net/netfilter/nf_conntrack_proto_tcp.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* (C) 1999-2001 Paul `Rusty' Russell
   3  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   4  * (C) 2002-2013 Jozsef Kadlecsik <kadlec@netfilter.org>
   5  * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
   6  */
   7
   8 #include <linux/types.h>
   9 #include <linux/timer.h>
  10 #include <linux/module.h>
  11 #include <linux/in.h>
  12 #include <linux/tcp.h>
  13 #include <linux/spinlock.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/ipv6.h>
  16 #include <net/ip6_checksum.h>
  17 #include <asm/unaligned.h>
  18
  19 #include <net/tcp.h>
  20
  21 #include <linux/netfilter.h>
  22 #include <linux/netfilter_ipv4.h>
  23 #include <linux/netfilter_ipv6.h>
  24 #include <net/netfilter/nf_conntrack.h>
  25 #include <net/netfilter/nf_conntrack_l4proto.h>
  26 #include <net/netfilter/nf_conntrack_ecache.h>
  27 #include <net/netfilter/nf_conntrack_seqadj.h>
  28 #include <net/netfilter/nf_conntrack_synproxy.h>
  29 #include <net/netfilter/nf_conntrack_timeout.h>
  30 #include <net/netfilter/nf_log.h>
  31 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
  32 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
  33
  34   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  35      closely.  They're more complex. --RR */
  36
  37 static const char *const tcp_conntrack_names[] = {
  38         "NONE",
  39         "SYN_SENT",
  40         "SYN_RECV",
  41         "ESTABLISHED",
  42         "FIN_WAIT",
  43         "CLOSE_WAIT",
  44         "LAST_ACK",
  45         "TIME_WAIT",
  46         "CLOSE",
  47         "SYN_SENT2",
  48 };
  49
  50 #define SECS * HZ
  51 #define MINS * 60 SECS
  52 #define HOURS * 60 MINS
  53 #define DAYS * 24 HOURS
  54
  55 static const unsigned int tcp_timeouts[TCP_CONNTRACK_TIMEOUT_MAX] = {
  56         [TCP_CONNTRACK_SYN_SENT]        = 2 MINS,
  57         [TCP_CONNTRACK_SYN_RECV]        = 60 SECS,
  58         [TCP_CONNTRACK_ESTABLISHED]     = 5 DAYS,
  59         [TCP_CONNTRACK_FIN_WAIT]        = 2 MINS,
  60         [TCP_CONNTRACK_CLOSE_WAIT]      = 60 SECS,
  61         [TCP_CONNTRACK_LAST_ACK]        = 30 SECS,
  62         [TCP_CONNTRACK_TIME_WAIT]       = 2 MINS,
  63         [TCP_CONNTRACK_CLOSE]           = 10 SECS,
  64         [TCP_CONNTRACK_SYN_SENT2]       = 2 MINS,
  65 /* RFC1122 says the R2 limit should be at least 100 seconds.
  66    Linux uses 15 packets as limit, which corresponds
  67    to ~13-30min depending on RTO. */
  68         [TCP_CONNTRACK_RETRANS]         = 5 MINS,
  69         [TCP_CONNTRACK_UNACK]           = 5 MINS,
  70 };
  71
  72 #define sNO TCP_CONNTRACK_NONE
  73 #define sSS TCP_CONNTRACK_SYN_SENT
  74 #define sSR TCP_CONNTRACK_SYN_RECV
  75 #define sES TCP_CONNTRACK_ESTABLISHED
  76 #define sFW TCP_CONNTRACK_FIN_WAIT
  77 #define sCW TCP_CONNTRACK_CLOSE_WAIT
  78 #define sLA TCP_CONNTRACK_LAST_ACK
  79 #define sTW TCP_CONNTRACK_TIME_WAIT
  80 #define sCL TCP_CONNTRACK_CLOSE
  81 #define sS2 TCP_CONNTRACK_SYN_SENT2
  82 #define sIV TCP_CONNTRACK_MAX
  83 #define sIG TCP_CONNTRACK_IGNORE
  84
  85 /* What TCP flags are set from RST/SYN/FIN/ACK. */
  86 enum tcp_bit_set {
  87         TCP_SYN_SET,
  88         TCP_SYNACK_SET,
  89         TCP_FIN_SET,
  90         TCP_ACK_SET,
  91         TCP_RST_SET,
  92         TCP_NONE_SET,
  93 };
  94
  95 /*
  96  * The TCP state transition table needs a few words...
  97  *
  98  * We are the man in the middle. All the packets go through us
  99  * but might get lost in transit to the destination.
 100  * It is assumed that the destinations can't receive segments
 101  * we haven't seen.
 102  *
 103  * The checked segment is in window, but our windows are *not*
 104  * equivalent with the ones of the sender/receiver. We always
 105  * try to guess the state of the current sender.
 106  *
 107  * The meaning of the states are:
 108  *
 109  * NONE:        initial state
 110  * SYN_SENT:    SYN-only packet seen
 111  * SYN_SENT2:   SYN-only packet seen from reply dir, simultaneous open
 112  * SYN_RECV:    SYN-ACK packet seen
 113  * ESTABLISHED: ACK packet seen
 114  * FIN_WAIT:    FIN packet seen
 115  * CLOSE_WAIT:  ACK seen (after FIN)
 116  * LAST_ACK:    FIN seen (after FIN)
 117  * TIME_WAIT:   last ACK seen
 118  * CLOSE:       closed connection (RST)
 119  *
 120  * Packets marked as IGNORED (sIG):
 121  *      if they may be either invalid or valid
 122  *      and the receiver may send back a connection
 123  *      closing RST or a SYN/ACK.
 124  *
 125  * Packets marked as INVALID (sIV):
 126  *      if we regard them as truly invalid packets
 127  */
 128 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 129         {
 130 /* ORIGINAL */
 131 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 132 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sS2 },
 133 /*
 134  *      sNO -> sSS      Initialize a new connection
 135  *      sSS -> sSS      Retransmitted SYN
 136  *      sS2 -> sS2      Late retransmitted SYN
 137  *      sSR -> sIG
 138  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 139  *                      are errors. Receiver will reply with RST
 140  *                      and close the connection.
 141  *                      Or we are not in sync and hold a dead connection.
 142  *      sFW -> sIG
 143  *      sCW -> sIG
 144  *      sLA -> sIG
 145  *      sTW -> sSS      Reopened connection (RFC 1122).
