net/netfilter/nf_conntrack_proto_tcp.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* (C) 1999-2001 Paul `Rusty' Russell
   3  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   4  * (C) 2002-2013 Jozsef Kadlecsik <kadlec@netfilter.org>
   5  * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
   6  */
   7
   8 #include <linux/types.h>
   9 #include <linux/timer.h>
  10 #include <linux/module.h>
  11 #include <linux/in.h>
  12 #include <linux/tcp.h>
  13 #include <linux/spinlock.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/ipv6.h>
  16 #include <net/ip6_checksum.h>
  17 #include <asm/unaligned.h>
  18
  19 #include <net/tcp.h>
  20
  21 #include <linux/netfilter.h>
  22 #include <linux/netfilter_ipv4.h>
  23 #include <linux/netfilter_ipv6.h>
  24 #include <net/netfilter/nf_conntrack.h>
  25 #include <net/netfilter/nf_conntrack_l4proto.h>
  26 #include <net/netfilter/nf_conntrack_ecache.h>
  27 #include <net/netfilter/nf_conntrack_seqadj.h>
  28 #include <net/netfilter/nf_conntrack_synproxy.h>
  29 #include <net/netfilter/nf_conntrack_timeout.h>
  30 #include <net/netfilter/nf_log.h>
  31 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
  32 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
  33
  34   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  35      closely.  They're more complex. --RR */
  36
  37 static const char *const tcp_conntrack_names[] = {
  38         "NONE",
  39         "SYN_SENT",
  40         "SYN_RECV",
  41         "ESTABLISHED",
  42         "FIN_WAIT",
  43         "CLOSE_WAIT",
  44         "LAST_ACK",
  45         "TIME_WAIT",
  46         "CLOSE",
  47         "SYN_SENT2",
  48 };
  49
  50 #define SECS * HZ
  51 #define MINS * 60 SECS
  52 #define HOURS * 60 MINS
  53 #define DAYS * 24 HOURS
  54
  55 static const unsigned int tcp_timeouts[TCP_CONNTRACK_TIMEOUT_MAX] = {
  56         [TCP_CONNTRACK_SYN_SENT]        = 2 MINS,
  57         [TCP_CONNTRACK_SYN_RECV]        = 60 SECS,
  58         [TCP_CONNTRACK_ESTABLISHED]     = 5 DAYS,
  59         [TCP_CONNTRACK_FIN_WAIT]        = 2 MINS,
  60         [TCP_CONNTRACK_CLOSE_WAIT]      = 60 SECS,
  61         [TCP_CONNTRACK_LAST_ACK]        = 30 SECS,
  62         [TCP_CONNTRACK_TIME_WAIT]       = 2 MINS,
  63         [TCP_CONNTRACK_CLOSE]           = 10 SECS,
  64         [TCP_CONNTRACK_SYN_SENT2]       = 2 MINS,
  65 /* RFC1122 says the R2 limit should be at least 100 seconds.
  66    Linux uses 15 packets as limit, which corresponds
  67    to ~13-30min depending on RTO. */
  68         [TCP_CONNTRACK_RETRANS]         = 5 MINS,
  69         [TCP_CONNTRACK_UNACK]           = 5 MINS,
  70 };
  71
  72 #define sNO TCP_CONNTRACK_NONE
  73 #define sSS TCP_CONNTRACK_SYN_SENT
  74 #define sSR TCP_CONNTRACK_SYN_RECV
  75 #define sES TCP_CONNTRACK_ESTABLISHED
  76 #define sFW TCP_CONNTRACK_FIN_WAIT
  77 #define sCW TCP_CONNTRACK_CLOSE_WAIT
  78 #define sLA TCP_CONNTRACK_LAST_ACK
  79 #define sTW TCP_CONNTRACK_TIME_WAIT
  80 #define sCL TCP_CONNTRACK_CLOSE
  81 #define sS2 TCP_CONNTRACK_SYN_SENT2
  82 #define sIV TCP_CONNTRACK_MAX
  83 #define sIG TCP_CONNTRACK_IGNORE
  84
  85 /* What TCP flags are set from RST/SYN/FIN/ACK. */
  86 enum tcp_bit_set {
  87         TCP_SYN_SET,
  88         TCP_SYNACK_SET,
  89         TCP_FIN_SET,
  90         TCP_ACK_SET,
  91         TCP_RST_SET,
  92         TCP_NONE_SET,
  93 };
  94
  95 /*
  96  * The TCP state transition table needs a few words...
  97  *
  98  * We are the man in the middle. All the packets go through us
  99  * but might get lost in transit to the destination.
 100  * It is assumed that the destinations can't receive segments
 101  * we haven't seen.
 102  *
 103  * The checked segment is in window, but our windows are *not*
 104  * equivalent with the ones of the sender/receiver. We always
 105  * try to guess the state of the current sender.
 106  *
 107  * The meaning of the states are:
 108  *
 109  * NONE:        initial state
 110  * SYN_SENT:    SYN-only packet seen
 111  * SYN_SENT2:   SYN-only packet seen from reply dir, simultaneous open
 112  * SYN_RECV:    SYN-ACK packet seen
 113  * ESTABLISHED: ACK packet seen
 114  * FIN_WAIT:    FIN packet seen
 115  * CLOSE_WAIT:  ACK seen (after FIN)
 116  * LAST_ACK:    FIN seen (after FIN)
 117  * TIME_WAIT:   last ACK seen
 118  * CLOSE:       closed connection (RST)
 119  *
 120  * Packets marked as IGNORED (sIG):
 121  *      if they may be either invalid or valid
 122  *      and the receiver may send back a connection
 123  *      closing RST or a SYN/ACK.
 124  *
 125  * Packets marked as INVALID (sIV):
 126  *      if we regard them as truly invalid packets
 127  */
 128 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 129         {
 130 /* ORIGINAL */
 131 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 132 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sS2 },
 133 /*
 134  *      sNO -> sSS      Initialize a new connection
 135  *      sSS -> sSS      Retransmitted SYN
 136  *      sS2 -> sS2      Late retransmitted SYN
 137  *      sSR -> sIG
 138  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 139  *                      are errors. Receiver will reply with RST
 140  *                      and close the connection.
 141  *                      Or we are not in sync and hold a dead connection.
 142  *      sFW -> sIG
 143  *      sCW -> sIG
 144  *      sLA -> sIG
 145  *      sTW -> sSS      Reopened connection (RFC 1122).
