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 #ifdef CONFIG_NF_CONNTRACK_PROCFS
 305 /* Print out the private part of the conntrack. */
 306 static void tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
 307 {
 308         seq_printf(s, "%s ", tcp_conntrack_names[ct->proto.tcp.state]);
 309 }
 310 #endif
 311
 312 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 313 {
 314         if (tcph->rst) return TCP_RST_SET;
 315         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 316         else if (tcph->fin) return TCP_FIN_SET;
 317         else if (tcph->ack) return TCP_ACK_SET;
 318         else return TCP_NONE_SET;
 319 }
 320
 321 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 322    in IP Filter' by Guido van Rooij.
 323
 324    http://www.sane.nl/events/sane2000/papers.html
 325    http://www.darkart.com/mirrors/www.obfuscation.org/ipf/
 326
 327    The boundaries and the conditions are changed according to RFC793:
 328    the packet must intersect the window (i.e. segments may be
 329    after the right or before the left edge) and thus receivers may ACK
 330    segments after the right edge of the window.
 331
 332         td_maxend = max(sack + max(win,1)) seen in reply packets
 333         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 334         td_maxwin += seq + len - sender.td_maxend
 335                         if seq + len > sender.td_maxend
 336         td_end    = max(seq + len) seen in sent packets
 337
 338    I.   Upper bound for valid data:     seq <= sender.td_maxend
 339    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 340    III. Upper bound for valid (s)ack:   sack <= receiver.td_end
 341    IV.  Lower bound for valid (s)ack:   sack >= receiver.td_end - MAXACKWINDOW
 342
 343    where sack is the highest right edge of sack block found in the packet
 344    or ack in the case of packet without SACK option.
 345
 346    The upper bound limit for a valid (s)ack is not ignored -
 347    we doesn't have to deal with fragments.
 348 */
 349
 350 static inline __u32 segment_seq_plus_len(__u32 seq,
 351                                          size_t len,
 352                                          unsigned int dataoff,
 353                                          const struct tcphdr *tcph)
 354 {
 355         /* XXX Should I use payload length field in IP/IPv6 header ?
 356          * - YK */
 357         return (seq + len - dataoff - tcph->doff*4
 358                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 359 }
 360
 361 /* Fixme: what about big packets? */
 362 #define MAXACKWINCONST                  66000
 363 #define MAXACKWINDOW(sender)                                            \
 364         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 365                                               : MAXACKWINCONST)
 366
 367 /*
 368  * Simplified tcp_parse_options routine from tcp_input.c
 369  */
 370 static void tcp_options(const struct sk_buff *skb,
 371                         unsigned int dataoff,
 372                         const struct tcphdr *tcph,
 373                         struct ip_ct_tcp_state *state)
 374 {
 375         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 376         const unsigned char *ptr;
 377         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 378
 379         if (!length)
 380                 return;
 381
 382         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 383                                  length, buff);
 384         BUG_ON(ptr == NULL);
 385
 386         state->td_scale =
 387         state->flags = 0;
 388
 389         while (length > 0) {
 390                 int opcode=*ptr++;
 391                 int opsize;
 392
 393                 switch (opcode) {
 394                 case TCPOPT_EOL:
 395                         return;
 396                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 397                         length--;
 398                         continue;
 399                 default:
 400                         if (length < 2)
 401                                 return;
 402                         opsize=*ptr++;
 403                         if (opsize < 2) /* "silly options" */
 404                                 return;
 405                         if (opsize > length)
 406                                 return; /* don't parse partial options */
 407
 408                         if (opcode == TCPOPT_SACK_PERM
 409                             && opsize == TCPOLEN_SACK_PERM)
 410                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 411                         else if (opcode == TCPOPT_WINDOW
 412                                  && opsize == TCPOLEN_WINDOW) {
 413                                 state->td_scale = *(u_int8_t *)ptr;
 414
 415                                 if (state->td_scale > TCP_MAX_WSCALE)
 416                                         state->td_scale = TCP_MAX_WSCALE;
 417
 418                                 state->flags |=
 419                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 420                         }
 421                         ptr += opsize - 2;
 422                         length -= opsize;
 423                 }
 424         }
 425 }
 426
 427 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 428                      const struct tcphdr *tcph, __u32 *sack)
 429 {
 430         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 431         const unsigned char *ptr;
 432         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 433         __u32 tmp;
 434
 435         if (!length)
 436                 return;
 437
 438         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 439                                  length, buff);
 440         BUG_ON(ptr == NULL);
 441
 442         /* Fast path for timestamp-only option */
 443         if (length == TCPOLEN_TSTAMP_ALIGNED
 444             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
 445                                        | (TCPOPT_NOP << 16)
 446                                        | (TCPOPT_TIMESTAMP << 8)
 447                                        | TCPOLEN_TIMESTAMP))
 448                 return;
 449
 450         while (length > 0) {
 451                 int opcode = *ptr++;
 452                 int opsize, i;
 453
 454                 switch (opcode) {
 455                 case TCPOPT_EOL:
 456                         return;
 457                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 458                         length--;
 459                         continue;
 460                 default:
 461                         if (length < 2)
 462                                 return;
 463                         opsize = *ptr++;
 464                         if (opsize < 2) /* "silly options" */
 465                                 return;
 466                         if (opsize > length)
 467                                 return; /* don't parse partial options */
 468
 469                         if (opcode == TCPOPT_SACK
 470                             && opsize >= (TCPOLEN_SACK_BASE
 471                                           + TCPOLEN_SACK_PERBLOCK)
 472                             && !((opsize - TCPOLEN_SACK_BASE)
 473                                  % TCPOLEN_SACK_PERBLOCK)) {
 474                                 for (i = 0;
 475                                      i < (opsize - TCPOLEN_SACK_BASE);
 476                                      i += TCPOLEN_SACK_PERBLOCK) {
 477                                         tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
 478
 479                                         if (after(tmp, *sack))
 480                                                 *sack = tmp;
 481                                 }
 482                                 return;
 483                         }
 484                         ptr += opsize - 2;
 485                         length -= opsize;
 486                 }
 487         }
 488 }
 489
 490 static bool tcp_in_window(const struct nf_conn *ct,
 491                           struct ip_ct_tcp *state,
 492                           enum ip_conntrack_dir dir,
 493                           unsigned int index,
 494                           const struct sk_buff *skb,
 495                           unsigned int dataoff,
 496                           const struct tcphdr *tcph,
 497                           u_int8_t pf)
 498 {
 499         struct net *net = nf_ct_net(ct);
 500         struct nf_tcp_net *tn = tcp_pernet(net);
 501         struct ip_ct_tcp_state *sender = &state->seen[dir];
 502         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 503         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 504         __u32 seq, ack, sack, end, win, swin;
 505         s32 receiver_offset;
 506         bool res, in_recv_win;
 507
 508         /*
 509          * Get the required data from the packet.
