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  * (C) 2002-2013 Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>
   4  * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/types.h>
  12 #include <linux/timer.h>
  13 #include <linux/module.h>
  14 #include <linux/in.h>
  15 #include <linux/tcp.h>
  16 #include <linux/spinlock.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/ipv6.h>
  19 #include <net/ip6_checksum.h>
  20 #include <asm/unaligned.h>
  21
  22 #include <net/tcp.h>
  23
  24 #include <linux/netfilter.h>
  25 #include <linux/netfilter_ipv4.h>
  26 #include <linux/netfilter_ipv6.h>
  27 #include <net/netfilter/nf_conntrack.h>
  28 #include <net/netfilter/nf_conntrack_l4proto.h>
  29 #include <net/netfilter/nf_conntrack_ecache.h>
  30 #include <net/netfilter/nf_conntrack_seqadj.h>
  31 #include <net/netfilter/nf_conntrack_synproxy.h>
  32 #include <net/netfilter/nf_log.h>
  33 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
  34 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
  35
  36 /* "Be conservative in what you do,
  37     be liberal in what you accept from others."
  38     If it's non-zero, we mark only out of window RST segments as INVALID. */
  39 static int nf_ct_tcp_be_liberal __read_mostly = 0;
  40
  41 /* If it is set to zero, we disable picking up already established
  42    connections. */
  43 static int nf_ct_tcp_loose __read_mostly = 1;
  44
  45 /* Max number of the retransmitted packets without receiving an (acceptable)
  46    ACK from the destination. If this number is reached, a shorter timer
  47    will be started. */
  48 static int nf_ct_tcp_max_retrans __read_mostly = 3;
  49
  50   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  51      closely.  They're more complex. --RR */
  52
  53 static const char *const tcp_conntrack_names[] = {
  54         "NONE",
  55         "SYN_SENT",
  56         "SYN_RECV",
  57         "ESTABLISHED",
  58         "FIN_WAIT",
  59         "CLOSE_WAIT",
  60         "LAST_ACK",
  61         "TIME_WAIT",
  62         "CLOSE",
  63         "SYN_SENT2",
  64 };
  65
  66 #define SECS * HZ
  67 #define MINS * 60 SECS
  68 #define HOURS * 60 MINS
  69 #define DAYS * 24 HOURS
  70
  71 static unsigned int tcp_timeouts[TCP_CONNTRACK_TIMEOUT_MAX] __read_mostly = {
  72         [TCP_CONNTRACK_SYN_SENT]        = 2 MINS,
  73         [TCP_CONNTRACK_SYN_RECV]        = 60 SECS,
  74         [TCP_CONNTRACK_ESTABLISHED]     = 5 DAYS,
  75         [TCP_CONNTRACK_FIN_WAIT]        = 2 MINS,
  76         [TCP_CONNTRACK_CLOSE_WAIT]      = 60 SECS,
  77         [TCP_CONNTRACK_LAST_ACK]        = 30 SECS,
  78         [TCP_CONNTRACK_TIME_WAIT]       = 2 MINS,
  79         [TCP_CONNTRACK_CLOSE]           = 10 SECS,
  80         [TCP_CONNTRACK_SYN_SENT2]       = 2 MINS,
  81 /* RFC1122 says the R2 limit should be at least 100 seconds.
  82    Linux uses 15 packets as limit, which corresponds
  83    to ~13-30min depending on RTO. */
  84         [TCP_CONNTRACK_RETRANS]         = 5 MINS,
  85         [TCP_CONNTRACK_UNACK]           = 5 MINS,
  86 };
  87
  88 #define sNO TCP_CONNTRACK_NONE
  89 #define sSS TCP_CONNTRACK_SYN_SENT
  90 #define sSR TCP_CONNTRACK_SYN_RECV
  91 #define sES TCP_CONNTRACK_ESTABLISHED
  92 #define sFW TCP_CONNTRACK_FIN_WAIT
  93 #define sCW TCP_CONNTRACK_CLOSE_WAIT
  94 #define sLA TCP_CONNTRACK_LAST_ACK
  95 #define sTW TCP_CONNTRACK_TIME_WAIT
  96 #define sCL TCP_CONNTRACK_CLOSE
  97 #define sS2 TCP_CONNTRACK_SYN_SENT2
  98 #define sIV TCP_CONNTRACK_MAX
  99 #define sIG TCP_CONNTRACK_IGNORE
 100
 101 /* What TCP flags are set from RST/SYN/FIN/ACK. */
 102 enum tcp_bit_set {
 103         TCP_SYN_SET,
 104         TCP_SYNACK_SET,
 105         TCP_FIN_SET,
 106         TCP_ACK_SET,
 107         TCP_RST_SET,
 108         TCP_NONE_SET,
 109 };
 110
 111 /*
 112  * The TCP state transition table needs a few words...
 113  *
 114  * We are the man in the middle. All the packets go through us
 115  * but might get lost in transit to the destination.
 116  * It is assumed that the destinations can't receive segments
 117  * we haven't seen.
 118  *
 119  * The checked segment is in window, but our windows are *not*
 120  * equivalent with the ones of the sender/receiver. We always
 121  * try to guess the state of the current sender.
 122  *
 123  * The meaning of the states are:
 124  *
 125  * NONE:        initial state
 126  * SYN_SENT:    SYN-only packet seen
 127  * SYN_SENT2:   SYN-only packet seen from reply dir, simultaneous open
 128  * SYN_RECV:    SYN-ACK packet seen
 129  * ESTABLISHED: ACK packet seen
 130  * FIN_WAIT:    FIN packet seen
 131  * CLOSE_WAIT:  ACK seen (after FIN)
 132  * LAST_ACK:    FIN seen (after FIN)
 133  * TIME_WAIT:   last ACK seen
 134  * CLOSE:       closed connection (RST)
 135  *
 136  * Packets marked as IGNORED (sIG):
 137  *      if they may be either invalid or valid
 138  *      and the receiver may send back a connection
 139  *      closing RST or a SYN/ACK.
 140  *
 141  * Packets marked as INVALID (sIV):
 142  *      if we regard them as truly invalid packets
 143  */
 144 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 145         {
 146 /* ORIGINAL */
 147 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 148 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sS2 },
 149 /*
 150  *      sNO -> sSS      Initialize a new connection
 151  *      sSS -> sSS      Retransmitted SYN
 152  *      sS2 -> sS2      Late retransmitted SYN
 153  *      sSR -> sIG
 154  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 155  *                      are errors. Receiver will reply with RST
 156  *                      and close the connection.
 157  *                      Or we are not in sync and hold a dead connection.
 158  *      sFW -> sIG
 159  *      sCW -> sIG
 160  *      sLA -> sIG
 161  *      sTW -> sSS      Reopened connection (RFC 1122).
