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

   1 /* (C) 1999-2001 Paul `Rusty' Russell
   2  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  */
   8
   9 #include <linux/types.h>
  10 #include <linux/timer.h>
  11 #include <linux/module.h>
  12 #include <linux/in.h>
  13 #include <linux/tcp.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/ipv6.h>
  17 #include <net/ip6_checksum.h>
  18
  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 ack:      sack <= receiver.td_end
 335    IV.  Lower bound for valid ack:      ack >= receiver.td_end - MAXACKWINDOW
 336
 337    where sack is the highest right edge of sack block found in the packet.
 338
 339    The upper bound limit for a valid ack is not ignored -
 340    we doesn't have to deal with fragments.
 341 */
 342
 343 static inline __u32 segment_seq_plus_len(__u32 seq,
 344                                          size_t len,
 345                                          unsigned int dataoff,
 346                                          const struct tcphdr *tcph)
 347 {
 348         /* XXX Should I use payload length field in IP/IPv6 header ?
 349          * - YK */
 350         return (seq + len - dataoff - tcph->doff*4
 351                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 352 }
 353
 354 /* Fixme: what about big packets? */
 355 #define MAXACKWINCONST                  66000
 356 #define MAXACKWINDOW(sender)                                            \
 357         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 358                                               : MAXACKWINCONST)
 359
 360 /*
 361  * Simplified tcp_parse_options routine from tcp_input.c
 362  */
 363 static void tcp_options(const struct sk_buff *skb,
 364                         unsigned int dataoff,
 365                         const struct tcphdr *tcph,
 366                         struct ip_ct_tcp_state *state)
 367 {
 368         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 369         const unsigned char *ptr;
 370         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 371
 372         if (!length)
 373                 return;
 374
 375         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 376                                  length, buff);
 377         BUG_ON(ptr == NULL);
 378
 379         state->td_scale =
 380         state->flags = 0;
 381
 382         while (length > 0) {
 383                 int opcode=*ptr++;
 384                 int opsize;
 385
 386                 switch (opcode) {
 387                 case TCPOPT_EOL:
 388                         return;
 389                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 390                         length--;
 391                         continue;
 392                 default:
 393                         opsize=*ptr++;
 394                         if (opsize < 2) /* "silly options" */
 395                                 return;
 396                         if (opsize > length)
 397                                 break;  /* don't parse partial options */
 398
 399                         if (opcode == TCPOPT_SACK_PERM
 400                             && opsize == TCPOLEN_SACK_PERM)
 401                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 402                         else if (opcode == TCPOPT_WINDOW
 403                                  && opsize == TCPOLEN_WINDOW) {
 404                                 state->td_scale = *(u_int8_t *)ptr;
 405
 406                                 if (state->td_scale > 14) {
 407                                         /* See RFC1323 */
 408                                         state->td_scale = 14;
 409                                 }
 410                                 state->flags |=
 411                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 412                         }
 413                         ptr += opsize - 2;
 414                         length -= opsize;
 415                 }
 416         }
 417 }
 418
 419 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 420                      const struct tcphdr *tcph, __u32 *sack)
 421 {
 422         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 423         const unsigned char *ptr;
 424         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 425         __u32 tmp;
 426
 427         if (!length)
 428                 return;
 429
 430         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 431                                  length, buff);
 432         BUG_ON(ptr == NULL);
 433
 434         /* Fast path for timestamp-only option */
 435         if (length == TCPOLEN_TSTAMP_ALIGNED*4
 436             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
 437                                        | (TCPOPT_NOP << 16)
 438                                        | (TCPOPT_TIMESTAMP << 8)
 439                                        | TCPOLEN_TIMESTAMP))
 440                 return;
 441
 442         while (length > 0) {
 443                 int opcode = *ptr++;
 444                 int opsize, i;
 445
 446                 switch (opcode) {
 447                 case TCPOPT_EOL:
 448                         return;
 449                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 450                         length--;
 451                         continue;
 452                 default:
 453                         opsize = *ptr++;
 454                         if (opsize < 2) /* "silly options" */
 455                                 return;
 456                         if (opsize > length)
 457                                 break;  /* don't parse partial options */
 458
 459                         if (opcode == TCPOPT_SACK
 460                             && opsize >= (TCPOLEN_SACK_BASE
 461                                           + TCPOLEN_SACK_PERBLOCK)
 462                             && !((opsize - TCPOLEN_SACK_BASE)
 463                                  % TCPOLEN_SACK_PERBLOCK)) {
 464                                 for (i = 0;
 465                                      i < (opsize - TCPOLEN_SACK_BASE);
 466                                      i += TCPOLEN_SACK_PERBLOCK) {
 467                                         tmp = ntohl(*((__be32 *)(ptr+i)+1));
 468
 469                                         if (after(tmp, *sack))
 470                                                 *sack = tmp;
 471                                 }
 472                                 return;
 473                         }
 474                         ptr += opsize - 2;
 475                         length -= opsize;
 476                 }
 477         }
 478 }
 479
 480 static bool tcp_in_window(const struct nf_conn *ct,
 481                           struct ip_ct_tcp *state,
 482                           enum ip_conntrack_dir dir,
 483                           unsigned int index,
 484                           const struct sk_buff *skb,
 485                           unsigned int dataoff,
 486                           const struct tcphdr *tcph,
 487                           int pf)
 488 {
 489         struct ip_ct_tcp_state *sender = &state->seen[dir];
 490         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 491         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 492         __u32 seq, ack, sack, end, win, swin;
 493         bool res;
 494
 495         /*
 496          * Get the required data from the packet.