 510          */
 511         seq = ntohl(tcph->seq);
 512         ack = sack = ntohl(tcph->ack_seq);
 513         win = ntohs(tcph->window);
 514         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 515
 516         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 517                 tcp_sack(skb, dataoff, tcph, &sack);
 518
 519         /* Take into account NAT sequence number mangling */
 520         receiver_offset = nf_ct_seq_offset(ct, !dir, ack - 1);
 521         ack -= receiver_offset;
 522         sack -= receiver_offset;
 523
 524         pr_debug("tcp_in_window: START\n");
 525         pr_debug("tcp_in_window: ");
 526         nf_ct_dump_tuple(tuple);
 527         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
 528                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
 529         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 530                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 531                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 532                  sender->td_scale,
 533                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 534                  receiver->td_scale);
 535
 536         if (sender->td_maxwin == 0) {
 537                 /*
 538                  * Initialize sender data.
 539                  */
 540                 if (tcph->syn) {
 541                         /*
 542                          * SYN-ACK in reply to a SYN
 543                          * or SYN from reply direction in simultaneous open.
 544                          */
 545                         sender->td_end =
 546                         sender->td_maxend = end;
 547                         sender->td_maxwin = (win == 0 ? 1 : win);
 548
 549                         tcp_options(skb, dataoff, tcph, sender);
 550                         /*
 551                          * RFC 1323:
 552                          * Both sides must send the Window Scale option
 553                          * to enable window scaling in either direction.
 554                          */
 555                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 556                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 557                                 sender->td_scale =
 558                                 receiver->td_scale = 0;
 559                         if (!tcph->ack)
 560                                 /* Simultaneous open */
 561                                 return true;
 562                 } else {
 563                         /*
 564                          * We are in the middle of a connection,
 565                          * its history is lost for us.
 566                          * Let's try to use the data from the packet.
 567                          */
 568                         sender->td_end = end;
 569                         swin = win << sender->td_scale;
 570                         sender->td_maxwin = (swin == 0 ? 1 : swin);
 571                         sender->td_maxend = end + sender->td_maxwin;
 572                         /*
 573                          * We haven't seen traffic in the other direction yet
 574                          * but we have to tweak window tracking to pass III
 575                          * and IV until that happens.
 576                          */
 577                         if (receiver->td_maxwin == 0)
 578                                 receiver->td_end = receiver->td_maxend = sack;
 579                 }
 580         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 581                      && dir == IP_CT_DIR_ORIGINAL)
 582                    || (state->state == TCP_CONNTRACK_SYN_RECV
 583                      && dir == IP_CT_DIR_REPLY))
 584                    && after(end, sender->td_end)) {
 585                 /*
 586                  * RFC 793: "if a TCP is reinitialized ... then it need
 587                  * not wait at all; it must only be sure to use sequence
 588                  * numbers larger than those recently used."
 589                  */
 590                 sender->td_end =
 591                 sender->td_maxend = end;
 592                 sender->td_maxwin = (win == 0 ? 1 : win);
 593
 594                 tcp_options(skb, dataoff, tcph, sender);
 595         }
 596
 597         if (!(tcph->ack)) {
 598                 /*
 599                  * If there is no ACK, just pretend it was set and OK.
 600                  */
 601                 ack = sack = receiver->td_end;
 602         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 603                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 604                    && (ack == 0)) {
 605                 /*
 606                  * Broken TCP stacks, that set ACK in RST packets as well
 607                  * with zero ack value.
 608                  */
 609                 ack = sack = receiver->td_end;
 610         }
 611
 612         if (tcph->rst && seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)
 613                 /*
 614                  * RST sent answering SYN.
 615                  */
 616                 seq = end = sender->td_end;
 617
 618         pr_debug("tcp_in_window: ");
 619         nf_ct_dump_tuple(tuple);
 620         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
 621                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
 622         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 623                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 624                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 625                  sender->td_scale,
 626                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 627                  receiver->td_scale);
 628
 629         /* Is the ending sequence in the receive window (if available)? */
 630         in_recv_win = !receiver->td_maxwin ||
 631                       after(end, sender->td_end - receiver->td_maxwin - 1);
 632
 633         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 634                  before(seq, sender->td_maxend + 1),
 635                  (in_recv_win ? 1 : 0),
 636                  before(sack, receiver->td_end + 1),
 637                  after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 638
 639         if (before(seq, sender->td_maxend + 1) &&
 640             in_recv_win &&
 641             before(sack, receiver->td_end + 1) &&
 642             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
 643                 /*
 644                  * Take into account window scaling (RFC 1323).
 645                  */
 646                 if (!tcph->syn)
 647                         win <<= sender->td_scale;
 648
 649                 /*
 650                  * Update sender data.