 162  *      sCL -> sSS
 163  */
 164 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 165 /*synack*/ { sIV, sIV, sSR, sIV, sIV, sIV, sIV, sIV, sIV, sSR },
 166 /*
 167  *      sNO -> sIV      Too late and no reason to do anything
 168  *      sSS -> sIV      Client can't send SYN and then SYN/ACK
 169  *      sS2 -> sSR      SYN/ACK sent to SYN2 in simultaneous open
 170  *      sSR -> sSR      Late retransmitted SYN/ACK in simultaneous open
 171  *      sES -> sIV      Invalid SYN/ACK packets sent by the client
 172  *      sFW -> sIV
 173  *      sCW -> sIV
 174  *      sLA -> sIV
 175  *      sTW -> sIV
 176  *      sCL -> sIV
 177  */
 178 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 179 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 180 /*
 181  *      sNO -> sIV      Too late and no reason to do anything...
 182  *      sSS -> sIV      Client migth not send FIN in this state:
 183  *                      we enforce waiting for a SYN/ACK reply first.
 184  *      sS2 -> sIV
 185  *      sSR -> sFW      Close started.
 186  *      sES -> sFW
 187  *      sFW -> sLA      FIN seen in both directions, waiting for
 188  *                      the last ACK.
 189  *                      Migth be a retransmitted FIN as well...
 190  *      sCW -> sLA
 191  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 192  *      sTW -> sTW
 193  *      sCL -> sCL
 194  */
 195 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 196 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 197 /*
 198  *      sNO -> sES      Assumed.
 199  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 200  *      sS2 -> sIV
 201  *      sSR -> sES      Established state is reached.
 202  *      sES -> sES      :-)
 203  *      sFW -> sCW      Normal close request answered by ACK.
 204  *      sCW -> sCW
 205  *      sLA -> sTW      Last ACK detected (RFC5961 challenged)
 206  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 207  *      sCL -> sCL
 208  */
 209 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 210 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
 211 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 212         },
 213         {
 214 /* REPLY */
 215 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 216 /*syn*/    { sIV, sS2, sIV, sIV, sIV, sIV, sIV, sSS, sIV, sS2 },
 217 /*
 218  *      sNO -> sIV      Never reached.
 219  *      sSS -> sS2      Simultaneous open
 220  *      sS2 -> sS2      Retransmitted simultaneous SYN
 221  *      sSR -> sIV      Invalid SYN packets sent by the server
 222  *      sES -> sIV
 223  *      sFW -> sIV
 224  *      sCW -> sIV
 225  *      sLA -> sIV
 226  *      sTW -> sSS      Reopened connection, but server may have switched role
 227  *      sCL -> sIV
 228  */
 229 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 230 /*synack*/ { sIV, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
 231 /*
 232  *      sSS -> sSR      Standard open.
 233  *      sS2 -> sSR      Simultaneous open
 234  *      sSR -> sIG      Retransmitted SYN/ACK, ignore it.
 235  *      sES -> sIG      Late retransmitted SYN/ACK?
 236  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 237  *      sCW -> sIG
 238  *      sLA -> sIG
 239  *      sTW -> sIG
 240  *      sCL -> sIG
 241  */
 242 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 243 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 244 /*
 245  *      sSS -> sIV      Server might not send FIN in this state.
 246  *      sS2 -> sIV
 247  *      sSR -> sFW      Close started.
 248  *      sES -> sFW
 249  *      sFW -> sLA      FIN seen in both directions.
 250  *      sCW -> sLA
 251  *      sLA -> sLA      Retransmitted FIN.
 252  *      sTW -> sTW
 253  *      sCL -> sCL
 254  */
 255 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 256 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIG },
 257 /*
 258  *      sSS -> sIG      Might be a half-open connection.
 259  *      sS2 -> sIG
 260  *      sSR -> sSR      Might answer late resent SYN.
 261  *      sES -> sES      :-)
 262  *      sFW -> sCW      Normal close request answered by ACK.
 263  *      sCW -> sCW
 264  *      sLA -> sTW      Last ACK detected (RFC5961 challenged)
 265  *      sTW -> sTW      Retransmitted last ACK.
 266  *      sCL -> sCL
 267  */
 268 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 269 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
 270 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 271         }
 272 };
 273
 274 static inline struct nf_tcp_net *tcp_pernet(struct net *net)
 275 {
 276         return &net->ct.nf_ct_proto.tcp;
 277 }
 278
 279 static bool tcp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
 280                              struct net *net, struct nf_conntrack_tuple *tuple)
 281 {
 282         const struct tcphdr *hp;
 283         struct tcphdr _hdr;
 284
 285         /* Actually only need first 4 bytes to get ports. */
 286         hp = skb_header_pointer(skb, dataoff, 4, &_hdr);
 287         if (hp == NULL)
 288                 return false;
 289
 290         tuple->src.u.tcp.port = hp->source;
 291         tuple->dst.u.tcp.port = hp->dest;
 292
 293         return true;
 294 }
 295
 296 static bool tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
 297                              const struct nf_conntrack_tuple *orig)
 298 {
 299         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
 300         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
 301         return true;
 302 }
 303
 304 /* Print out the per-protocol part of the tuple. */
 305 static void tcp_print_tuple(struct seq_file *s,
 306                             const struct nf_conntrack_tuple *tuple)
 307 {
 308         seq_printf(s, "sport=%hu dport=%hu ",
 309                    ntohs(tuple->src.u.tcp.port),
 310                    ntohs(tuple->dst.u.tcp.port));
 311 }
 312
 313 /* Print out the private part of the conntrack. */
 314 static void tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
 315 {
 316         seq_printf(s, "%s ", tcp_conntrack_names[ct->proto.tcp.state]);
 317 }
 318
 319 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 320 {
 321         if (tcph->rst) return TCP_RST_SET;
 322         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 323         else if (tcph->fin) return TCP_FIN_SET;
 324         else if (tcph->ack) return TCP_ACK_SET;
 325         else return TCP_NONE_SET;
 326 }
 327
 328 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 329    in IP Filter' by Guido van Rooij.
 330
 331    http://www.sane.nl/events/sane2000/papers.html
 332    http://www.darkart.com/mirrors/www.obfuscation.org/ipf/
 333
 334    The boundaries and the conditions are changed according to RFC793:
 335    the packet must intersect the window (i.e. segments may be
 336    after the right or before the left edge) and thus receivers may ACK
 337    segments after the right edge of the window.
 338
 339         td_maxend = max(sack + max(win,1)) seen in reply packets
 340         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 341         td_maxwin += seq + len - sender.td_maxend
 342                         if seq + len > sender.td_maxend
 343         td_end    = max(seq + len) seen in sent packets
 344
 345    I.   Upper bound for valid data:     seq <= sender.td_maxend
 346    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 347    III. Upper bound for valid (s)ack:   sack <= receiver.td_end
 348    IV.  Lower bound for valid (s)ack:   sack >= receiver.td_end - MAXACKWINDOW
 349
 350    where sack is the highest right edge of sack block found in the packet
 351    or ack in the case of packet without SACK option.
 352
 353    The upper bound limit for a valid (s)ack is not ignored -
 354    we doesn't have to deal with fragments.
 355 */
 356
 357 static inline __u32 segment_seq_plus_len(__u32 seq,
 358                                          size_t len,
 359                                          unsigned int dataoff,
 360                                          const struct tcphdr *tcph)
 361 {
 362         /* XXX Should I use payload length field in IP/IPv6 header ?