 146  *      sCL -> sSS
 147  */
 148 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 149 /*synack*/ { sIV, sIV, sSR, sIV, sIV, sIV, sIV, sIV, sIV, sSR },
 150 /*
 151  *      sNO -> sIV      Too late and no reason to do anything
 152  *      sSS -> sIV      Client can't send SYN and then SYN/ACK
 153  *      sS2 -> sSR      SYN/ACK sent to SYN2 in simultaneous open
 154  *      sSR -> sSR      Late retransmitted SYN/ACK in simultaneous open
 155  *      sES -> sIV      Invalid SYN/ACK packets sent by the client
 156  *      sFW -> sIV
 157  *      sCW -> sIV
 158  *      sLA -> sIV
 159  *      sTW -> sIV
 160  *      sCL -> sIV
 161  */
 162 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 163 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 164 /*
 165  *      sNO -> sIV      Too late and no reason to do anything...
 166  *      sSS -> sIV      Client migth not send FIN in this state:
 167  *                      we enforce waiting for a SYN/ACK reply first.
 168  *      sS2 -> sIV
 169  *      sSR -> sFW      Close started.
 170  *      sES -> sFW
 171  *      sFW -> sLA      FIN seen in both directions, waiting for
 172  *                      the last ACK.
 173  *                      Migth be a retransmitted FIN as well...
 174  *      sCW -> sLA
 175  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 176  *      sTW -> sTW
 177  *      sCL -> sCL
 178  */
 179 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 180 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 181 /*
 182  *      sNO -> sES      Assumed.
 183  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 184  *      sS2 -> sIV
 185  *      sSR -> sES      Established state is reached.
 186  *      sES -> sES      :-)
 187  *      sFW -> sCW      Normal close request answered by ACK.
 188  *      sCW -> sCW
 189  *      sLA -> sTW      Last ACK detected (RFC5961 challenged)
 190  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 191  *      sCL -> sCL
 192  */
 193 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 194 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
 195 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 196         },
 197         {
 198 /* REPLY */
 199 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 200 /*syn*/    { sIV, sS2, sIV, sIV, sIV, sIV, sIV, sSS, sIV, sS2 },
 201 /*
 202  *      sNO -> sIV      Never reached.
 203  *      sSS -> sS2      Simultaneous open
 204  *      sS2 -> sS2      Retransmitted simultaneous SYN
 205  *      sSR -> sIV      Invalid SYN packets sent by the server
 206  *      sES -> sIV
 207  *      sFW -> sIV
 208  *      sCW -> sIV
 209  *      sLA -> sIV
 210  *      sTW -> sSS      Reopened connection, but server may have switched role
 211  *      sCL -> sIV
 212  */
 213 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 214 /*synack*/ { sIV, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
 215 /*
 216  *      sSS -> sSR      Standard open.
 217  *      sS2 -> sSR      Simultaneous open
 218  *      sSR -> sIG      Retransmitted SYN/ACK, ignore it.
 219  *      sES -> sIG      Late retransmitted SYN/ACK?
 220  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 221  *      sCW -> sIG
 222  *      sLA -> sIG
 223  *      sTW -> sIG
 224  *      sCL -> sIG
 225  */
 226 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 227 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 228 /*
 229  *      sSS -> sIV      Server might not send FIN in this state.
 230  *      sS2 -> sIV
 231  *      sSR -> sFW      Close started.
 232  *      sES -> sFW
 233  *      sFW -> sLA      FIN seen in both directions.
 234  *      sCW -> sLA
 235  *      sLA -> sLA      Retransmitted FIN.
 236  *      sTW -> sTW
 237  *      sCL -> sCL
 238  */
 239 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 240 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIG },
 241 /*
 242  *      sSS -> sIG      Might be a half-open connection.
 243  *      sS2 -> sIG
 244  *      sSR -> sSR      Might answer late resent SYN.
 245  *      sES -> sES      :-)
 246  *      sFW -> sCW      Normal close request answered by ACK.
 247  *      sCW -> sCW
 248  *      sLA -> sTW      Last ACK detected (RFC5961 challenged)
 249  *      sTW -> sTW      Retransmitted last ACK.
 250  *      sCL -> sCL
 251  */
 252 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 253 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
 254 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 255         }
 256 };
 257
 258 #ifdef CONFIG_NF_CONNTRACK_PROCFS
 259 /* Print out the private part of the conntrack. */
 260 static void tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
 261 {
 262         if (test_bit(IPS_OFFLOAD_BIT, &ct->status))
 263                 return;
 264
 265         seq_printf(s, "%s ", tcp_conntrack_names[ct->proto.tcp.state]);
 266 }
 267 #endif
 268
 269 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 270 {
 271         if (tcph->rst) return TCP_RST_SET;
 272         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 273         else if (tcph->fin) return TCP_FIN_SET;
 274         else if (tcph->ack) return TCP_ACK_SET;
 275         else return TCP_NONE_SET;
 276 }
 277
 278 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 279    in IP Filter' by Guido van Rooij.
 280
 281    http://www.sane.nl/events/sane2000/papers.html
 282    http://www.darkart.com/mirrors/www.obfuscation.org/ipf/
 283
 284    The boundaries and the conditions are changed according to RFC793:
 285    the packet must intersect the window (i.e. segments may be
 286    after the right or before the left edge) and thus receivers may ACK
 287    segments after the right edge of the window.
 288
 289         td_maxend = max(sack + max(win,1)) seen in reply packets
 290         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 291         td_maxwin += seq + len - sender.td_maxend
 292                         if seq + len > sender.td_maxend
 293         td_end    = max(seq + len) seen in sent packets
 294
 295    I.   Upper bound for valid data:     seq <= sender.td_maxend
 296    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 297    III. Upper bound for valid (s)ack:   sack <= receiver.td_end
 298    IV.  Lower bound for valid (s)ack:   sack >= receiver.td_end - MAXACKWINDOW
 299
 300    where sack is the highest right edge of sack block found in the packet
 301    or ack in the case of packet without SACK option.
 302
 303    The upper bound limit for a valid (s)ack is not ignored -
 304    we doesn't have to deal with fragments.
 305 */
 306
 307 static inline __u32 segment_seq_plus_len(__u32 seq,
 308                                          size_t len,
 309                                          unsigned int dataoff,
 310                                          const struct tcphdr *tcph)
 311 {
 312         /* XXX Should I use payload length field in IP/IPv6 header ?