 651                  */
 652                 swin = win + (sack - ack);
 653                 if (sender->td_maxwin < swin)
 654                         sender->td_maxwin = swin;
 655                 if (after(end, sender->td_end)) {
 656                         sender->td_end = end;
 657                         sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 658                 }
 659                 if (tcph->ack) {
 660                         if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
 661                                 sender->td_maxack = ack;
 662                                 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
 663                         } else if (after(ack, sender->td_maxack))
 664                                 sender->td_maxack = ack;
 665                 }
 666
 667                 /*
 668                  * Update receiver data.
 669                  */
 670                 if (receiver->td_maxwin != 0 && after(end, sender->td_maxend))
 671                         receiver->td_maxwin += end - sender->td_maxend;
 672                 if (after(sack + win, receiver->td_maxend - 1)) {
 673                         receiver->td_maxend = sack + win;
 674                         if (win == 0)
 675                                 receiver->td_maxend++;
 676                 }
 677                 if (ack == receiver->td_end)
 678                         receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 679
 680                 /*
 681                  * Check retransmissions.
 682                  */
 683                 if (index == TCP_ACK_SET) {
 684                         if (state->last_dir == dir
 685                             && state->last_seq == seq
 686                             && state->last_ack == ack
 687                             && state->last_end == end
 688                             && state->last_win == win)
 689                                 state->retrans++;
 690                         else {
 691                                 state->last_dir = dir;
 692                                 state->last_seq = seq;
 693                                 state->last_ack = ack;
 694                                 state->last_end = end;
 695                                 state->last_win = win;
 696                                 state->retrans = 0;
 697                         }
 698                 }
 699                 res = true;
 700         } else {
 701                 res = false;
 702                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 703                     tn->tcp_be_liberal)
 704                         res = true;
 705                 if (!res && LOG_INVALID(net, IPPROTO_TCP))
 706                         nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 707                         "nf_ct_tcp: %s ",
 708                         before(seq, sender->td_maxend + 1) ?
 709                         in_recv_win ?
 710                         before(sack, receiver->td_end + 1) ?
 711                         after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
 712                         : "ACK is under the lower bound (possible overly delayed ACK)"
 713                         : "ACK is over the upper bound (ACKed data not seen yet)"
 714                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 715                         : "SEQ is over the upper bound (over the window of the receiver)");
 716         }
 717
 718         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 719                  "receiver end=%u maxend=%u maxwin=%u\n",
 720                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 721                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 722
 723         return res;
 724 }
 725
 726 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 727 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
 728                                  TCPHDR_URG) + 1] =
 729 {
 730         [TCPHDR_SYN]                            = 1,
 731         [TCPHDR_SYN|TCPHDR_URG]                 = 1,
 732         [TCPHDR_SYN|TCPHDR_ACK]                 = 1,
 733         [TCPHDR_RST]                            = 1,
 734         [TCPHDR_RST|TCPHDR_ACK]                 = 1,
 735         [TCPHDR_FIN|TCPHDR_ACK]                 = 1,
 736         [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG]      = 1,
 737         [TCPHDR_ACK]                            = 1,
 738         [TCPHDR_ACK|TCPHDR_URG]                 = 1,
 739 };
 740
 741 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 742 static int tcp_error(struct net *net, struct nf_conn *tmpl,
 743                      struct sk_buff *skb,
 744                      unsigned int dataoff,
 745                      u_int8_t pf,
 746                      unsigned int hooknum)
 747 {
 748         const struct tcphdr *th;
 749         struct tcphdr _tcph;
 750         unsigned int tcplen = skb->len - dataoff;
 751         u_int8_t tcpflags;
 752
 753         /* Smaller that minimal TCP header? */
 754         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 755         if (th == NULL) {
 756                 if (LOG_INVALID(net, IPPROTO_TCP))
 757                         nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 758                                 "nf_ct_tcp: short packet ");
 759                 return -NF_ACCEPT;
 760         }
 761
 762         /* Not whole TCP header or malformed packet */
 763         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 764                 if (LOG_INVALID(net, IPPROTO_TCP))
 765                         nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 766                                 "nf_ct_tcp: truncated/malformed packet ");
 767                 return -NF_ACCEPT;
 768         }
 769
 770         /* Checksum invalid? Ignore.
 771          * We skip checking packets on the outgoing path
 772          * because the checksum is assumed to be correct.
 773          */
 774         /* FIXME: Source route IP option packets --RR */
 775         if (net->ct.sysctl_checksum && hooknum == NF_INET_PRE_ROUTING &&
 776             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
 777                 if (LOG_INVALID(net, IPPROTO_TCP))
 778                         nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 779                                   "nf_ct_tcp: bad TCP checksum ");
 780                 return -NF_ACCEPT;
 781         }
 782
 783         /* Check TCP flags. */
 784         tcpflags = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
 785         if (!tcp_valid_flags[tcpflags]) {
 786                 if (LOG_INVALID(net, IPPROTO_TCP))
 787                         nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 788                                   "nf_ct_tcp: invalid TCP flag combination ");
 789                 return -NF_ACCEPT;
 790         }
 791
 792         return NF_ACCEPT;
 793 }
 794
 795 static unsigned int *tcp_get_timeouts(struct net *net)
 796 {
 797         return tcp_pernet(net)->timeouts;
 798 }
 799
 800 /* Returns verdict for packet, or -1 for invalid. */
 801 static int tcp_packet(struct nf_conn *ct,
 802                       const struct sk_buff *skb,
 803                       unsigned int dataoff,
 804                       enum ip_conntrack_info ctinfo,
 805                       u_int8_t pf,
 806                       unsigned int *timeouts)
 807 {
 808         struct net *net = nf_ct_net(ct);
 809         struct nf_tcp_net *tn = tcp_pernet(net);
 810         struct nf_conntrack_tuple *tuple;
 811         enum tcp_conntrack new_state, old_state;
 812         enum ip_conntrack_dir dir;
 813         const struct tcphdr *th;
 814         struct tcphdr _tcph;
 815         unsigned long timeout;
 816         unsigned int index;
 817
 818         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 819         BUG_ON(th == NULL);
 820
 821         spin_lock_bh(&ct->lock);
 822         old_state = ct->proto.tcp.state;
 823         dir = CTINFO2DIR(ctinfo);
 824         index = get_conntrack_index(th);
 825         new_state = tcp_conntracks[dir][index][old_state];
 826         tuple = &ct->tuplehash[dir].tuple;
 827
 828         switch (new_state) {
 829         case TCP_CONNTRACK_SYN_SENT:
 830                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 831                         break;
 832                 /* RFC 1122: "When a connection is closed actively,
 833                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 834                  * (Maximum Segment Lifetime). However, it MAY accept
 835                  * a new SYN from the remote TCP to reopen the connection
 836                  * directly from TIME-WAIT state, if..."