 363          * - YK */
 364         return (seq + len - dataoff - tcph->doff*4
 365                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 366 }
 367
 368 /* Fixme: what about big packets? */
 369 #define MAXACKWINCONST                  66000
 370 #define MAXACKWINDOW(sender)                                            \
 371         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 372                                               : MAXACKWINCONST)
 373
 374 /*
 375  * Simplified tcp_parse_options routine from tcp_input.c
 376  */
 377 static void tcp_options(const struct sk_buff *skb,
 378                         unsigned int dataoff,
 379                         const struct tcphdr *tcph,
 380                         struct ip_ct_tcp_state *state)
 381 {
 382         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 383         const unsigned char *ptr;
 384         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 385
 386         if (!length)
 387                 return;
 388
 389         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 390                                  length, buff);
 391         BUG_ON(ptr == NULL);
 392
 393         state->td_scale =
 394         state->flags = 0;
 395
 396         while (length > 0) {
 397                 int opcode=*ptr++;
 398                 int opsize;
 399
 400                 switch (opcode) {
 401                 case TCPOPT_EOL:
 402                         return;
 403                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 404                         length--;
 405                         continue;
 406                 default:
 407                         if (length < 2)
 408                                 return;
 409                         opsize=*ptr++;
 410                         if (opsize < 2) /* "silly options" */
 411                                 return;
 412                         if (opsize > length)
 413                                 return; /* don't parse partial options */
 414
 415                         if (opcode == TCPOPT_SACK_PERM
 416                             && opsize == TCPOLEN_SACK_PERM)
 417                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 418                         else if (opcode == TCPOPT_WINDOW
 419                                  && opsize == TCPOLEN_WINDOW) {
 420                                 state->td_scale = *(u_int8_t *)ptr;
 421
 422                                 if (state->td_scale > 14) {
 423                                         /* See RFC1323 */
 424                                         state->td_scale = 14;
 425                                 }
 426                                 state->flags |=
 427                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 428                         }
 429                         ptr += opsize - 2;
 430                         length -= opsize;
 431                 }
 432         }
 433 }
 434
 435 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 436                      const struct tcphdr *tcph, __u32 *sack)
 437 {
 438         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 439         const unsigned char *ptr;
 440         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 441         __u32 tmp;
 442
 443         if (!length)
 444                 return;
 445
 446         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 447                                  length, buff);
 448         BUG_ON(ptr == NULL);
 449
 450         /* Fast path for timestamp-only option */
 451         if (length == TCPOLEN_TSTAMP_ALIGNED
 452             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
 453                                        | (TCPOPT_NOP << 16)
 454                                        | (TCPOPT_TIMESTAMP << 8)
 455                                        | TCPOLEN_TIMESTAMP))
 456                 return;
 457
 458         while (length > 0) {
 459                 int opcode = *ptr++;
 460                 int opsize, i;
 461
 462                 switch (opcode) {
 463                 case TCPOPT_EOL:
 464                         return;
 465                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 466                         length--;
 467                         continue;
 468                 default:
 469                         if (length < 2)
 470                                 return;
 471                         opsize = *ptr++;
 472                         if (opsize < 2) /* "silly options" */
 473                                 return;
 474                         if (opsize > length)
 475                                 return; /* don't parse partial options */
 476
 477                         if (opcode == TCPOPT_SACK
 478                             && opsize >= (TCPOLEN_SACK_BASE
 479                                           + TCPOLEN_SACK_PERBLOCK)
 480                             && !((opsize - TCPOLEN_SACK_BASE)
 481                                  % TCPOLEN_SACK_PERBLOCK)) {
 482                                 for (i = 0;
 483                                      i < (opsize - TCPOLEN_SACK_BASE);
 484                                      i += TCPOLEN_SACK_PERBLOCK) {
 485                                         tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
 486
 487                                         if (after(tmp, *sack))
 488                                                 *sack = tmp;
 489                                 }
 490                                 return;
 491                         }
 492                         ptr += opsize - 2;
 493                         length -= opsize;
 494                 }
 495         }
 496 }
 497
 498 static bool tcp_in_window(const struct nf_conn *ct,
 499                           struct ip_ct_tcp *state,
 500                           enum ip_conntrack_dir dir,
 501                           unsigned int index,
 502                           const struct sk_buff *skb,
 503                           unsigned int dataoff,
 504                           const struct tcphdr *tcph,
 505                           u_int8_t pf)
 506 {
 507         struct net *net = nf_ct_net(ct);
 508         struct nf_tcp_net *tn = tcp_pernet(net);
 509         struct ip_ct_tcp_state *sender = &state->seen[dir];
 510         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 511         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 512         __u32 seq, ack, sack, end, win, swin;
 513         s32 receiver_offset;
 514         bool res, in_recv_win;
 515
 516         /*
 517          * Get the required data from the packet.
 518          */
 519         seq = ntohl(tcph->seq);
 520         ack = sack = ntohl(tcph->ack_seq);
 521         win = ntohs(tcph->window);
 522         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 523
 524         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 525                 tcp_sack(skb, dataoff, tcph, &sack);
 526
 527         /* Take into account NAT sequence number mangling */
 528         receiver_offset = nf_ct_seq_offset(ct, !dir, ack - 1);
 529         ack -= receiver_offset;
 530         sack -= receiver_offset;
 531
 532         pr_debug("tcp_in_window: START\n");
 533         pr_debug("tcp_in_window: ");
 534         nf_ct_dump_tuple(tuple);
 535         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
 536                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
 537         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 538                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 539                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 540                  sender->td_scale,
 541                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 542                  receiver->td_scale);
 543
 544         if (sender->td_maxwin == 0) {
 545                 /*
 546                  * Initialize sender data.
 547                  */
 548                 if (tcph->syn) {
 549                         /*
 550                          * SYN-ACK in reply to a SYN
 551                          * or SYN from reply direction in simultaneous open.
 552                          */
 553                         sender->td_end =
 554                         sender->td_maxend = end;
 555                         sender->td_maxwin = (win == 0 ? 1 : win);
 556
 557                         tcp_options(skb, dataoff, tcph, sender);
 558                         /*
 559                          * RFC 1323:
 560                          * Both sides must send the Window Scale option
 561                          * to enable window scaling in either direction.
 562                          */
 563                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 564                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 565                                 sender->td_scale =
 566                                 receiver->td_scale = 0;
 567                         if (!tcph->ack)
 568                                 /* Simultaneous open */
 569                                 return true;
 570                 } else {
 571                         /*
 572                          * We are in the middle of a connection,
 573                          * its history is lost for us.
 574                          * Let's try to use the data from the packet.
 575                          */
 576                         sender->td_end = end;
 577                         swin = win << sender->td_scale;
 578                         sender->td_maxwin = (swin == 0 ? 1 : swin);
 579                         sender->td_maxend = end + sender->td_maxwin;
 580                         /*
 581                          * We haven't seen traffic in the other direction yet
 582                          * but we have to tweak window tracking to pass III
 583                          * and IV until that happens.
