1 /* (C) 1999-2001 Paul `Rusty' Russell
2 * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>:
9 * - Real stateful connection tracking
10 * - Modified state transitions table
11 * - Window scaling support added
12 * - SACK support added
15 * - State table bugfixes
16 * - More robust state changes
17 * - Tuning timer parameters
22 #include <linux/config.h>
23 #include <linux/types.h>
24 #include <linux/sched.h>
25 #include <linux/timer.h>
26 #include <linux/netfilter.h>
27 #include <linux/module.h>
30 #include <linux/tcp.h>
31 #include <linux/spinlock.h>
35 #include <linux/netfilter.h>
36 #include <linux/netfilter_ipv4.h>
37 #include <linux/netfilter_ipv4/ip_conntrack.h>
38 #include <linux/netfilter_ipv4/ip_conntrack_protocol.h>
39 #include <linux/netfilter_ipv4/lockhelp.h>
45 #define DEBUGP(format, args...)
48 /* Protects conntrack->proto.tcp */
49 static DECLARE_RWLOCK(tcp_lock);
51 /* "Be conservative in what you do,
52 be liberal in what you accept from others."
53 If it's non-zero, we mark only out of window RST segments as INVALID. */
54 int ip_ct_tcp_be_liberal = 0;
56 /* When connection is picked up from the middle, how many packets are required
57 to pass in each direction when we assume we are in sync - if any side uses
58 window scaling, we lost the game.
59 If it is set to zero, we disable picking up already established
61 int ip_ct_tcp_loose = 3;
63 /* Max number of the retransmitted packets without receiving an (acceptable)
64 ACK from the destination. If this number is reached, a shorter timer
66 int ip_ct_tcp_max_retrans = 3;
68 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
69 closely. They're more complex. --RR */
71 static const char *tcp_conntrack_names[] = {
85 #define MINS * 60 SECS
86 #define HOURS * 60 MINS
87 #define DAYS * 24 HOURS
89 unsigned long ip_ct_tcp_timeout_syn_sent = 2 MINS;
90 unsigned long ip_ct_tcp_timeout_syn_recv = 60 SECS;
91 unsigned long ip_ct_tcp_timeout_established = 2 DAYS;
92 unsigned long ip_ct_tcp_timeout_fin_wait = 2 MINS;
93 unsigned long ip_ct_tcp_timeout_close_wait = 60 SECS;
94 unsigned long ip_ct_tcp_timeout_last_ack = 30 SECS;
95 unsigned long ip_ct_tcp_timeout_time_wait = 2 MINS;
96 unsigned long ip_ct_tcp_timeout_close = 10 SECS;
98 /* RFC1122 says the R2 limit should be at least 100 seconds.
99 Linux uses 15 packets as limit, which corresponds
100 to ~13-30min depending on RTO. */
101 unsigned long ip_ct_tcp_timeout_max_retrans = 5 MINS;
103 static unsigned long * tcp_timeouts[]
104 = { NULL, /* TCP_CONNTRACK_NONE */
105 &ip_ct_tcp_timeout_syn_sent, /* TCP_CONNTRACK_SYN_SENT, */
106 &ip_ct_tcp_timeout_syn_recv, /* TCP_CONNTRACK_SYN_RECV, */
107 &ip_ct_tcp_timeout_established, /* TCP_CONNTRACK_ESTABLISHED, */
108 &ip_ct_tcp_timeout_fin_wait, /* TCP_CONNTRACK_FIN_WAIT, */
109 &ip_ct_tcp_timeout_close_wait, /* TCP_CONNTRACK_CLOSE_WAIT, */
110 &ip_ct_tcp_timeout_last_ack, /* TCP_CONNTRACK_LAST_ACK, */
111 &ip_ct_tcp_timeout_time_wait, /* TCP_CONNTRACK_TIME_WAIT, */
112 &ip_ct_tcp_timeout_close, /* TCP_CONNTRACK_CLOSE, */
113 NULL, /* TCP_CONNTRACK_LISTEN */
116 #define sNO TCP_CONNTRACK_NONE
117 #define sSS TCP_CONNTRACK_SYN_SENT
118 #define sSR TCP_CONNTRACK_SYN_RECV
119 #define sES TCP_CONNTRACK_ESTABLISHED
120 #define sFW TCP_CONNTRACK_FIN_WAIT
121 #define sCW TCP_CONNTRACK_CLOSE_WAIT
122 #define sLA TCP_CONNTRACK_LAST_ACK
123 #define sTW TCP_CONNTRACK_TIME_WAIT
124 #define sCL TCP_CONNTRACK_CLOSE
125 #define sLI TCP_CONNTRACK_LISTEN
126 #define sIV TCP_CONNTRACK_MAX
127 #define sIG TCP_CONNTRACK_IGNORE
129 /* What TCP flags are set from RST/SYN/FIN/ACK. */
140 * The TCP state transition table needs a few words...
