCP_CONNTRACK_CLOSE
  92 #define sLI TCP_CONNTRACK_LISTEN
  93 #define sIV TCP_CONNTRACK_MAX
  94 #define sIG TCP_CONNTRACK_IGNORE
  95
  96 /* What TCP flags are set from RST/SYN/FIN/ACK. */
  97 enum tcp_bit_set {
  98         TCP_SYN_SET,
  99         TCP_SYNACK_SET,
 100         TCP_FIN_SET,
 101         TCP_ACK_SET,
 102         TCP_RST_SET,
 103         TCP_NONE_SET,
 104 };
 105
 106 /*
 107  * The TCP state transition table needs a few words...
 108  *
 109  * We are the man in the middle. All the packets go through us
 110  * but might get lost in transit to the destination.
 111  * It is assumed that the destinations can't receive segments
 112  * we haven't seen.
 113  *
 114  * The checked segment is in window, but our windows are *not*
 115  * equivalent with the ones of the sender/receiver. We always
 116  * try to guess the state of the current sender.
 117  *
 118  * The meaning of the states are:
 119  *
 120  * NONE:        initial state
 121  * SYN_SENT:    SYN-only packet seen
 122  * SYN_RECV:    SYN-ACK packet seen
 123  * ESTABLISHED: ACK packet seen
 124  * FIN_WAIT:    FIN packet seen
 125  * CLOSE_WAIT:  ACK seen (after FIN)
 126  * LAST_ACK:    FIN seen (after FIN)
 127  * TIME_WAIT:   last ACK seen
 128  * CLOSE:       closed connection (RST)
 129  *
 130  * LISTEN state is not used.
 131  *
 132  * Packets marked as IGNORED (sIG):
 133  *      if they may be either invalid or valid
 134  *      and the receiver may send back a connection
 135  *      closing RST or a SYN/ACK.
 136  *
 137  * Packets marked as INVALID (sIV):
 138  *      if they are invalid
 139  *      or we do not support the request (simultaneous open)
 140  */
 141 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 142         {
 143 /* ORIGINAL */
 144 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 145 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
 146 /*
 147  *      sNO -> sSS      Initialize a new connection
 148  *      sSS -> sSS      Retransmitted SYN
 149  *      sSR -> sIG      Late retransmitted SYN?
 150  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 151  *                      are errors. Receiver will reply with RST
 152  *                      and close the connection.
 153  *                      Or we are not in sync and hold a dead connection.
 154  *      sFW -> sIG
 155  *      sCW -> sIG
 156  *      sLA -> sIG
 157  *      sTW -> sSS      Reopened connection (RFC 1122).
 158  *      sCL -> sSS
 159  */
 160 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 161 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 162 /*
 163  * A SYN/ACK from the client is always invalid:
 164  *      - either it tries to set up a simultaneous open, which is
 165  *        not supported;
 166  *      - or the firewall has just been inserted between the two hosts
 167  *        during the session set-up. The SYN will be retransmitted
 168  *        by the true client (or it'll time out).
 169  */
 170 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 171 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 172 /*
 173  *      sNO -> sIV      Too late and no reason to do anything...
 174  *      sSS -> sIV      Client migth not send FIN in this state:
 175  *                      we enforce waiting for a SYN/ACK reply first.
 176  *      sSR -> sFW      Close started.
 177  *      sES -> sFW
 178  *      sFW -> sLA      FIN seen in both directions, waiting for
 179  *                      the last ACK.
 180  *                      Migth be a retransmitted FIN as well...
 181  *      sCW -> sLA
 182  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 183  *      sTW -> sTW
 184  *      sCL -> sCL
 185  */
 186 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 187 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 188 /*
 189  *      sNO -> sES      Assumed.
 190  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 191  *      sSR -> sES      Established state is reached.
 192  *      sES -> sES      :-)
 193  *      sFW -> sCW      Normal close request answered by ACK.
 194  *      sCW -> sCW
 195  *      sLA -> sTW      Last ACK detected.
 196  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 197  *      sCL -> sCL
 198  */
 199 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 200 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 201 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 202         },
 203         {
 204 /* REPLY */
 205 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 206 /*syn*/    { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
 207 /*
 208  *      sNO -> sIV      Never reached.
 209  *      sSS -> sIV      Simultaneous open, not supported
 210  *      sSR -> sIV      Simultaneous open, not supported.
 211  *      sES -> sIV      Server may not initiate a connection.
 212  *      sFW -> sIV
 213  *      sCW -> sIV
 214  *      sLA -> sIV
 215  *      sTW -> sIV      Reopened connection, but server may not do it.
 216  *      sCL -> sIV
 217  */
 218 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 219 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
 220 /*
 221  *      sSS -> sSR      Standard open.
 222  *      sSR -> sSR      Retransmitted SYN/ACK.
 223  *      sES -> sIG      Late retransmitted SYN/ACK?
 224  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 225  *      sCW -> sIG
 226  *      sLA -> sIG
 227  *      sTW -> sIG
 228  *      sCL -> sIG
 229  */
 230 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 231 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 232 /*
 233  *      sSS -> sIV      Server might not send FIN in this state.
 234  *      sSR -> sFW      Close started.
 235  *      sES -> sFW
 236  *      sFW -> sLA      FIN seen in both directions.
 237  *      sCW -> sLA
 238  *      sLA -> sLA      Retransmitted FIN.
 239  *      sTW -> sTW
 240  *      sCL -> sCL
 241  */
 242 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 243 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 244 /*
 245  *      sSS -> sIG      Might be a half-open connection.
 246  *      sSR -> sSR      Might answer late resent SYN.
 247  *      sES -> sES      :-)
 248  *      sFW -> sCW      Normal close request answered by ACK.