 146  *      sCL -> sSS
 147  */
 148 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 149 /*synack*/ { sIV, sIV, sSR, sIV, sIV, sIV, sIV, sIV, sIV, sSR },
 150 /*
 151  *      sNO -> sIV      Too late and no reason to do anything
 152  *      sSS -> sIV      Client can't send SYN and then SYN/ACK
 153  *      sS2 -> sSR      SYN/ACK sent to SYN2 in simultaneous open
 154  *      sSR -> sSR      Late retransmitted SYN/ACK in simultaneous open
 155  *      sES -> sIV      Invalid SYN/ACK packets sent by the client
 156  *      sFW -> sIV
 157  *      sCW -> sIV
 158  *      sLA -> sIV
 159  *      sTW -> sIV
 160  *      sCL -> sIV
 161  */
 162 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 163 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 164 /*
 165  *      sNO -> sIV      Too late and no reason to do anything...
 166  *      sSS -> sIV      Client migth not send FIN in this state:
 167  *                      we enforce waiting for a SYN/ACK reply first.
 168  *      sS2 -> sIV
 169  *      sSR -> sFW      Close started.
 170  *      sES -> sFW
 171  *      sFW -> sLA      FIN seen in both directions, waiting for
 172  *                      the last ACK.
 173  *                      Migth be a retransmitted FIN as well...
 174  *      sCW -> sLA
 175  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 176  *      sTW -> sTW
 177  *      sCL -> sCL
 178  */
 179 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 180 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 181 /*
 182  *      sNO -> sES      Assumed.
 183  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 184  *      sS2 -> sIV
 185  *      sSR -> sES      Established state is reached.
 186  *      sES -> sES      :-)
 187  *      sFW -> sCW      Normal close request answered by ACK.
 188  *      sCW -> sCW
 189  *      sLA -> sTW      Last ACK detected (RFC5961 challenged)
 190  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 191  *      sCL -> sCL
 192  */
 193 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 194 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
 195 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 196         },
 197         {
 198 /* REPLY */
 199 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 200 /*syn*/    { sIV, sS2, sIV, sIV, sIV, sIV, sIV, sSS, sIV, sS2 },
 201 /*
 202  *      sNO -> sIV      Never reached.
 203  *      sSS -> sS2      Simultaneous open
 204  *      sS2 -> sS2      Retransmitted simultaneous SYN
 205  *      sSR -> sIV      Invalid SYN packets sent by the server
 206  *      sES -> sIV
 207  *      sFW -> sIV
 208  *      sCW -> sIV
 209  *      sLA -> sIV
 210  *      sTW -> sSS      Reopened connection, but server may have switched role
 211  *      sCL -> sIV
 212  */
 213 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 214 /*synack*/ { sIV, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
 215 /*
 216  *      sSS -> sSR      Standard open.
 217  *      sS2 -> sSR      Simultaneous open
 218  *      sSR -> sIG      Retransmitted SYN/ACK, ignore it.
 219  *      sES -> sIG      Late retransmitted SYN/ACK?
 220  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 221  *      sCW -> sIG
 222  *      sLA -> sIG
 223  *      sTW -> sIG
 224  *      sCL -> sIG
 225  */
 226 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 227 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 228 /*
 229  *      sSS -> sIV      Server might not send FIN in this state.
 230  *      sS2 -> sIV
 231  *      sSR -> sFW      Close started.
 232  *      sES -> sFW
 233  *      sFW -> sLA      FIN seen in both directions.
 234  *      sCW -> sLA
 235  *      sLA -> sLA      Retransmitted FIN.
 236  *      sTW -> sTW
 237  *      sCL -> sCL
 238  */
 239 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 240 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIG },
 241 /*
 242  *      sSS -> sIG      Might be a half-open connection.
 243  *      sS2 -> sIG
 244  *      sSR -> sSR      Might answer late resent SYN.
 245  *      sES -> sES      :-)
 246  *      sFW -> sCW      Normal close request answered by ACK.
 247  *      sCW -> sCW
 248  *      sLA -> sTW      Last ACK detected (RFC5961 challenged)
 249  *      sTW -> sTW      Retransmitted last ACK.
 250  *      sCL -> sCL
 251  */
 252 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 253 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
 254 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 255         }
 256 };
 257
 258 #ifdef CONFIG_NF_CONNTRACK_PROCFS
 259 /* Print out the private part of the conntrack. */
 260 static void tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
 261 {
 262         if (test_bit(IPS_OFFLOAD_BIT, &ct->status))
 263                 return;
 264
 265         seq_printf(s, "%s ", tcp_conntrack_names[ct->proto.tcp.state]);
 266 }
 267 #endif
 268
 269 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 270 {
 271         if (tcph->rst) return TCP_RST_SET;
 272         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 273         else if (tcph->fin) return TCP_FIN_SET;
 274         else if (tcph->ack) return TCP_ACK_SET;
 275         else return TCP_NONE_SET;
 276 }
 277
 278 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 279    in IP Filter' by Guido van Rooij.
 280
 281    http://www.sane.nl/events/sane2000/papers.html
 282    http://www.darkart.com/mirrors/www.obfuscation.org/ipf/
 283
 284    The boundaries and the conditions are changed according to RFC793:
 285    the packet must intersect the window (i.e. segments may be
 286    after the right or before the left edge) and thus receivers may ACK
 287    segments after the right edge of the window.
 288
 289         td_maxend = max(sack + max(win,1)) seen in reply packets
 290         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 291         td_maxwin += seq + len - sender.td_maxend
 292                         if seq + len > sender.td_maxend
 293         td_end    = max(seq + len) seen in sent packets
 294
 295    I.   Upper bound for valid data:     seq <= sender.td_maxend
 296    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 297    III. Upper bound for valid (s)ack:   sack <= receiver.td_end
 298    IV.  Lower bound for valid (s)ack:   sack >= receiver.td_end - MAXACKWINDOW
 299
 300    where sack is the highest right edge of sack block found in the packet
 301    or ack in the case of packet without SACK option.
 302
 303    The upper bound limit for a valid (s)ack is not ignored -
 304    we doesn't have to deal with fragments.
 305 */
 306
 307 static inline __u32 segment_seq_plus_len(__u32 seq,
 308                                          size_t len,
 309                                          unsigned int dataoff,
 310                                          const struct tcphdr *tcph)
 311 {
 312         /* XXX Should I use payload length field in IP/IPv6 header ?
