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                         /*
 545                          * We haven't seen traffic in the other direction yet
 546                          * but we have to tweak window tracking to pass III
 547                          * and IV until that happens.
 548                          */
 549                         if (receiver->td_maxwin == 0)
 550                                 receiver->td_end = receiver->td_maxend = sack;
 551                 }
 552         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 553                      && dir == IP_CT_DIR_ORIGINAL)
 554                    || (state->state == TCP_CONNTRACK_SYN_RECV
 555                      && dir == IP_CT_DIR_REPLY))
 556                    && after(end, sender->td_end)) {
 557                 /*
 558                  * RFC 793: "if a TCP is reinitialized ... then it need
 559                  * not wait at all; it must only be sure to use sequence
 560                  * numbers larger than those recently used."
 561                  */
 562                 sender->td_end =
 563                 sender->td_maxend = end;
 564                 sender->td_maxwin = (win == 0 ? 1 : win);
 565
 566                 tcp_options(skb, dataoff, tcph, sender);
 567         }
 568
 569         if (!(tcph->ack)) {
 570                 /*
 571                  * If there is no ACK, just pretend it was set and OK.
 572                  */
 573                 ack = sack = receiver->td_end;
 574         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 575                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 576                    && (ack == 0)) {
 577                 /*
 578                  * Broken TCP stacks, that set ACK in RST packets as well
 579                  * with zero ack value.
 580                  */
 581                 ack = sack = receiver->td_end;
 582         }
 583
 584         if (tcph->rst && seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)
 585                 /*
 586                  * RST sent answering SYN.
 587                  */
 588                 seq = end = sender->td_end;
 589
 590         pr_debug("tcp_in_window: ");
 591         nf_ct_dump_tuple(tuple);
 592         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
 593                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
 594         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 595                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 596                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 597                  sender->td_scale,
 598                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 599                  receiver->td_scale);
 600
 601         /* Is the ending sequence in the receive window (if available)? */
 602         in_recv_win = !receiver->td_maxwin ||
 603                       after(end, sender->td_end - receiver->td_maxwin - 1);
 604
 605         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 606                  before(seq, sender->td_maxend + 1),
 607                  (in_recv_win ? 1 : 0),
 608                  before(sack, receiver->td_end + 1),
 609                  after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 610
 611         if (before(seq, sender->td_maxend + 1) &&
 612             in_recv_win &&
 613             before(sack, receiver->td_end + 1) &&
 614             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
 615                 /*
 616                  * Take into account window scaling (RFC 1323).
 617                  */
 618                 if (!tcph->syn)
 619                         win <<= sender->td_scale;
 620
 621                 /*
 622                  * Update sender data.
 623                  */
 624                 swin = win + (sack - ack);
 625                 if (sender->td_maxwin < swin)
 626                         sender->td_maxwin = swin;
 627                 if (after(end, sender->td_end)) {
 628                         sender->td_end = end;
 629                         sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 630                 }
 631                 if (tcph->ack) {
 632                         if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
 633                                 sender->td_maxack = ack;
 634                                 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
 635                         } else if (after(ack, sender->td_maxack))
 636                                 sender->td_maxack = ack;
 637                 }
 638
 639                 /*
 640                  * Update receiver data.
 641                  */
 642                 if (receiver->td_maxwin != 0 && after(end, sender->td_maxend))
 643                         receiver->td_maxwin += end - sender->td_maxend;
 644                 if (after(sack + win, receiver->td_maxend - 1)) {
 645                         receiver->td_maxend = sack + win;
 646                         if (win == 0)
 647                                 receiver->td_maxend++;
 648                 }
 649                 if (ack == receiver->td_end)
 650                         receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 651
 652                 /*
 653                  * Check retransmissions.
 654                  */
 655                 if (index == TCP_ACK_SET) {
 656                         if (state->last_dir == dir
 657                             && state->last_seq == seq
 658                             && state->last_ack == ack
 659                             && state->last_end == end
 660                             && state->last_win == win_raw)
 661                                 state->retrans++;
 662                         else {
 663                                 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 = 0;
 669                         }
 670                 }
 671                 res = true;
 672         } else {
 673                 res = false;
 674                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 675                     tn->tcp_be_liberal)
 676                         res = true;
 677                 if (!res) {
 678                         nf_ct_l4proto_log_invalid(skb, ct,
 679                         "%s",
 680                         before(seq, sender->td_maxend + 1) ?