 249  *      sCW -> sCW
 250  *      sLA -> sTW      Last ACK detected.
 251  *      sTW -> sTW      Retransmitted last ACK.
 252  *      sCL -> sCL
 253  */
 254 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
 255 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
 256 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 257         }
 258 };
 259
 260 static bool tcp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
 261                              struct nf_conntrack_tuple *tuple)
 262 {
 263         const struct tcphdr *hp;
 264         struct tcphdr _hdr;
 265
 266         /* Actually only need first 8 bytes. */
 267         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
 268         if (hp == NULL)
 269                 return false;
 270
 271         tuple->src.u.tcp.port = hp->source;
 272         tuple->dst.u.tcp.port = hp->dest;
 273
 274         return true;
 275 }
 276
 277 static bool tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
 278                              const struct nf_conntrack_tuple *orig)
 279 {
 280         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
 281         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
 282         return true;
 283 }
 284
 285 /* Print out the per-protocol part of the tuple. */
 286 static int tcp_print_tuple(struct seq_file *s,
 287                            const struct nf_conntrack_tuple *tuple)
 288 {
 289         return seq_printf(s, "sport=%hu dport=%hu ",
 290                           ntohs(tuple->src.u.tcp.port),
 291                           ntohs(tuple->dst.u.tcp.port));
 292 }
 293
 294 /* Print out the private part of the conntrack. */
 295 static int tcp_print_conntrack(struct seq_file *s, const struct nf_conn *ct)
 296 {
 297         enum tcp_conntrack state;
 298
 299         read_lock_bh(&tcp_lock);
 300         state = ct->proto.tcp.state;
 301         read_unlock_bh(&tcp_lock);
 302
 303         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
 304 }
 305
 306 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 307 {
 308         if (tcph->rst) return TCP_RST_SET;
 309         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 310         else if (tcph->fin) return TCP_FIN_SET;
 311         else if (tcph->ack) return TCP_ACK_SET;
 312         else return TCP_NONE_SET;
 313 }
 314
 315 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 316    in IP Filter' by Guido van Rooij.
 317
 318    http://www.nluug.nl/events/sane2000/papers.html
 319    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
 320
 321    The boundaries and the conditions are changed according to RFC793:
 322    the packet must intersect the window (i.e. segments may be
 323    after the right or before the left edge) and thus receivers may ACK
 324    segments after the right edge of the window.
 325
 326         td_maxend = max(sack + max(win,1)) seen in reply packets
 327         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 328         td_maxwin += seq + len - sender.td_maxend
 329                         if seq + len > sender.td_maxend
 330         td_end    = max(seq + len) seen in sent packets
 331
 332    I.   Upper bound for valid data:     seq <= sender.td_maxend
 333    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 334    III. Upper bound for valid (s)ack:   sack <= receiver.td_end
 335    IV.  Lower bound for valid (s)ack:   sack >= receiver.td_end - MAXACKWINDOW
 336
 337    where sack is the highest right edge of sack block found in the packet
 338    or ack in the case of packet without SACK option.
 339
 340    The upper bound limit for a valid (s)ack is not ignored -
 341    we doesn't have to deal with fragments.
 342 */
 343
 344 static inline __u32 segment_seq_plus_len(__u32 seq,
 345                                          size_t len,
 346                                          unsigned int dataoff,
 347                                          const struct tcphdr *tcph)
 348 {
 349         /* XXX Should I use payload length field in IP/IPv6 header ?
 350          * - YK */
 351         return (seq + len - dataoff - tcph->doff*4
 352                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 353 }
 354
 355 /* Fixme: what about big packets? */
 356 #define MAXACKWINCONST                  66000
 357 #define MAXACKWINDOW(sender)                                            \
 358         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 359                                               : MAXACKWINCONST)
 360
 361 /*
 362  * Simplified tcp_parse_options routine from tcp_input.c
 363  */
 364 static void tcp_options(const struct sk_buff *skb,
 365                         unsigned int dataoff,
 366                         const struct tcphdr *tcph,
 367                         struct ip_ct_tcp_state *state)
 368 {
 369         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 370         const unsigned char *ptr;
 371         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 372
 373         if (!length)
 374                 return;
 375
 376         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 377                                  length, buff);
 378         BUG_ON(ptr == NULL);
 379
 380         state->td_scale =
 381         state->flags = 0;
 382
 383         while (length > 0) {
 384                 int opcode=*ptr++;
 385                 int opsize;
 386
 387                 switch (opcode) {
 388                 case TCPOPT_EOL:
 389                         return;
 390                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 391                         length--;
 392                         continue;
 393                 default:
 394                         opsize=*ptr++;
 395                         if (opsize < 2) /* "silly options" */
 396                                 return;
 397                         if (opsize > length)
 398                                 break;  /* don't parse partial options */
 399
 400                         if (opcode == TCPOPT_SACK_PERM
 401                             && opsize == TCPOLEN_SACK_PERM)
 402                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 403                         else if (opcode == TCPOPT_WINDOW
 404                                  && opsize == TCPOLEN_WINDOW) {
 405                                 state->td_scale = *(u_int8_t *)ptr;
 406
 407                                 if (state->td_scale > 14) {
 408                                         /* See RFC1323 */
 409                                         state->td_scale = 14;
 410                                 }
 411                                 state->flags |=
 412                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 413                         }
 414                         ptr += opsize - 2;
 415                         length -= opsize;
 416                 }
 417         }
 418 }
 419
 420 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 421                      const struct tcphdr *tcph, __u32 *sack)
 422 {
 423         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 424         const unsigned char *ptr;
 425         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 426         __u32 tmp;
 427
 428         if (!length)
 429                 return;
 430
 431         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 432                                  length, buff);