 313          * - YK */
 314         return (seq + len - dataoff - tcph->doff*4
 315                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 316 }
 317
 318 /* Fixme: what about big packets? */
 319 #define MAXACKWINCONST                  66000
 320 #define MAXACKWINDOW(sender)                                            \
 321         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 322                                               : MAXACKWINCONST)
 323
 324 /*
 325  * Simplified tcp_parse_options routine from tcp_input.c
 326  */
 327 static void tcp_options(const struct sk_buff *skb,
 328                         unsigned int dataoff,
 329                         const struct tcphdr *tcph,
 330                         struct ip_ct_tcp_state *state)
 331 {
 332         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 333         const unsigned char *ptr;
 334         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 335
 336         if (!length)
 337                 return;
 338
 339         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 340                                  length, buff);
 341         if (!ptr)
 342                 return;
 343
 344         state->td_scale =
 345         state->flags = 0;
 346
 347         while (length > 0) {
 348                 int opcode=*ptr++;
 349                 int opsize;
 350
 351                 switch (opcode) {
 352                 case TCPOPT_EOL:
 353                         return;
 354                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 355                         length--;
 356                         continue;
 357                 default:
 358                         if (length < 2)
 359                                 return;
 360                         opsize=*ptr++;
 361                         if (opsize < 2) /* "silly options" */
 362                                 return;
 363                         if (opsize > length)
 364                                 return; /* don't parse partial options */
 365
 366                         if (opcode == TCPOPT_SACK_PERM
 367                             && opsize == TCPOLEN_SACK_PERM)
 368                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 369                         else if (opcode == TCPOPT_WINDOW
 370                                  && opsize == TCPOLEN_WINDOW) {
 371                                 state->td_scale = *(u_int8_t *)ptr;
 372
 373                                 if (state->td_scale > TCP_MAX_WSCALE)
 374                                         state->td_scale = TCP_MAX_WSCALE;
 375
 376                                 state->flags |=
 377                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 378                         }
 379                         ptr += opsize - 2;
 380                         length -= opsize;
 381                 }
 382         }
 383 }
 384
 385 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 386                      const struct tcphdr *tcph, __u32 *sack)
 387 {
 388         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 389         const unsigned char *ptr;
 390         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 391         __u32 tmp;
 392
 393         if (!length)
 394                 return;
 395
 396         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 397                                  length, buff);
 398         if (!ptr)
 399                 return;
 400
 401         /* Fast path for timestamp-only option */
 402         if (length == TCPOLEN_TSTAMP_ALIGNED
 403             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
 404                                        | (TCPOPT_NOP << 16)
 405                                        | (TCPOPT_TIMESTAMP << 8)
 406                                        | TCPOLEN_TIMESTAMP))
 407                 return;
 408
 409         while (length > 0) {
 410                 int opcode = *ptr++;
 411                 int opsize, i;
 412
 413                 switch (opcode) {
 414                 case TCPOPT_EOL:
 415                         return;
 416                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 417                         length--;
 418                         continue;
 419                 default:
 420                         if (length < 2)
 421                                 return;
 422                         opsize = *ptr++;
 423                         if (opsize < 2) /* "silly options" */
 424                                 return;
 425                         if (opsize > length)
 426                                 return; /* don't parse partial options */
 427
 428                         if (opcode == TCPOPT_SACK
 429                             && opsize >= (TCPOLEN_SACK_BASE
 430                                           + TCPOLEN_SACK_PERBLOCK)
 431                             && !((opsize - TCPOLEN_SACK_BASE)
 432                                  % TCPOLEN_SACK_PERBLOCK)) {
 433                                 for (i = 0;
 434                                      i < (opsize - TCPOLEN_SACK_BASE);
 435                                      i += TCPOLEN_SACK_PERBLOCK) {
 436                                         tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
 437
 438                                         if (after(tmp, *sack))
 439                                                 *sack = tmp;
 440                                 }
 441                                 return;
 442                         }
 443                         ptr += opsize - 2;
 444                         length -= opsize;
 445                 }
 446         }
 447 }
 448
 449 static bool tcp_in_window(const struct nf_conn *ct,
 450                           struct ip_ct_tcp *state,
 451                           enum ip_conntrack_dir dir,
 452                           unsigned int index,
 453                           const struct sk_buff *skb,
 454                           unsigned int dataoff,
 455                           const struct tcphdr *tcph)
 456 {
 457         struct net *net = nf_ct_net(ct);
 458         struct nf_tcp_net *tn = nf_tcp_pernet(net);
 459         struct ip_ct_tcp_state *sender = &state->seen[dir];
 460         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 461         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 462         __u32 seq, ack, sack, end, win, swin;
 463         u16 win_raw;
 464         s32 receiver_offset;
 465         bool res, in_recv_win;
 466
 467         /*
 468          * Get the required data from the packet.
 469          */
 470         seq = ntohl(tcph->seq);
 471         ack = sack = ntohl(tcph->ack_seq);
 472         win_raw = ntohs(tcph->window);
 473         win = win_raw;
 474         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 475
 476         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 477                 tcp_sack(skb, dataoff, tcph, &sack);
 478
 479         /* Take into account NAT sequence number mangling */
 480         receiver_offset = nf_ct_seq_offset(ct, !dir, ack - 1);
 481         ack -= receiver_offset;
 482         sack -= receiver_offset;
 483
 484         pr_debug("tcp_in_window: START\n");
 485         pr_debug("tcp_in_window: ");
 486         nf_ct_dump_tuple(tuple);
 487         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
 488                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
 489         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 490                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 491                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 492                  sender->td_scale,
 493                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 494                  receiver->td_scale);
 495
 496         if (sender->td_maxwin == 0) {
 497                 /*
 498                  * Initialize sender data.
 499                  */
 500                 if (tcph->syn) {
 501                         /*
 502                          * SYN-ACK in reply to a SYN
 503                          * or SYN from reply direction in simultaneous open.
 504                          */
 505                         sender->td_end =
 506                         sender->td_maxend = end;
 507                         sender->td_maxwin = (win == 0 ? 1 : win);
 508
 509                         tcp_options(skb, dataoff, tcph, sender);
 510                         /*
 511                          * RFC 1323:
 512                          * Both sides must send the Window Scale option
 513                          * to enable window scaling in either direction.
 514                          */
 515                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 516                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 517                                 sender->td_scale =
 518                                 receiver->td_scale = 0;
 519                         if (!tcph->ack)
 520                                 /* Simultaneous open */
 521                                 return true;
 522                 } else {
 523                         /*
 524                          * We are in the middle of a connection,
 525                          * its history is lost for us.
 526                          * Let's try to use the data from the packet.
 527                          */
 528                         sender->td_end = end;
 529                         swin = win << sender->td_scale;
 530                         sender->td_maxwin = (swin == 0 ? 1 : swin);
 531                         sender->td_maxend = end + sender->td_maxwin;
 532                         if (receiver->td_maxwin == 0) {
 533                                 /* We haven't seen traffic in the other
 534                                  * direction yet but we have to tweak window
 535                                  * tracking to pass III and IV until that
 536                                  * happens.