 313          * - YK */
 314         return (seq + len - dataoff - tcph->doff*4
 315                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 316 }
 317
 318 /* Fixme: what about big packets? */
 319 #define MAXACKWINCONST                  66000
 320 #define MAXACKWINDOW(sender)                                            \
 321         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 322                                               : MAXACKWINCONST)
 323
 324 /*
 325  * Simplified tcp_parse_options routine from tcp_input.c
 326  */
 327 static void tcp_options(const struct sk_buff *skb,
 328                         unsigned int dataoff,
 329                         const struct tcphdr *tcph,
 330                         struct ip_ct_tcp_state *state)
 331 {
 332         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 333         const unsigned char *ptr;
 334         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 335
 336         if (!length)
 337                 return;
 338
 339         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 340                                  length, buff);
 341         if (!ptr)
 342                 return;
 343
 344         state->td_scale =
 345         state->flags = 0;
 346
 347         while (length > 0) {
 348                 int opcode=*ptr++;
 349                 int opsize;
 350
 351                 switch (opcode) {
 352                 case TCPOPT_EOL:
 353                         return;
 354                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 355                         length--;
 356                         continue;
 357                 default:
 358                         if (length < 2)
 359                                 return;
 360                         opsize=*ptr++;
 361                         if (opsize < 2) /* "silly options" */
 362                                 return;
 363                         if (opsize > length)
 364                                 return; /* don't parse partial options */
 365
 366                         if (opcode == TCPOPT_SACK_PERM
 367                             && opsize == TCPOLEN_SACK_PERM)
 368                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 369                         else if (opcode == TCPOPT_WINDOW
 370                                  && opsize == TCPOLEN_WINDOW) {
 371                                 state->td_scale = *(u_int8_t *)ptr;
 372
 373                                 if (state->td_scale > TCP_MAX_WSCALE)
 374                                         state->td_scale = TCP_MAX_WSCALE;
 375
 376                                 state->flags |=
 377                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 378                         }
 379                         ptr += opsize - 2;
 380                         length -= opsize;
 381                 }
 382         }
 383 }
 384
 385 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 386                      const struct tcphdr *tcph, __u32 *sack)
 387 {
 388         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 389         const unsigned char *ptr;
 390         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 391         __u32 tmp;
 392
 393         if (!length)
 394                 return;
 395
 396         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 397                                  length, buff);
 398         if (!ptr)
 399                 return;
 400
 401         /* Fast path for timestamp-only option */
 402         if (length == TCPOLEN_TSTAMP_ALIGNED
 403             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
 404                                        | (TCPOPT_NOP << 16)
 405                                        | (TCPOPT_TIMESTAMP << 8)
 406                                        | TCPOLEN_TIMESTAMP))
 407                 return;
 408
 409         while (length > 0) {
 410                 int opcode = *ptr++;
 411                 int opsize, i;
 412
 413                 switch (opcode) {
 414                 case TCPOPT_EOL:
 415                         return;
 416                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 417                         length--;
 418                         continue;
 419                 default:
 420                         if (length < 2)
 421                                 return;
 422                         opsize = *ptr++;
 423                         if (opsize < 2) /* "silly options" */
 424                                 return;
 425                         if (opsize > length)
 426                                 return; /* don't parse partial options */
 427
 428                         if (opcode == TCPOPT_SACK
 429                             && opsize >= (TCPOLEN_SACK_BASE
 430                                           + TCPOLEN_SACK_PERBLOCK)
 431                             && !((opsize - TCPOLEN_SACK_BASE)
 432                                  % TCPOLEN_SACK_PERBLOCK)) {
 433                                 for (i = 0;
 434                                      i < (opsize - TCPOLEN_SACK_BASE);
 435                                      i += TCPOLEN_SACK_PERBLOCK) {
 436                                         tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
 437
 438                                         if (after(tmp, *sack))
 439                                                 *sack = tmp;
 440                                 }
 441                                 return;
 442                         }
 443                         ptr += opsize - 2;
 444                         length -= opsize;
 445                 }
 446         }
 447 }
 448
 449 static bool tcp_in_window(struct nf_conn *ct,
 450                           enum ip_conntrack_dir dir,
 451                           unsigned int index,
 452                           const struct sk_buff *skb,
 453                           unsigned int dataoff,
 454                           const struct tcphdr *tcph,
 455                           const struct nf_hook_state *hook_state)
 456 {
 457         struct ip_ct_tcp *state = &ct->proto.tcp;
 458         struct net *net = nf_ct_net(ct);
 459         struct nf_tcp_net *tn = nf_tcp_pernet(net);
 460         struct ip_ct_tcp_state *sender = &state->seen[dir];
 461         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 462         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 463         __u32 seq, ack, sack, end, win, swin;
 464         u16 win_raw;
 465         s32 receiver_offset;
 466         bool res, in_recv_win;
 467
 468         /*
 469          * Get the required data from the packet.
 470          */
 471         seq = ntohl(tcph->seq);
 472         ack = sack = ntohl(tcph->ack_seq);
 473         win_raw = ntohs(tcph->window);
 474         win = win_raw;
 475         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 476
 477         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 478                 tcp_sack(skb, dataoff, tcph, &sack);
 479
 480         /* Take into account NAT sequence number mangling */
 481         receiver_offset = nf_ct_seq_offset(ct, !dir, ack - 1);
 482         ack -= receiver_offset;
 483         sack -= receiver_offset;
 484
 485         pr_debug("tcp_in_window: START\n");
 486         pr_debug("tcp_in_window: ");
 487         nf_ct_dump_tuple(tuple);
 488         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
 489                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
 490         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 491                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 492                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 493                  sender->td_scale,
 494                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 495                  receiver->td_scale);
 496
 497         if (sender->td_maxwin == 0) {
 498                 /*
 499                  * Initialize sender data.
 500                  */
 501                 if (tcph->syn) {
 502                         /*
 503                          * SYN-ACK in reply to a SYN
 504                          * or SYN from reply direction in simultaneous open.
 505                          */
 506                         sender->td_end =
 507                         sender->td_maxend = end;
 508                         sender->td_maxwin = (win == 0 ? 1 : win);
 509
 510                         tcp_options(skb, dataoff, tcph, sender);
 511                         /*
 512                          * RFC 1323:
 513                          * Both sides must send the Window Scale option
 514                          * to enable window scaling in either direction.
 515                          */
 516                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 517                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 518                                 sender->td_scale =
 519                                 receiver->td_scale = 0;
 520                         if (!tcph->ack)
 521                                 /* Simultaneous open */
 522                                 return true;
 523                 } else {
 524                         /*
 525                          * We are in the middle of a connection,
 526                          * its history is lost for us.
 527                          * Let's try to use the data from the packet.
 528                          */
 529                         sender->td_end = end;
 530                         swin = win << sender->td_scale;
 531                         sender->td_maxwin = (swin == 0 ? 1 : swin);
 532                         sender->td_maxend = end + sender->td_maxwin;
 533                         if (receiver->td_maxwin == 0) {
 534                                 /* We haven't seen traffic in the other
 535                                  * direction yet but we have to tweak window
 536                                  * tracking to pass III and IV until that
 537                                  * happens.