 681                         in_recv_win ?
 682                         before(sack, receiver->td_end + 1) ?
 683                         after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
 684                         : "ACK is under the lower bound (possible overly delayed ACK)"
 685                         : "ACK is over the upper bound (ACKed data not seen yet)"
 686                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 687                         : "SEQ is over the upper bound (over the window of the receiver)");
 688                 }
 689         }
 690
 691         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 692                  "receiver end=%u maxend=%u maxwin=%u\n",
 693                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 694                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 695
 696         return res;
 697 }
 698
 699 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 700 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
 701                                  TCPHDR_URG) + 1] =
 702 {
 703         [TCPHDR_SYN]                            = 1,
 704         [TCPHDR_SYN|TCPHDR_URG]                 = 1,
 705         [TCPHDR_SYN|TCPHDR_ACK]                 = 1,
 706         [TCPHDR_RST]                            = 1,
 707         [TCPHDR_RST|TCPHDR_ACK]                 = 1,
 708         [TCPHDR_FIN|TCPHDR_ACK]                 = 1,
 709         [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG]      = 1,
 710         [TCPHDR_ACK]                            = 1,
 711         [TCPHDR_ACK|TCPHDR_URG]                 = 1,
 712 };
 713
 714 static void tcp_error_log(const struct sk_buff *skb,
 715                           const struct nf_hook_state *state,
 716                           const char *msg)
 717 {
 718         nf_l4proto_log_invalid(skb, state->net, state->pf, IPPROTO_TCP, "%s", msg);
 719 }
 720
 721 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 722 static bool tcp_error(const struct tcphdr *th,
 723                       struct sk_buff *skb,
 724                       unsigned int dataoff,
 725                       const struct nf_hook_state *state)
 726 {
 727         unsigned int tcplen = skb->len - dataoff;
 728         u8 tcpflags;
 729
 730         /* Not whole TCP header or malformed packet */
 731         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 732                 tcp_error_log(skb, state, "truncated packet");
 733                 return true;
 734         }
 735
 736         /* Checksum invalid? Ignore.
 737          * We skip checking packets on the outgoing path
 738          * because the checksum is assumed to be correct.
 739          */
 740         /* FIXME: Source route IP option packets --RR */
 741         if (state->net->ct.sysctl_checksum &&
 742             state->hook == NF_INET_PRE_ROUTING &&
 743             nf_checksum(skb, state->hook, dataoff, IPPROTO_TCP, state->pf)) {
 744                 tcp_error_log(skb, state, "bad checksum");
 745                 return true;
 746         }
 747
 748         /* Check TCP flags. */
 749         tcpflags = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
 750         if (!tcp_valid_flags[tcpflags]) {
 751                 tcp_error_log(skb, state, "invalid tcp flag combination");
 752                 return true;
 753         }
 754
 755         return false;
 756 }
 757
 758 static noinline bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
 759                              unsigned int dataoff,
 760                              const struct tcphdr *th)
 761 {
 762         enum tcp_conntrack new_state;
 763         struct net *net = nf_ct_net(ct);
 764         const struct nf_tcp_net *tn = nf_tcp_pernet(net);
 765         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
 766         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
 767
 768         /* Don't need lock here: this conntrack not in circulation yet */
 769         new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
 770
 771         /* Invalid: delete conntrack */
 772         if (new_state >= TCP_CONNTRACK_MAX) {
 773                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
 774                 return false;
 775         }
 776
 777         if (new_state == TCP_CONNTRACK_SYN_SENT) {
 778                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
 779                 /* SYN packet */
 780                 ct->proto.tcp.seen[0].td_end =
 781                         segment_seq_plus_len(ntohl(th->seq), skb->len,
 782                                              dataoff, th);
 783                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
 784                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
 785                         ct->proto.tcp.seen[0].td_maxwin = 1;
 786                 ct->proto.tcp.seen[0].td_maxend =
 787                         ct->proto.tcp.seen[0].td_end;
 788
 789                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
 790         } else if (tn->tcp_loose == 0) {
 791                 /* Don't try to pick up connections. */
 792                 return false;
 793         } else {
 794                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
 795                 /*
 796                  * We are in the middle of a connection,
 797                  * its history is lost for us.