142 * We are the man in the middle. All the packets go through us
143 * but might get lost in transit to the destination.
144 * It is assumed that the destinations can't receive segments
147 * The checked segment is in window, but our windows are *not*
148 * equivalent with the ones of the sender/receiver. We always
149 * try to guess the state of the current sender.
151 * The meaning of the states are:
153 * NONE: initial state
154 * SYN_SENT: SYN-only packet seen
155 * SYN_RECV: SYN-ACK packet seen
156 * ESTABLISHED: ACK packet seen
157 * FIN_WAIT: FIN packet seen
158 * CLOSE_WAIT: ACK seen (after FIN)
159 * LAST_ACK: FIN seen (after FIN)
160 * TIME_WAIT: last ACK seen
161 * CLOSE: closed connection
163 * LISTEN state is not used.
165 * Packets marked as IGNORED (sIG):
166 * if they may be either invalid or valid
167 * and the receiver may send back a connection
168 * closing RST or a SYN/ACK.
170 * Packets marked as INVALID (sIV):
171 * if they are invalid
172 * or we do not support the request (simultaneous open)
174 static enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
177 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
178 /*syn*/ { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
180 * sNO -> sSS Initialize a new connection
181 * sSS -> sSS Retransmitted SYN
182 * sSR -> sIG Late retransmitted SYN?
183 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
184 * are errors. Receiver will reply with RST
185 * and close the connection.
186 * Or we are not in sync and hold a dead connection.
190 * sTW -> sSS Reopened connection (RFC 1122).
193 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
194 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
196 * A SYN/ACK from the client is always invalid:
197 * - either it tries to set up a simultaneous open, which is
199 * - or the firewall has just been inserted between the two hosts
200 * during the session set-up. The SYN will be retransmitted
201 * by the true client (or it'll time out).
203 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
204 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
206 * sNO -> sIV Too late and no reason to do anything...
207 * sSS -> sIV Client migth not send FIN in this state:
208 * we enforce waiting for a SYN/ACK reply first.
209 * sSR -> sFW Close started.
211 * sFW -> sLA FIN seen in both directions, waiting for
213 * Migth be a retransmitted FIN as well...
215 * sLA -> sLA Retransmitted FIN. Remain in the same state.
219 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
220 /*ack*/ { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
222 * sNO -> sES Assumed.
223 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
224 * sSR -> sES Established state is reached.
226 * sFW -> sCW Normal close request answered by ACK.
228 * sLA -> sTW Last ACK detected.
229 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
232 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
233 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
234 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
238 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
239 /*syn*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
241 * sNO -> sIV Never reached.
242 * sSS -> sIV Simultaneous open, not supported
243 * sSR -> sIV Simultaneous open, not supported.
244 * sES -> sIV Server may not initiate a connection.
248 * sTW -> sIV Reopened connection, but server may not do it.
251 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
252 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
254 * sSS -> sSR Standard open.
255 * sSR -> sSR Retransmitted SYN/ACK.
256 * sES -> sIG Late retransmitted SYN/ACK?
257 * sFW -> sIG Might be SYN/ACK answering ignored SYN
263 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
264 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
266 * sSS -> sIV Server might not send FIN in this state.
267 * sSR -> sFW Close started.
269 * sFW -> sLA FIN seen in both directions.
271 * sLA -> sLA Retransmitted FIN.
275 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
276 /*ack*/ { sIV, sIV, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
278 * sSS -> sIV Might be a half-open connection.
279 * sSR -> sSR Might answer late resent SYN.
281 * sFW -> sCW Normal close request answered by ACK.
283 * sLA -> sTW Last ACK detected.
284 * sTW -> sTW Retransmitted last ACK.