 538                                  */
 539                                 receiver->td_end = receiver->td_maxend = sack;
 540                         } else if (sack == receiver->td_end + 1) {
 541                                 /* Likely a reply to a keepalive.
 542                                  * Needed for III.
 543                                  */
 544                                 receiver->td_end++;
 545                         }
 546
 547                 }
 548         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 549                      && dir == IP_CT_DIR_ORIGINAL)
 550                    || (state->state == TCP_CONNTRACK_SYN_RECV
 551                      && dir == IP_CT_DIR_REPLY))
 552                    && after(end, sender->td_end)) {
 553                 /*
 554                  * RFC 793: "if a TCP is reinitialized ... then it need
 555                  * not wait at all; it must only be sure to use sequence
 556                  * numbers larger than those recently used."
 557                  */
 558                 sender->td_end =
 559                 sender->td_maxend = end;
 560                 sender->td_maxwin = (win == 0 ? 1 : win);
 561
 562                 tcp_options(skb, dataoff, tcph, sender);
 563         }
 564
 565         if (!(tcph->ack)) {
 566                 /*
 567                  * If there is no ACK, just pretend it was set and OK.
 568                  */
 569                 ack = sack = receiver->td_end;
 570         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 571                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 572                    && (ack == 0)) {
 573                 /*
 574                  * Broken TCP stacks, that set ACK in RST packets as well
 575                  * with zero ack value.
 576                  */
 577                 ack = sack = receiver->td_end;
 578         }
 579
 580         if (tcph->rst && seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)
 581                 /*
 582                  * RST sent answering SYN.
 583                  */
 584                 seq = end = sender->td_end;
 585
 586         pr_debug("tcp_in_window: ");
 587         nf_ct_dump_tuple(tuple);
 588         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
 589                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
 590         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 591                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 592                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 593                  sender->td_scale,
 594                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 595                  receiver->td_scale);
 596
 597         /* Is the ending sequence in the receive window (if available)? */
 598         in_recv_win = !receiver->td_maxwin ||
 599                       after(end, sender->td_end - receiver->td_maxwin - 1);
 600
 601         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 602                  before(seq, sender->td_maxend + 1),
 603                  (in_recv_win ? 1 : 0),
 604                  before(sack, receiver->td_end + 1),
 605                  after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 606
 607         if (before(seq, sender->td_maxend + 1) &&
 608             in_recv_win &&
 609             before(sack, receiver->td_end + 1) &&
 610             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
 611                 /*
 612                  * Take into account window scaling (RFC 1323).
 613                  */
 614                 if (!tcph->syn)
 615                         win <<= sender->td_scale;
 616
 617                 /*
 618                  * Update sender data.
 619                  */
 620                 swin = win + (sack - ack);
 621                 if (sender->td_maxwin < swin)
 622                         sender->td_maxwin = swin;
 623                 if (after(end, sender->td_end)) {
 624                         sender->td_end = end;
 625                         sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 626                 }
 627                 if (tcph->ack) {
 628                         if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
 629                                 sender->td_maxack = ack;
 630                                 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
 631                         } else if (after(ack, sender->td_maxack))
 632                                 sender->td_maxack = ack;
 633                 }
 634
 635                 /*
 636                  * Update receiver data.
 637                  */
 638                 if (receiver->td_maxwin != 0 && after(end, sender->td_maxend))
 639                         receiver->td_maxwin += end - sender->td_maxend;
 640                 if (after(sack + win, receiver->td_maxend - 1)) {
 641                         receiver->td_maxend = sack + win;
 642                         if (win == 0)
 643                                 receiver->td_maxend++;
 644                 }
 645                 if (ack == receiver->td_end)
 646                         receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 647
 648                 /*
 649                  * Check retransmissions.
 650                  */
 651                 if (index == TCP_ACK_SET) {
 652                         if (state->last_dir == dir
 653                             && state->last_seq == seq
 654                             && state->last_ack == ack
 655                             && state->last_end == end
 656                             && state->last_win == win_raw)
 657                                 state->retrans++;
 658                         else {
 659                                 state->last_dir = dir;
 660                                 state->last_seq = seq;
 661                                 state->last_ack = ack;
 662                                 state->last_end = end;
 663                                 state->last_win = win_raw;
 664                                 state->retrans = 0;
 665                         }
 666                 }
 667                 res = true;
 668         } else {
 669                 res = false;
 670                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 671                     tn->tcp_be_liberal)
 672                         res = true;
 673                 if (!res) {
 674                         nf_ct_l4proto_log_invalid(skb, ct, hook_state,
 675                         "%s",
 676                         before(seq, sender->td_maxend + 1) ?
 677                         in_recv_win ?
 678                         before(sack, receiver->td_end + 1) ?
 679                         after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
 680                         : "ACK is under the lower bound (possible overly delayed ACK)"
 681                         : "ACK is over the upper bound (ACKed data not seen yet)"
 682                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 683                         : "SEQ is over the upper bound (over the window of the receiver)");
 684                 }
 685         }
 686
 687         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 688                  "receiver end=%u maxend=%u maxwin=%u\n",
 689                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 690                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 691
 692         return res;
 693 }
 694
 695 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 696 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
 697                                  TCPHDR_URG) + 1] =
 698 {
 699         [TCPHDR_SYN]                            = 1,
 700         [TCPHDR_SYN|TCPHDR_URG]                 = 1,
 701         [TCPHDR_SYN|TCPHDR_ACK]                 = 1,
 702         [TCPHDR_RST]                            = 1,
 703         [TCPHDR_RST|TCPHDR_ACK]                 = 1,
 704         [TCPHDR_FIN|TCPHDR_ACK]                 = 1,
 705         [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG]      = 1,
 706         [TCPHDR_ACK]                            = 1,
 707         [TCPHDR_ACK|TCPHDR_URG]                 = 1,
 708 };
 709
 710 static void tcp_error_log(const struct sk_buff *skb,
 711                           const struct nf_hook_state *state,
 712                           const char *msg)
 713 {
 714         nf_l4proto_log_invalid(skb, state, IPPROTO_TCP, "%s", msg);
 715 }
 716
 717 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 718 static bool tcp_error(const struct tcphdr *th,
 719                       struct sk_buff *skb,
 720                       unsigned int dataoff,
 721                       const struct nf_hook_state *state)
 722 {
 723         unsigned int tcplen = skb->len - dataoff;
 724         u8 tcpflags;
 725
 726         /* Not whole TCP header or malformed packet */
 727         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 728                 tcp_error_log(skb, state, "truncated packet");
 729                 return true;
 730         }
 731
 732         /* Checksum invalid? Ignore.
