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