 433         BUG_ON(ptr == NULL);
 434
 435         /* Fast path for timestamp-only option */
 436         if (length == TCPOLEN_TSTAMP_ALIGNED*4
 437             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
 438                                        | (TCPOPT_NOP << 16)
 439                                        | (TCPOPT_TIMESTAMP << 8)
 440                                        | TCPOLEN_TIMESTAMP))
 441                 return;
 442
 443         while (length > 0) {
 444                 int opcode = *ptr++;
 445                 int opsize, i;
 446
 447                 switch (opcode) {
 448                 case TCPOPT_EOL:
 449                         return;
 450                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 451                         length--;
 452                         continue;
 453                 default:
 454                         opsize = *ptr++;
 455                         if (opsize < 2) /* "silly options" */
 456                                 return;
 457                         if (opsize > length)
 458                                 break;  /* don't parse partial options */
 459
 460                         if (opcode == TCPOPT_SACK
 461                             && opsize >= (TCPOLEN_SACK_BASE
 462                                           + TCPOLEN_SACK_PERBLOCK)
 463                             && !((opsize - TCPOLEN_SACK_BASE)
 464                                  % TCPOLEN_SACK_PERBLOCK)) {
 465                                 for (i = 0;
 466                                      i < (opsize - TCPOLEN_SACK_BASE);
 467                                      i += TCPOLEN_SACK_PERBLOCK) {
 468                                         tmp = ntohl(*((__be32 *)(ptr+i)+1));
 469
 470                                         if (after(tmp, *sack))
 471                                                 *sack = tmp;
 472                                 }
 473                                 return;
 474                         }
 475                         ptr += opsize - 2;
 476                         length -= opsize;
 477                 }
 478         }
 479 }
 480
 481 static bool tcp_in_window(const struct nf_conn *ct,
 482                           struct ip_ct_tcp *state,
 483                           enum ip_conntrack_dir dir,
 484                           unsigned int index,
 485                           const struct sk_buff *skb,
 486                           unsigned int dataoff,
 487                           const struct tcphdr *tcph,
 488                           int pf)
 489 {
 490         struct ip_ct_tcp_state *sender = &state->seen[dir];
 491         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 492         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 493         __u32 seq, ack, sack, end, win, swin;
 494         bool res;
 495
 496         /*
 497          * Get the required data from the packet.
 498          */
 499         seq = ntohl(tcph->seq);
 500         ack = sack = ntohl(tcph->ack_seq);
 501         win = ntohs(tcph->window);
 502         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 503
 504         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 505                 tcp_sack(skb, dataoff, tcph, &sack);
 506
 507         pr_debug("tcp_in_window: START\n");
 508         pr_debug("tcp_in_window: ");
 509         nf_ct_dump_tuple(tuple);
 510         pr_debug("seq=%u ack=%u sack=%u win=%u end=%u\n",
 511                  seq, ack, sack, win, end);
 512         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 513                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 514                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 515                  sender->td_scale,
 516                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 517                  receiver->td_scale);
 518
 519         if (sender->td_end == 0) {
 520                 /*
 521                  * Initialize sender data.
 522                  */
 523                 if (tcph->syn && tcph->ack) {
 524                         /*
 525                          * Outgoing SYN-ACK in reply to a SYN.
 526                          */
 527                         sender->td_end =
 528                         sender->td_maxend = end;
 529                         sender->td_maxwin = (win == 0 ? 1 : win);
 530
 531                         tcp_options(skb, dataoff, tcph, sender);
 532                         /*
 533                          * RFC 1323:
 534                          * Both sides must send the Window Scale option
 535                          * to enable window scaling in either direction.
 536                          */
 537                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 538                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 539                                 sender->td_scale =
 540                                 receiver->td_scale = 0;
 541                 } else {
 542                         /*
 543                          * We are in the middle of a connection,
 544                          * its history is lost for us.
 545                          * Let's try to use the data from the packet.
 546                          */
 547                         sender->td_end = end;
 548                         sender->td_maxwin = (win == 0 ? 1 : win);
 549                         sender->td_maxend = end + sender->td_maxwin;
 550                 }
 551         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 552                      && dir == IP_CT_DIR_ORIGINAL)
 553                    || (state->state == TCP_CONNTRACK_SYN_RECV
 554                      && dir == IP_CT_DIR_REPLY))
 555                    && after(end, sender->td_end)) {
 556                 /*
 557                  * RFC 793: "if a TCP is reinitialized ... then it need
 558                  * not wait at all; it must only be sure to use sequence
 559                  * numbers larger than those recently used."
 560                  */
 561                 sender->td_end =
 562                 sender->td_maxend = end;
 563                 sender->td_maxwin = (win == 0 ? 1 : win);
 564
 565                 tcp_options(skb, dataoff, tcph, sender);
 566         }
 567
 568         if (!(tcph->ack)) {
 569                 /*
 570                  * If there is no ACK, just pretend it was set and OK.
 571                  */
 572                 ack = sack = receiver->td_end;
 573         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 574                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 575                    && (ack == 0)) {
 576                 /*
 577                  * Broken TCP stacks, that set ACK in RST packets as well
 578                  * with zero ack value.
 579                  */
 580                 ack = sack = receiver->td_end;
 581         }
 582
 583         if (seq == end
 584             && (!tcph->rst
 585                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
 586                 /*
 587                  * Packets contains no data: we assume it is valid
 588                  * and check the ack value only.
 589                  * However RST segments are always validated by their
 590                  * SEQ number, except when seq == 0 (reset sent answering
 591                  * SYN.
