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