 584                          */
 585                         if (receiver->td_maxwin == 0)
 586                                 receiver->td_end = receiver->td_maxend = sack;
 587                 }
 588         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 589                      && dir == IP_CT_DIR_ORIGINAL)
 590                    || (state->state == TCP_CONNTRACK_SYN_RECV
 591                      && dir == IP_CT_DIR_REPLY))
 592                    && after(end, sender->td_end)) {
 593                 /*
 594                  * RFC 793: "if a TCP is reinitialized ... then it need
 595                  * not wait at all; it must only be sure to use sequence
 596                  * numbers larger than those recently used."
 597                  */
 598                 sender->td_end =
 599                 sender->td_maxend = end;
 600                 sender->td_maxwin = (win == 0 ? 1 : win);
 601
 602                 tcp_options(skb, dataoff, tcph, sender);
 603         }
 604
 605         if (!(tcph->ack)) {
 606                 /*
 607                  * If there is no ACK, just pretend it was set and OK.
 608                  */
 609                 ack = sack = receiver->td_end;
 610         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 611                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 612                    && (ack == 0)) {
 613                 /*
 614                  * Broken TCP stacks, that set ACK in RST packets as well
 615                  * with zero ack value.
 616                  */
 617                 ack = sack = receiver->td_end;
 618         }
 619
 620         if (tcph->rst && seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)
 621                 /*
 622                  * RST sent answering SYN.
 623                  */
 624                 seq = end = sender->td_end;
 625
 626         pr_debug("tcp_in_window: ");
 627         nf_ct_dump_tuple(tuple);
 628         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
 629                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
 630         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 631                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 632                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 633                  sender->td_scale,
 634                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 635                  receiver->td_scale);
 636
 637         /* Is the ending sequence in the receive window (if available)? */
 638         in_recv_win = !receiver->td_maxwin ||
 639                       after(end, sender->td_end - receiver->td_maxwin - 1);
 640
 641         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 642                  before(seq, sender->td_maxend + 1),
 643                  (in_recv_win ? 1 : 0),
 644                  before(sack, receiver->td_end + 1),
 645                  after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 646
 647         if (before(seq, sender->td_maxend + 1) &&
 648             in_recv_win &&
 649             before(sack, receiver->td_end + 1) &&
 650             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
 651                 /*
 652                  * Take into account window scaling (RFC 1323).
 653                  */
 654                 if (!tcph->syn)
 655                         win <<= sender->td_scale;
 656
 657                 /*
 658                  * Update sender data.
 659                  */
 660                 swin = win + (sack - ack);
 661                 if (sender->td_maxwin < swin)
 662                         sender->td_maxwin = swin;
 663                 if (after(end, sender->td_end)) {
 664                         sender->td_end = end;
 665                         sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 666                 }
 667                 if (tcph->ack) {
 668                         if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
 669                                 sender->td_maxack = ack;
 670                                 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
 671                         } else if (after(ack, sender->td_maxack))
 672                                 sender->td_maxack = ack;
 673                 }
 674
 675                 /*
 676                  * Update receiver data.
 677                  */
 678                 if (receiver->td_maxwin != 0 && after(end, sender->td_maxend))
 679                         receiver->td_maxwin += end - sender->td_maxend;
 680                 if (after(sack + win, receiver->td_maxend - 1)) {
 681                         receiver->td_maxend = sack + win;
 682                         if (win == 0)
 683                                 receiver->td_maxend++;
 684                 }
 685                 if (ack == receiver->td_end)
 686                         receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 687
 688                 /*
 689                  * Check retransmissions.
 690                  */
 691                 if (index == TCP_ACK_SET) {
 692                         if (state->last_dir == dir
 693                             && state->last_seq == seq
 694                             && state->last_ack == ack
 695                             && state->last_end == end
 696                             && state->last_win == win)
 697                                 state->retrans++;
 698                         else {
 699                                 state->last_dir = dir;
 700                                 state->last_seq = seq;
 701                                 state->last_ack = ack;
 702                                 state->last_end = end;
 703                                 state->last_win = win;
 704                                 state->retrans = 0;
 705                         }
 706                 }
 707                 res = true;
 708         } else {
 709                 res = false;
 710                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 711                     tn->tcp_be_liberal)
 712                         res = true;
 713                 if (!res && LOG_INVALID(net, IPPROTO_TCP))
 714                         nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 715                         "nf_ct_tcp: %s ",
 716                         before(seq, sender->td_maxend + 1) ?
 717                         in_recv_win ?
 718                         before(sack, receiver->td_end + 1) ?
 719                         after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
 720                         : "ACK is under the lower bound (possible overly delayed ACK)"
 721                         : "ACK is over the upper bound (ACKed data not seen yet)"
 722                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 723                         : "SEQ is over the upper bound (over the window of the receiver)");
 724         }
 725
 726         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 727                  "receiver end=%u maxend=%u maxwin=%u\n",
 728                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 729                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 730
 731         return res;
 732 }
 733
 734 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 735 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
 736                                  TCPHDR_URG) + 1] =
 737 {
 738         [TCPHDR_SYN]                            = 1,
 739         [TCPHDR_SYN|TCPHDR_URG]                 = 1,
 740         [TCPHDR_SYN|TCPHDR_ACK]                 = 1,
 741         [TCPHDR_RST]                            = 1,
 742         [TCPHDR_RST|TCPHDR_ACK]                 = 1,
 743         [TCPHDR_FIN|TCPHDR_ACK]                 = 1,
 744         [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG]      = 1,
 745         [TCPHDR_ACK]                            = 1,
 746         [TCPHDR_ACK|TCPHDR_URG]                 = 1,
 747 };
 748
 749 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 750 static int tcp_error(struct net *net, struct nf_conn *tmpl,
 751                      struct sk_buff *skb,
 752                      unsigned int dataoff,
 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(net, IPPROTO_TCP))
 765                         nf_log_packet(net, 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(net, IPPROTO_TCP))
 773                         nf_log_packet(net, 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(net, IPPROTO_TCP))
 786                         nf_log_packet(net, 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 = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
 793         if (!tcp_valid_flags[tcpflags]) {
 794                 if (LOG_INVALID(net, IPPROTO_TCP))
 795                         nf_log_packet(net, 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 static unsigned int *tcp_get_timeouts(struct net *net)
 804 {
 805         return tcp_pernet(net)->timeouts;
 806 }
 807
 808 /* Returns verdict for packet, or -1 for invalid. */
 809 static int tcp_packet(struct nf_conn *ct,
 810                       const struct sk_buff *skb,
 811                       unsigned int dataoff,
 812                       enum ip_conntrack_info ctinfo,
 813                       u_int8_t pf,
 814                       unsigned int hooknum,
 815                       unsigned int *timeouts)
 816 {
 817         struct net *net = nf_ct_net(ct);
 818         struct nf_tcp_net *tn = tcp_pernet(net);
 819         struct nf_conntrack_tuple *tuple;
 820         enum tcp_conntrack new_state, old_state;
 821         enum ip_conntrack_dir dir;
 822         const struct tcphdr *th;
 823         struct tcphdr _tcph;
 824         unsigned long timeout;
 825         unsigned int index;
 826
 827         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 828         BUG_ON(th == NULL);
 829
 830         spin_lock_bh(&ct->lock);
 831         old_state = ct->proto.tcp.state;
 832         dir = CTINFO2DIR(ctinfo);
 833         index = get_conntrack_index(th);
 834         new_state = tcp_conntracks[dir][index][old_state];
 835         tuple = &ct->tuplehash[dir].tuple;
 836
 837         switch (new_state) {
 838         case TCP_CONNTRACK_SYN_SENT:
 839                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 840                         break;
 841                 /* RFC 1122: "When a connection is closed actively,
 842                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 843                  * (Maximum Segment Lifetime). However, it MAY accept
 844                  * a new SYN from the remote TCP to reopen the connection
 845                  * directly from TIME-WAIT state, if..."