 592                  */
 593                 seq = end = sender->td_end;
 594
 595         pr_debug("tcp_in_window: ");
 596         nf_ct_dump_tuple(tuple);
 597         pr_debug("seq=%u ack=%u sack =%u win=%u end=%u\n",
 598                  seq, ack, sack, win, end);
 599         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 600                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 601                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 602                  sender->td_scale,
 603                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 604                  receiver->td_scale);
 605
 606         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 607                  before(seq, sender->td_maxend + 1),
 608                  after(end, sender->td_end - receiver->td_maxwin - 1),
 609                  before(sack, receiver->td_end + 1),
 610                  after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 611
 612         if (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                  * Take into account window scaling (RFC 1323).
 618                  */
 619                 if (!tcph->syn)
 620                         win <<= sender->td_scale;
 621
 622                 /*
 623                  * Update sender data.
 624                  */
 625                 swin = win + (sack - ack);
 626                 if (sender->td_maxwin < swin)
 627                         sender->td_maxwin = swin;
 628                 if (after(end, sender->td_end))
 629                         sender->td_end = end;
 630                 /*
 631                  * Update receiver data.
 632                  */
 633                 if (after(end, sender->td_maxend))
 634                         receiver->td_maxwin += end - sender->td_maxend;
 635                 if (after(sack + win, receiver->td_maxend - 1)) {
 636                         receiver->td_maxend = sack + win;
 637                         if (win == 0)
 638                                 receiver->td_maxend++;
 639                 }
 640
 641                 /*
 642                  * Check retransmissions.
 643                  */
 644                 if (index == TCP_ACK_SET) {
 645                         if (state->last_dir == dir
 646                             && state->last_seq == seq
 647                             && state->last_ack == ack
 648                             && state->last_end == end
 649                             && state->last_win == win)
 650                                 state->retrans++;
 651                         else {
 652                                 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 = 0;
 658                         }
 659                 }
 660                 res = true;
 661         } else {
 662                 res = false;
 663                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 664                     nf_ct_tcp_be_liberal)
 665                         res = true;
 666                 if (!res && LOG_INVALID(IPPROTO_TCP))
 667                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 668                         "nf_ct_tcp: %s ",
 669                         before(seq, sender->td_maxend + 1) ?
 670                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
 671                         before(sack, receiver->td_end + 1) ?
 672                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
 673                         : "ACK is under the lower bound (possible overly delayed ACK)"
 674                         : "ACK is over the upper bound (ACKed data not seen yet)"
 675                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 676                         : "SEQ is over the upper bound (over the window of the receiver)");
 677         }
 678
 679         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 680                  "receiver end=%u maxend=%u maxwin=%u\n",
 681                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 682                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 683
 684         return res;
 685 }
 686
 687 #ifdef CONFIG_NF_NAT_NEEDED
 688 /* Update sender->td_end after NAT successfully mangled the packet */
 689 /* Caller must linearize skb at tcp header. */
 690 void nf_conntrack_tcp_update(const struct sk_buff *skb,
 691                              unsigned int dataoff,
 692                              struct nf_conn *ct,
 693                              int dir)
 694 {
 695         const struct tcphdr *tcph = (const void *)skb->data + dataoff;
 696         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[dir];
 697         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[!dir];
 698         __u32 end;
 699
 700         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
 701
 702         write_lock_bh(&tcp_lock);
 703         /*
 704          * We have to worry for the ack in the reply packet only...
 705          */
 706         if (after(end, ct->proto.tcp.seen[dir].td_end))
 707                 ct->proto.tcp.seen[dir].td_end = end;
 708         ct->proto.tcp.last_end = end;
 709         write_unlock_bh(&tcp_lock);
 710         pr_debug("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
 711                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 712                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 713                  sender->td_scale,
 714                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 715                  receiver->td_scale);
 716 }
 717 EXPORT_SYMBOL_GPL(nf_conntrack_tcp_update);
 718 #endif
 719
 720 #define TH_FIN  0x01
 721 #define TH_SYN  0x02
 722 #define TH_RST  0x04
 723 #define TH_PUSH 0x08
 724 #define TH_ACK  0x10
 725 #define TH_URG  0x20
 726 #define TH_ECE  0x40
 727 #define TH_CWR  0x80
 728
 729 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 730 static const u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_ACK|TH_URG) + 1] =
 731 {
 732         [TH_SYN]                        = 1,
 733         [TH_SYN|TH_URG]                 = 1,
 734         [TH_SYN|TH_ACK]                 = 1,
 735         [TH_RST]                        = 1,
 736         [TH_RST|TH_ACK]                 = 1,
 737         [TH_FIN|TH_ACK]                 = 1,
 738         [TH_FIN|TH_ACK|TH_URG]          = 1,
 739         [TH_ACK]                        = 1,
 740         [TH_ACK|TH_URG]                 = 1,
 741 };
 742
 743 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 744 static int tcp_error(struct sk_buff *skb,
 745                      unsigned int dataoff,
 746                      enum ip_conntrack_info *ctinfo,
 747                      int pf,
 748                      unsigned int hooknum)
 749 {
 750         const struct tcphdr *th;
 751         struct tcphdr _tcph;
 752         unsigned int tcplen = skb->len - dataoff;
 753         u_int8_t tcpflags;
 754
 755         /* Smaller that minimal TCP header? */
 756         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 757         if (th == NULL) {
 758                 if (LOG_INVALID(IPPROTO_TCP))
 759                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 760                                 "nf_ct_tcp: short packet ");
 761                 return -NF_ACCEPT;
 762         }
 763
 764         /* Not whole TCP header or malformed packet */
 765         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 766                 if (LOG_INVALID(IPPROTO_TCP))
 767                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 768                                 "nf_ct_tcp: truncated/malformed packet ");
 769                 return -NF_ACCEPT;
 770         }
 771
 772         /* Checksum invalid? Ignore.
 773          * We skip checking packets on the outgoing path
 774          * because the checksum is assumed to be correct.