287 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
288 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
289 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
293 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
294 unsigned int dataoff,
295 struct ip_conntrack_tuple *tuple)
297 struct tcphdr _hdr, *hp;
299 /* Actually only need first 8 bytes. */
300 hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
304 tuple->src.u.tcp.port = hp->source;
305 tuple->dst.u.tcp.port = hp->dest;
310 static int tcp_invert_tuple(struct ip_conntrack_tuple *tuple,
311 const struct ip_conntrack_tuple *orig)
313 tuple->src.u.tcp.port = orig->dst.u.tcp.port;
314 tuple->dst.u.tcp.port = orig->src.u.tcp.port;
318 /* Print out the per-protocol part of the tuple. */
319 static int tcp_print_tuple(struct seq_file *s,
320 const struct ip_conntrack_tuple *tuple)
322 return seq_printf(s, "sport=%hu dport=%hu ",
323 ntohs(tuple->src.u.tcp.port),
324 ntohs(tuple->dst.u.tcp.port));
327 /* Print out the private part of the conntrack. */
328 static int tcp_print_conntrack(struct seq_file *s,
329 const struct ip_conntrack *conntrack)
331 enum tcp_conntrack state;
333 READ_LOCK(&tcp_lock);
334 state = conntrack->proto.tcp.state;
335 READ_UNLOCK(&tcp_lock);
337 return seq_printf(s, "%s ", tcp_conntrack_names[state]);
340 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
342 if (tcph->rst) return TCP_RST_SET;
343 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
344 else if (tcph->fin) return TCP_FIN_SET;
345 else if (tcph->ack) return TCP_ACK_SET;
346 else return TCP_NONE_SET;
349 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
350 in IP Filter' by Guido van Rooij.
352 http://www.nluug.nl/events/sane2000/papers.html
353 http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
355 The boundaries and the conditions are changed according to RFC793:
356 the packet must intersect the window (i.e. segments may be
357 after the right or before the left edge) and thus receivers may ACK
358 segments after the right edge of the window.
360 td_maxend = max(sack + max(win,1)) seen in reply packets
361 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
362 td_maxwin += seq + len - sender.td_maxend
363 if seq + len > sender.td_maxend
364 td_end = max(seq + len) seen in sent packets
366 I. Upper bound for valid data: seq <= sender.td_maxend
367 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
368 III. Upper bound for valid ack: sack <= receiver.td_end
369 IV. Lower bound for valid ack: ack >= receiver.td_end - MAXACKWINDOW
371 where sack is the highest right edge of sack block found in the packet.
373 The upper bound limit for a valid ack is not ignored -
374 we doesn't have to deal with fragments.
377 static inline __u32 segment_seq_plus_len(__u32 seq,
382 return (seq + len - (iph->ihl + tcph->doff)*4
383 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
386 /* Fixme: what about big packets? */
387 #define MAXACKWINCONST 66000
388 #define MAXACKWINDOW(sender) \
389 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
393 * Simplified tcp_parse_options routine from tcp_input.c
395 static void tcp_options(const struct sk_buff *skb,
398 struct ip_ct_tcp_state *state)
400 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
402 int length = (tcph->doff*4) - sizeof(struct tcphdr);
407 ptr = skb_header_pointer(skb,
408 (iph->ihl * 4) + sizeof(struct tcphdr),
422 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
427 if (opsize < 2) /* "silly options" */
430 break; /* don't parse partial options */
432 if (opcode == TCPOPT_SACK_PERM
433 && opsize == TCPOLEN_SACK_PERM)
434 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
435 else if (opcode == TCPOPT_WINDOW
436 && opsize == TCPOLEN_WINDOW) {
437 state->td_scale = *(u_int8_t *)ptr;
439 if (state->td_scale > 14) {
441 state->td_scale = 14;
444 IP_CT_TCP_FLAG_WINDOW_SCALE;
452 static void tcp_sack(const struct sk_buff *skb,
457 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
459 int length = (tcph->doff*4) - sizeof(struct tcphdr);
465 ptr = skb_header_pointer(skb,
466 (iph->ihl * 4) + sizeof(struct tcphdr),
470 /* Fast path for timestamp-only option */
471 if (length == TCPOLEN_TSTAMP_ALIGNED*4
473 __constant_ntohl((TCPOPT_NOP << 24)
475 | (TCPOPT_TIMESTAMP << 8)
476 | TCPOLEN_TIMESTAMP))
486 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
491 if (opsize < 2) /* "silly options" */
494 break; /* don't parse partial options */
496 if (opcode == TCPOPT_SACK
497 && opsize >= (TCPOLEN_SACK_BASE
498 + TCPOLEN_SACK_PERBLOCK)
499 && !((opsize - TCPOLEN_SACK_BASE)
500 % TCPOLEN_SACK_PERBLOCK)) {
502 i < (opsize - TCPOLEN_SACK_BASE);
503 i += TCPOLEN_SACK_PERBLOCK) {
504 tmp = ntohl(*((u_int32_t *)(ptr+i)+1));
506 if (after(tmp, *sack))
517 static int tcp_in_window(struct ip_ct_tcp *state,
518 enum ip_conntrack_dir dir,
520 const struct sk_buff *skb,
524 struct ip_ct_tcp_state *sender = &state->seen[dir];
525 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
526 __u32 seq, ack, sack, end, win, swin;
530 * Get the required data from the packet.