 846                  * We ignore the conditions because we are in the
 847                  * TIME-WAIT state anyway.
 848                  *
 849                  * Handle aborted connections: we and the server
 850                  * think there is an existing connection but the client
 851                  * aborts it and starts a new one.
 852                  */
 853                 if (((ct->proto.tcp.seen[dir].flags
 854                       | ct->proto.tcp.seen[!dir].flags)
 855                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 856                     || (ct->proto.tcp.last_dir == dir
 857                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 858                         /* Attempt to reopen a closed/aborted connection.
 859                          * Delete this connection and look up again. */
 860                         spin_unlock_bh(&ct->lock);
 861
 862                         /* Only repeat if we can actually remove the timer.
 863                          * Destruction may already be in progress in process
 864                          * context and we must give it a chance to terminate.
 865                          */
 866                         if (nf_ct_kill(ct))
 867                                 return -NF_REPEAT;
 868                         return NF_DROP;
 869                 }
 870                 /* Fall through */
 871         case TCP_CONNTRACK_IGNORE:
 872                 /* Ignored packets:
 873                  *
 874                  * Our connection entry may be out of sync, so ignore
 875                  * packets which may signal the real connection between
 876                  * the client and the server.
 877                  *
 878                  * a) SYN in ORIGINAL
 879                  * b) SYN/ACK in REPLY
 880                  * c) ACK in reply direction after initial SYN in original.
 881                  *
 882                  * If the ignored packet is invalid, the receiver will send
 883                  * a RST we'll catch below.
 884                  */
 885                 if (index == TCP_SYNACK_SET
 886                     && ct->proto.tcp.last_index == TCP_SYN_SET
 887                     && ct->proto.tcp.last_dir != dir
 888                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 889                         /* b) This SYN/ACK acknowledges a SYN that we earlier
 890                          * ignored as invalid. This means that the client and
 891                          * the server are both in sync, while the firewall is
 892                          * not. We get in sync from the previously annotated
 893                          * values.
 894                          */
 895                         old_state = TCP_CONNTRACK_SYN_SENT;
 896                         new_state = TCP_CONNTRACK_SYN_RECV;
 897                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
 898                                 ct->proto.tcp.last_end;
 899                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
 900                                 ct->proto.tcp.last_end;
 901                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
 902                                 ct->proto.tcp.last_win == 0 ?
 903                                         1 : ct->proto.tcp.last_win;
 904                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
 905                                 ct->proto.tcp.last_wscale;
 906                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
 907                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
 908                                 ct->proto.tcp.last_flags;
 909                         memset(&ct->proto.tcp.seen[dir], 0,
 910                                sizeof(struct ip_ct_tcp_state));
 911                         break;
 912                 }
 913                 ct->proto.tcp.last_index = index;
 914                 ct->proto.tcp.last_dir = dir;
 915                 ct->proto.tcp.last_seq = ntohl(th->seq);
 916                 ct->proto.tcp.last_end =
 917                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 918                 ct->proto.tcp.last_win = ntohs(th->window);
 919
 920                 /* a) This is a SYN in ORIGINAL. The client and the server
 921                  * may be in sync but we are not. In that case, we annotate
 922                  * the TCP options and let the packet go through. If it is a
 923                  * valid SYN packet, the server will reply with a SYN/ACK, and
 924                  * then we'll get in sync. Otherwise, the server potentially
 925                  * responds with a challenge ACK if implementing RFC5961.
 926                  */
 927                 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
 928                         struct ip_ct_tcp_state seen = {};
 929
 930                         ct->proto.tcp.last_flags =
 931                         ct->proto.tcp.last_wscale = 0;
 932                         tcp_options(skb, dataoff, th, &seen);
 933                         if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
 934                                 ct->proto.tcp.last_flags |=
 935                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 936                                 ct->proto.tcp.last_wscale = seen.td_scale;
 937                         }
 938                         if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
 939                                 ct->proto.tcp.last_flags |=
 940                                         IP_CT_TCP_FLAG_SACK_PERM;
 941                         }
 942                         /* Mark the potential for RFC5961 challenge ACK,
 943                          * this pose a special problem for LAST_ACK state
 944                          * as ACK is intrepretated as ACKing last FIN.
 945                          */
 946                         if (old_state == TCP_CONNTRACK_LAST_ACK)
 947                                 ct->proto.tcp.last_flags |=
 948                                         IP_CT_EXP_CHALLENGE_ACK;
 949                 }
 950                 spin_unlock_bh(&ct->lock);
 951                 if (LOG_INVALID(net, IPPROTO_TCP))
 952                         nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 953                                   "nf_ct_tcp: invalid packet ignored in "
 954                                   "state %s ", tcp_conntrack_names[old_state]);
 955                 return NF_ACCEPT;
 956         case TCP_CONNTRACK_MAX:
 957                 /* Special case for SYN proxy: when the SYN to the server or
 958                  * the SYN/ACK from the server is lost, the client may transmit
 959                  * a keep-alive packet while in SYN_SENT state. This needs to
 960                  * be associated with the original conntrack entry in order to
 961                  * generate a new SYN with the correct sequence number.
 962                  */
 963                 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
 964                     index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
 965                     ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
 966                     ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
 967                         pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
 968                         spin_unlock_bh(&ct->lock);
 969                         return NF_ACCEPT;
 970                 }
 971
 972                 /* Invalid packet */
 973                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 974                          dir, get_conntrack_index(th), old_state);
 975                 spin_unlock_bh(&ct->lock);
 976                 if (LOG_INVALID(net, IPPROTO_TCP))
 977                         nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 978                                   "nf_ct_tcp: invalid state ");
 979                 return -NF_ACCEPT;
 980         case TCP_CONNTRACK_TIME_WAIT:
 981                 /* RFC5961 compliance cause stack to send "challenge-ACK"
 982                  * e.g. in response to spurious SYNs.  Conntrack MUST
 983                  * not believe this ACK is acking last FIN.