 798                  * Let's try to use the data from the packet.
 799                  */
 800                 ct->proto.tcp.seen[0].td_end =
 801                         segment_seq_plus_len(ntohl(th->seq), skb->len,
 802                                              dataoff, th);
 803                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
 804                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
 805                         ct->proto.tcp.seen[0].td_maxwin = 1;
 806                 ct->proto.tcp.seen[0].td_maxend =
 807                         ct->proto.tcp.seen[0].td_end +
 808                         ct->proto.tcp.seen[0].td_maxwin;
 809
 810                 /* We assume SACK and liberal window checking to handle
 811                  * window scaling */
 812                 ct->proto.tcp.seen[0].flags =
 813                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
 814                                               IP_CT_TCP_FLAG_BE_LIBERAL;
 815         }
 816
 817         /* tcp_packet will set them */
 818         ct->proto.tcp.last_index = TCP_NONE_SET;
 819
 820         pr_debug("%s: sender end=%u maxend=%u maxwin=%u scale=%i "
 821                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 822                  __func__,
 823                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 824                  sender->td_scale,
 825                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 826                  receiver->td_scale);
 827         return true;
 828 }
 829
 830 static bool nf_conntrack_tcp_established(const struct nf_conn *ct)
 831 {
 832         return ct->proto.tcp.state == TCP_CONNTRACK_ESTABLISHED &&
 833                test_bit(IPS_ASSURED_BIT, &ct->status);
 834 }
 835
 836 /* Returns verdict for packet, or -1 for invalid. */
 837 int nf_conntrack_tcp_packet(struct nf_conn *ct,
 838                             struct sk_buff *skb,
 839                             unsigned int dataoff,
 840                             enum ip_conntrack_info ctinfo,
 841                             const struct nf_hook_state *state)
 842 {
 843         struct net *net = nf_ct_net(ct);
 844         struct nf_tcp_net *tn = nf_tcp_pernet(net);
 845         struct nf_conntrack_tuple *tuple;
 846         enum tcp_conntrack new_state, old_state;
 847         unsigned int index, *timeouts;
 848         enum ip_conntrack_dir dir;
 849         const struct tcphdr *th;
 850         struct tcphdr _tcph;
 851         unsigned long timeout;
 852
 853         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 854         if (th == NULL)
 855                 return -NF_ACCEPT;
 856
 857         if (tcp_error(th, skb, dataoff, state))
 858                 return -NF_ACCEPT;
 859
 860         if (!nf_ct_is_confirmed(ct) && !tcp_new(ct, skb, dataoff, th))
 861                 return -NF_ACCEPT;
 862
 863         spin_lock_bh(&ct->lock);
 864         old_state = ct->proto.tcp.state;
 865         dir = CTINFO2DIR(ctinfo);
 866         index = get_conntrack_index(th);
 867         new_state = tcp_conntracks[dir][index][old_state];
 868         tuple = &ct->tuplehash[dir].tuple;
 869
 870         switch (new_state) {
 871         case TCP_CONNTRACK_SYN_SENT:
 872                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 873                         break;
 874                 /* RFC 1122: "When a connection is closed actively,
 875                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 876                  * (Maximum Segment Lifetime). However, it MAY accept
 877                  * a new SYN from the remote TCP to reopen the connection
 878                  * directly from TIME-WAIT state, if..."
 879                  * We ignore the conditions because we are in the
 880                  * TIME-WAIT state anyway.
 881                  *
 882                  * Handle aborted connections: we and the server
 883                  * think there is an existing connection but the client
 884                  * aborts it and starts a new one.
 885                  */
 886                 if (((ct->proto.tcp.seen[dir].flags
 887                       | ct->proto.tcp.seen[!dir].flags)
 888                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 889                     || (ct->proto.tcp.last_dir == dir
 890                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 891                         /* Attempt to reopen a closed/aborted connection.