 775          */
 776         /* FIXME: Source route IP option packets --RR */
 777         if (nf_conntrack_checksum && hooknum == NF_INET_PRE_ROUTING &&
 778             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
 779                 if (LOG_INVALID(IPPROTO_TCP))
 780                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 781                                   "nf_ct_tcp: bad TCP checksum ");
 782                 return -NF_ACCEPT;
 783         }
 784
 785         /* Check TCP flags. */
 786         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR|TH_PUSH));
 787         if (!tcp_valid_flags[tcpflags]) {
 788                 if (LOG_INVALID(IPPROTO_TCP))
 789                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 790                                   "nf_ct_tcp: invalid TCP flag combination ");
 791                 return -NF_ACCEPT;
 792         }
 793
 794         return NF_ACCEPT;
 795 }
 796
 797 /* Returns verdict for packet, or -1 for invalid. */
 798 static int tcp_packet(struct nf_conn *ct,
 799                       const struct sk_buff *skb,
 800                       unsigned int dataoff,
 801                       enum ip_conntrack_info ctinfo,
 802                       int pf,
 803                       unsigned int hooknum)
 804 {
 805         struct nf_conntrack_tuple *tuple;
 806         enum tcp_conntrack new_state, old_state;
 807         enum ip_conntrack_dir dir;
 808         const struct tcphdr *th;
 809         struct tcphdr _tcph;
 810         unsigned long timeout;
 811         unsigned int index;
 812
 813         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 814         BUG_ON(th == NULL);
 815
 816         write_lock_bh(&tcp_lock);
 817         old_state = ct->proto.tcp.state;
 818         dir = CTINFO2DIR(ctinfo);
 819         index = get_conntrack_index(th);
 820         new_state = tcp_conntracks[dir][index][old_state];
 821         tuple = &ct->tuplehash[dir].tuple;
 822
 823         switch (new_state) {
 824         case TCP_CONNTRACK_SYN_SENT:
 825                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 826                         break;
 827                 /* RFC 1122: "When a connection is closed actively,
 828                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 829                  * (Maximum Segment Lifetime). However, it MAY accept
 830                  * a new SYN from the remote TCP to reopen the connection
 831                  * directly from TIME-WAIT state, if..."
 832                  * We ignore the conditions because we are in the
 833                  * TIME-WAIT state anyway.
 834                  *
 835                  * Handle aborted connections: we and the server
 836                  * think there is an existing connection but the client
 837                  * aborts it and starts a new one.
 838                  */
 839                 if (((ct->proto.tcp.seen[dir].flags
 840                       | ct->proto.tcp.seen[!dir].flags)
 841                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 842                     || (ct->proto.tcp.last_dir == dir
 843                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 844                         /* Attempt to reopen a closed/aborted connection.
 845                          * Delete this connection and look up again. */
 846                         write_unlock_bh(&tcp_lock);
 847                         nf_ct_kill(ct);
 848                         return -NF_REPEAT;
 849                 }
 850                 /* Fall through */
 851         case TCP_CONNTRACK_IGNORE:
 852                 /* Ignored packets:
 853                  *
 854                  * Our connection entry may be out of sync, so ignore
 855                  * packets which may signal the real connection between
 856                  * the client and the server.
 857                  *
 858                  * a) SYN in ORIGINAL
 859                  * b) SYN/ACK in REPLY
 860                  * c) ACK in reply direction after initial SYN in original.
 861                  *
 862                  * If the ignored packet is invalid, the receiver will send
 863                  * a RST we'll catch below.
 864                  */
 865                 if (index == TCP_SYNACK_SET
 866                     && ct->proto.tcp.last_index == TCP_SYN_SET
 867                     && ct->proto.tcp.last_dir != dir
 868                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 869                         /* b) This SYN/ACK acknowledges a SYN that we earlier
 870                          * ignored as invalid. This means that the client and
 871                          * the server are both in sync, while the firewall is
 872                          * not. We kill this session and block the SYN/ACK so
 873                          * that the client cannot but retransmit its SYN and
 874                          * thus initiate a clean new session.
 875                          */
 876                         write_unlock_bh(&tcp_lock);
 877                         if (LOG_INVALID(IPPROTO_TCP))
 878                                 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 879                                           "nf_ct_tcp: killing out of sync session ");
 880                         nf_ct_kill(ct);
 881                         return -NF_DROP;
 882                 }
 883                 ct->proto.tcp.last_index = index;
 884                 ct->proto.tcp.last_dir = dir;
 885                 ct->proto.tcp.last_seq = ntohl(th->seq);
 886                 ct->proto.tcp.last_end =
 887                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 888
 889                 write_unlock_bh(&tcp_lock);
 890                 if (LOG_INVALID(IPPROTO_TCP))
 891                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 892                                   "nf_ct_tcp: invalid packet ignored ");
 893                 return NF_ACCEPT;
 894         case TCP_CONNTRACK_MAX:
 895                 /* Invalid packet */
 896                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 897                          dir, get_conntrack_index(th), old_state);
 898                 write_unlock_bh(&tcp_lock);
 899                 if (LOG_INVALID(IPPROTO_TCP))
 900                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 901                                   "nf_ct_tcp: invalid state ");
 902                 return -NF_ACCEPT;
 903         case TCP_CONNTRACK_CLOSE:
 904                 if (index == TCP_RST_SET
 905                     && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
 906                          && ct->proto.tcp.last_index == TCP_SYN_SET)
 907                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
 908                             && ct->proto.tcp.last_index == TCP_ACK_SET))
 909                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 910                         /* RST sent to invalid SYN or ACK we had let through
 911                          * at a) and c) above:
 912                          *
 913                          * a) SYN was in window then
 914                          * c) we hold a half-open connection.
 915                          *
 916                          * Delete our connection entry.