 837                  * We ignore the conditions because we are in the
 838                  * TIME-WAIT state anyway.
 839                  *
 840                  * Handle aborted connections: we and the server
 841                  * think there is an existing connection but the client
 842                  * aborts it and starts a new one.
 843                  */
 844                 if (((ct->proto.tcp.seen[dir].flags
 845                       | ct->proto.tcp.seen[!dir].flags)
 846                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 847                     || (ct->proto.tcp.last_dir == dir
 848                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 849                         /* Attempt to reopen a closed/aborted connection.
 850                          * Delete this connection and look up again. */
 851                         spin_unlock_bh(&ct->lock);
 852
 853                         /* Only repeat if we can actually remove the timer.
 854                          * Destruction may already be in progress in process
 855                          * context and we must give it a chance to terminate.
 856                          */
 857                         if (nf_ct_kill(ct))
 858                                 return -NF_REPEAT;
 859                         return NF_DROP;
 860                 }
 861                 /* Fall through */
 862         case TCP_CONNTRACK_IGNORE:
 863                 /* Ignored packets:
 864                  *
 865                  * Our connection entry may be out of sync, so ignore
 866                  * packets which may signal the real connection between
 867                  * the client and the server.
 868                  *
 869                  * a) SYN in ORIGINAL
 870                  * b) SYN/ACK in REPLY
 871                  * c) ACK in reply direction after initial SYN in original.
 872                  *
 873                  * If the ignored packet is invalid, the receiver will send
 874                  * a RST we'll catch below.
 875                  */
 876                 if (index == TCP_SYNACK_SET
 877                     && ct->proto.tcp.last_index == TCP_SYN_SET
 878                     && ct->proto.tcp.last_dir != dir
 879                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 880                         /* b) This SYN/ACK acknowledges a SYN that we earlier
 881                          * ignored as invalid. This means that the client and
 882                          * the server are both in sync, while the firewall is
 883                          * not. We get in sync from the previously annotated
 884                          * values.
 885                          */
 886                         old_state = TCP_CONNTRACK_SYN_SENT;
 887                         new_state = TCP_CONNTRACK_SYN_RECV;
 888                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
 889                                 ct->proto.tcp.last_end;
 890                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
 891                                 ct->proto.tcp.last_end;
 892                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
 893                                 ct->proto.tcp.last_win == 0 ?
 894                                         1 : ct->proto.tcp.last_win;
 895                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
 896                                 ct->proto.tcp.last_wscale;
 897                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
 898                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
 899                                 ct->proto.tcp.last_flags;
 900                         memset(&ct->proto.tcp.seen[dir], 0,
 901                                sizeof(struct ip_ct_tcp_state));
 902                         break;
 903                 }
 904                 ct->proto.tcp.last_index = index;
 905                 ct->proto.tcp.last_dir = dir;
 906                 ct->proto.tcp.last_seq = ntohl(th->seq);
 907                 ct->proto.tcp.last_end =
 908                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 909                 ct->proto.tcp.last_win = ntohs(th->window);
 910
 911                 /* a) This is a SYN in ORIGINAL. The client and the server
 912                  * may be in sync but we are not. In that case, we annotate
 913                  * the TCP options and let the packet go through. If it is a
 914                  * valid SYN packet, the server will reply with a SYN/ACK, and
 915                  * then we'll get in sync. Otherwise, the server potentially
 916                  * responds with a challenge ACK if implementing RFC5961.
 917                  */
 918                 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
 919                         struct ip_ct_tcp_state seen = {};
 920
 921                         ct->proto.tcp.last_flags =
 922                         ct->proto.tcp.last_wscale = 0;
 923                         tcp_options(skb, dataoff, th, &seen);
 924                         if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
 925                                 ct->proto.tcp.last_flags |=
 926                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 927                                 ct->proto.tcp.last_wscale = seen.td_scale;
 928                         }
 929                         if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
 930                                 ct->proto.tcp.last_flags |=
 931                                         IP_CT_TCP_FLAG_SACK_PERM;
 932                         }
 933                         /* Mark the potential for RFC5961 challenge ACK,
 934                          * this pose a special problem for LAST_ACK state
 935                          * as ACK is intrepretated as ACKing last FIN.
 936                          */
 937                         if (old_state == TCP_CONNTRACK_LAST_ACK)
 938                                 ct->proto.tcp.last_flags |=
 939                                         IP_CT_EXP_CHALLENGE_ACK;
 940                 }
 941                 spin_unlock_bh(&ct->lock);
 942                 if (LOG_INVALID(net, IPPROTO_TCP))
 943                         nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 944                                   "nf_ct_tcp: invalid packet ignored in "
 945                                   "state %s ", tcp_conntrack_names[old_state]);
 946                 return NF_ACCEPT;
 947         case TCP_CONNTRACK_MAX:
 948                 /* Special case for SYN proxy: when the SYN to the server or
 949                  * the SYN/ACK from the server is lost, the client may transmit
 950                  * a keep-alive packet while in SYN_SENT state. This needs to
 951                  * be associated with the original conntrack entry in order to
 952                  * generate a new SYN with the correct sequence number.