 892                          * Delete this connection and look up again. */
 893                         spin_unlock_bh(&ct->lock);
 894
 895                         /* Only repeat if we can actually remove the timer.
 896                          * Destruction may already be in progress in process
 897                          * context and we must give it a chance to terminate.
 898                          */
 899                         if (nf_ct_kill(ct))
 900                                 return -NF_REPEAT;
 901                         return NF_DROP;
 902                 }
 903                 fallthrough;
 904         case TCP_CONNTRACK_IGNORE:
 905                 /* Ignored packets:
 906                  *
 907                  * Our connection entry may be out of sync, so ignore
 908                  * packets which may signal the real connection between
 909                  * the client and the server.
 910                  *
 911                  * a) SYN in ORIGINAL
 912                  * b) SYN/ACK in REPLY
 913                  * c) ACK in reply direction after initial SYN in original.
 914                  *
 915                  * If the ignored packet is invalid, the receiver will send
 916                  * a RST we'll catch below.
 917                  */
 918                 if (index == TCP_SYNACK_SET
 919                     && ct->proto.tcp.last_index == TCP_SYN_SET
 920                     && ct->proto.tcp.last_dir != dir
 921                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 922                         /* b) This SYN/ACK acknowledges a SYN that we earlier
 923                          * ignored as invalid. This means that the client and
 924                          * the server are both in sync, while the firewall is
 925                          * not. We get in sync from the previously annotated
 926                          * values.
 927                          */
 928                         old_state = TCP_CONNTRACK_SYN_SENT;
 929                         new_state = TCP_CONNTRACK_SYN_RECV;
 930                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
 931                                 ct->proto.tcp.last_end;
 932                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
 933                                 ct->proto.tcp.last_end;
 934                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
 935                                 ct->proto.tcp.last_win == 0 ?
 936                                         1 : ct->proto.tcp.last_win;
 937                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
 938                                 ct->proto.tcp.last_wscale;
 939                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
 940                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
 941                                 ct->proto.tcp.last_flags;
 942                         memset(&ct->proto.tcp.seen[dir], 0,
 943                                sizeof(struct ip_ct_tcp_state));
 944                         break;
 945                 }
 946                 ct->proto.tcp.last_index = index;
 947                 ct->proto.tcp.last_dir = dir;
 948                 ct->proto.tcp.last_seq = ntohl(th->seq);
 949                 ct->proto.tcp.last_end =
 950                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 951                 ct->proto.tcp.last_win = ntohs(th->window);
 952
 953                 /* a) This is a SYN in ORIGINAL. The client and the server
 954                  * may be in sync but we are not. In that case, we annotate
 955                  * the TCP options and let the packet go through. If it is a
 956                  * valid SYN packet, the server will reply with a SYN/ACK, and
 957                  * then we'll get in sync. Otherwise, the server potentially
 958                  * responds with a challenge ACK if implementing RFC5961.
 959                  */
 960                 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
 961                         struct ip_ct_tcp_state seen = {};
 962
 963                         ct->proto.tcp.last_flags =
 964                         ct->proto.tcp.last_wscale = 0;
 965                         tcp_options(skb, dataoff, th, &seen);
 966                         if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
 967                                 ct->proto.tcp.last_flags |=
 968                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 969                                 ct->proto.tcp.last_wscale = seen.td_scale;
 970                         }
 971                         if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
 972                                 ct->proto.tcp.last_flags |=
 973                                         IP_CT_TCP_FLAG_SACK_PERM;
 974                         }
 975                         /* Mark the potential for RFC5961 challenge ACK,
 976                          * this pose a special problem for LAST_ACK state
 977                          * as ACK is intrepretated as ACKing last FIN.