 917                          * We skip window checking, because packet might ACK
 918                          * segments we ignored. */
 919                         goto in_window;
 920                 }
 921                 /* Just fall through */
 922         default:
 923                 /* Keep compilers happy. */
 924                 break;
 925         }
 926
 927         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
 928                            skb, dataoff, th, pf)) {
 929                 write_unlock_bh(&tcp_lock);
 930                 return -NF_ACCEPT;
 931         }
 932      in_window:
 933         /* From now on we have got in-window packets */
 934         ct->proto.tcp.last_index = index;
 935         ct->proto.tcp.last_dir = dir;
 936
 937         pr_debug("tcp_conntracks: ");
 938         nf_ct_dump_tuple(tuple);
 939         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
 940                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
 941                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
 942                  old_state, new_state);
 943
 944         ct->proto.tcp.state = new_state;
 945         if (old_state != new_state
 946             && new_state == TCP_CONNTRACK_FIN_WAIT)
 947                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
 948         timeout = ct->proto.tcp.retrans >= nf_ct_tcp_max_retrans
 949                   && tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans
 950                   ? nf_ct_tcp_timeout_max_retrans : tcp_timeouts[new_state];
 951         write_unlock_bh(&tcp_lock);
 952
 953         nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
 954         if (new_state != old_state)
 955                 nf_conntrack_event_cache(IPCT_PROTOINFO, skb);
 956
 957         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 958                 /* If only reply is a RST, we can consider ourselves not to
 959                    have an established connection: this is a fairly common
 960                    problem case, so we can delete the conntrack
 961                    immediately.  --RR */
 962                 if (th->rst) {
 963                         nf_ct_kill_acct(ct, ctinfo, skb);
 964                         return NF_ACCEPT;
 965                 }
 966         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
 967                    && (old_state == TCP_CONNTRACK_SYN_RECV
 968                        || old_state == TCP_CONNTRACK_ESTABLISHED)
 969                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
 970                 /* Set ASSURED if we see see valid ack in ESTABLISHED
 971                    after SYN_RECV or a valid answer for a picked up
 972                    connection. */
 973                 set_bit(IPS_ASSURED_BIT, &ct->status);
 974                 nf_conntrack_event_cache(IPCT_STATUS, skb);
 975         }
 976         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
 977
 978         return NF_ACCEPT;
 979 }
 980
 981 /* Called when a new connection for this protocol found. */
 982 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
 983                     unsigned int dataoff)
 984 {
 985         enum tcp_conntrack new_state;
 986         const struct tcphdr *th;
 987         struct tcphdr _tcph;
 988         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
 989         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
 990
 991         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 992         BUG_ON(th == NULL);
 993
 994         /* Don't need lock here: this conntrack not in circulation yet */
 995         new_state
 996                 = tcp_conntracks[0][get_conntrack_index(th)]
 997                 [TCP_CONNTRACK_NONE];
 998
 999         /* Invalid: delete conntrack */
1000         if (new_state >= TCP_CONNTRACK_MAX) {
1001                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1002                 return false;
1003         }
1004
1005         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1006                 /* SYN packet */
1007                 ct->proto.tcp.seen[0].td_end =
1008                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1009                                              dataoff, th);
1010                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1011                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1012                         ct->proto.tcp.seen[0].td_maxwin = 1;
1013                 ct->proto.tcp.seen[0].td_maxend =
1014                         ct->proto.tcp.seen[0].td_end;
1015
1016                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1017                 ct->proto.tcp.seen[1].flags = 0;
1018         } else if (nf_ct_tcp_loose == 0) {
1019                 /* Don't try to pick up connections. */
1020                 return false;
1021         } else {
1022                 /*
1023                  * We are in the middle of a connection,
1024                  * its history is lost for us.
1025                  * Let's try to use the data from the packet.
1026                  */
1027                 ct->proto.tcp.seen[0].td_end =
1028                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1029                                              dataoff, th);
1030                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1031                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1032                         ct->proto.tcp.seen[0].td_maxwin = 1;
1033                 ct->proto.tcp.seen[0].td_maxend =
1034                         ct->proto.tcp.seen[0].td_end +
1035                         ct->proto.tcp.seen[0].td_maxwin;
1036                 ct->proto.tcp.seen[0].td_scale = 0;
1037
1038                 /* We assume SACK and liberal window checking to handle
1039                  * window scaling */
1040                 ct->proto.tcp.seen[0].flags =
1041                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1042                                               IP_CT_TCP_FLAG_BE_LIBERAL;
1043         }
1044
1045         ct->proto.tcp.seen[1].td_end = 0;
1046         ct->proto.tcp.seen[1].td_maxend = 0;
1047         ct->proto.tcp.seen[1].td_maxwin = 1;
1048         ct->proto.tcp.seen[1].td_scale = 0;
1049
1050         /* tcp_packet will set them */
1051         ct->proto.tcp.state = TCP_CONNTRACK_NONE;
1052         ct->proto.tcp.last_index = TCP_NONE_SET;
1053
1054         pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1055                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1056                  sender->td_end, sender->td_maxend, sender->td_maxwin,
1057                  sender->td_scale,
1058                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1059                  receiver->td_scale);
1060         return true;
1061 }
1062
1063 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1064
1065 #include <linux/netfilter/nfnetlink.h>
1066 #include <linux/netfilter/nfnetlink_conntrack.h>
1067
1068 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1069                          const struct nf_conn *ct)
1070 {
1071         struct nlattr *nest_parms;
1072         struct nf_ct_tcp_flags tmp = {};
1073
1074         read_lock_bh(&tcp_lock);
1075         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1076         if (!nest_parms)
1077                 goto nla_put_failure;
1078
1079         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state);
1080
1081         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1082                    ct->proto.tcp.seen[0].td_scale);