 497          */
 498         seq = ntohl(tcph->seq);
 499         ack = sack = ntohl(tcph->ack_seq);
 500         win = ntohs(tcph->window);
 501         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 502
 503         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 504                 tcp_sack(skb, dataoff, tcph, &sack);
 505
 506         pr_debug("tcp_in_window: START\n");
 507         pr_debug("tcp_in_window: ");
 508         NF_CT_DUMP_TUPLE(tuple);
 509         pr_debug("seq=%u ack=%u sack=%u win=%u end=%u\n",
 510                  seq, ack, sack, win, end);
 511         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 512                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 513                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 514                  sender->td_scale,
 515                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 516                  receiver->td_scale);
 517
 518         if (sender->td_end == 0) {
 519                 /*
 520                  * Initialize sender data.
 521                  */
 522                 if (tcph->syn && tcph->ack) {
 523                         /*
 524                          * Outgoing SYN-ACK in reply to a SYN.
 525                          */
 526                         sender->td_end =
 527                         sender->td_maxend = end;
 528                         sender->td_maxwin = (win == 0 ? 1 : win);
 529
 530                         tcp_options(skb, dataoff, tcph, sender);
 531                         /*
 532                          * RFC 1323:
 533                          * Both sides must send the Window Scale option
 534                          * to enable window scaling in either direction.
 535                          */
 536                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 537                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 538                                 sender->td_scale =
 539                                 receiver->td_scale = 0;
 540                 } else {
 541                         /*
 542                          * We are in the middle of a connection,
 543                          * its history is lost for us.
 544                          * Let's try to use the data from the packet.
 545                          */
 546                         sender->td_end = end;
 547                         sender->td_maxwin = (win == 0 ? 1 : win);
 548                         sender->td_maxend = end + sender->td_maxwin;
 549                 }
 550         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 551                      && dir == IP_CT_DIR_ORIGINAL)
 552                    || (state->state == TCP_CONNTRACK_SYN_RECV
 553                      && dir == IP_CT_DIR_REPLY))
 554                    && after(end, sender->td_end)) {
 555                 /*
 556                  * RFC 793: "if a TCP is reinitialized ... then it need
 557                  * not wait at all; it must only be sure to use sequence
 558                  * numbers larger than those recently used."
 559                  */
 560                 sender->td_end =
 561                 sender->td_maxend = end;
 562                 sender->td_maxwin = (win == 0 ? 1 : win);
 563
 564                 tcp_options(skb, dataoff, tcph, sender);
 565         }
 566
 567         if (!(tcph->ack)) {
 568                 /*
 569                  * If there is no ACK, just pretend it was set and OK.
 570                  */
 571                 ack = sack = receiver->td_end;
 572         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 573                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 574                    && (ack == 0)) {
 575                 /*
 576                  * Broken TCP stacks, that set ACK in RST packets as well
 577                  * with zero ack value.
 578                  */
 579                 ack = sack = receiver->td_end;
 580         }
 581
 582         if (seq == end
 583             && (!tcph->rst
 584                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
 585                 /*
 586                  * Packets contains no data: we assume it is valid
 587                  * and check the ack value only.
 588                  * However RST segments are always validated by their
 589                  * SEQ number, except when seq == 0 (reset sent answering
 590                  * SYN.