 537                                  */
 538                                 receiver->td_end = receiver->td_maxend = sack;
 539                         } else if (sack == receiver->td_end + 1) {
 540                                 /* Likely a reply to a keepalive.
 541                                  * Needed for III.
 542                                  */
 543                                 receiver->td_end++;
 544                         }
 545
 546                 }
 547         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 548                      && dir == IP_CT_DIR_ORIGINAL)
 549                    || (state->state == TCP_CONNTRACK_SYN_RECV
 550                      && dir == IP_CT_DIR_REPLY))
 551                    && after(end, sender->td_end)) {
 552                 /*
 553                  * RFC 793: "if a TCP is reinitialized ... then it need
 554                  * not wait at all; it must only be sure to use sequence
 555                  * numbers larger than those recently used."
 556                  */
 557                 sender->td_end =
 558                 sender->td_maxend = end;
 559                 sender->td_maxwin = (win == 0 ? 1 : win);
 560
 561                 tcp_options(skb, dataoff, tcph, sender);
 562         }
 563
 564         if (!(tcph->ack)) {
 565                 /*
 566                  * If there is no ACK, just pretend it was set and OK.
 567                  */
 568                 ack = sack = receiver->td_end;
 569         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 570                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 571                    && (ack == 0)) {
 572                 /*
 573                  * Broken TCP stacks, that set ACK in RST packets as well
 574                  * with zero ack value.
 575                  */
 576                 ack = sack = receiver->td_end;
 577         }
 578
 579         if (tcph->rst && seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)
 580                 /*
 581                  * RST sent answering SYN.
 582                  */
 583                 seq = end = sender->td_end;
 584
 585         pr_debug("tcp_in_window: ");
 586         nf_ct_dump_tuple(tuple);
 587         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
 588                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
 589         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 590                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 591                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 592                  sender->td_scale,
 593                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 594                  receiver->td_scale);
 595
 596         /* Is the ending sequence in the receive window (if available)? */
 597         in_recv_win = !receiver->td_maxwin ||
 598                       after(end, sender->td_end - receiver->td_maxwin - 1);
 599
 600         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 601                  before(seq, sender->td_maxend + 1),
 602                  (in_recv_win ? 1 : 0),
 603                  before(sack, receiver->td_end + 1),
 604                  after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 605
 606         if (before(seq, sender->td_maxend + 1) &&
 607             in_recv_win &&
 608             before(sack, receiver->td_end + 1) &&
 609             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
 610                 /*
 611                  * Take into account window scaling (RFC 1323).
 612                  */
 613                 if (!tcph->syn)
 614                         win <<= sender->td_scale;
 615
 616                 /*
 617                  * Update sender data.
 618                  */
 619                 swin = win + (sack - ack);
 620                 if (sender->td_maxwin < swin)
 621                         sender->td_maxwin = swin;
 622                 if (after(end, sender->td_end)) {
 623                         sender->td_end = end;
 624                         sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 625                 }
 626                 if (tcph->ack) {
 627                         if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
 628                                 sender->td_maxack = ack;
 629                                 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
 630                         } else if (after(ack, sender->td_maxack))
 631                                 sender->td_maxack = ack;
 632                 }
 633
 634                 /*
 635                  * Update receiver data.
 636                  */
 637                 if (receiver->td_maxwin != 0 && after(end, sender->td_maxend))
 638                         receiver->td_maxwin += end - sender->td_maxend;
 639                 if (after(sack + win, receiver->td_maxend - 1)) {
 640                         receiver->td_maxend = sack + win;
 641                         if (win == 0)
 642                                 receiver->td_maxend++;
 643                 }
 644                 if (ack == receiver->td_end)
 645                         receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 646
 647                 /*
 648                  * Check retransmissions.
 649                  */
 650                 if (index == TCP_ACK_SET) {
 651                         if (state->last_dir == dir
 652                             && state->last_seq == seq
 653                             && state->last_ack == ack
 654                             && state->last_end == end
 655                             && state->last_win == win_raw)
 656                                 state->retrans++;
 657                         else {
 658                                 state->last_dir = dir;
 659                                 state->last_seq = seq;
 660                                 state->last_ack = ack;
 661                                 state->last_end = end;
 662                                 state->last_win = win_raw;
 663                                 state->retrans = 0;
 664                         }
 665                 }
 666                 res = true;
 667         } else {
 668                 res = false;
 669                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 670                     tn->tcp_be_liberal)
 671                         res = true;
 672                 if (!res) {
 673                         nf_ct_l4proto_log_invalid(skb, ct,
 674                         "%s",
 675                         before(seq, sender->td_maxend + 1) ?
 676                         in_recv_win ?
 677                         before(sack, receiver->td_end + 1) ?
 678                         after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
 679                         : "ACK is under the lower bound (possible overly delayed ACK)"
 680                         : "ACK is over the upper bound (ACKed data not seen yet)"
 681                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 682                         : "SEQ is over the upper bound (over the window of the receiver)");
 683                 }
 684         }
 685
 686         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 687                  "receiver end=%u maxend=%u maxwin=%u\n",
 688                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 689                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 690
 691         return res;
 692 }
 693
 694 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 695 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
 696                                  TCPHDR_URG) + 1] =
 697 {
 698         [TCPHDR_SYN]                            = 1,
 699         [TCPHDR_SYN|TCPHDR_URG]                 = 1,
 700         [TCPHDR_SYN|TCPHDR_ACK]                 = 1,
 701         [TCPHDR_RST]                            = 1,
 702         [TCPHDR_RST|TCPHDR_ACK]                 = 1,
 703         [TCPHDR_FIN|TCPHDR_ACK]                 = 1,
 704         [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG]      = 1,
 705         [TCPHDR_ACK]                            = 1,
 706         [TCPHDR_ACK|TCPHDR_URG]                 = 1,
 707 };
 708
 709 static void tcp_error_log(const struct sk_buff *skb,
 710                           const struct nf_hook_state *state,
 711                           const char *msg)
 712 {
 713         nf_l4proto_log_invalid(skb, state->net, state->pf, IPPROTO_TCP, "%s", msg);
 714 }
 715
 716 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 717 static bool tcp_error(const struct tcphdr *th,
 718                       struct sk_buff *skb,
 719                       unsigned int dataoff,
 720                       const struct nf_hook_state *state)
 721 {
 722         unsigned int tcplen = skb->len - dataoff;
 723         u8 tcpflags;
 724
 725         /* Not whole TCP header or malformed packet */
 726         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 727                 tcp_error_log(skb, state, "truncated packet");
 728                 return true;
 729         }
 730
 731         /* Checksum invalid? Ignore.