 733          * We skip checking packets on the outgoing path
 734          * because the checksum is assumed to be correct.
 735          */
 736         /* FIXME: Source route IP option packets --RR */
 737         if (state->net->ct.sysctl_checksum &&
 738             state->hook == NF_INET_PRE_ROUTING &&
 739             nf_checksum(skb, state->hook, dataoff, IPPROTO_TCP, state->pf)) {
 740                 tcp_error_log(skb, state, "bad checksum");
 741                 return true;
 742         }
 743
 744         /* Check TCP flags. */
 745         tcpflags = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
 746         if (!tcp_valid_flags[tcpflags]) {
 747                 tcp_error_log(skb, state, "invalid tcp flag combination");
 748                 return true;
 749         }
 750
 751         return false;
 752 }
 753
 754 static noinline bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
 755                              unsigned int dataoff,
 756                              const struct tcphdr *th)
 757 {
 758         enum tcp_conntrack new_state;
 759         struct net *net = nf_ct_net(ct);
 760         const struct nf_tcp_net *tn = nf_tcp_pernet(net);
 761         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
 762         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
 763
 764         /* Don't need lock here: this conntrack not in circulation yet */
 765         new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
 766
 767         /* Invalid: delete conntrack */
 768         if (new_state >= TCP_CONNTRACK_MAX) {
 769                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
 770                 return false;
 771         }
 772
 773         if (new_state == TCP_CONNTRACK_SYN_SENT) {
 774                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
 775                 /* SYN packet */
 776                 ct->proto.tcp.seen[0].td_end =
 777                         segment_seq_plus_len(ntohl(th->seq), skb->len,
 778                                              dataoff, th);
 779                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
 780                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
 781                         ct->proto.tcp.seen[0].td_maxwin = 1;
 782                 ct->proto.tcp.seen[0].td_maxend =
 783                         ct->proto.tcp.seen[0].td_end;
 784
 785                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
 786         } else if (tn->tcp_loose == 0) {
 787                 /* Don't try to pick up connections. */
 788                 return false;
 789         } else {
 790                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
 791                 /*
 792                  * We are in the middle of a connection,
 793                  * its history is lost for us.
 794                  * Let's try to use the data from the packet.
 795                  */
 796                 ct->proto.tcp.seen[0].td_end =
 797                         segment_seq_plus_len(ntohl(th->seq), skb->len,
 798                                              dataoff, th);
 799                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
 800                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
 801                         ct->proto.tcp.seen[0].td_maxwin = 1;
 802                 ct->proto.tcp.seen[0].td_maxend =
 803                         ct->proto.tcp.seen[0].td_end +
 804                         ct->proto.tcp.seen[0].td_maxwin;
 805
 806                 /* We assume SACK and liberal window checking to handle
 807                  * window scaling */
 808                 ct->proto.tcp.seen[0].flags =
 809                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
 810                                               IP_CT_TCP_FLAG_BE_LIBERAL;
 811         }
 812
 813         /* tcp_packet will set them */
 814         ct->proto.tcp.last_index = TCP_NONE_SET;
 815
 816         pr_debug("%s: sender end=%u maxend=%u maxwin=%u scale=%i "
 817                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 818                  __func__,
 819                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 820                  sender->td_scale,
 821                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 822                  receiver->td_scale);
 823         return true;
 824 }
 825
 826 /* Returns verdict for packet, or -1 for invalid. */
 827 int nf_conntrack_tcp_packet(struct nf_conn *ct,
 828                             struct sk_buff *skb,
 829                             unsigned int dataoff,
 830                             enum ip_conntrack_info ctinfo,
 831                             const struct nf_hook_state *state)
 832 {
 833         struct net *net = nf_ct_net(ct);
 834         struct nf_tcp_net *tn = nf_tcp_pernet(net);
 835         struct nf_conntrack_tuple *tuple;
 836         enum tcp_conntrack new_state, old_state;
 837         unsigned int index, *timeouts;
 838         enum ip_conntrack_dir dir;
 839         const struct tcphdr *th;
 840         struct tcphdr _tcph;
 841         unsigned long timeout;
 842
 843         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 844         if (th == NULL)
 845                 return -NF_ACCEPT;
 846
 847         if (tcp_error(th, skb, dataoff, state))
 848                 return -NF_ACCEPT;
 849
 850         if (!nf_ct_is_confirmed(ct) && !tcp_new(ct, skb, dataoff, th))
 851                 return -NF_ACCEPT;
 852
 853         spin_lock_bh(&ct->lock);
 854         old_state = ct->proto.tcp.state;
 855         dir = CTINFO2DIR(ctinfo);
 856         index = get_conntrack_index(th);
 857         new_state = tcp_conntracks[dir][index][old_state];
 858         tuple = &ct->tuplehash[dir].tuple;
 859
 860         switch (new_state) {
 861         case TCP_CONNTRACK_SYN_SENT:
 862                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 863                         break;
 864                 /* RFC 1122: "When a connection is closed actively,
 865                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 866                  * (Maximum Segment Lifetime). However, it MAY accept
 867                  * a new SYN from the remote TCP to reopen the connection
 868                  * directly from TIME-WAIT state, if..."
 869                  * We ignore the conditions because we are in the
 870                  * TIME-WAIT state anyway.
 871                  *
 872                  * Handle aborted connections: we and the server
 873                  * think there is an existing connection but the client
 874                  * aborts it and starts a new one.
 875                  */
 876                 if (((ct->proto.tcp.seen[dir].flags
 877                       | ct->proto.tcp.seen[!dir].flags)
 878                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 879                     || (ct->proto.tcp.last_dir == dir
 880                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 881                         /* Attempt to reopen a closed/aborted connection.
 882                          * Delete this connection and look up again. */
 883                         spin_unlock_bh(&ct->lock);
 884
 885                         /* Only repeat if we can actually remove the timer.
 886                          * Destruction may already be in progress in process
 887                          * context and we must give it a chance to terminate.
 888                          */
 889                         if (nf_ct_kill(ct))
 890                                 return -NF_REPEAT;
 891                         return NF_DROP;
 892                 }
 893                 fallthrough;
 894         case TCP_CONNTRACK_IGNORE:
 895                 /* Ignored packets:
 896                  *
 897                  * Our connection entry may be out of sync, so ignore
 898                  * packets which may signal the real connection between
 899                  * the client and the server.
 900                  *
 901                  * a) SYN in ORIGINAL
 902                  * b) SYN/ACK in REPLY
 903                  * c) ACK in reply direction after initial SYN in original.
 904                  *
 905                  * If the ignored packet is invalid, the receiver will send
 906                  * a RST we'll catch below.