 591                  */
 592                 seq = end = sender->td_end;
 593
 594         pr_debug("tcp_in_window: ");
 595         NF_CT_DUMP_TUPLE(tuple);
 596         pr_debug("seq=%u ack=%u sack =%u win=%u end=%u\n",
 597                  seq, ack, sack, win, end);
 598         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 599                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 600                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 601                  sender->td_scale,
 602                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 603                  receiver->td_scale);
 604
 605         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 606                  before(seq, sender->td_maxend + 1),
 607                  after(end, sender->td_end - receiver->td_maxwin - 1),
 608                  before(sack, receiver->td_end + 1),
 609                  after(ack, receiver->td_end - MAXACKWINDOW(sender)));
 610
 611         if (before(seq, sender->td_maxend + 1) &&
 612             after(end, sender->td_end - receiver->td_maxwin - 1) &&
 613             before(sack, receiver->td_end + 1) &&
 614             after(ack, receiver->td_end - MAXACKWINDOW(sender))) {
 615                 /*
 616                  * Take into account window scaling (RFC 1323).
 617                  */
 618                 if (!tcph->syn)
 619                         win <<= sender->td_scale;
 620
 621                 /*
 622                  * Update sender data.
 623                  */
 624                 swin = win + (sack - ack);
 625                 if (sender->td_maxwin < swin)
 626                         sender->td_maxwin = swin;
 627                 if (after(end, sender->td_end))
 628                         sender->td_end = end;
 629                 /*
 630                  * Update receiver data.
 631                  */
 632                 if (after(end, sender->td_maxend))
 633                         receiver->td_maxwin += end - sender->td_maxend;
 634                 if (after(sack + win, receiver->td_maxend - 1)) {
 635                         receiver->td_maxend = sack + win;
 636                         if (win == 0)
 637                                 receiver->td_maxend++;
 638                 }
 639
 640                 /*
 641                  * Check retransmissions.
 642                  */
 643                 if (index == TCP_ACK_SET) {
 644                         if (state->last_dir == dir
 645                             && state->last_seq == seq
 646                             && state->last_ack == ack
 647                             && state->last_end == end
 648                             && state->last_win == win)
 649                                 state->retrans++;
 650                         else {
 651                                 state->last_dir = dir;
 652                                 state->last_seq = seq;
 653                                 state->last_ack = ack;
 654                                 state->last_end = end;
 655                                 state->last_win = win;
 656                                 state->retrans = 0;
 657                         }
 658                 }
 659                 res = true;
 660         } else {
 661                 res = false;
 662                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 663                     nf_ct_tcp_be_liberal)
 664                         res = true;
 665                 if (!res && LOG_INVALID(IPPROTO_TCP))
 666                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 667                         "nf_ct_tcp: %s ",
 668                         before(seq, sender->td_maxend + 1) ?
 669                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
 670                         before(sack, receiver->td_end + 1) ?
 671                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
 672                         : "ACK is under the lower bound (possible overly delayed ACK)"
 673                         : "ACK is over the upper bound (ACKed data not seen yet)"
 674                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 675                         : "SEQ is over the upper bound (over the window of the receiver)");
 676         }
 677
 678         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 679                  "receiver end=%u maxend=%u maxwin=%u\n",
 680                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 681                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 682
 683         return res;
 684 }
 685
 686 #ifdef CONFIG_NF_NAT_NEEDED
 687 /* Update sender->td_end after NAT successfully mangled the packet */
 688 /* Caller must linearize skb at tcp header. */
 689 void nf_conntrack_tcp_update(const struct sk_buff *skb,
 690                              unsigned int dataoff,
 691                              struct nf_conn *ct,
 692                              int dir)
 693 {
 694         const struct tcphdr *tcph = (const void *)skb->data + dataoff;
 695         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[dir];
 696         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[!dir];
 697         __u32 end;
 698
 699         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
 700
 701         write_lock_bh(&tcp_lock);
 702         /*
 703          * We have to worry for the ack in the reply packet only...