 984                  */
 985                 if (old_state == TCP_CONNTRACK_LAST_ACK &&
 986                     index == TCP_ACK_SET &&
 987                     ct->proto.tcp.last_dir != dir &&
 988                     ct->proto.tcp.last_index == TCP_SYN_SET &&
 989                     (ct->proto.tcp.last_flags & IP_CT_EXP_CHALLENGE_ACK)) {
 990                         /* Detected RFC5961 challenge ACK */
 991                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
 992                         spin_unlock_bh(&ct->lock);
 993                         if (LOG_INVALID(net, IPPROTO_TCP))
 994                                 nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 995                                       "nf_ct_tcp: challenge-ACK ignored ");
 996                         return NF_ACCEPT; /* Don't change state */
 997                 }
 998                 break;
 999         case TCP_CONNTRACK_CLOSE:
1000                 if (index == TCP_RST_SET
1001                     && (ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET)
1002                     && before(ntohl(th->seq), ct->proto.tcp.seen[!dir].td_maxack)) {
1003                         /* Invalid RST  */
1004                         spin_unlock_bh(&ct->lock);
1005                         if (LOG_INVALID(net, IPPROTO_TCP))
1006                                 nf_log_packet(net, pf, 0, skb, NULL, NULL,
1007                                               NULL, "nf_ct_tcp: invalid RST ");
1008                         return -NF_ACCEPT;
1009                 }
1010                 if (index == TCP_RST_SET
1011                     && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
1012                          && ct->proto.tcp.last_index == TCP_SYN_SET)
1013                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
1014                             && ct->proto.tcp.last_index == TCP_ACK_SET))
1015                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
1016                         /* RST sent to invalid SYN or ACK we had let through
1017                          * at a) and c) above:
1018                          *
1019                          * a) SYN was in window then
1020                          * c) we hold a half-open connection.
1021                          *
1022                          * Delete our connection entry.
1023                          * We skip window checking, because packet might ACK
1024                          * segments we ignored. */
1025                         goto in_window;
1026                 }
1027                 /* Just fall through */
1028         default:
1029                 /* Keep compilers happy. */
1030                 break;
1031         }
1032
1033         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
1034                            skb, dataoff, th, pf)) {
1035                 spin_unlock_bh(&ct->lock);
1036                 return -NF_ACCEPT;
1037         }
1038      in_window:
1039         /* From now on we have got in-window packets */
1040         ct->proto.tcp.last_index = index;
1041         ct->proto.tcp.last_dir = dir;
1042
1043         pr_debug("tcp_conntracks: ");
1044         nf_ct_dump_tuple(tuple);
1045         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1046                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1047                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1048                  old_state, new_state);
1049
1050         ct->proto.tcp.state = new_state;
1051         if (old_state != new_state
1052             && new_state == TCP_CONNTRACK_FIN_WAIT)
1053                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1054
1055         if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1056             timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1057                 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1058         else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1059                  IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1060                  timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1061                 timeout = timeouts[TCP_CONNTRACK_UNACK];
1062         else
1063                 timeout = timeouts[new_state];
1064         spin_unlock_bh(&ct->lock);
1065
1066         if (new_state != old_state)
1067                 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1068
1069         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1070                 /* If only reply is a RST, we can consider ourselves not to
1071                    have an established connection: this is a fairly common
1072                    problem case, so we can delete the conntrack
1073                    immediately.  --RR */
1074                 if (th->rst) {
1075                         nf_ct_kill_acct(ct, ctinfo, skb);
1076                         return NF_ACCEPT;
1077                 }
1078                 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1079                  * pickup with loose=1. Avoid large ESTABLISHED timeout.
1080                  */
1081                 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1082                     timeout > timeouts[TCP_CONNTRACK_UNACK])
1083                         timeout = timeouts[TCP_CONNTRACK_UNACK];
1084         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1085                    && (old_state == TCP_CONNTRACK_SYN_RECV
1086                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1087                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1088                 /* Set ASSURED if we see see valid ack in ESTABLISHED
1089                    after SYN_RECV or a valid answer for a picked up
1090                    connection. */
1091                 set_bit(IPS_ASSURED_BIT, &ct->status);
1092                 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1093         }
1094         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1095
1096         return NF_ACCEPT;
1097 }
1098
1099 /* Called when a new connection for this protocol found. */
1100 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
1101                     unsigned int dataoff, unsigned int *timeouts)
1102 {
1103         enum tcp_conntrack new_state;
1104         const struct tcphdr *th;
1105         struct tcphdr _tcph;
1106         struct net *net = nf_ct_net(ct);
1107         struct nf_tcp_net *tn = tcp_pernet(net);
1108         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
1109         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
1110
1111         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1112         BUG_ON(th == NULL);
1113
1114         /* Don't need lock here: this conntrack not in circulation yet */
1115         new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
1116
1117         /* Invalid: delete conntrack */
1118         if (new_state >= TCP_CONNTRACK_MAX) {
1119                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1120                 return false;
1121         }
1122
1123         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1124                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
1125                 /* SYN packet */
1126                 ct->proto.tcp.seen[0].td_end =
1127                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1128                                              dataoff, th);
1129                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1130                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1131                         ct->proto.tcp.seen[0].td_maxwin = 1;
1132                 ct->proto.tcp.seen[0].td_maxend =
1133                         ct->proto.tcp.seen[0].td_end;
1134
1135                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1136         } else if (tn->tcp_loose == 0) {
1137                 /* Don't try to pick up connections. */
1138                 return false;
1139         } else {
1140                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
1141                 /*
1142                  * We are in the middle of a connection,
1143                  * its history is lost for us.
1144                  * Let's try to use the data from the packet.