 953                  */
 954                 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
 955                     index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
 956                     ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
 957                     ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
 958                         pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
 959                         spin_unlock_bh(&ct->lock);
 960                         return NF_ACCEPT;
 961                 }
 962
 963                 /* Invalid packet */
 964                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 965                          dir, get_conntrack_index(th), old_state);
 966                 spin_unlock_bh(&ct->lock);
 967                 if (LOG_INVALID(net, IPPROTO_TCP))
 968                         nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 969                                   "nf_ct_tcp: invalid state ");
 970                 return -NF_ACCEPT;
 971         case TCP_CONNTRACK_TIME_WAIT:
 972                 /* RFC5961 compliance cause stack to send "challenge-ACK"
 973                  * e.g. in response to spurious SYNs.  Conntrack MUST
 974                  * not believe this ACK is acking last FIN.
 975                  */
 976                 if (old_state == TCP_CONNTRACK_LAST_ACK &&
 977                     index == TCP_ACK_SET &&
 978                     ct->proto.tcp.last_dir != dir &&
 979                     ct->proto.tcp.last_index == TCP_SYN_SET &&
 980                     (ct->proto.tcp.last_flags & IP_CT_EXP_CHALLENGE_ACK)) {
 981                         /* Detected RFC5961 challenge ACK */
 982                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
 983                         spin_unlock_bh(&ct->lock);
 984                         if (LOG_INVALID(net, IPPROTO_TCP))
 985                                 nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
 986                                       "nf_ct_tcp: challenge-ACK ignored ");
 987                         return NF_ACCEPT; /* Don't change state */
 988                 }
 989                 break;
 990         case TCP_CONNTRACK_CLOSE:
 991                 if (index == TCP_RST_SET
 992                     && (ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET)
 993                     && before(ntohl(th->seq), ct->proto.tcp.seen[!dir].td_maxack)) {
 994                         /* Invalid RST  */
 995                         spin_unlock_bh(&ct->lock);
 996                         if (LOG_INVALID(net, IPPROTO_TCP))
 997                                 nf_log_packet(net, pf, 0, skb, NULL, NULL,
 998                                               NULL, "nf_ct_tcp: invalid RST ");
 999                         return -NF_ACCEPT;
1000                 }
1001                 if (index == TCP_RST_SET
1002                     && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
1003                          && ct->proto.tcp.last_index == TCP_SYN_SET)
1004                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
1005                             && ct->proto.tcp.last_index == TCP_ACK_SET))
1006                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
1007                         /* RST sent to invalid SYN or ACK we had let through
1008                          * at a) and c) above:
1009                          *
1010                          * a) SYN was in window then
1011                          * c) we hold a half-open connection.
1012                          *
1013                          * Delete our connection entry.
1014                          * We skip window checking, because packet might ACK
1015                          * segments we ignored. */
1016                         goto in_window;
1017                 }
1018                 /* Just fall through */
1019         default:
1020                 /* Keep compilers happy. */
1021                 break;
1022         }
1023
1024         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
1025                            skb, dataoff, th, pf)) {
1026                 spin_unlock_bh(&ct->lock);
1027                 return -NF_ACCEPT;
1028         }
1029      in_window:
1030         /* From now on we have got in-window packets */
1031         ct->proto.tcp.last_index = index;
1032         ct->proto.tcp.last_dir = dir;
1033
1034         pr_debug("tcp_conntracks: ");
1035         nf_ct_dump_tuple(tuple);
1036         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1037                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1038                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1039                  old_state, new_state);
1040
1041         ct->proto.tcp.state = new_state;
1042         if (old_state != new_state
1043             && new_state == TCP_CONNTRACK_FIN_WAIT)
1044                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1045
1046         if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1047             timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1048                 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1049         else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1050                  IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1051                  timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1052                 timeout = timeouts[TCP_CONNTRACK_UNACK];
1053         else
1054                 timeout = timeouts[new_state];
1055         spin_unlock_bh(&ct->lock);
1056
1057         if (new_state != old_state)
1058                 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1059
1060         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1061                 /* If only reply is a RST, we can consider ourselves not to
1062                    have an established connection: this is a fairly common
1063                    problem case, so we can delete the conntrack
1064                    immediately.  --RR */
1065                 if (th->rst) {
1066                         nf_ct_kill_acct(ct, ctinfo, skb);
1067                         return NF_ACCEPT;
1068                 }
1069                 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1070                  * pickup with loose=1. Avoid large ESTABLISHED timeout.
1071                  */
1072                 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1073                     timeout > timeouts[TCP_CONNTRACK_UNACK])
1074                         timeout = timeouts[TCP_CONNTRACK_UNACK];
1075         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1076                    && (old_state == TCP_CONNTRACK_SYN_RECV
1077                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1078                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1079                 /* Set ASSURED if we see see valid ack in ESTABLISHED
1080                    after SYN_RECV or a valid answer for a picked up
1081                    connection. */
1082                 set_bit(IPS_ASSURED_BIT, &ct->status);
1083                 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1084         }
1085         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1086
1087         return NF_ACCEPT;
1088 }
1089
1090 /* Called when a new connection for this protocol found. */
1091 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
1092                     unsigned int dataoff, unsigned int *timeouts)
1093 {
1094         enum tcp_conntrack new_state;
1095         const struct tcphdr *th;
1096         struct tcphdr _tcph;
1097         struct net *net = nf_ct_net(ct);
1098         struct nf_tcp_net *tn = tcp_pernet(net);
1099         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
1100         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
1101
1102         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1103         BUG_ON(th == NULL);
1104
1105         /* Don't need lock here: this conntrack not in circulation yet */
1106         new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
1107
1108         /* Invalid: delete conntrack */
1109         if (new_state >= TCP_CONNTRACK_MAX) {
1110                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1111                 return false;
1112         }
1113
1114         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1115                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
1116                 /* SYN packet */
1117                 ct->proto.tcp.seen[0].td_end =
1118                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1119                                              dataoff, th);
1120                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1121                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1122                         ct->proto.tcp.seen[0].td_maxwin = 1;
1123                 ct->proto.tcp.seen[0].td_maxend =
1124                         ct->proto.tcp.seen[0].td_end;
1125
1126                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1127         } else if (tn->tcp_loose == 0) {
1128                 /* Don't try to pick up connections. */
1129                 return false;
1130         } else {
1131                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
1132                 /*
1133                  * We are in the middle of a connection,
1134                  * its history is lost for us.