 704          */
 705         if (after(end, ct->proto.tcp.seen[dir].td_end))
 706                 ct->proto.tcp.seen[dir].td_end = end;
 707         ct->proto.tcp.last_end = end;
 708         write_unlock_bh(&tcp_lock);
 709         pr_debug("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
 710                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 711                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 712                  sender->td_scale,
 713                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 714                  receiver->td_scale);
 715 }
 716 EXPORT_SYMBOL_GPL(nf_conntrack_tcp_update);
 717 #endif
 718
 719 #define TH_FIN  0x01
 720 #define TH_SYN  0x02
 721 #define TH_RST  0x04
 722 #define TH_PUSH 0x08
 723 #define TH_ACK  0x10
 724 #define TH_URG  0x20
 725 #define TH_ECE  0x40
 726 #define TH_CWR  0x80
 727
 728 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 729 static const u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_ACK|TH_URG) + 1] =
 730 {
 731         [TH_SYN]                        = 1,
 732         [TH_SYN|TH_URG]                 = 1,
 733         [TH_SYN|TH_ACK]                 = 1,
 734         [TH_RST]                        = 1,
 735         [TH_RST|TH_ACK]                 = 1,
 736         [TH_FIN|TH_ACK]                 = 1,
 737         [TH_FIN|TH_ACK|TH_URG]          = 1,
 738         [TH_ACK]                        = 1,
 739         [TH_ACK|TH_URG]                 = 1,
 740 };
 741
 742 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 743 static int tcp_error(struct sk_buff *skb,
 744                      unsigned int dataoff,
 745                      enum ip_conntrack_info *ctinfo,
 746                      int pf,
 747                      unsigned int hooknum)
 748 {
 749         const struct tcphdr *th;
 750         struct tcphdr _tcph;
 751         unsigned int tcplen = skb->len - dataoff;
 752         u_int8_t tcpflags;
 753
 754         /* Smaller that minimal TCP header? */
 755         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 756         if (th == NULL) {
 757                 if (LOG_INVALID(IPPROTO_TCP))
 758                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 759                                 "nf_ct_tcp: short packet ");
 760                 return -NF_ACCEPT;
 761         }
 762
 763         /* Not whole TCP header or malformed packet */
 764         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 765                 if (LOG_INVALID(IPPROTO_TCP))
 766                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 767                                 "nf_ct_tcp: truncated/malformed packet ");
 768                 return -NF_ACCEPT;
 769         }
 770
 771         /* Checksum invalid? Ignore.
 772          * We skip checking packets on the outgoing path
 773          * because the checksum is assumed to be correct.
 774          */
 775         /* FIXME: Source route IP option packets --RR */
 776         if (nf_conntrack_checksum && hooknum == NF_INET_PRE_ROUTING &&
 777             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
 778                 if (LOG_INVALID(IPPROTO_TCP))
 779                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 780                                   "nf_ct_tcp: bad TCP checksum ");
 781                 return -NF_ACCEPT;
 782         }
 783
 784         /* Check TCP flags. */
 785         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR|TH_PUSH));
 786         if (!tcp_valid_flags[tcpflags]) {
 787                 if (LOG_INVALID(IPPROTO_TCP))
 788                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 789                                   "nf_ct_tcp: invalid TCP flag combination ");
 790                 return -NF_ACCEPT;
 791         }
 792
 793         return NF_ACCEPT;
 794 }
 795
 796 /* Returns verdict for packet, or -1 for invalid. */
 797 static int tcp_packet(struct nf_conn *ct,
 798                       const struct sk_buff *skb,
 799                       unsigned int dataoff,
 800                       enum ip_conntrack_info ctinfo,
 801                       int pf,
 802                       unsigned int hooknum)
 803 {
 804         struct nf_conntrack_tuple *tuple;
 805         enum tcp_conntrack new_state, old_state;
 806         enum ip_conntrack_dir dir;
 807         const struct tcphdr *th;
 808         struct tcphdr _tcph;
 809         unsigned long timeout;
 810         unsigned int index;
 811
 812         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 813         BUG_ON(th == NULL);
 814
 815         write_lock_bh(&tcp_lock);
 816         old_state = ct->proto.tcp.state;
 817         dir = CTINFO2DIR(ctinfo);
 818         index = get_conntrack_index(th);
 819         new_state = tcp_conntracks[dir][index][old_state];
 820         tuple = &ct->tuplehash[dir].tuple;
 821
 822         switch (new_state) {
 823         case TCP_CONNTRACK_SYN_SENT:
 824                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 825                         break;
 826                 /* RFC 1122: "When a connection is closed actively,
 827                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 828                  * (Maximum Segment Lifetime). However, it MAY accept
 829                  * a new SYN from the remote TCP to reopen the connection
 830                  * directly from TIME-WAIT state, if..."
 831                  * We ignore the conditions because we are in the
 832                  * TIME-WAIT state anyway.
 833                  *
 834                  * Handle aborted connections: we and the server
 835                  * think there is an existing connection but the client
 836                  * aborts it and starts a new one.
 837                  */
 838                 if (((ct->proto.tcp.seen[dir].flags
 839                       | ct->proto.tcp.seen[!dir].flags)
 840                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 841                     || (ct->proto.tcp.last_dir == dir
 842                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 843                         /* Attempt to reopen a closed/aborted connection.