532 seq = ntohl(tcph->seq);
533 ack = sack = ntohl(tcph->ack_seq);
534 win = ntohs(tcph->window);
535 end = segment_seq_plus_len(seq, skb->len, iph, tcph);
537 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
538 tcp_sack(skb, iph, tcph, &sack);
540 DEBUGP("tcp_in_window: START\n");
541 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
542 "seq=%u ack=%u sack=%u win=%u end=%u\n",
543 NIPQUAD(iph->saddr), ntohs(tcph->source),
544 NIPQUAD(iph->daddr), ntohs(tcph->dest),
545 seq, ack, sack, win, end);
546 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
547 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
548 sender->td_end, sender->td_maxend, sender->td_maxwin,
550 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
553 if (sender->td_end == 0) {
555 * Initialize sender data.
557 if (tcph->syn && tcph->ack) {
559 * Outgoing SYN-ACK in reply to a SYN.
562 sender->td_maxend = end;
563 sender->td_maxwin = (win == 0 ? 1 : win);
565 tcp_options(skb, iph, tcph, sender);
568 * Both sides must send the Window Scale option
569 * to enable window scaling in either direction.
571 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
572 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
574 receiver->td_scale = 0;
577 * We are in the middle of a connection,
578 * its history is lost for us.
579 * Let's try to use the data from the packet.
581 sender->td_end = end;
582 sender->td_maxwin = (win == 0 ? 1 : win);
583 sender->td_maxend = end + sender->td_maxwin;
585 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
586 && dir == IP_CT_DIR_ORIGINAL)
587 || (state->state == TCP_CONNTRACK_SYN_RECV
588 && dir == IP_CT_DIR_REPLY))
589 && after(end, sender->td_end)) {
591 * RFC 793: "if a TCP is reinitialized ... then it need
592 * not wait at all; it must only be sure to use sequence
593 * numbers larger than those recently used."
596 sender->td_maxend = end;
597 sender->td_maxwin = (win == 0 ? 1 : win);
599 tcp_options(skb, iph, tcph, sender);
604 * If there is no ACK, just pretend it was set and OK.
606 ack = sack = receiver->td_end;
607 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
608 (TCP_FLAG_ACK|TCP_FLAG_RST))
611 * Broken TCP stacks, that set ACK in RST packets as well
612 * with zero ack value.
614 ack = sack = receiver->td_end;
619 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
621 * Packets contains no data: we assume it is valid
622 * and check the ack value only.
623 * However RST segments are always validated by their
624 * SEQ number, except when seq == 0 (reset sent answering
627 seq = end = sender->td_end;
629 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
630 "seq=%u ack=%u sack =%u win=%u end=%u\n",
631 NIPQUAD(iph->saddr), ntohs(tcph->source),
632 NIPQUAD(iph->daddr), ntohs(tcph->dest),
633 seq, ack, sack, win, end);
634 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
635 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
636 sender->td_end, sender->td_maxend, sender->td_maxwin,
638 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
641 DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
642 before(seq, sender->td_maxend + 1),
643 after(end, sender->td_end - receiver->td_maxwin - 1),
644 before(sack, receiver->td_end + 1),
645 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
647 if (sender->loose || receiver->loose ||
648 (before(seq, sender->td_maxend + 1) &&
649 after(end, sender->td_end - receiver->td_maxwin - 1) &&
650 before(sack, receiver->td_end + 1) &&
651 after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
653 * Take into account window scaling (RFC 1323).
656 win <<= sender->td_scale;
659 * Update sender data.
661 swin = win + (sack - ack);
662 if (sender->td_maxwin < swin)
663 sender->td_maxwin = swin;
664 if (after(end, sender->td_end))
665 sender->td_end = end;
667 * Update receiver data.
669 if (after(end, sender->td_maxend))
670 receiver->td_maxwin += end - sender->td_maxend;
671 if (after(sack + win, receiver->td_maxend - 1)) {
672 receiver->td_maxend = sack + win;
674 receiver->td_maxend++;
678 * Check retransmissions.