 978                          */
 979                         if (old_state == TCP_CONNTRACK_LAST_ACK)
 980                                 ct->proto.tcp.last_flags |=
 981                                         IP_CT_EXP_CHALLENGE_ACK;
 982                 }
 983                 spin_unlock_bh(&ct->lock);
 984                 nf_ct_l4proto_log_invalid(skb, ct, "invalid packet ignored in "
 985                                           "state %s ", tcp_conntrack_names[old_state]);
 986                 return NF_ACCEPT;
 987         case TCP_CONNTRACK_MAX:
 988                 /* Special case for SYN proxy: when the SYN to the server or
 989                  * the SYN/ACK from the server is lost, the client may transmit
 990                  * a keep-alive packet while in SYN_SENT state. This needs to
 991                  * be associated with the original conntrack entry in order to
 992                  * generate a new SYN with the correct sequence number.
 993                  */
 994                 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
 995                     index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
 996                     ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
 997                     ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
 998                         pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
 999                         spin_unlock_bh(&ct->lock);
1000                         return NF_ACCEPT;
1001                 }
1002
1003                 /* Invalid packet */
1004                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
1005                          dir, get_conntrack_index(th), old_state);
1006                 spin_unlock_bh(&ct->lock);
1007                 nf_ct_l4proto_log_invalid(skb, ct, "invalid state");
1008                 return -NF_ACCEPT;
1009         case TCP_CONNTRACK_TIME_WAIT:
1010                 /* RFC5961 compliance cause stack to send "challenge-ACK"
1011                  * e.g. in response to spurious SYNs.  Conntrack MUST
1012                  * not believe this ACK is acking last FIN.
1013                  */
1014                 if (old_state == TCP_CONNTRACK_LAST_ACK &&
1015                     index == TCP_ACK_SET &&
1016                     ct->proto.tcp.last_dir != dir &&
1017                     ct->proto.tcp.last_index == TCP_SYN_SET &&
1018                     (ct->proto.tcp.last_flags & IP_CT_EXP_CHALLENGE_ACK)) {
1019                         /* Detected RFC5961 challenge ACK */
1020                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
1021                         spin_unlock_bh(&ct->lock);
1022                         nf_ct_l4proto_log_invalid(skb, ct, "challenge-ack ignored");
1023                         return NF_ACCEPT; /* Don't change state */
1024                 }
1025                 break;
1026         case TCP_CONNTRACK_SYN_SENT2:
1027                 /* tcp_conntracks table is not smart enough to handle
1028                  * simultaneous open.
1029                  */
1030                 ct->proto.tcp.last_flags |= IP_CT_TCP_SIMULTANEOUS_OPEN;
1031                 break;
1032         case TCP_CONNTRACK_SYN_RECV:
1033                 if (dir == IP_CT_DIR_REPLY && index == TCP_ACK_SET &&
1034                     ct->proto.tcp.last_flags & IP_CT_TCP_SIMULTANEOUS_OPEN)
1035                         new_state = TCP_CONNTRACK_ESTABLISHED;
1036                 break;
1037         case TCP_CONNTRACK_CLOSE:
1038                 if (index != TCP_RST_SET)
1039                         break;
1040
1041                 if (ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET) {
1042                         u32 seq = ntohl(th->seq);
1043
1044                         if (before(seq, ct->proto.tcp.seen[!dir].td_maxack)) {
1045                                 /* Invalid RST  */
1046                                 spin_unlock_bh(&ct->lock);
1047                                 nf_ct_l4proto_log_invalid(skb, ct, "invalid rst");
1048                                 return -NF_ACCEPT;
1049                         }
1050
1051                         if (!nf_conntrack_tcp_established(ct) ||
1052                             seq == ct->proto.tcp.seen[!dir].td_maxack)
1053                                 break;
1054
1055                         /* Check if rst is part of train, such as
1056                          *   foo:80 > bar:4379: P, 235946583:235946602(19) ack 42
1057                          *   foo:80 > bar:4379: R, 235946602:235946602(0)  ack 42
1058                          */
1059                         if (ct->proto.tcp.last_index == TCP_ACK_SET &&
1060                             ct->proto.tcp.last_dir == dir &&
1061                             seq == ct->proto.tcp.last_end)
1062                                 break;
1063
1064                         /* ... RST sequence number doesn't match exactly, keep
1065                          * established state to allow a possible challenge ACK.