1145                  */
1146                 ct->proto.tcp.seen[0].td_end =
1147                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1148                                              dataoff, th);
1149                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1150                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1151                         ct->proto.tcp.seen[0].td_maxwin = 1;
1152                 ct->proto.tcp.seen[0].td_maxend =
1153                         ct->proto.tcp.seen[0].td_end +
1154                         ct->proto.tcp.seen[0].td_maxwin;
1155
1156                 /* We assume SACK and liberal window checking to handle
1157                  * window scaling */
1158                 ct->proto.tcp.seen[0].flags =
1159                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1160                                               IP_CT_TCP_FLAG_BE_LIBERAL;
1161         }
1162
1163         /* tcp_packet will set them */
1164         ct->proto.tcp.last_index = TCP_NONE_SET;
1165
1166         pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1167                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1168                  sender->td_end, sender->td_maxend, sender->td_maxwin,
1169                  sender->td_scale,
1170                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1171                  receiver->td_scale);
1172         return true;
1173 }
1174
1175 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1176
1177 #include <linux/netfilter/nfnetlink.h>
1178 #include <linux/netfilter/nfnetlink_conntrack.h>
1179
1180 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1181                          struct nf_conn *ct)
1182 {
1183         struct nlattr *nest_parms;
1184         struct nf_ct_tcp_flags tmp = {};
1185
1186         spin_lock_bh(&ct->lock);
1187         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1188         if (!nest_parms)
1189                 goto nla_put_failure;
1190
1191         if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state) ||
1192             nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1193                        ct->proto.tcp.seen[0].td_scale) ||
1194             nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1195                        ct->proto.tcp.seen[1].td_scale))
1196                 goto nla_put_failure;
1197
1198         tmp.flags = ct->proto.tcp.seen[0].flags;
1199         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1200                     sizeof(struct nf_ct_tcp_flags), &tmp))
1201                 goto nla_put_failure;
1202
1203         tmp.flags = ct->proto.tcp.seen[1].flags;
1204         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1205                     sizeof(struct nf_ct_tcp_flags), &tmp))
1206                 goto nla_put_failure;
1207         spin_unlock_bh(&ct->lock);
1208
1209         nla_nest_end(skb, nest_parms);
1210
1211         return 0;
1212
1213 nla_put_failure:
1214         spin_unlock_bh(&ct->lock);
1215         return -1;
1216 }
1217
1218 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1219         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1220         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1221         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1222         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1223         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len =  sizeof(struct nf_ct_tcp_flags) },
1224 };
1225
1226 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1227 {
1228         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1229         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1230         int err;
1231
1232         /* updates could not contain anything about the private
1233          * protocol info, in that case skip the parsing */
1234         if (!pattr)
1235                 return 0;
1236
1237         err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr, tcp_nla_policy);
1238         if (err < 0)
1239                 return err;
1240
1241         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1242             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1243                 return -EINVAL;
1244
1245         spin_lock_bh(&ct->lock);
1246         if (tb[CTA_PROTOINFO_TCP_STATE])
1247                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1248
1249         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1250                 struct nf_ct_tcp_flags *attr =
1251                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1252                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1253                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1254         }
1255
1256         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1257                 struct nf_ct_tcp_flags *attr =
1258                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1259                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1260                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1261         }
1262
1263         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1264             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1265             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1266             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1267                 ct->proto.tcp.seen[0].td_scale =
1268                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1269                 ct->proto.tcp.seen[1].td_scale =
1270                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1271         }
1272         spin_unlock_bh(&ct->lock);
1273
1274         return 0;
1275 }
1276
1277 static int tcp_nlattr_size(void)
1278 {
1279         return nla_total_size(0)           /* CTA_PROTOINFO_TCP */
1280                 + nla_policy_len(tcp_nla_policy, CTA_PROTOINFO_TCP_MAX + 1);
1281 }
1282
1283 static int tcp_nlattr_tuple_size(void)
1284 {
1285         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1286 }
1287 #endif
1288
1289 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1290
1291 #include <linux/netfilter/nfnetlink.h>
1292 #include <linux/netfilter/nfnetlink_cttimeout.h>
1293
1294 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1295                                      struct net *net, void *data)
1296 {
1297         unsigned int *timeouts = data;
1298         struct nf_tcp_net *tn = tcp_pernet(net);
1299         int i;
1300
1301         /* set default TCP timeouts. */
1302         for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1303                 timeouts[i] = tn->timeouts[i];
1304
1305         if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1306                 timeouts[TCP_CONNTRACK_SYN_SENT] =
1307                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1308         }
1309         if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1310                 timeouts[TCP_CONNTRACK_SYN_RECV] =
1311                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1312         }
1313         if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1314                 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1315                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1316         }
1317         if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1318                 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1319                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1320         }
1321         if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1322                 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1323                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1324         }
1325         if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1326                 timeouts[TCP_CONNTRACK_LAST_ACK] =
1327                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1328         }
1329         if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1330                 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1331                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1332         }
1333         if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1334                 timeouts[TCP_CONNTRACK_CLOSE] =
1335                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1336         }
1337         if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1338                 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1339                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1340         }
1341         if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1342                 timeouts[TCP_CONNTRACK_RETRANS] =
1343                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1344         }
1345         if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1346                 timeouts[TCP_CONNTRACK_UNACK] =
1347                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1348         }
1349         return 0;
1350 }
1351
1352 static int
1353 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1354 {
1355         const unsigned int *timeouts = data;
1356
1357         if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1358                         htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1359             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1360                          htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1361             nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1362                          htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1363             nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1364                          htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1365             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1366                          htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1367             nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1368                          htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1369             nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1370                          htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1371             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1372                          htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1373             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1374                          htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1375             nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1376                          htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1377             nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1378                          htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1379                 goto nla_put_failure;