 844                          * Delete this connection and look up again. */
 845                         write_unlock_bh(&tcp_lock);
 846                         if (del_timer(&ct->timeout))
 847                                 ct->timeout.function((unsigned long)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                         if (del_timer(&ct->timeout))
 881                                 ct->timeout.function((unsigned long)ct);
 882                         return -NF_DROP;
 883                 }
 884                 ct->proto.tcp.last_index = index;
 885                 ct->proto.tcp.last_dir = dir;
 886                 ct->proto.tcp.last_seq = ntohl(th->seq);
 887                 ct->proto.tcp.last_end =
 888                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 889
 890                 write_unlock_bh(&tcp_lock);
 891                 if (LOG_INVALID(IPPROTO_TCP))
 892                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 893                                   "nf_ct_tcp: invalid packet ignored ");
 894                 return NF_ACCEPT;
 895         case TCP_CONNTRACK_MAX:
 896                 /* Invalid packet */
 897                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 898                          dir, get_conntrack_index(th), old_state);
 899                 write_unlock_bh(&tcp_lock);
 900                 if (LOG_INVALID(IPPROTO_TCP))
 901                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 902                                   "nf_ct_tcp: invalid state ");
 903                 return -NF_ACCEPT;
 904         case TCP_CONNTRACK_CLOSE:
 905                 if (index == TCP_RST_SET
 906                     && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
 907                          && ct->proto.tcp.last_index == TCP_SYN_SET)
 908                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
 909                             && ct->proto.tcp.last_index == TCP_ACK_SET))
 910                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 911                         /* RST sent to invalid SYN or ACK we had let through
 912                          * at a) and c) above:
 913                          *
 914                          * a) SYN was in window then
 915                          * c) we hold a half-open connection.
 916                          *
 917                          * Delete our connection entry.
 918                          * We skip window checking, because packet might ACK
 919                          * segments we ignored. */
 920                         goto in_window;
 921                 }
 922                 /* Just fall through */
 923         default:
 924                 /* Keep compilers happy. */
 925                 break;
 926         }
 927
 928         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
 929                            skb, dataoff, th, pf)) {
 930                 write_unlock_bh(&tcp_lock);
 931                 return -NF_ACCEPT;
 932         }
 933      in_window:
 934         /* From now on we have got in-window packets */
 935         ct->proto.tcp.last_index = index;
 936         ct->proto.tcp.last_dir = dir;
 937
 938         pr_debug("tcp_conntracks: ");
 939         NF_CT_DUMP_TUPLE(tuple);
 940         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
 941                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
 942                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
 943                  old_state, new_state);
 944
 945         ct->proto.tcp.state = new_state;
 946         if (old_state != new_state
 947             && new_state == TCP_CONNTRACK_FIN_WAIT)
 948                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
 949         timeout = ct->proto.tcp.retrans >= nf_ct_tcp_max_retrans
 950                   && tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans
 951                   ? nf_ct_tcp_timeout_max_retrans : tcp_timeouts[new_state];
 952         write_unlock_bh(&tcp_lock);
 953
 954         nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
 955         if (new_state != old_state)
 956                 nf_conntrack_event_cache(IPCT_PROTOINFO, skb);
 957
 958         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 959                 /* If only reply is a RST, we can consider ourselves not to
 960                    have an established connection: this is a fairly common
 961                    problem case, so we can delete the conntrack
 962                    immediately.  --RR */
 963                 if (th->rst) {
 964                         if (del_timer(&ct->timeout))
 965                                 ct->timeout.function((unsigned long)ct);
 966                         return NF_ACCEPT;
 967                 }
 968         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
 969                    && (old_state == TCP_CONNTRACK_SYN_RECV
 970                        || old_state == TCP_CONNTRACK_ESTABLISHED)
 971                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
 972                 /* Set ASSURED if we see see valid ack in ESTABLISHED
 973                    after SYN_RECV or a valid answer for a picked up
 974                    connection. */
 975                 set_bit(IPS_ASSURED_BIT, &ct->status);
 976                 nf_conntrack_event_cache(IPCT_STATUS, skb);
 977         }
 978         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
 979
 980         return NF_ACCEPT;
 981 }
 982
 983 /* Called when a new connection for this protocol found. */
 984 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
 985                     unsigned int dataoff)
 986 {
 987         enum tcp_conntrack new_state;
 988         const struct tcphdr *th;
 989         struct tcphdr _tcph;
 990         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
 991         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
 992
 993         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 994         BUG_ON(th == NULL);
 995
 996         /* Don't need lock here: this conntrack not in circulation yet */
 997         new_state
 998                 = tcp_conntracks[0][get_conntrack_index(th)]
 999                 [TCP_CONNTRACK_NONE];
1000
1001         /* Invalid: delete conntrack */
1002         if (new_state >= TCP_CONNTRACK_MAX) {
1003                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1004                 return false;
1005         }
1006
1007         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1008                 /* SYN packet */
1009                 ct->proto.tcp.seen[0].td_end =
1010                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1011                                              dataoff, th);
1012                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1013                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1014                         ct->proto.tcp.seen[0].td_maxwin = 1;
1015                 ct->proto.tcp.seen[0].td_maxend =
1016                         ct->proto.tcp.seen[0].td_end;
1017
1018                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1019                 ct->proto.tcp.seen[1].flags = 0;
1020         } else if (nf_ct_tcp_loose == 0) {
1021                 /* Don't try to pick up connections. */
1022                 return false;
1023         } else {
1024                 /*
1025                  * We are in the middle of a connection,
1026                  * its history is lost for us.