1066                          */
1067                         new_state = old_state;
1068                 }
1069                 if (((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
1070                          && ct->proto.tcp.last_index == TCP_SYN_SET)
1071                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
1072                             && ct->proto.tcp.last_index == TCP_ACK_SET))
1073                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
1074                         /* RST sent to invalid SYN or ACK we had let through
1075                          * at a) and c) above:
1076                          *
1077                          * a) SYN was in window then
1078                          * c) we hold a half-open connection.
1079                          *
1080                          * Delete our connection entry.
1081                          * We skip window checking, because packet might ACK
1082                          * segments we ignored. */
1083                         goto in_window;
1084                 }
1085                 break;
1086         default:
1087                 /* Keep compilers happy. */
1088                 break;
1089         }
1090
1091         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
1092                            skb, dataoff, th)) {
1093                 spin_unlock_bh(&ct->lock);
1094                 return -NF_ACCEPT;
1095         }
1096      in_window:
1097         /* From now on we have got in-window packets */
1098         ct->proto.tcp.last_index = index;
1099         ct->proto.tcp.last_dir = dir;
1100
1101         pr_debug("tcp_conntracks: ");
1102         nf_ct_dump_tuple(tuple);
1103         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1104                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1105                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1106                  old_state, new_state);
1107
1108         ct->proto.tcp.state = new_state;
1109         if (old_state != new_state
1110             && new_state == TCP_CONNTRACK_FIN_WAIT)
1111                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1112
1113         timeouts = nf_ct_timeout_lookup(ct);
1114         if (!timeouts)
1115                 timeouts = tn->timeouts;
1116
1117         if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1118             timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1119                 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1120         else if (unlikely(index == TCP_RST_SET))
1121                 timeout = timeouts[TCP_CONNTRACK_CLOSE];
1122         else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1123                  IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1124                  timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1125                 timeout = timeouts[TCP_CONNTRACK_UNACK];
1126         else if (ct->proto.tcp.last_win == 0 &&
1127                  timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1128                 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1129         else
1130                 timeout = timeouts[new_state];
1131         spin_unlock_bh(&ct->lock);
1132
1133         if (new_state != old_state)
1134                 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1135
1136         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1137                 /* If only reply is a RST, we can consider ourselves not to
1138                    have an established connection: this is a fairly common
1139                    problem case, so we can delete the conntrack
1140                    immediately.  --RR */
1141                 if (th->rst) {
1142                         nf_ct_kill_acct(ct, ctinfo, skb);
1143                         return NF_ACCEPT;
1144                 }
1145                 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1146                  * pickup with loose=1. Avoid large ESTABLISHED timeout.
1147                  */
1148                 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1149                     timeout > timeouts[TCP_CONNTRACK_UNACK])
1150                         timeout = timeouts[TCP_CONNTRACK_UNACK];
1151         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1152                    && (old_state == TCP_CONNTRACK_SYN_RECV
1153                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1154                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1155                 /* Set ASSURED if we see valid ack in ESTABLISHED
1156                    after SYN_RECV or a valid answer for a picked up
1157                    connection. */
1158                 set_bit(IPS_ASSURED_BIT, &ct->status);
1159                 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1160         }
1161         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1162
1163         return NF_ACCEPT;
1164 }
1165
1166 static bool tcp_can_early_drop(const struct nf_conn *ct)
1167 {
1168         switch (ct->proto.tcp.state) {
1169         case TCP_CONNTRACK_FIN_WAIT:
1170         case TCP_CONNTRACK_LAST_ACK:
1171         case TCP_CONNTRACK_TIME_WAIT:
1172         case TCP_CONNTRACK_CLOSE:
1173         case TCP_CONNTRACK_CLOSE_WAIT:
1174                 return true;
1175         default:
1176                 break;
1177         }
1178
1179         return false;
1180 }
1181
1182 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1183
1184 #include <linux/netfilter/nfnetlink.h>
1185 #include <linux/netfilter/nfnetlink_conntrack.h>
1186
1187 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1188                          struct nf_conn *ct)
1189 {
1190         struct nlattr *nest_parms;
1191         struct nf_ct_tcp_flags tmp = {};
1192
1193         spin_lock_bh(&ct->lock);
1194         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP);
1195         if (!nest_parms)
1196                 goto nla_put_failure;
1197
1198         if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state) ||
1199             nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1200                        ct->proto.tcp.seen[0].td_scale) ||