1083
1084         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1085                    ct->proto.tcp.seen[1].td_scale);
1086
1087         tmp.flags = ct->proto.tcp.seen[0].flags;
1088         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1089                 sizeof(struct nf_ct_tcp_flags), &tmp);
1090
1091         tmp.flags = ct->proto.tcp.seen[1].flags;
1092         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1093                 sizeof(struct nf_ct_tcp_flags), &tmp);
1094         read_unlock_bh(&tcp_lock);
1095
1096         nla_nest_end(skb, nest_parms);
1097
1098         return 0;
1099
1100 nla_put_failure:
1101         read_unlock_bh(&tcp_lock);
1102         return -1;
1103 }
1104
1105 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1106         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1107         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1108         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1109         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1110         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len =  sizeof(struct nf_ct_tcp_flags) },
1111 };
1112
1113 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1114 {
1115         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1116         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1117         int err;
1118
1119         /* updates could not contain anything about the private
1120          * protocol info, in that case skip the parsing */
1121         if (!pattr)
1122                 return 0;
1123
1124         err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr, tcp_nla_policy);
1125         if (err < 0)
1126                 return err;
1127
1128         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1129             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1130                 return -EINVAL;
1131
1132         write_lock_bh(&tcp_lock);
1133         if (tb[CTA_PROTOINFO_TCP_STATE])
1134                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1135
1136         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1137                 struct nf_ct_tcp_flags *attr =
1138                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1139                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1140                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1141         }
1142
1143         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1144                 struct nf_ct_tcp_flags *attr =
1145                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1146                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1147                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1148         }
1149
1150         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1151             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1152             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1153             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1154                 ct->proto.tcp.seen[0].td_scale =
1155                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1156                 ct->proto.tcp.seen[1].td_scale =
1157                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1158         }
1159         write_unlock_bh(&tcp_lock);
1160
1161         return 0;
1162 }
1163 #endif
1164
1165 #ifdef CONFIG_SYSCTL
1166 static unsigned int tcp_sysctl_table_users;
1167 static struct ctl_table_header *tcp_sysctl_header;
1168 static struct ctl_table tcp_sysctl_table[] = {
1169         {
1170                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1171                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1172                 .maxlen         = sizeof(unsigned int),
1173                 .mode           = 0644,
1174                 .proc_handler   = &proc_dointvec_jiffies,
1175         },
1176         {
1177                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1178                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1179                 .maxlen         = sizeof(unsigned int),
1180                 .mode           = 0644,
1181                 .proc_handler   = &proc_dointvec_jiffies,
1182         },
1183         {
1184                 .procname       = "nf_conntrack_tcp_timeout_established",
1185                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1186                 .maxlen         = sizeof(unsigned int),
1187                 .mode           = 0644,
1188                 .proc_handler   = &proc_dointvec_jiffies,
1189         },
1190         {
1191                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1192                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1193                 .maxlen         = sizeof(unsigned int),
1194                 .mode           = 0644,
1195                 .proc_handler   = &proc_dointvec_jiffies,
1196         },
1197         {
1198                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1199                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1200                 .maxlen         = sizeof(unsigned int),
1201                 .mode           = 0644,
1202                 .proc_handler   = &proc_dointvec_jiffies,
1203         },
1204         {
1205                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1206                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1207                 .maxlen         = sizeof(unsigned int),
1208                 .mode           = 0644,
1209                 .proc_handler   = &proc_dointvec_jiffies,
1210         },
1211         {
1212                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1213                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1214                 .maxlen         = sizeof(unsigned int),
1215                 .mode           = 0644,
1216                 .proc_handler   = &proc_dointvec_jiffies,
1217         },
1218         {
1219                 .procname       = "nf_conntrack_tcp_timeout_close",
1220                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1221                 .maxlen         = sizeof(unsigned int),
1222                 .mode           = 0644,
1223                 .proc_handler   = &proc_dointvec_jiffies,
1224         },
1225         {
1226                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1227                 .data           = &nf_ct_tcp_timeout_max_retrans,
1228                 .maxlen         = sizeof(unsigned int),
1229                 .mode           = 0644,
1230                 .proc_handler   = &proc_dointvec_jiffies,
1231         },
1232         {
1233                 .ctl_name       = NET_NF_CONNTRACK_TCP_LOOSE,
1234                 .procname       = "nf_conntrack_tcp_loose",
1235                 .data           = &nf_ct_tcp_loose,
1236                 .maxlen         = sizeof(unsigned int),
1237                 .mode           = 0644,
1238                 .proc_handler   = &proc_dointvec,
1239         },
1240         {
1241                 .ctl_name       = NET_NF_CONNTRACK_TCP_BE_LIBERAL,
1242                 .procname       = "nf_conntrack_tcp_be_liberal",
1243                 .data           = &nf_ct_tcp_be_liberal,
1244                 .maxlen         = sizeof(unsigned int),
1245                 .mode           = 0644,
1246                 .proc_handler   = &proc_dointvec,
1247         },
1248         {
1249                 .ctl_name       = NET_NF_CONNTRACK_TCP_MAX_RETRANS,
1250                 .procname       = "nf_conntrack_tcp_max_retrans",