 732          * We skip checking packets on the outgoing path
 733          * because the checksum is assumed to be correct.
 734          */
 735         /* FIXME: Source route IP option packets --RR */
 736         if (state->net->ct.sysctl_checksum &&
 737             state->hook == NF_INET_PRE_ROUTING &&
 738             nf_checksum(skb, state->hook, dataoff, IPPROTO_TCP, state->pf)) {
 739                 tcp_error_log(skb, state, "bad checksum");
 740                 return true;
 741         }
 742
 743         /* Check TCP flags. */
 744         tcpflags = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
 745         if (!tcp_valid_flags[tcpflags]) {
 746                 tcp_error_log(skb, state, "invalid tcp flag combination");
 747                 return true;
 748         }
 749
 750         return false;
 751 }
 752
 753 static noinline bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
 754                              unsigned int dataoff,
 755                              const struct tcphdr *th)
 756 {
 757         enum tcp_conntrack new_state;
 758         struct net *net = nf_ct_net(ct);
 759         const struct nf_tcp_net *tn = nf_tcp_pernet(net);
 760         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
 761         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
 762
 763         /* Don't need lock here: this conntrack not in circulation yet */
 764         new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
 765
 766         /* Invalid: delete conntrack */
 767         if (new_state >= TCP_CONNTRACK_MAX) {
 768                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
 769                 return false;
 770         }
 771
 772         if (new_state == TCP_CONNTRACK_SYN_SENT) {
 773                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
 774                 /* SYN packet */
 775                 ct->proto.tcp.seen[0].td_end =
 776                         segment_seq_plus_len(ntohl(th->seq), skb->len,
 777                                              dataoff, th);
 778                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
 779                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
 780                         ct->proto.tcp.seen[0].td_maxwin = 1;
 781                 ct->proto.tcp.seen[0].td_maxend =
 782                         ct->proto.tcp.seen[0].td_end;
 783
 784                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
 785         } else if (tn->tcp_loose == 0) {
 786                 /* Don't try to pick up connections. */
 787                 return false;
 788         } else {
 789                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
 790                 /*
 791                  * We are in the middle of a connection,
 792                  * its history is lost for us.
 793                  * Let's try to use the data from the packet.
 794                  */
 795                 ct->proto.tcp.seen[0].td_end =
 796                         segment_seq_plus_len(ntohl(th->seq), skb->len,
 797                                              dataoff, th);
 798                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
 799                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
 800                         ct->proto.tcp.seen[0].td_maxwin = 1;
 801                 ct->proto.tcp.seen[0].td_maxend =
 802                         ct->proto.tcp.seen[0].td_end +
 803                         ct->proto.tcp.seen[0].td_maxwin;
 804
 805                 /* We assume SACK and liberal window checking to handle
 806                  * window scaling */
 807                 ct->proto.tcp.seen[0].flags =
 808                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
 809                                               IP_CT_TCP_FLAG_BE_LIBERAL;
 810         }
 811
 812         /* tcp_packet will set them */
 813         ct->proto.tcp.last_index = TCP_NONE_SET;
 814
 815         pr_debug("%s: sender end=%u maxend=%u maxwin=%u scale=%i "
 816                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 817                  __func__,
 818                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 819                  sender->td_scale,
 820                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 821                  receiver->td_scale);
 822         return true;
 823 }
 824
 825 /* Returns verdict for packet, or -1 for invalid. */
 826 int nf_conntrack_tcp_packet(struct nf_conn *ct,
 827                             struct sk_buff *skb,
 828                             unsigned int dataoff,
 829                             enum ip_conntrack_info ctinfo,
 830                             const struct nf_hook_state *state)
 831 {
 832         struct net *net = nf_ct_net(ct);
 833         struct nf_tcp_net *tn = nf_tcp_pernet(net);
 834         struct nf_conntrack_tuple *tuple;
 835         enum tcp_conntrack new_state, old_state;
 836         unsigned int index, *timeouts;
 837         enum ip_conntrack_dir dir;
 838         const struct tcphdr *th;
 839         struct tcphdr _tcph;
 840         unsigned long timeout;
 841
 842         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 843         if (th == NULL)
 844                 return -NF_ACCEPT;
 845
 846         if (tcp_error(th, skb, dataoff, state))
 847                 return -NF_ACCEPT;
 848
 849         if (!nf_ct_is_confirmed(ct) && !tcp_new(ct, skb, dataoff, th))
 850                 return -NF_ACCEPT;
 851
 852         spin_lock_bh(&ct->lock);
 853         old_state = ct->proto.tcp.state;
 854         dir = CTINFO2DIR(ctinfo);
 855         index = get_conntrack_index(th);
 856         new_state = tcp_conntracks[dir][index][old_state];
 857         tuple = &ct->tuplehash[dir].tuple;
 858
 859         switch (new_state) {
 860         case TCP_CONNTRACK_SYN_SENT:
 861                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 862                         break;
 863                 /* RFC 1122: "When a connection is closed actively,
 864                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 865                  * (Maximum Segment Lifetime). However, it MAY accept
 866                  * a new SYN from the remote TCP to reopen the connection
 867                  * directly from TIME-WAIT state, if..."
 868                  * We ignore the conditions because we are in the
 869                  * TIME-WAIT state anyway.
 870                  *
 871                  * Handle aborted connections: we and the server
 872                  * think there is an existing connection but the client
 873                  * aborts it and starts a new one.
 874                  */
 875                 if (((ct->proto.tcp.seen[dir].flags
 876                       | ct->proto.tcp.seen[!dir].flags)
 877                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 878                     || (ct->proto.tcp.last_dir == dir
 879                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 880                         /* Attempt to reopen a closed/aborted connection.
 881                          * Delete this connection and look up again. */
 882                         spin_unlock_bh(&ct->lock);
 883
 884                         /* Only repeat if we can actually remove the timer.
 885                          * Destruction may already be in progress in process
 886                          * context and we must give it a chance to terminate.
 887                          */
 888                         if (nf_ct_kill(ct))
 889                                 return -NF_REPEAT;
 890                         return NF_DROP;
 891                 }
 892                 fallthrough;
 893         case TCP_CONNTRACK_IGNORE:
 894                 /* Ignored packets:
 895                  *
 896                  * Our connection entry may be out of sync, so ignore
 897                  * packets which may signal the real connection between
 898                  * the client and the server.
 899                  *
 900                  * a) SYN in ORIGINAL
 901                  * b) SYN/ACK in REPLY
 902                  * c) ACK in reply direction after initial SYN in original.
 903                  *
 904                  * If the ignored packet is invalid, the receiver will send
 905                  * a RST we'll catch below.