 907                  */
 908                 if (index == TCP_SYNACK_SET
 909                     && ct->proto.tcp.last_index == TCP_SYN_SET
 910                     && ct->proto.tcp.last_dir != dir
 911                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 912                         /* b) This SYN/ACK acknowledges a SYN that we earlier
 913                          * ignored as invalid. This means that the client and
 914                          * the server are both in sync, while the firewall is
 915                          * not. We get in sync from the previously annotated
 916                          * values.
 917                          */
 918                         old_state = TCP_CONNTRACK_SYN_SENT;
 919                         new_state = TCP_CONNTRACK_SYN_RECV;
 920                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
 921                                 ct->proto.tcp.last_end;
 922                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
 923                                 ct->proto.tcp.last_end;
 924                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
 925                                 ct->proto.tcp.last_win == 0 ?
 926                                         1 : ct->proto.tcp.last_win;
 927                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
 928                                 ct->proto.tcp.last_wscale;
 929                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
 930                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
 931                                 ct->proto.tcp.last_flags;
 932                         memset(&ct->proto.tcp.seen[dir], 0,
 933                                sizeof(struct ip_ct_tcp_state));
 934                         break;
 935                 }
 936                 ct->proto.tcp.last_index = index;
 937                 ct->proto.tcp.last_dir = dir;
 938                 ct->proto.tcp.last_seq = ntohl(th->seq);
 939                 ct->proto.tcp.last_end =
 940                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 941                 ct->proto.tcp.last_win = ntohs(th->window);
 942
 943                 /* a) This is a SYN in ORIGINAL. The client and the server
 944                  * may be in sync but we are not. In that case, we annotate
 945                  * the TCP options and let the packet go through. If it is a
 946                  * valid SYN packet, the server will reply with a SYN/ACK, and
 947                  * then we'll get in sync. Otherwise, the server potentially
 948                  * responds with a challenge ACK if implementing RFC5961.
 949                  */
 950                 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
 951                         struct ip_ct_tcp_state seen = {};
 952
 953                         ct->proto.tcp.last_flags =
 954                         ct->proto.tcp.last_wscale = 0;
 955                         tcp_options(skb, dataoff, th, &seen);
 956                         if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
 957                                 ct->proto.tcp.last_flags |=
 958                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 959                                 ct->proto.tcp.last_wscale = seen.td_scale;
 960                         }
 961                         if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
 962                                 ct->proto.tcp.last_flags |=
 963                                         IP_CT_TCP_FLAG_SACK_PERM;
 964                         }
 965                         /* Mark the potential for RFC5961 challenge ACK,
 966                          * this pose a special problem for LAST_ACK state
 967                          * as ACK is intrepretated as ACKing last FIN.
 968                          */
 969                         if (old_state == TCP_CONNTRACK_LAST_ACK)
 970                                 ct->proto.tcp.last_flags |=
 971                                         IP_CT_EXP_CHALLENGE_ACK;
 972                 }
 973                 spin_unlock_bh(&ct->lock);
 974                 nf_ct_l4proto_log_invalid(skb, ct, state,
 975                                           "packet (index %d) in dir %d ignored, state %s",
 976                                           index, dir,
 977                                           tcp_conntrack_names[old_state]);
 978                 return NF_ACCEPT;
 979         case TCP_CONNTRACK_MAX:
 980                 /* Special case for SYN proxy: when the SYN to the server or
 981                  * the SYN/ACK from the server is lost, the client may transmit
 982                  * a keep-alive packet while in SYN_SENT state. This needs to
 983                  * be associated with the original conntrack entry in order to
 984                  * generate a new SYN with the correct sequence number.
 985                  */
 986                 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
 987                     index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
 988                     ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
 989                     ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
 990                         pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
 991                         spin_unlock_bh(&ct->lock);
 992                         return NF_ACCEPT;
 993                 }
 994
 995                 /* Invalid packet */
 996                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 997                          dir, get_conntrack_index(th), old_state);
 998                 spin_unlock_bh(&ct->lock);
 999                 nf_ct_l4proto_log_invalid(skb, ct, state, "invalid state");
1000                 return -NF_ACCEPT;
1001         case TCP_CONNTRACK_TIME_WAIT:
1002                 /* RFC5961 compliance cause stack to send "challenge-ACK"
1003                  * e.g. in response to spurious SYNs.  Conntrack MUST
1004                  * not believe this ACK is acking last FIN.
1005                  */
1006                 if (old_state == TCP_CONNTRACK_LAST_ACK &&
1007                     index == TCP_ACK_SET &&
1008                     ct->proto.tcp.last_dir != dir &&
1009                     ct->proto.tcp.last_index == TCP_SYN_SET &&
1010                     (ct->proto.tcp.last_flags & IP_CT_EXP_CHALLENGE_ACK)) {
1011                         /* Detected RFC5961 challenge ACK */
1012                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
1013                         spin_unlock_bh(&ct->lock);
1014                         nf_ct_l4proto_log_invalid(skb, ct, state, "challenge-ack ignored");
1015                         return NF_ACCEPT; /* Don't change state */
1016                 }
1017                 break;
1018         case TCP_CONNTRACK_SYN_SENT2:
1019                 /* tcp_conntracks table is not smart enough to handle
1020                  * simultaneous open.
1021                  */
1022                 ct->proto.tcp.last_flags |= IP_CT_TCP_SIMULTANEOUS_OPEN;
1023                 break;
1024         case TCP_CONNTRACK_SYN_RECV:
1025                 if (dir == IP_CT_DIR_REPLY && index == TCP_ACK_SET &&
1026                     ct->proto.tcp.last_flags & IP_CT_TCP_SIMULTANEOUS_OPEN)
1027                         new_state = TCP_CONNTRACK_ESTABLISHED;
1028                 break;
1029         case TCP_CONNTRACK_CLOSE:
1030                 if (index != TCP_RST_SET)
1031                         break;
1032
1033                 if (ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET) {
1034                         u32 seq = ntohl(th->seq);
1035
1036                         if (before(seq, ct->proto.tcp.seen[!dir].td_maxack)) {
1037                                 /* Invalid RST  */
1038                                 spin_unlock_bh(&ct->lock);
1039                                 nf_ct_l4proto_log_invalid(skb, ct, state, "invalid rst");
1040                                 return -NF_ACCEPT;
1041                         }
1042
1043                         if (!nf_conntrack_tcp_established(ct) ||
1044                             seq == ct->proto.tcp.seen[!dir].td_maxack)
1045                                 break;
1046
1047                         /* Check if rst is part of train, such as
1048                          *   foo:80 > bar:4379: P, 235946583:235946602(19) ack 42
1049                          *   foo:80 > bar:4379: R, 235946602:235946602(0)  ack 42
1050                          */
1051                         if (ct->proto.tcp.last_index == TCP_ACK_SET &&
1052                             ct->proto.tcp.last_dir == dir &&
1053                             seq == ct->proto.tcp.last_end)
1054                                 break;
1055
1056                         /* ... RST sequence number doesn't match exactly, keep
1057                          * established state to allow a possible challenge ACK.