1027                  * Let's try to use the data from the packet.
1028                  */
1029                 ct->proto.tcp.seen[0].td_end =
1030                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1031                                              dataoff, th);
1032                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1033                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1034                         ct->proto.tcp.seen[0].td_maxwin = 1;
1035                 ct->proto.tcp.seen[0].td_maxend =
1036                         ct->proto.tcp.seen[0].td_end +
1037                         ct->proto.tcp.seen[0].td_maxwin;
1038                 ct->proto.tcp.seen[0].td_scale = 0;
1039
1040                 /* We assume SACK and liberal window checking to handle
1041                  * window scaling */
1042                 ct->proto.tcp.seen[0].flags =
1043                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1044                                               IP_CT_TCP_FLAG_BE_LIBERAL;
1045         }
1046
1047         ct->proto.tcp.seen[1].td_end = 0;
1048         ct->proto.tcp.seen[1].td_maxend = 0;
1049         ct->proto.tcp.seen[1].td_maxwin = 1;
1050         ct->proto.tcp.seen[1].td_scale = 0;
1051
1052         /* tcp_packet will set them */
1053         ct->proto.tcp.state = TCP_CONNTRACK_NONE;
1054         ct->proto.tcp.last_index = TCP_NONE_SET;
1055
1056         pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1057                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1058                  sender->td_end, sender->td_maxend, sender->td_maxwin,
1059                  sender->td_scale,
1060                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1061                  receiver->td_scale);
1062         return true;
1063 }
1064
1065 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1066
1067 #include <linux/netfilter/nfnetlink.h>
1068 #include <linux/netfilter/nfnetlink_conntrack.h>
1069
1070 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1071                          const struct nf_conn *ct)
1072 {
1073         struct nlattr *nest_parms;
1074         struct nf_ct_tcp_flags tmp = {};
1075
1076         read_lock_bh(&tcp_lock);
1077         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1078         if (!nest_parms)
1079                 goto nla_put_failure;
1080
1081         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state);
1082
1083         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1084                    ct->proto.tcp.seen[0].td_scale);
1085
1086         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1087                    ct->proto.tcp.seen[1].td_scale);
1088
1089         tmp.flags = ct->proto.tcp.seen[0].flags;
1090         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1091                 sizeof(struct nf_ct_tcp_flags), &tmp);
1092
1093         tmp.flags = ct->proto.tcp.seen[1].flags;
1094         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1095                 sizeof(struct nf_ct_tcp_flags), &tmp);
1096         read_unlock_bh(&tcp_lock);
1097
1098         nla_nest_end(skb, nest_parms);
1099
1100         return 0;
1101
1102 nla_put_failure:
1103         read_unlock_bh(&tcp_lock);
1104         return -1;
1105 }
1106
1107 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1108         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1109         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1110         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1111         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1112         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len =  sizeof(struct nf_ct_tcp_flags) },
1113 };
1114
1115 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1116 {
1117         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1118         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1119         int err;
1120
1121         /* updates could not contain anything about the private
1122          * protocol info, in that case skip the parsing */
1123         if (!pattr)
1124                 return 0;
1125
1126         err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr, tcp_nla_policy);
1127         if (err < 0)
1128                 return err;
1129
1130         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1131             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1132                 return -EINVAL;
1133
1134         write_lock_bh(&tcp_lock);
1135         if (tb[CTA_PROTOINFO_TCP_STATE])
1136                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1137
1138         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1139                 struct nf_ct_tcp_flags *attr =
1140                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1141                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1142                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1143         }
1144
1145         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1146                 struct nf_ct_tcp_flags *attr =
1147                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1148                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1149                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1150         }
1151
1152         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1153             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1154             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1155             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1156                 ct->proto.tcp.seen[0].td_scale =
1157                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1158                 ct->proto.tcp.seen[1].td_scale =
1159                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1160         }
1161         write_unlock_bh(&tcp_lock);