 906                  */
 907                 if (index == TCP_SYNACK_SET
 908                     && ct->proto.tcp.last_index == TCP_SYN_SET
 909                     && ct->proto.tcp.last_dir != dir
 910                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 911                         /* b) This SYN/ACK acknowledges a SYN that we earlier
 912                          * ignored as invalid. This means that the client and
 913                          * the server are both in sync, while the firewall is
 914                          * not. We get in sync from the previously annotated
 915                          * values.
 916                          */
 917                         old_state = TCP_CONNTRACK_SYN_SENT;
 918                         new_state = TCP_CONNTRACK_SYN_RECV;
 919                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
 920                                 ct->proto.tcp.last_end;
 921                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
 922                                 ct->proto.tcp.last_end;
 923                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
 924                                 ct->proto.tcp.last_win == 0 ?
 925                                         1 : ct->proto.tcp.last_win;
 926                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
 927                                 ct->proto.tcp.last_wscale;
 928                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
 929                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
 930                                 ct->proto.tcp.last_flags;
 931                         memset(&ct->proto.tcp.seen[dir], 0,
 932                                sizeof(struct ip_ct_tcp_state));
 933                         break;
 934                 }
 935                 ct->proto.tcp.last_index = index;
 936                 ct->proto.tcp.last_dir = dir;
 937                 ct->proto.tcp.last_seq = ntohl(th->seq);
 938                 ct->proto.tcp.last_end =
 939                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 940                 ct->proto.tcp.last_win = ntohs(th->window);
 941
 942                 /* a) This is a SYN in ORIGINAL. The client and the server
 943                  * may be in sync but we are not. In that case, we annotate
 944                  * the TCP options and let the packet go through. If it is a
 945                  * valid SYN packet, the server will reply with a SYN/ACK, and
 946                  * then we'll get in sync. Otherwise, the server potentially
 947                  * responds with a challenge ACK if implementing RFC5961.
 948                  */
 949                 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
 950                         struct ip_ct_tcp_state seen = {};
 951
 952                         ct->proto.tcp.last_flags =
 953                         ct->proto.tcp.last_wscale = 0;
 954                         tcp_options(skb, dataoff, th, &seen);
 955                         if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
 956                                 ct->proto.tcp.last_flags |=
 957                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 958                                 ct->proto.tcp.last_wscale = seen.td_scale;
 959                         }
 960                         if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
 961                                 ct->proto.tcp.last_flags |=
 962                                         IP_CT_TCP_FLAG_SACK_PERM;
 963                         }
 964                         /* Mark the potential for RFC5961 challenge ACK,
 965                          * this pose a special problem for LAST_ACK state
 966                          * as ACK is intrepretated as ACKing last FIN.
 967                          */
 968                         if (old_state == TCP_CONNTRACK_LAST_ACK)
 969                                 ct->proto.tcp.last_flags |=
 970                                         IP_CT_EXP_CHALLENGE_ACK;
 971                 }
 972                 spin_unlock_bh(&ct->lock);
 973                 nf_ct_l4proto_log_invalid(skb, ct,
 974                                           "packet (index %d) in dir %d ignored, state %s",
 975                                           index, dir,
 976                                           tcp_conntrack_names[old_state]);
 977                 return NF_ACCEPT;
 978         case TCP_CONNTRACK_MAX:
 979                 /* Special case for SYN proxy: when the SYN to the server or
 980                  * the SYN/ACK from the server is lost, the client may transmit
 981                  * a keep-alive packet while in SYN_SENT state. This needs to
 982                  * be associated with the original conntrack entry in order to
 983                  * generate a new SYN with the correct sequence number.
 984                  */
 985                 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
 986                     index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
 987                     ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
 988                     ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
 989                         pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
 990                         spin_unlock_bh(&ct->lock);
 991                         return NF_ACCEPT;
 992                 }
 993
 994                 /* Invalid packet */
 995                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 996                          dir, get_conntrack_index(th), old_state);
 997                 spin_unlock_bh(&ct->lock);
 998                 nf_ct_l4proto_log_invalid(skb, ct, "invalid state");
 999                 return -NF_ACCEPT;
1000         case TCP_CONNTRACK_TIME_WAIT:
1001                 /* RFC5961 compliance cause stack to send "challenge-ACK"
1002                  * e.g. in response to spurious SYNs.  Conntrack MUST
1003                  * not believe this ACK is acking last FIN.
1004                  */
1005                 if (old_state == TCP_CONNTRACK_LAST_ACK &&
1006                     index == TCP_ACK_SET &&
1007                     ct->proto.tcp.last_dir != dir &&
1008                     ct->proto.tcp.last_index == TCP_SYN_SET &&
1009                     (ct->proto.tcp.last_flags & IP_CT_EXP_CHALLENGE_ACK)) {
1010                         /* Detected RFC5961 challenge ACK */
1011                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
1012                         spin_unlock_bh(&ct->lock);
1013                         nf_ct_l4proto_log_invalid(skb, ct, "challenge-ack ignored");
1014                         return NF_ACCEPT; /* Don't change state */
1015                 }
1016                 break;
1017         case TCP_CONNTRACK_SYN_SENT2:
1018                 /* tcp_conntracks table is not smart enough to handle
1019                  * simultaneous open.
1020                  */
1021                 ct->proto.tcp.last_flags |= IP_CT_TCP_SIMULTANEOUS_OPEN;
1022                 break;
1023         case TCP_CONNTRACK_SYN_RECV:
1024                 if (dir == IP_CT_DIR_REPLY && index == TCP_ACK_SET &&
1025                     ct->proto.tcp.last_flags & IP_CT_TCP_SIMULTANEOUS_OPEN)
1026                         new_state = TCP_CONNTRACK_ESTABLISHED;
1027                 break;
1028         case TCP_CONNTRACK_CLOSE:
1029                 if (index != TCP_RST_SET)
1030                         break;
1031
1032                 if (ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET) {
1033                         u32 seq = ntohl(th->seq);
1034
1035                         if (before(seq, ct->proto.tcp.seen[!dir].td_maxack)) {
1036                                 /* Invalid RST  */
1037                                 spin_unlock_bh(&ct->lock);
1038                                 nf_ct_l4proto_log_invalid(skb, ct, "invalid rst");
1039                                 return -NF_ACCEPT;
1040                         }
1041
1042                         if (!nf_conntrack_tcp_established(ct) ||
1043                             seq == ct->proto.tcp.seen[!dir].td_maxack)
1044                                 break;
1045
1046                         /* Check if rst is part of train, such as
1047                          *   foo:80 > bar:4379: P, 235946583:235946602(19) ack 42
1048                          *   foo:80 > bar:4379: R, 235946602:235946602(0)  ack 42
1049                          */
1050                         if (ct->proto.tcp.last_index == TCP_ACK_SET &&
1051                             ct->proto.tcp.last_dir == dir &&
1052                             seq == ct->proto.tcp.last_end)
1053                                 break;
1054
1055                         /* ... RST sequence number doesn't match exactly, keep
1056                          * established state to allow a possible challenge ACK.