1058                          */
1059                         new_state = old_state;
1060                 }
1061                 if (((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
1062                          && ct->proto.tcp.last_index == TCP_SYN_SET)
1063                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
1064                             && ct->proto.tcp.last_index == TCP_ACK_SET))
1065                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
1066                         /* RST sent to invalid SYN or ACK we had let through
1067                          * at a) and c) above:
1068                          *
1069                          * a) SYN was in window then
1070                          * c) we hold a half-open connection.
1071                          *
1072                          * Delete our connection entry.
1073                          * We skip window checking, because packet might ACK
1074                          * segments we ignored. */
1075                         goto in_window;
1076                 }
1077                 break;
1078         default:
1079                 /* Keep compilers happy. */
1080                 break;
1081         }
1082
1083         if (!tcp_in_window(ct, dir, index,
1084                            skb, dataoff, th, state)) {
1085                 spin_unlock_bh(&ct->lock);
1086                 return -NF_ACCEPT;
1087         }
1088      in_window:
1089         /* From now on we have got in-window packets */
1090         ct->proto.tcp.last_index = index;
1091         ct->proto.tcp.last_dir = dir;
1092
1093         pr_debug("tcp_conntracks: ");
1094         nf_ct_dump_tuple(tuple);
1095         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1096                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1097                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1098                  old_state, new_state);
1099
1100         ct->proto.tcp.state = new_state;
1101         if (old_state != new_state
1102             && new_state == TCP_CONNTRACK_FIN_WAIT)
1103                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1104
1105         timeouts = nf_ct_timeout_lookup(ct);
1106         if (!timeouts)
1107                 timeouts = tn->timeouts;
1108
1109         if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1110             timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1111                 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1112         else if (unlikely(index == TCP_RST_SET))
1113                 timeout = timeouts[TCP_CONNTRACK_CLOSE];
1114         else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1115                  IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1116                  timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1117                 timeout = timeouts[TCP_CONNTRACK_UNACK];
1118         else if (ct->proto.tcp.last_win == 0 &&
1119                  timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1120                 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1121         else
1122                 timeout = timeouts[new_state];
1123         spin_unlock_bh(&ct->lock);
1124
1125         if (new_state != old_state)
1126                 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1127
1128         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1129                 /* If only reply is a RST, we can consider ourselves not to
1130                    have an established connection: this is a fairly common
1131                    problem case, so we can delete the conntrack
1132                    immediately.  --RR */
1133                 if (th->rst) {
1134                         nf_ct_kill_acct(ct, ctinfo, skb);
1135                         return NF_ACCEPT;
1136                 }
1137                 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1138                  * pickup with loose=1. Avoid large ESTABLISHED timeout.
1139                  */
1140                 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1141                     timeout > timeouts[TCP_CONNTRACK_UNACK])
1142                         timeout = timeouts[TCP_CONNTRACK_UNACK];
1143         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1144                    && (old_state == TCP_CONNTRACK_SYN_RECV
1145                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1146                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1147                 /* Set ASSURED if we see valid ack in ESTABLISHED
1148                    after SYN_RECV or a valid answer for a picked up
1149                    connection. */
1150                 set_bit(IPS_ASSURED_BIT, &ct->status);
1151                 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1152         }
1153         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1154
1155         return NF_ACCEPT;
1156 }
1157
1158 static bool tcp_can_early_drop(const struct nf_conn *ct)
1159 {
1160         switch (ct->proto.tcp.state) {
1161         case TCP_CONNTRACK_FIN_WAIT:
1162         case TCP_CONNTRACK_LAST_ACK:
1163         case TCP_CONNTRACK_TIME_WAIT:
1164         case TCP_CONNTRACK_CLOSE:
1165         case TCP_CONNTRACK_CLOSE_WAIT:
1166                 return true;
1167         default:
1168                 break;
1169         }
1170
1171         return false;
1172 }
1173
1174 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1175
1176 #include <linux/netfilter/nfnetlink.h>
1177 #include <linux/netfilter/nfnetlink_conntrack.h>
1178
1179 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1180                          struct nf_conn *ct, bool destroy)
1181 {
1182         struct nlattr *nest_parms;
1183         struct nf_ct_tcp_flags tmp = {};
1184
1185         spin_lock_bh(&ct->lock);
1186         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP);
1187         if (!nest_parms)
1188                 goto nla_put_failure;
1189
1190         if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state))
1191                 goto nla_put_failure;
1192
1193         if (destroy)
1194                 goto skip_state;
1195
1196         if (nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1197                        ct->proto.tcp.seen[0].td_scale) ||
1198             nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1199                        ct->proto.tcp.seen[1].td_scale))
1200                 goto nla_put_failure;
1201
1202         tmp.flags = ct->proto.tcp.seen[0].flags;
1203         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1204                     sizeof(struct nf_ct_tcp_flags), &tmp))
1205                 goto nla_put_failure;
1206
1207         tmp.flags = ct->proto.tcp.seen[1].flags;
1208         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1209                     sizeof(struct nf_ct_tcp_flags), &tmp))
1210                 goto nla_put_failure;
1211 skip_state:
1212         spin_unlock_bh(&ct->lock);
1213         nla_nest_end(skb, nest_parms);
1214
1215         return 0;
1216
1217 nla_put_failure:
1218         spin_unlock_bh(&ct->lock);
1219         return -1;
1220 }
1221
1222 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1223         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1224         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1225         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1226         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1227         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len = sizeof(struct nf_ct_tcp_flags) },
1228 };
1229
1230 #define TCP_NLATTR_SIZE ( \
1231         NLA_ALIGN(NLA_HDRLEN + 1) + \
1232         NLA_ALIGN(NLA_HDRLEN + 1) + \
1233         NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)) + \
1234         NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)))
1235
1236 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1237 {
1238         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1239         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1240         int err;
1241
1242         /* updates could not contain anything about the private
1243          * protocol info, in that case skip the parsing */
1244         if (!pattr)
1245                 return 0;
1246
1247         err = nla_parse_nested_deprecated(tb, CTA_PROTOINFO_TCP_MAX, pattr,
1248                                           tcp_nla_policy, NULL);
1249         if (err < 0)
1250                 return err;
1251
1252         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1253             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1254                 return -EINVAL;
1255
1256         spin_lock_bh(&ct->lock);
1257         if (tb[CTA_PROTOINFO_TCP_STATE])
1258                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1259
1260         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1261                 struct nf_ct_tcp_flags *attr =
1262                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1263                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1264                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1265         }
1266
1267         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1268                 struct nf_ct_tcp_flags *attr =
1269                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1270                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1271                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1272         }
1273