1135                  * Let's try to use the data from the packet.
1136                  */
1137                 ct->proto.tcp.seen[0].td_end =
1138                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1139                                              dataoff, th);
1140                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1141                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1142                         ct->proto.tcp.seen[0].td_maxwin = 1;
1143                 ct->proto.tcp.seen[0].td_maxend =
1144                         ct->proto.tcp.seen[0].td_end +
1145                         ct->proto.tcp.seen[0].td_maxwin;
1146
1147                 /* We assume SACK and liberal window checking to handle
1148                  * window scaling */
1149                 ct->proto.tcp.seen[0].flags =
1150                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1151                                               IP_CT_TCP_FLAG_BE_LIBERAL;
1152         }
1153
1154         /* tcp_packet will set them */
1155         ct->proto.tcp.last_index = TCP_NONE_SET;
1156
1157         pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1158                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1159                  sender->td_end, sender->td_maxend, sender->td_maxwin,
1160                  sender->td_scale,
1161                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1162                  receiver->td_scale);
1163         return true;
1164 }
1165
1166 static bool tcp_can_early_drop(const struct nf_conn *ct)
1167 {
1168         switch (ct->proto.tcp.state) {
1169         case TCP_CONNTRACK_FIN_WAIT:
1170         case TCP_CONNTRACK_LAST_ACK:
1171         case TCP_CONNTRACK_TIME_WAIT:
1172         case TCP_CONNTRACK_CLOSE:
1173         case TCP_CONNTRACK_CLOSE_WAIT:
1174                 return true;
1175         default:
1176                 break;
1177         }
1178
1179         return false;
1180 }
1181
1182 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1183
1184 #include <linux/netfilter/nfnetlink.h>
1185 #include <linux/netfilter/nfnetlink_conntrack.h>
1186
1187 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1188                          struct nf_conn *ct)
1189 {
1190         struct nlattr *nest_parms;
1191         struct nf_ct_tcp_flags tmp = {};
1192
1193         spin_lock_bh(&ct->lock);
1194         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1195         if (!nest_parms)
1196                 goto nla_put_failure;
1197
1198         if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state) ||
1199             nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1200                        ct->proto.tcp.seen[0].td_scale) ||
1201             nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1202                        ct->proto.tcp.seen[1].td_scale))
1203                 goto nla_put_failure;
1204
1205         tmp.flags = ct->proto.tcp.seen[0].flags;
1206         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1207                     sizeof(struct nf_ct_tcp_flags), &tmp))
1208                 goto nla_put_failure;
1209
1210         tmp.flags = ct->proto.tcp.seen[1].flags;
1211         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1212                     sizeof(struct nf_ct_tcp_flags), &tmp))
1213                 goto nla_put_failure;
1214         spin_unlock_bh(&ct->lock);
1215
1216         nla_nest_end(skb, nest_parms);
1217
1218         return 0;
1219
1220 nla_put_failure:
1221         spin_unlock_bh(&ct->lock);
1222         return -1;
1223 }
1224
1225 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1226         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1227         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1228         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1229         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1230         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len =  sizeof(struct nf_ct_tcp_flags) },
1231 };
1232
1233 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1234 {
1235         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1236         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1237         int err;
1238
1239         /* updates could not contain anything about the private
1240          * protocol info, in that case skip the parsing */
1241         if (!pattr)
1242                 return 0;
1243
1244         err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr,
1245                                tcp_nla_policy, NULL);
1246         if (err < 0)
1247                 return err;
1248
1249         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1250             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1251                 return -EINVAL;
1252
1253         spin_lock_bh(&ct->lock);
1254         if (tb[CTA_PROTOINFO_TCP_STATE])
1255                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1256
1257         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1258                 struct nf_ct_tcp_flags *attr =
1259                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1260                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1261                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1262         }
1263
1264         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1265                 struct nf_ct_tcp_flags *attr =
1266                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1267                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1268                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1269         }
1270
1271         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1272             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1273             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1274             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1275                 ct->proto.tcp.seen[0].td_scale =
1276                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1277                 ct->proto.tcp.seen[1].td_scale =
1278                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1279         }
1280         spin_unlock_bh(&ct->lock);
1281
1282         return 0;
1283 }
1284
1285 static int tcp_nlattr_size(void)
1286 {
1287         return nla_total_size(0)           /* CTA_PROTOINFO_TCP */
1288                 + nla_policy_len(tcp_nla_policy, CTA_PROTOINFO_TCP_MAX + 1);
1289 }
1290
1291 static int tcp_nlattr_tuple_size(void)
1292 {
1293         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1294 }
1295 #endif
1296
1297 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1298
1299 #include <linux/netfilter/nfnetlink.h>
1300 #include <linux/netfilter/nfnetlink_cttimeout.h>
1301
1302 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1303                                      struct net *net, void *data)
1304 {
1305         unsigned int *timeouts = data;
1306         struct nf_tcp_net *tn = tcp_pernet(net);
1307         int i;
1308
1309         /* set default TCP timeouts. */
1310         for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1311                 timeouts[i] = tn->timeouts[i];
1312
1313         if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1314                 timeouts[TCP_CONNTRACK_SYN_SENT] =