1251                 .data           = &nf_ct_tcp_max_retrans,
1252                 .maxlen         = sizeof(unsigned int),
1253                 .mode           = 0644,
1254                 .proc_handler   = &proc_dointvec,
1255         },
1256         {
1257                 .ctl_name       = 0
1258         }
1259 };
1260
1261 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1262 static struct ctl_table tcp_compat_sysctl_table[] = {
1263         {
1264                 .procname       = "ip_conntrack_tcp_timeout_syn_sent",
1265                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1266                 .maxlen         = sizeof(unsigned int),
1267                 .mode           = 0644,
1268                 .proc_handler   = &proc_dointvec_jiffies,
1269         },
1270         {
1271                 .procname       = "ip_conntrack_tcp_timeout_syn_recv",
1272                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1273                 .maxlen         = sizeof(unsigned int),
1274                 .mode           = 0644,
1275                 .proc_handler   = &proc_dointvec_jiffies,
1276         },
1277         {
1278                 .procname       = "ip_conntrack_tcp_timeout_established",
1279                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1280                 .maxlen         = sizeof(unsigned int),
1281                 .mode           = 0644,
1282                 .proc_handler   = &proc_dointvec_jiffies,
1283         },
1284         {
1285                 .procname       = "ip_conntrack_tcp_timeout_fin_wait",
1286                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1287                 .maxlen         = sizeof(unsigned int),
1288                 .mode           = 0644,
1289                 .proc_handler   = &proc_dointvec_jiffies,
1290         },
1291         {
1292                 .procname       = "ip_conntrack_tcp_timeout_close_wait",
1293                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1294                 .maxlen         = sizeof(unsigned int),
1295                 .mode           = 0644,
1296                 .proc_handler   = &proc_dointvec_jiffies,
1297         },
1298         {
1299                 .procname       = "ip_conntrack_tcp_timeout_last_ack",
1300                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1301                 .maxlen         = sizeof(unsigned int),
1302                 .mode           = 0644,
1303                 .proc_handler   = &proc_dointvec_jiffies,
1304         },
1305         {
1306                 .procname       = "ip_conntrack_tcp_timeout_time_wait",
1307                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1308                 .maxlen         = sizeof(unsigned int),
1309                 .mode           = 0644,
1310                 .proc_handler   = &proc_dointvec_jiffies,
1311         },
1312         {
1313                 .procname       = "ip_conntrack_tcp_timeout_close",
1314                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1315                 .maxlen         = sizeof(unsigned int),
1316                 .mode           = 0644,
1317                 .proc_handler   = &proc_dointvec_jiffies,
1318         },
1319         {
1320                 .procname       = "ip_conntrack_tcp_timeout_max_retrans",
1321                 .data           = &nf_ct_tcp_timeout_max_retrans,
1322                 .maxlen         = sizeof(unsigned int),
1323                 .mode           = 0644,
1324                 .proc_handler   = &proc_dointvec_jiffies,
1325         },
1326         {
1327                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_LOOSE,
1328                 .procname       = "ip_conntrack_tcp_loose",
1329                 .data           = &nf_ct_tcp_loose,
1330                 .maxlen         = sizeof(unsigned int),
1331                 .mode           = 0644,
1332                 .proc_handler   = &proc_dointvec,
1333         },
1334         {
1335                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_BE_LIBERAL,
1336                 .procname       = "ip_conntrack_tcp_be_liberal",
1337                 .data           = &nf_ct_tcp_be_liberal,
1338                 .maxlen         = sizeof(unsigned int),
1339                 .mode           = 0644,
1340                 .proc_handler   = &proc_dointvec,
1341         },
1342         {
1343                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_MAX_RETRANS,
1344                 .procname       = "ip_conntrack_tcp_max_retrans",
1345                 .data           = &nf_ct_tcp_max_retrans,
1346                 .maxlen         = sizeof(unsigned int),
1347                 .mode           = 0644,
1348                 .proc_handler   = &proc_dointvec,
1349         },
1350         {
1351                 .ctl_name       = 0
1352         }
1353 };
1354 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
1355 #endif /* CONFIG_SYSCTL */
1356
1357 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1358 {
1359         .l3proto                = PF_INET,
1360         .l4proto                = IPPROTO_TCP,
1361         .name                   = "tcp",
1362         .pkt_to_tuple           = tcp_pkt_to_tuple,
1363         .invert_tuple           = tcp_invert_tuple,
1364         .print_tuple            = tcp_print_tuple,
1365         .print_conntrack        = tcp_print_conntrack,
1366         .packet                 = tcp_packet,
1367         .new                    = tcp_new,
1368         .error                  = tcp_error,
1369 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1370         .to_nlattr              = tcp_to_nlattr,
1371         .from_nlattr            = nlattr_to_tcp,
1372         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1373         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1374         .nla_policy             = nf_ct_port_nla_policy,
1375 #endif
1376 #ifdef CONFIG_SYSCTL
1377         .ctl_table_users        = &tcp_sysctl_table_users,
1378         .ctl_table_header       = &tcp_sysctl_header,
1379         .ctl_table              = tcp_sysctl_table,
1380 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1381         .ctl_compat_table       = tcp_compat_sysctl_table,
1382 #endif
1383 #endif
1384 };
1385 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1386
1387 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1388 {
1389         .l3proto                = PF_INET6,
1390         .l4proto                = IPPROTO_TCP,
1391         .name                   = "tcp",
1392         .pkt_to_tuple           = tcp_pkt_to_tuple,
1393         .invert_tuple           = tcp_invert_tuple,
1394         .print_tuple            = tcp_print_tuple,
1395         .print_conntrack        = tcp_print_conntrack,
1396         .packet                 = tcp_packet,
1397         .new                    = tcp_new,
1398         .error                  = tcp_error,
1399 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1400         .to_nlattr              = tcp_to_nlattr,
1401         .from_nlattr            = nlattr_to_tcp,
1402         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1403         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1404         .nla_policy             = nf_ct_port_nla_policy,
1405 #endif
1406 #ifdef CONFIG_SYSCTL
1407         .ctl_table_users        = &tcp_sysctl_table_users,
1408         .ctl_table_header       = &tcp_sysctl_header,
1409         .ctl_table              = tcp_sysctl_table,
1410 #endif
1411 };
1412 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);