1162
1163         return 0;
1164 }
1165 #endif
1166
1167 #ifdef CONFIG_SYSCTL
1168 static unsigned int tcp_sysctl_table_users;
1169 static struct ctl_table_header *tcp_sysctl_header;
1170 static struct ctl_table tcp_sysctl_table[] = {
1171         {
1172                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1173                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1174                 .maxlen         = sizeof(unsigned int),
1175                 .mode           = 0644,
1176                 .proc_handler   = &proc_dointvec_jiffies,
1177         },
1178         {
1179                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1180                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1181                 .maxlen         = sizeof(unsigned int),
1182                 .mode           = 0644,
1183                 .proc_handler   = &proc_dointvec_jiffies,
1184         },
1185         {
1186                 .procname       = "nf_conntrack_tcp_timeout_established",
1187                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1188                 .maxlen         = sizeof(unsigned int),
1189                 .mode           = 0644,
1190                 .proc_handler   = &proc_dointvec_jiffies,
1191         },
1192         {
1193                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1194                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1195                 .maxlen         = sizeof(unsigned int),
1196                 .mode           = 0644,
1197                 .proc_handler   = &proc_dointvec_jiffies,
1198         },
1199         {
1200                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1201                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1202                 .maxlen         = sizeof(unsigned int),
1203                 .mode           = 0644,
1204                 .proc_handler   = &proc_dointvec_jiffies,
1205         },
1206         {
1207                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1208                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1209                 .maxlen         = sizeof(unsigned int),
1210                 .mode           = 0644,
1211                 .proc_handler   = &proc_dointvec_jiffies,
1212         },
1213         {
1214                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1215                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1216                 .maxlen         = sizeof(unsigned int),
1217                 .mode           = 0644,
1218                 .proc_handler   = &proc_dointvec_jiffies,
1219         },
1220         {
1221                 .procname       = "nf_conntrack_tcp_timeout_close",
1222                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1223                 .maxlen         = sizeof(unsigned int),
1224                 .mode           = 0644,
1225                 .proc_handler   = &proc_dointvec_jiffies,
1226         },
1227         {
1228                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1229                 .data           = &nf_ct_tcp_timeout_max_retrans,
1230                 .maxlen         = sizeof(unsigned int),
1231                 .mode           = 0644,
1232                 .proc_handler   = &proc_dointvec_jiffies,
1233         },
1234         {
1235                 .ctl_name       = NET_NF_CONNTRACK_TCP_LOOSE,
1236                 .procname       = "nf_conntrack_tcp_loose",
1237                 .data           = &nf_ct_tcp_loose,
1238                 .maxlen         = sizeof(unsigned int),
1239                 .mode           = 0644,
1240                 .proc_handler   = &proc_dointvec,
1241         },
1242         {
1243                 .ctl_name       = NET_NF_CONNTRACK_TCP_BE_LIBERAL,
1244                 .procname       = "nf_conntrack_tcp_be_liberal",
1245                 .data           = &nf_ct_tcp_be_liberal,
1246                 .maxlen         = sizeof(unsigned int),
1247                 .mode           = 0644,
1248                 .proc_handler   = &proc_dointvec,
1249         },
1250         {
1251                 .ctl_name       = NET_NF_CONNTRACK_TCP_MAX_RETRANS,
1252                 .procname       = "nf_conntrack_tcp_max_retrans",
1253                 .data           = &nf_ct_tcp_max_retrans,
1254                 .maxlen         = sizeof(unsigned int),
1255                 .mode           = 0644,
1256                 .proc_handler   = &proc_dointvec,
1257         },
1258         {
1259                 .ctl_name       = 0
1260         }
1261 };
1262
1263 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1264 static struct ctl_table tcp_compat_sysctl_table[] = {
1265         {
1266                 .procname       = "ip_conntrack_tcp_timeout_syn_sent",
1267                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1268                 .maxlen         = sizeof(unsigned int),
1269                 .mode           = 0644,
1270                 .proc_handler   = &proc_dointvec_jiffies,
1271         },
1272         {
1273                 .procname       = "ip_conntrack_tcp_timeout_syn_recv",
1274                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1275                 .maxlen         = sizeof(unsigned int),
1276                 .mode           = 0644,
1277                 .proc_handler   = &proc_dointvec_jiffies,
1278         },
1279         {
1280                 .procname       = "ip_conntrack_tcp_timeout_established",
1281                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1282                 .maxlen         = sizeof(unsigned int),
1283                 .mode           = 0644,
1284                 .proc_handler   = &proc_dointvec_jiffies,
1285         },
1286         {
1287                 .procname       = "ip_conntrack_tcp_timeout_fin_wait",
1288                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1289                 .maxlen         = sizeof(unsigned int),