1057                          */
1058                         new_state = old_state;
1059                 }
1060                 if (((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
1061                          && ct->proto.tcp.last_index == TCP_SYN_SET)
1062                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
1063                             && ct->proto.tcp.last_index == TCP_ACK_SET))
1064                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
1065                         /* RST sent to invalid SYN or ACK we had let through
1066                          * at a) and c) above:
1067                          *
1068                          * a) SYN was in window then
1069                          * c) we hold a half-open connection.
1070                          *
1071                          * Delete our connection entry.
1072                          * We skip window checking, because packet might ACK
1073                          * segments we ignored. */
1074                         goto in_window;
1075                 }
1076                 break;
1077         default:
1078                 /* Keep compilers happy. */
1079                 break;
1080         }
1081
1082         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
1083                            skb, dataoff, th)) {
1084                 spin_unlock_bh(&ct->lock);
1085                 return -NF_ACCEPT;
1086         }
1087      in_window:
1088         /* From now on we have got in-window packets */
1089         ct->proto.tcp.last_index = index;
1090         ct->proto.tcp.last_dir = dir;
1091
1092         pr_debug("tcp_conntracks: ");
1093         nf_ct_dump_tuple(tuple);
1094         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1095                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1096                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1097                  old_state, new_state);
1098
1099         ct->proto.tcp.state = new_state;
1100         if (old_state != new_state
1101             && new_state == TCP_CONNTRACK_FIN_WAIT)
1102                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1103
1104         timeouts = nf_ct_timeout_lookup(ct);
1105         if (!timeouts)
1106                 timeouts = tn->timeouts;
1107
1108         if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1109             timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1110                 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1111         else if (unlikely(index == TCP_RST_SET))
1112                 timeout = timeouts[TCP_CONNTRACK_CLOSE];
1113         else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1114                  IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1115                  timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1116                 timeout = timeouts[TCP_CONNTRACK_UNACK];
1117         else if (ct->proto.tcp.last_win == 0 &&
1118                  timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1119                 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1120         else
1121                 timeout = timeouts[new_state];
1122         spin_unlock_bh(&ct->lock);
1123
1124         if (new_state != old_state)
1125                 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1126
1127         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1128                 /* If only reply is a RST, we can consider ourselves not to
1129                    have an established connection: this is a fairly common
1130                    problem case, so we can delete the conntrack
1131                    immediately.  --RR */
1132                 if (th->rst) {
1133                         nf_ct_kill_acct(ct, ctinfo, skb);
1134                         return NF_ACCEPT;
1135                 }
1136                 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1137                  * pickup with loose=1. Avoid large ESTABLISHED timeout.
1138                  */
1139                 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1140                     timeout > timeouts[TCP_CONNTRACK_UNACK])
1141                         timeout = timeouts[TCP_CONNTRACK_UNACK];
1142         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1143                    && (old_state == TCP_CONNTRACK_SYN_RECV
1144                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1145                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1146                 /* Set ASSURED if we see valid ack in ESTABLISHED
1147                    after SYN_RECV or a valid answer for a picked up
1148                    connection. */
1149                 set_bit(IPS_ASSURED_BIT, &ct->status);
1150                 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1151         }
1152         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1153
1154         return NF_ACCEPT;
1155 }
1156
1157 static bool tcp_can_early_drop(const struct nf_conn *ct)
1158 {
1159         switch (ct->proto.tcp.state) {
1160         case TCP_CONNTRACK_FIN_WAIT:
1161         case TCP_CONNTRACK_LAST_ACK:
1162         case TCP_CONNTRACK_TIME_WAIT:
1163         case TCP_CONNTRACK_CLOSE:
1164         case TCP_CONNTRACK_CLOSE_WAIT:
1165                 return true;
1166         default:
1167                 break;
1168         }
1169
1170         return false;
1171 }
1172
1173 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1174
1175 #include <linux/netfilter/nfnetlink.h>
1176 #include <linux/netfilter/nfnetlink_conntrack.h>
1177
1178 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1179                          struct nf_conn *ct, bool destroy)
1180 {
1181         struct nlattr *nest_parms;
1182         struct nf_ct_tcp_flags tmp = {};
1183
1184         spin_lock_bh(&ct->lock);
1185         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP);
1186         if (!nest_parms)
1187                 goto nla_put_failure;
1188
1189         if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state))
1190                 goto nla_put_failure;
1191
1192         if (destroy)
1193                 goto skip_state;
1194
1195         if (nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1196                        ct->proto.tcp.seen[0].td_scale) ||
1197             nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1198                        ct->proto.tcp.seen[1].td_scale))
1199                 goto nla_put_failure;
1200
1201         tmp.flags = ct->proto.tcp.seen[0].flags;
1202         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1203                     sizeof(struct nf_ct_tcp_flags), &tmp))
1204                 goto nla_put_failure;
1205
1206         tmp.flags = ct->proto.tcp.seen[1].flags;
1207         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1208                     sizeof(struct nf_ct_tcp_flags), &tmp))
1209                 goto nla_put_failure;
1210 skip_state:
1211         spin_unlock_bh(&ct->lock);
1212         nla_nest_end(skb, nest_parms);
1213
1214         return 0;
1215
1216 nla_put_failure:
1217         spin_unlock_bh(&ct->lock);
1218         return -1;
1219 }
1220
1221 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1222         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1223         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1224         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1225         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1226         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len = sizeof(struct nf_ct_tcp_flags) },
1227 };
1228
1229 #define TCP_NLATTR_SIZE ( \
1230         NLA_ALIGN(NLA_HDRLEN + 1) + \
1231         NLA_ALIGN(NLA_HDRLEN + 1) + \
1232         NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)) + \
1233         NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)))
1234
1235 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1236 {
1237         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1238         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1239         int err;
1240
1241         /* updates could not contain anything about the private
1242          * protocol info, in that case skip the parsing */
1243         if (!pattr)
1244                 return 0;
1245
1246         err = nla_parse_nested_deprecated(tb, CTA_PROTOINFO_TCP_MAX, pattr,
1247                                           tcp_nla_policy, NULL);
1248         if (err < 0)
1249                 return err;
1250
1251         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1252             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1253                 return -EINVAL;
1254
1255         spin_lock_bh(&ct->lock);
1256         if (tb[CTA_PROTOINFO_TCP_STATE])
1257                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1258
1259         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1260                 struct nf_ct_tcp_flags *attr =
1261                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1262                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1263                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1264         }
1265
1266         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1267                 struct nf_ct_tcp_flags *attr =
1268                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1269                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1270                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1271         }