1315                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1316         }
1317         if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1318                 timeouts[TCP_CONNTRACK_SYN_RECV] =
1319                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1320         }
1321         if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1322                 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1323                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1324         }
1325         if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1326                 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1327                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1328         }
1329         if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1330                 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1331                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1332         }
1333         if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1334                 timeouts[TCP_CONNTRACK_LAST_ACK] =
1335                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1336         }
1337         if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1338                 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1339                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1340         }
1341         if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1342                 timeouts[TCP_CONNTRACK_CLOSE] =
1343                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1344         }
1345         if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1346                 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1347                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1348         }
1349         if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1350                 timeouts[TCP_CONNTRACK_RETRANS] =
1351                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1352         }
1353         if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1354                 timeouts[TCP_CONNTRACK_UNACK] =
1355                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1356         }
1357         return 0;
1358 }
1359
1360 static int
1361 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1362 {
1363         const unsigned int *timeouts = data;
1364
1365         if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1366                         htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1367             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1368                          htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1369             nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1370                          htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1371             nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1372                          htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1373             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1374                          htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1375             nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1376                          htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1377             nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1378                          htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1379             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1380                          htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1381             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1382                          htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1383             nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1384                          htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1385             nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1386                          htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1387                 goto nla_put_failure;
1388         return 0;
1389
1390 nla_put_failure:
1391         return -ENOSPC;
1392 }
1393
1394 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1395         [CTA_TIMEOUT_TCP_SYN_SENT]      = { .type = NLA_U32 },
1396         [CTA_TIMEOUT_TCP_SYN_RECV]      = { .type = NLA_U32 },
1397         [CTA_TIMEOUT_TCP_ESTABLISHED]   = { .type = NLA_U32 },
1398         [CTA_TIMEOUT_TCP_FIN_WAIT]      = { .type = NLA_U32 },
1399         [CTA_TIMEOUT_TCP_CLOSE_WAIT]    = { .type = NLA_U32 },
1400         [CTA_TIMEOUT_TCP_LAST_ACK]      = { .type = NLA_U32 },
1401         [CTA_TIMEOUT_TCP_TIME_WAIT]     = { .type = NLA_U32 },
1402         [CTA_TIMEOUT_TCP_CLOSE]         = { .type = NLA_U32 },
1403         [CTA_TIMEOUT_TCP_SYN_SENT2]     = { .type = NLA_U32 },
1404         [CTA_TIMEOUT_TCP_RETRANS]       = { .type = NLA_U32 },
1405         [CTA_TIMEOUT_TCP_UNACK]         = { .type = NLA_U32 },
1406 };
1407 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1408
1409 #ifdef CONFIG_SYSCTL
1410 static struct ctl_table tcp_sysctl_table[] = {
1411         {
1412                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1413                 .maxlen         = sizeof(unsigned int),
1414                 .mode           = 0644,
1415                 .proc_handler   = proc_dointvec_jiffies,
1416         },
1417         {
1418                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1419                 .maxlen         = sizeof(unsigned int),
1420                 .mode           = 0644,
1421                 .proc_handler   = proc_dointvec_jiffies,
1422         },
1423         {
1424                 .procname       = "nf_conntrack_tcp_timeout_established",
1425                 .maxlen         = sizeof(unsigned int),
1426                 .mode           = 0644,
1427                 .proc_handler   = proc_dointvec_jiffies,
1428         },
1429         {
1430                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1431                 .maxlen         = sizeof(unsigned int),
1432                 .mode           = 0644,
1433                 .proc_handler   = proc_dointvec_jiffies,
1434         },
1435         {
1436                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1437                 .maxlen         = sizeof(unsigned int),
1438                 .mode           = 0644,
1439                 .proc_handler   = proc_dointvec_jiffies,
1440         },
1441         {
1442                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1443                 .maxlen         = sizeof(unsigned int),
1444                 .mode           = 0644,
1445                 .proc_handler   = proc_dointvec_jiffies,
1446         },
1447         {
1448                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1449                 .maxlen         = sizeof(unsigned int),
1450                 .mode           = 0644,
1451                 .proc_handler   = proc_dointvec_jiffies,
1452         },
1453         {
1454                 .procname       = "nf_conntrack_tcp_timeout_close",
1455                 .maxlen         = sizeof(unsigned int),
1456                 .mode           = 0644,
1457                 .proc_handler   = proc_dointvec_jiffies,
1458         },
1459         {