1290                 .mode           = 0644,
1291                 .proc_handler   = &proc_dointvec_jiffies,
1292         },
1293         {
1294                 .procname       = "ip_conntrack_tcp_timeout_close_wait",
1295                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1296                 .maxlen         = sizeof(unsigned int),
1297                 .mode           = 0644,
1298                 .proc_handler   = &proc_dointvec_jiffies,
1299         },
1300         {
1301                 .procname       = "ip_conntrack_tcp_timeout_last_ack",
1302                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1303                 .maxlen         = sizeof(unsigned int),
1304                 .mode           = 0644,
1305                 .proc_handler   = &proc_dointvec_jiffies,
1306         },
1307         {
1308                 .procname       = "ip_conntrack_tcp_timeout_time_wait",
1309                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1310                 .maxlen         = sizeof(unsigned int),
1311                 .mode           = 0644,
1312                 .proc_handler   = &proc_dointvec_jiffies,
1313         },
1314         {
1315                 .procname       = "ip_conntrack_tcp_timeout_close",
1316                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1317                 .maxlen         = sizeof(unsigned int),
1318                 .mode           = 0644,
1319                 .proc_handler   = &proc_dointvec_jiffies,
1320         },
1321         {
1322                 .procname       = "ip_conntrack_tcp_timeout_max_retrans",
1323                 .data           = &nf_ct_tcp_timeout_max_retrans,
1324                 .maxlen         = sizeof(unsigned int),
1325                 .mode           = 0644,
1326                 .proc_handler   = &proc_dointvec_jiffies,
1327         },
1328         {
1329                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_LOOSE,
1330                 .procname       = "ip_conntrack_tcp_loose",
1331                 .data           = &nf_ct_tcp_loose,
1332                 .maxlen         = sizeof(unsigned int),
1333                 .mode           = 0644,
1334                 .proc_handler   = &proc_dointvec,
1335         },
1336         {
1337                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_BE_LIBERAL,
1338                 .procname       = "ip_conntrack_tcp_be_liberal",
1339                 .data           = &nf_ct_tcp_be_liberal,
1340                 .maxlen         = sizeof(unsigned int),
1341                 .mode           = 0644,
1342                 .proc_handler   = &proc_dointvec,
1343         },
1344         {
1345                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_MAX_RETRANS,
1346                 .procname       = "ip_conntrack_tcp_max_retrans",
1347                 .data           = &nf_ct_tcp_max_retrans,
1348                 .maxlen         = sizeof(unsigned int),
1349                 .mode           = 0644,
1350                 .proc_handler   = &proc_dointvec,
1351         },
1352         {
1353                 .ctl_name       = 0
1354         }
1355 };
1356 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
1357 #endif /* CONFIG_SYSCTL */
1358
1359 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1360 {
1361         .l3proto                = PF_INET,
1362         .l4proto                = IPPROTO_TCP,
1363         .name                   = "tcp",
1364         .pkt_to_tuple           = tcp_pkt_to_tuple,
1365         .invert_tuple           = tcp_invert_tuple,
1366         .print_tuple            = tcp_print_tuple,
1367         .print_conntrack        = tcp_print_conntrack,
1368         .packet                 = tcp_packet,
1369         .new                    = tcp_new,
1370         .error                  = tcp_error,
1371 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1372         .to_nlattr              = tcp_to_nlattr,
1373         .from_nlattr            = nlattr_to_tcp,
1374         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1375         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1376         .nla_policy             = nf_ct_port_nla_policy,
1377 #endif
1378 #ifdef CONFIG_SYSCTL
1379         .ctl_table_users        = &tcp_sysctl_table_users,
1380         .ctl_table_header       = &tcp_sysctl_header,
1381         .ctl_table              = tcp_sysctl_table,
1382 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1383         .ctl_compat_table       = tcp_compat_sysctl_table,
1384 #endif
1385 #endif
1386 };
1387 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1388
1389 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1390 {
1391         .l3proto                = PF_INET6,
1392         .l4proto                = IPPROTO_TCP,
1393         .name                   = "tcp",
1394         .pkt_to_tuple           = tcp_pkt_to_tuple,
1395         .invert_tuple           = tcp_invert_tuple,
1396         .print_tuple            = tcp_print_tuple,
1397         .print_conntrack        = tcp_print_conntrack,
1398         .packet                 = tcp_packet,
1399         .new                    = tcp_new,
1400         .error                  = tcp_error,
1401 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1402         .to_nlattr              = tcp_to_nlattr,
1403         .from_nlattr            = nlattr_to_tcp,
1404         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1405         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1406         .nla_policy             = nf_ct_port_nla_policy,
1407 #endif
1408 #ifdef CONFIG_SYSCTL
1409         .ctl_table_users        = &tcp_sysctl_table_users,
1410         .ctl_table_header       = &tcp_sysctl_header,
1411         .ctl_table              = tcp_sysctl_table,
1412 #endif
1413 };
1414 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);