1272
1273         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1274             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1275             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1276             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1277                 ct->proto.tcp.seen[0].td_scale =
1278                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1279                 ct->proto.tcp.seen[1].td_scale =
1280                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1281         }
1282         spin_unlock_bh(&ct->lock);
1283
1284         return 0;
1285 }
1286
1287 static unsigned int tcp_nlattr_tuple_size(void)
1288 {
1289         static unsigned int size __read_mostly;
1290
1291         if (!size)
1292                 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1293
1294         return size;
1295 }
1296 #endif
1297
1298 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1299
1300 #include <linux/netfilter/nfnetlink.h>
1301 #include <linux/netfilter/nfnetlink_cttimeout.h>
1302
1303 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1304                                      struct net *net, void *data)
1305 {
1306         struct nf_tcp_net *tn = nf_tcp_pernet(net);
1307         unsigned int *timeouts = data;
1308         int i;
1309
1310         if (!timeouts)
1311                 timeouts = tn->timeouts;
1312         /* set default TCP timeouts. */
1313         for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1314                 timeouts[i] = tn->timeouts[i];
1315
1316         if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1317                 timeouts[TCP_CONNTRACK_SYN_SENT] =
1318                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1319         }
1320
1321         if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1322                 timeouts[TCP_CONNTRACK_SYN_RECV] =
1323                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1324         }
1325         if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1326                 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1327                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1328         }
1329         if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1330                 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1331                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1332         }
1333         if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1334                 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1335                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1336         }
1337         if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1338                 timeouts[TCP_CONNTRACK_LAST_ACK] =
1339                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1340         }
1341         if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1342                 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1343                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1344         }
1345         if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1346                 timeouts[TCP_CONNTRACK_CLOSE] =
1347                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1348         }
1349         if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1350                 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1351                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1352         }
1353         if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1354                 timeouts[TCP_CONNTRACK_RETRANS] =
1355                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1356         }
1357         if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1358                 timeouts[TCP_CONNTRACK_UNACK] =
1359                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1360         }
1361
1362         timeouts[CTA_TIMEOUT_TCP_UNSPEC] = timeouts[CTA_TIMEOUT_TCP_SYN_SENT];
1363         return 0;
1364 }
1365
1366 static int
1367 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1368 {
1369         const unsigned int *timeouts = data;
1370
1371         if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1372                         htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1373             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1374                          htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1375             nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1376                          htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1377             nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1378                          htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1379             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1380                          htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1381             nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1382                          htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1383             nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1384                          htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1385             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1386                          htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1387             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1388                          htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1389             nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1390                          htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1391             nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1392                          htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1393                 goto nla_put_failure;
1394         return 0;
1395
1396 nla_put_failure:
1397         return -ENOSPC;
1398 }
1399
1400 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1401         [CTA_TIMEOUT_TCP_SYN_SENT]      = { .type = NLA_U32 },
1402         [CTA_TIMEOUT_TCP_SYN_RECV]      = { .type = NLA_U32 },
1403         [CTA_TIMEOUT_TCP_ESTABLISHED]   = { .type = NLA_U32 },
1404         [CTA_TIMEOUT_TCP_FIN_WAIT]      = { .type = NLA_U32 },
1405         [CTA_TIMEOUT_TCP_CLOSE_WAIT]    = { .type = NLA_U32 },
1406         [CTA_TIMEOUT_TCP_LAST_ACK]      = { .type = NLA_U32 },
1407         [CTA_TIMEOUT_TCP_TIME_WAIT]     = { .type = NLA_U32 },
1408         [CTA_TIMEOUT_TCP_CLOSE]         = { .type = NLA_U32 },
1409         [CTA_TIMEOUT_TCP_SYN_SENT2]     = { .type = NLA_U32 },
1410         [CTA_TIMEOUT_TCP_RETRANS]       = { .type = NLA_U32 },
1411         [CTA_TIMEOUT_TCP_UNACK]         = { .type = NLA_U32 },
1412 };
1413 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1414
1415 void nf_conntrack_tcp_init_net(struct net *net)
1416 {
1417         struct nf_tcp_net *tn = nf_tcp_pernet(net);
1418         int i;
1419
1420         for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1421                 tn->timeouts[i] = tcp_timeouts[i];
1422
1423         /* timeouts[0] is unused, make it same as SYN_SENT so
1424          * ->timeouts[0] contains 'new' timeout, like udp or icmp.
1425          */
1426         tn->timeouts[0] = tcp_timeouts[TCP_CONNTRACK_SYN_SENT];
1427
1428         /* If it is set to zero, we disable picking up already established
1429          * connections.
1430          */
1431         tn->tcp_loose = 1;
1432
1433         /* "Be conservative in what you do,
1434          *  be liberal in what you accept from others."
1435          * If it's non-zero, we mark only out of window RST segments as INVALID.
1436          */
1437         tn->tcp_be_liberal = 0;
1438
1439         /* Max number of the retransmitted packets without receiving an (acceptable)
1440          * ACK from the destination. If this number is reached, a shorter timer
1441          * will be started.
1442          */
1443         tn->tcp_max_retrans = 3;
1444 }
1445
1446 const struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp =
1447 {
1448         .l4proto                = IPPROTO_TCP,
1449 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1450         .print_conntrack        = tcp_print_conntrack,
1451 #endif
1452         .can_early_drop         = tcp_can_early_drop,
1453 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1454         .to_nlattr              = tcp_to_nlattr,
1455         .from_nlattr            = nlattr_to_tcp,
1456         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1457         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1458         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1459         .nlattr_size            = TCP_NLATTR_SIZE,
1460         .nla_policy             = nf_ct_port_nla_policy,
1461 #endif
1462 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1463         .ctnl_timeout           = {
1464                 .nlattr_to_obj  = tcp_timeout_nlattr_to_obj,
1465                 .obj_to_nlattr  = tcp_timeout_obj_to_nlattr,
1466                 .nlattr_max     = CTA_TIMEOUT_TCP_MAX,
1467                 .obj_size       = sizeof(unsigned int) *
1468                                         TCP_CONNTRACK_TIMEOUT_MAX,
1469                 .nla_policy     = tcp_timeout_nla_policy,
1470         },
1471 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1472 };