1201             nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1202                        ct->proto.tcp.seen[1].td_scale))
1203                 goto nla_put_failure;
1204
1205         tmp.flags = ct->proto.tcp.seen[0].flags;
1206         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1207                     sizeof(struct nf_ct_tcp_flags), &tmp))
1208                 goto nla_put_failure;
1209
1210         tmp.flags = ct->proto.tcp.seen[1].flags;
1211         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1212                     sizeof(struct nf_ct_tcp_flags), &tmp))
1213                 goto nla_put_failure;
1214         spin_unlock_bh(&ct->lock);
1215
1216         nla_nest_end(skb, nest_parms);
1217
1218         return 0;
1219
1220 nla_put_failure:
1221         spin_unlock_bh(&ct->lock);
1222         return -1;
1223 }
1224
1225 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1226         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1227         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1228         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1229         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1230         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len = sizeof(struct nf_ct_tcp_flags) },
1231 };
1232
1233 #define TCP_NLATTR_SIZE ( \
1234         NLA_ALIGN(NLA_HDRLEN + 1) + \
1235         NLA_ALIGN(NLA_HDRLEN + 1) + \
1236         NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)) + \
1237         NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)))
1238
1239 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1240 {
1241         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1242         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1243         int err;
1244
1245         /* updates could not contain anything about the private
1246          * protocol info, in that case skip the parsing */
1247         if (!pattr)
1248                 return 0;
1249
1250         err = nla_parse_nested_deprecated(tb, CTA_PROTOINFO_TCP_MAX, pattr,
1251                                           tcp_nla_policy, NULL);
1252         if (err < 0)
1253                 return err;
1254
1255         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1256             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1257                 return -EINVAL;
1258
1259         spin_lock_bh(&ct->lock);
1260         if (tb[CTA_PROTOINFO_TCP_STATE])
1261                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1262
1263         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1264                 struct nf_ct_tcp_flags *attr =
1265                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1266                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1267                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1268         }
1269
1270         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1271                 struct nf_ct_tcp_flags *attr =
1272                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1273                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1274                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1275         }
1276
1277         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1278             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1279             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1280             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1281                 ct->proto.tcp.seen[0].td_scale =
1282                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1283                 ct->proto.tcp.seen[1].td_scale =
1284                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1285         }
1286         spin_unlock_bh(&ct->lock);
1287
1288         return 0;
1289 }
1290
1291 static unsigned int tcp_nlattr_tuple_size(void)
1292 {
1293         static unsigned int size __read_mostly;
1294
1295         if (!size)
1296                 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1297
1298         return size;
1299 }
1300 #endif
1301
1302 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1303
1304 #include <linux/netfilter/nfnetlink.h>
1305 #include <linux/netfilter/nfnetlink_cttimeout.h>
1306
1307 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1308                                      struct net *net, void *data)
1309 {
1310         struct nf_tcp_net *tn = nf_tcp_pernet(net);
1311         unsigned int *timeouts = data;
1312         int i;
1313
1314         if (!timeouts)
1315                 timeouts = tn->timeouts;
1316         /* set default TCP timeouts. */
1317         for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1318                 timeouts[i] = tn->timeouts[i];
1319
1320         if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1321                 timeouts[TCP_CONNTRACK_SYN_SENT] =
1322                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1323         }
1324
1325         if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1326                 timeouts[TCP_CONNTRACK_SYN_RECV] =
1327                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1328         }
1329         if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1330                 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1331                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1332         }
1333         if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1334                 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1335                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1336         }
1337         if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1338                 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1339                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1340         }
1341         if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1342                 timeouts[TCP_CONNTRACK_LAST_ACK] =