1380         return 0;
1381
1382 nla_put_failure:
1383         return -ENOSPC;
1384 }
1385
1386 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1387         [CTA_TIMEOUT_TCP_SYN_SENT]      = { .type = NLA_U32 },
1388         [CTA_TIMEOUT_TCP_SYN_RECV]      = { .type = NLA_U32 },
1389         [CTA_TIMEOUT_TCP_ESTABLISHED]   = { .type = NLA_U32 },
1390         [CTA_TIMEOUT_TCP_FIN_WAIT]      = { .type = NLA_U32 },
1391         [CTA_TIMEOUT_TCP_CLOSE_WAIT]    = { .type = NLA_U32 },
1392         [CTA_TIMEOUT_TCP_LAST_ACK]      = { .type = NLA_U32 },
1393         [CTA_TIMEOUT_TCP_TIME_WAIT]     = { .type = NLA_U32 },
1394         [CTA_TIMEOUT_TCP_CLOSE]         = { .type = NLA_U32 },
1395         [CTA_TIMEOUT_TCP_SYN_SENT2]     = { .type = NLA_U32 },
1396         [CTA_TIMEOUT_TCP_RETRANS]       = { .type = NLA_U32 },
1397         [CTA_TIMEOUT_TCP_UNACK]         = { .type = NLA_U32 },
1398 };
1399 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1400
1401 #ifdef CONFIG_SYSCTL
1402 static struct ctl_table tcp_sysctl_table[] = {
1403         {
1404                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1405                 .maxlen         = sizeof(unsigned int),
1406                 .mode           = 0644,
1407                 .proc_handler   = proc_dointvec_jiffies,
1408         },
1409         {
1410                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1411                 .maxlen         = sizeof(unsigned int),
1412                 .mode           = 0644,
1413                 .proc_handler   = proc_dointvec_jiffies,
1414         },
1415         {
1416                 .procname       = "nf_conntrack_tcp_timeout_established",
1417                 .maxlen         = sizeof(unsigned int),
1418                 .mode           = 0644,
1419                 .proc_handler   = proc_dointvec_jiffies,
1420         },
1421         {
1422                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1423                 .maxlen         = sizeof(unsigned int),
1424                 .mode           = 0644,
1425                 .proc_handler   = proc_dointvec_jiffies,
1426         },
1427         {
1428                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1429                 .maxlen         = sizeof(unsigned int),
1430                 .mode           = 0644,
1431                 .proc_handler   = proc_dointvec_jiffies,
1432         },
1433         {
1434                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1435                 .maxlen         = sizeof(unsigned int),
1436                 .mode           = 0644,
1437                 .proc_handler   = proc_dointvec_jiffies,
1438         },
1439         {
1440                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1441                 .maxlen         = sizeof(unsigned int),
1442                 .mode           = 0644,
1443                 .proc_handler   = proc_dointvec_jiffies,
1444         },
1445         {
1446                 .procname       = "nf_conntrack_tcp_timeout_close",
1447                 .maxlen         = sizeof(unsigned int),
1448                 .mode           = 0644,
1449                 .proc_handler   = proc_dointvec_jiffies,
1450         },
1451         {
1452                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1453                 .maxlen         = sizeof(unsigned int),
1454                 .mode           = 0644,
1455                 .proc_handler   = proc_dointvec_jiffies,
1456         },
1457         {
1458                 .procname       = "nf_conntrack_tcp_timeout_unacknowledged",
1459                 .maxlen         = sizeof(unsigned int),
1460                 .mode           = 0644,
1461                 .proc_handler   = proc_dointvec_jiffies,
1462         },
1463         {
1464                 .procname       = "nf_conntrack_tcp_loose",
1465                 .maxlen         = sizeof(unsigned int),
1466                 .mode           = 0644,
1467                 .proc_handler   = proc_dointvec,
1468         },
1469         {
1470                 .procname       = "nf_conntrack_tcp_be_liberal",
1471                 .maxlen         = sizeof(unsigned int),
1472                 .mode           = 0644,
1473                 .proc_handler   = proc_dointvec,
1474         },
1475         {
1476                 .procname       = "nf_conntrack_tcp_max_retrans",
1477                 .maxlen         = sizeof(unsigned int),
1478                 .mode           = 0644,
1479                 .proc_handler   = proc_dointvec,
1480         },
1481         { }
1482 };
1483 #endif /* CONFIG_SYSCTL */
1484
1485 static int tcp_kmemdup_sysctl_table(struct nf_proto_net *pn,
1486                                     struct nf_tcp_net *tn)
1487 {
1488 #ifdef CONFIG_SYSCTL
1489         if (pn->ctl_table)
1490                 return 0;
1491
1492         pn->ctl_table = kmemdup(tcp_sysctl_table,
1493                                 sizeof(tcp_sysctl_table),
1494                                 GFP_KERNEL);
1495         if (!pn->ctl_table)
1496                 return -ENOMEM;
1497
1498         pn->ctl_table[0].data = &tn->timeouts[TCP_CONNTRACK_SYN_SENT];
1499         pn->ctl_table[1].data = &tn->timeouts[TCP_CONNTRACK_SYN_RECV];
1500         pn->ctl_table[2].data = &tn->timeouts[TCP_CONNTRACK_ESTABLISHED];
1501         pn->ctl_table[3].data = &tn->timeouts[TCP_CONNTRACK_FIN_WAIT];
1502         pn->ctl_table[4].data = &tn->timeouts[TCP_CONNTRACK_CLOSE_WAIT];
1503         pn->ctl_table[5].data = &tn->timeouts[TCP_CONNTRACK_LAST_ACK];
1504         pn->ctl_table[6].data = &tn->timeouts[TCP_CONNTRACK_TIME_WAIT];
1505         pn->ctl_table[7].data = &tn->timeouts[TCP_CONNTRACK_CLOSE];
1506         pn->ctl_table[8].data = &tn->timeouts[TCP_CONNTRACK_RETRANS];
1507         pn->ctl_table[9].data = &tn->timeouts[TCP_CONNTRACK_UNACK];
1508         pn->ctl_table[10].data = &tn->tcp_loose;
1509         pn->ctl_table[11].data = &tn->tcp_be_liberal;
1510         pn->ctl_table[12].data = &tn->tcp_max_retrans;
1511 #endif
1512         return 0;
1513 }
1514
1515 static int tcp_init_net(struct net *net, u_int16_t proto)
1516 {
1517         struct nf_tcp_net *tn = tcp_pernet(net);
1518         struct nf_proto_net *pn = &tn->pn;
1519
1520         if (!pn->users) {
1521                 int i;
1522
1523                 for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1524                         tn->timeouts[i] = tcp_timeouts[i];
1525
1526                 tn->tcp_loose = nf_ct_tcp_loose;
1527                 tn->tcp_be_liberal = nf_ct_tcp_be_liberal;
1528                 tn->tcp_max_retrans = nf_ct_tcp_max_retrans;
1529         }
1530
1531         return tcp_kmemdup_sysctl_table(pn, tn);
1532 }
1533
1534 static struct nf_proto_net *tcp_get_net_proto(struct net *net)
1535 {
1536         return &net->ct.nf_ct_proto.tcp.pn;
1537 }
1538
1539 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1540 {
1541         .l3proto                = PF_INET,
1542         .l4proto                = IPPROTO_TCP,
1543         .name                   = "tcp",
1544         .pkt_to_tuple           = tcp_pkt_to_tuple,
1545         .invert_tuple           = tcp_invert_tuple,
1546         .print_tuple            = tcp_print_tuple,
1547         .print_conntrack        = tcp_print_conntrack,
1548         .packet                 = tcp_packet,
1549         .get_timeouts           = tcp_get_timeouts,
1550         .new                    = tcp_new,
1551         .error                  = tcp_error,
1552 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1553         .to_nlattr              = tcp_to_nlattr,
1554         .nlattr_size            = tcp_nlattr_size,
1555         .from_nlattr            = nlattr_to_tcp,
1556         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1557         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1558         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1559         .nla_policy             = nf_ct_port_nla_policy,
1560 #endif
1561 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1562         .ctnl_timeout           = {
1563                 .nlattr_to_obj  = tcp_timeout_nlattr_to_obj,
1564                 .obj_to_nlattr  = tcp_timeout_obj_to_nlattr,
1565                 .nlattr_max     = CTA_TIMEOUT_TCP_MAX,
1566                 .obj_size       = sizeof(unsigned int) *
1567                                         TCP_CONNTRACK_TIMEOUT_MAX,
1568                 .nla_policy     = tcp_timeout_nla_policy,
1569         },
1570 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1571         .init_net               = tcp_init_net,
1572         .get_net_proto          = tcp_get_net_proto,
1573 };
1574 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1575
1576 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1577 {
1578         .l3proto                = PF_INET6,
1579         .l4proto                = IPPROTO_TCP,
1580         .name                   = "tcp",
1581         .pkt_to_tuple           = tcp_pkt_to_tuple,
1582         .invert_tuple           = tcp_invert_tuple,
1583         .print_tuple            = tcp_print_tuple,
1584         .print_conntrack        = tcp_print_conntrack,
1585         .packet                 = tcp_packet,
1586         .get_timeouts           = tcp_get_timeouts,
1587         .new                    = tcp_new,
1588         .error                  = tcp_error,
1589 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1590         .to_nlattr              = tcp_to_nlattr,
1591         .nlattr_size            = tcp_nlattr_size,
1592         .from_nlattr            = nlattr_to_tcp,
1593         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1594         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1595         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1596         .nla_policy             = nf_ct_port_nla_policy,
1597 #endif
1598 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1599         .ctnl_timeout           = {
1600                 .nlattr_to_obj  = tcp_timeout_nlattr_to_obj,
1601                 .obj_to_nlattr  = tcp_timeout_obj_to_nlattr,
1602                 .nlattr_max     = CTA_TIMEOUT_TCP_MAX,
1603                 .obj_size       = sizeof(unsigned int) *
1604                                         TCP_CONNTRACK_TIMEOUT_MAX,
1605                 .nla_policy     = tcp_timeout_nla_policy,
1606         },
1607 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1608         .init_net               = tcp_init_net,
1609         .get_net_proto          = tcp_get_net_proto,
1610 };
1611 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);