1274         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1275             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1276             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1277             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1278                 ct->proto.tcp.seen[0].td_scale =
1279                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1280                 ct->proto.tcp.seen[1].td_scale =
1281                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1282         }
1283         spin_unlock_bh(&ct->lock);
1284
1285         return 0;
1286 }
1287
1288 static unsigned int tcp_nlattr_tuple_size(void)
1289 {
1290         static unsigned int size __read_mostly;
1291
1292         if (!size)
1293                 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1294
1295         return size;
1296 }
1297 #endif
1298
1299 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1300
1301 #include <linux/netfilter/nfnetlink.h>
1302 #include <linux/netfilter/nfnetlink_cttimeout.h>
1303
1304 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1305                                      struct net *net, void *data)
1306 {
1307         struct nf_tcp_net *tn = nf_tcp_pernet(net);
1308         unsigned int *timeouts = data;
1309         int i;
1310
1311         if (!timeouts)
1312                 timeouts = tn->timeouts;
1313         /* set default TCP timeouts. */
1314         for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1315                 timeouts[i] = tn->timeouts[i];
1316
1317         if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1318                 timeouts[TCP_CONNTRACK_SYN_SENT] =
1319                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1320         }
1321
1322         if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1323                 timeouts[TCP_CONNTRACK_SYN_RECV] =
1324                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1325         }
1326         if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1327                 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1328                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1329         }
1330         if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1331                 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1332                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1333         }
1334         if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1335                 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1336                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1337         }
1338         if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1339                 timeouts[TCP_CONNTRACK_LAST_ACK] =
1340                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1341         }
1342         if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1343                 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1344                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1345         }
1346         if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1347                 timeouts[TCP_CONNTRACK_CLOSE] =
1348                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1349         }
1350         if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1351                 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1352                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1353         }
1354         if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1355                 timeouts[TCP_CONNTRACK_RETRANS] =
1356                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1357         }
1358         if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1359                 timeouts[TCP_CONNTRACK_UNACK] =
1360                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1361         }
1362
1363         timeouts[CTA_TIMEOUT_TCP_UNSPEC] = timeouts[CTA_TIMEOUT_TCP_SYN_SENT];
1364         return 0;
1365 }
1366
1367 static int
1368 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1369 {
1370         const unsigned int *timeouts = data;
1371
1372         if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1373                         htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1374             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1375                          htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1376             nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1377                          htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1378             nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1379                          htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1380             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1381                          htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1382             nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1383                          htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1384             nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1385                          htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1386             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1387                          htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1388             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1389                          htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1390             nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1391                          htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1392             nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1393                          htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1394                 goto nla_put_failure;
1395         return 0;
1396
1397 nla_put_failure:
1398         return -ENOSPC;
1399 }
1400
1401 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1402         [CTA_TIMEOUT_TCP_SYN_SENT]      = { .type = NLA_U32 },
1403         [CTA_TIMEOUT_TCP_SYN_RECV]      = { .type = NLA_U32 },
1404         [CTA_TIMEOUT_TCP_ESTABLISHED]   = { .type = NLA_U32 },
1405         [CTA_TIMEOUT_TCP_FIN_WAIT]      = { .type = NLA_U32 },
1406         [CTA_TIMEOUT_TCP_CLOSE_WAIT]    = { .type = NLA_U32 },
1407         [CTA_TIMEOUT_TCP_LAST_ACK]      = { .type = NLA_U32 },
1408         [CTA_TIMEOUT_TCP_TIME_WAIT]     = { .type = NLA_U32 },
1409         [CTA_TIMEOUT_TCP_CLOSE]         = { .type = NLA_U32 },
1410         [CTA_TIMEOUT_TCP_SYN_SENT2]     = { .type = NLA_U32 },
1411         [CTA_TIMEOUT_TCP_RETRANS]       = { .type = NLA_U32 },
1412         [CTA_TIMEOUT_TCP_UNACK]         = { .type = NLA_U32 },
1413 };
1414 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1415
1416 void nf_conntrack_tcp_init_net(struct net *net)
1417 {
1418         struct nf_tcp_net *tn = nf_tcp_pernet(net);
1419         int i;
1420
1421         for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1422                 tn->timeouts[i] = tcp_timeouts[i];
1423
1424         /* timeouts[0] is unused, make it same as SYN_SENT so
1425          * ->timeouts[0] contains 'new' timeout, like udp or icmp.
1426          */
1427         tn->timeouts[0] = tcp_timeouts[TCP_CONNTRACK_SYN_SENT];
1428
1429         /* If it is set to zero, we disable picking up already established
1430          * connections.
1431          */
1432         tn->tcp_loose = 1;
1433
1434         /* "Be conservative in what you do,
1435          *  be liberal in what you accept from others."
1436          * If it's non-zero, we mark only out of window RST segments as INVALID.
1437          */
1438         tn->tcp_be_liberal = 0;
1439
1440         /* Max number of the retransmitted packets without receiving an (acceptable)
1441          * ACK from the destination. If this number is reached, a shorter timer
1442          * will be started.
1443          */
1444         tn->tcp_max_retrans = 3;
1445
1446 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
1447         tn->offload_timeout = 30 * HZ;
1448         tn->offload_pickup = 120 * HZ;
1449 #endif
1450 }
1451
1452 const struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp =
1453 {
1454         .l4proto                = IPPROTO_TCP,
1455 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1456         .print_conntrack        = tcp_print_conntrack,
1457 #endif
1458         .can_early_drop         = tcp_can_early_drop,
1459 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1460         .to_nlattr              = tcp_to_nlattr,
1461         .from_nlattr            = nlattr_to_tcp,
1462         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1463         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1464         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1465         .nlattr_size            = TCP_NLATTR_SIZE,
1466         .nla_policy             = nf_ct_port_nla_policy,
1467 #endif
1468 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1469         .ctnl_timeout           = {
1470                 .nlattr_to_obj  = tcp_timeout_nlattr_to_obj,
1471                 .obj_to_nlattr  = tcp_timeout_obj_to_nlattr,
1472                 .nlattr_max     = CTA_TIMEOUT_TCP_MAX,
1473                 .obj_size       = sizeof(unsigned int) *
1474                                         TCP_CONNTRACK_TIMEOUT_MAX,
1475                 .nla_policy     = tcp_timeout_nla_policy,
1476         },
1477 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1478 };