1343                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1344         }
1345         if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1346                 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1347                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1348         }
1349         if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1350                 timeouts[TCP_CONNTRACK_CLOSE] =
1351                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1352         }
1353         if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1354                 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1355                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1356         }
1357         if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1358                 timeouts[TCP_CONNTRACK_RETRANS] =
1359                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1360         }
1361         if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1362                 timeouts[TCP_CONNTRACK_UNACK] =
1363                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1364         }
1365
1366         timeouts[CTA_TIMEOUT_TCP_UNSPEC] = timeouts[CTA_TIMEOUT_TCP_SYN_SENT];
1367         return 0;
1368 }
1369
1370 static int
1371 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1372 {
1373         const unsigned int *timeouts = data;
1374
1375         if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1376                         htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1377             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1378                          htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1379             nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1380                          htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1381             nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1382                          htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1383             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1384                          htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1385             nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1386                          htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1387             nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1388                          htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1389             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1390                          htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1391             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1392                          htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1393             nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1394                          htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1395             nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1396                          htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1397                 goto nla_put_failure;
1398         return 0;
1399
1400 nla_put_failure:
1401         return -ENOSPC;
1402 }
1403
1404 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1405         [CTA_TIMEOUT_TCP_SYN_SENT]      = { .type = NLA_U32 },
1406         [CTA_TIMEOUT_TCP_SYN_RECV]      = { .type = NLA_U32 },
1407         [CTA_TIMEOUT_TCP_ESTABLISHED]   = { .type = NLA_U32 },
1408         [CTA_TIMEOUT_TCP_FIN_WAIT]      = { .type = NLA_U32 },
1409         [CTA_TIMEOUT_TCP_CLOSE_WAIT]    = { .type = NLA_U32 },
1410         [CTA_TIMEOUT_TCP_LAST_ACK]      = { .type = NLA_U32 },
1411         [CTA_TIMEOUT_TCP_TIME_WAIT]     = { .type = NLA_U32 },
1412         [CTA_TIMEOUT_TCP_CLOSE]         = { .type = NLA_U32 },
1413         [CTA_TIMEOUT_TCP_SYN_SENT2]     = { .type = NLA_U32 },
1414         [CTA_TIMEOUT_TCP_RETRANS]       = { .type = NLA_U32 },
1415         [CTA_TIMEOUT_TCP_UNACK]         = { .type = NLA_U32 },
1416 };
1417 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1418
1419 void nf_conntrack_tcp_init_net(struct net *net)
1420 {
1421         struct nf_tcp_net *tn = nf_tcp_pernet(net);
1422         int i;
1423
1424         for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1425                 tn->timeouts[i] = tcp_timeouts[i];
1426
1427         /* timeouts[0] is unused, make it same as SYN_SENT so
1428          * ->timeouts[0] contains 'new' timeout, like udp or icmp.
1429          */
1430         tn->timeouts[0] = tcp_timeouts[TCP_CONNTRACK_SYN_SENT];
1431         tn->tcp_loose = nf_ct_tcp_loose;
1432         tn->tcp_be_liberal = nf_ct_tcp_be_liberal;
1433         tn->tcp_max_retrans = nf_ct_tcp_max_retrans;
1434 }
1435
1436 const struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp =
1437 {
1438         .l4proto                = IPPROTO_TCP,
1439 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1440         .print_conntrack        = tcp_print_conntrack,
1441 #endif
1442         .can_early_drop         = tcp_can_early_drop,
1443 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1444         .to_nlattr              = tcp_to_nlattr,
1445         .from_nlattr            = nlattr_to_tcp,
1446         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1447         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1448         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1449         .nlattr_size            = TCP_NLATTR_SIZE,
1450         .nla_policy             = nf_ct_port_nla_policy,
1451 #endif
1452 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1453         .ctnl_timeout           = {
1454                 .nlattr_to_obj  = tcp_timeout_nlattr_to_obj,
1455                 .obj_to_nlattr  = tcp_timeout_obj_to_nlattr,
1456                 .nlattr_max     = CTA_TIMEOUT_TCP_MAX,
1457                 .obj_size       = sizeof(unsigned int) *
1458                                         TCP_CONNTRACK_TIMEOUT_MAX,
1459                 .nla_policy     = tcp_timeout_nla_policy,
1460         },
1461 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1462 };