1460                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1461                 .maxlen         = sizeof(unsigned int),
1462                 .mode           = 0644,
1463                 .proc_handler   = proc_dointvec_jiffies,
1464         },
1465         {
1466                 .procname       = "nf_conntrack_tcp_timeout_unacknowledged",
1467                 .maxlen         = sizeof(unsigned int),
1468                 .mode           = 0644,
1469                 .proc_handler   = proc_dointvec_jiffies,
1470         },
1471         {
1472                 .procname       = "nf_conntrack_tcp_loose",
1473                 .maxlen         = sizeof(unsigned int),
1474                 .mode           = 0644,
1475                 .proc_handler   = proc_dointvec,
1476         },
1477         {
1478                 .procname       = "nf_conntrack_tcp_be_liberal",
1479                 .maxlen         = sizeof(unsigned int),
1480                 .mode           = 0644,
1481                 .proc_handler   = proc_dointvec,
1482         },
1483         {
1484                 .procname       = "nf_conntrack_tcp_max_retrans",
1485                 .maxlen         = sizeof(unsigned int),
1486                 .mode           = 0644,
1487                 .proc_handler   = proc_dointvec,
1488         },
1489         { }
1490 };
1491 #endif /* CONFIG_SYSCTL */
1492
1493 static int tcp_kmemdup_sysctl_table(struct nf_proto_net *pn,
1494                                     struct nf_tcp_net *tn)
1495 {
1496 #ifdef CONFIG_SYSCTL
1497         if (pn->ctl_table)
1498                 return 0;
1499
1500         pn->ctl_table = kmemdup(tcp_sysctl_table,
1501                                 sizeof(tcp_sysctl_table),
1502                                 GFP_KERNEL);
1503         if (!pn->ctl_table)
1504                 return -ENOMEM;
1505
1506         pn->ctl_table[0].data = &tn->timeouts[TCP_CONNTRACK_SYN_SENT];
1507         pn->ctl_table[1].data = &tn->timeouts[TCP_CONNTRACK_SYN_RECV];
1508         pn->ctl_table[2].data = &tn->timeouts[TCP_CONNTRACK_ESTABLISHED];
1509         pn->ctl_table[3].data = &tn->timeouts[TCP_CONNTRACK_FIN_WAIT];
1510         pn->ctl_table[4].data = &tn->timeouts[TCP_CONNTRACK_CLOSE_WAIT];
1511         pn->ctl_table[5].data = &tn->timeouts[TCP_CONNTRACK_LAST_ACK];
1512         pn->ctl_table[6].data = &tn->timeouts[TCP_CONNTRACK_TIME_WAIT];
1513         pn->ctl_table[7].data = &tn->timeouts[TCP_CONNTRACK_CLOSE];
1514         pn->ctl_table[8].data = &tn->timeouts[TCP_CONNTRACK_RETRANS];
1515         pn->ctl_table[9].data = &tn->timeouts[TCP_CONNTRACK_UNACK];
1516         pn->ctl_table[10].data = &tn->tcp_loose;
1517         pn->ctl_table[11].data = &tn->tcp_be_liberal;
1518         pn->ctl_table[12].data = &tn->tcp_max_retrans;
1519 #endif
1520         return 0;
1521 }
1522
1523 static int tcp_init_net(struct net *net, u_int16_t proto)
1524 {
1525         struct nf_tcp_net *tn = tcp_pernet(net);
1526         struct nf_proto_net *pn = &tn->pn;
1527
1528         if (!pn->users) {
1529                 int i;
1530
1531                 for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1532                         tn->timeouts[i] = tcp_timeouts[i];
1533
1534                 tn->tcp_loose = nf_ct_tcp_loose;
1535                 tn->tcp_be_liberal = nf_ct_tcp_be_liberal;
1536                 tn->tcp_max_retrans = nf_ct_tcp_max_retrans;
1537         }
1538
1539         return tcp_kmemdup_sysctl_table(pn, tn);
1540 }
1541
1542 static struct nf_proto_net *tcp_get_net_proto(struct net *net)
1543 {
1544         return &net->ct.nf_ct_proto.tcp.pn;
1545 }
1546
1547 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1548 {
1549         .l3proto                = PF_INET,
1550         .l4proto                = IPPROTO_TCP,
1551         .pkt_to_tuple           = tcp_pkt_to_tuple,
1552         .invert_tuple           = tcp_invert_tuple,
1553 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1554         .print_conntrack        = tcp_print_conntrack,
1555 #endif
1556         .packet                 = tcp_packet,
1557         .get_timeouts           = tcp_get_timeouts,
1558         .new                    = tcp_new,
1559         .error                  = tcp_error,
1560         .can_early_drop         = tcp_can_early_drop,
1561 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1562         .to_nlattr              = tcp_to_nlattr,
1563         .nlattr_size            = tcp_nlattr_size,
1564         .from_nlattr            = nlattr_to_tcp,
1565         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1566         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1567         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1568         .nla_policy             = nf_ct_port_nla_policy,
1569 #endif
1570 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1571         .ctnl_timeout           = {
1572                 .nlattr_to_obj  = tcp_timeout_nlattr_to_obj,
1573                 .obj_to_nlattr  = tcp_timeout_obj_to_nlattr,
1574                 .nlattr_max     = CTA_TIMEOUT_TCP_MAX,
1575                 .obj_size       = sizeof(unsigned int) *
1576                                         TCP_CONNTRACK_TIMEOUT_MAX,
1577                 .nla_policy     = tcp_timeout_nla_policy,
1578         },
1579 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1580         .init_net               = tcp_init_net,
1581         .get_net_proto          = tcp_get_net_proto,
1582 };
1583 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1584
1585 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1586 {
1587         .l3proto                = PF_INET6,
1588         .l4proto                = IPPROTO_TCP,
1589         .pkt_to_tuple           = tcp_pkt_to_tuple,
1590         .invert_tuple           = tcp_invert_tuple,
1591 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1592         .print_conntrack        = tcp_print_conntrack,
1593 #endif
1594         .packet                 = tcp_packet,
1595         .get_timeouts           = tcp_get_timeouts,
1596         .new                    = tcp_new,
1597         .error                  = tcp_error,
1598         .can_early_drop         = tcp_can_early_drop,
1599 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1600         .to_nlattr              = tcp_to_nlattr,
1601         .nlattr_size            = tcp_nlattr_size,
1602         .from_nlattr            = nlattr_to_tcp,
1603         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1604         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1605         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1606         .nla_policy             = nf_ct_port_nla_policy,
1607 #endif
1608 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1609         .ctnl_timeout           = {
1610                 .nlattr_to_obj  = tcp_timeout_nlattr_to_obj,
1611                 .obj_to_nlattr  = tcp_timeout_obj_to_nlattr,
1612                 .nlattr_max     = CTA_TIMEOUT_TCP_MAX,
1613                 .obj_size       = sizeof(unsigned int) *
1614                                         TCP_CONNTRACK_TIMEOUT_MAX,
1615                 .nla_policy     = tcp_timeout_nla_policy,
1616         },
1617 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1618         .init_net               = tcp_init_net,
1619         .get_net_proto          = tcp_get_net_proto,
1620 };
1621 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);