680 if (index == TCP_ACK_SET) {
681 if (state->last_dir == dir
682 && state->last_seq == seq
683 && state->last_ack == ack
684 && state->last_end == end)
687 state->last_dir = dir;
688 state->last_seq = seq;
689 state->last_ack = ack;
690 state->last_end = end;
695 * Close the window of disabled window tracking :-)
702 if (LOG_INVALID(IPPROTO_TCP))
703 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
705 before(seq, sender->td_maxend + 1) ?
706 after(end, sender->td_end - receiver->td_maxwin - 1) ?
707 before(sack, receiver->td_end + 1) ?
708 after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
709 : "ACK is under the lower bound (possible overly delayed ACK)"
710 : "ACK is over the upper bound (ACKed data not seen yet)"
711 : "SEQ is under the lower bound (already ACKed data retransmitted)"
712 : "SEQ is over the upper bound (over the window of the receiver)");
714 res = ip_ct_tcp_be_liberal;
717 DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
718 "receiver end=%u maxend=%u maxwin=%u\n",
719 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
720 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
725 #ifdef CONFIG_IP_NF_NAT_NEEDED
726 /* Update sender->td_end after NAT successfully mangled the packet */
727 void ip_conntrack_tcp_update(struct sk_buff *skb,
728 struct ip_conntrack *conntrack,
729 enum ip_conntrack_dir dir)
731 struct iphdr *iph = skb->nh.iph;
732 struct tcphdr *tcph = (void *)skb->nh.iph + skb->nh.iph->ihl*4;
735 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
736 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
739 end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, iph, tcph);
741 WRITE_LOCK(&tcp_lock);
743 * We have to worry for the ack in the reply packet only...
745 if (after(end, conntrack->proto.tcp.seen[dir].td_end))
746 conntrack->proto.tcp.seen[dir].td_end = end;
747 conntrack->proto.tcp.last_end = end;
748 WRITE_UNLOCK(&tcp_lock);
749 DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
750 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
751 sender->td_end, sender->td_maxend, sender->td_maxwin,
753 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
768 /* table of valid flag combinations - ECE and CWR are always valid */
769 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
773 [TH_SYN|TH_ACK|TH_PUSH] = 1,
776 [TH_RST|TH_ACK|TH_PUSH] = 1,
779 [TH_ACK|TH_PUSH] = 1,
781 [TH_ACK|TH_URG|TH_PUSH] = 1,
782 [TH_FIN|TH_ACK|TH_PUSH] = 1,
783 [TH_FIN|TH_ACK|TH_URG] = 1,
784 [TH_FIN|TH_ACK|TH_URG|TH_PUSH] = 1,
787 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
788 static int tcp_error(struct sk_buff *skb,
789 enum ip_conntrack_info *ctinfo,
790 unsigned int hooknum)
792 struct iphdr *iph = skb->nh.iph;
793 struct tcphdr _tcph, *th;
794 unsigned int tcplen = skb->len - iph->ihl * 4;
797 /* Smaller that minimal TCP header? */
798 th = skb_header_pointer(skb, iph->ihl * 4,
799 sizeof(_tcph), &_tcph);
801 if (LOG_INVALID(IPPROTO_TCP))
802 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
803 "ip_ct_tcp: short packet ");
807 /* Not whole TCP header or malformed packet */
808 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
809 if (LOG_INVALID(IPPROTO_TCP))
810 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
811 "ip_ct_tcp: truncated/malformed packet ");
815 /* Checksum invalid? Ignore.
816 * We skip checking packets on the outgoing path
817 * because the semantic of CHECKSUM_HW is different there
818 * and moreover root might send raw packets.
820 /* FIXME: Source route IP option packets --RR */
821 if (hooknum == NF_IP_PRE_ROUTING
822 && skb->ip_summed != CHECKSUM_UNNECESSARY
823 && csum_tcpudp_magic(iph->saddr, iph->daddr, tcplen, IPPROTO_TCP,
824 skb->ip_summed == CHECKSUM_HW ? skb->csum
825 : skb_checksum(skb, iph->ihl*4, tcplen, 0))) {
826 if (LOG_INVALID(IPPROTO_TCP))
827 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
828 "ip_ct_tcp: bad TCP checksum ");
832 /* Check TCP flags. */
833 tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
834 if (!tcp_valid_flags[tcpflags]) {
835 if (LOG_INVALID(IPPROTO_TCP))
836 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
837 "ip_ct_tcp: invalid TCP flag combination ");
844 /* Returns verdict for packet, or -1 for invalid. */
845 static int tcp_packet(struct ip_conntrack *conntrack,
846 const struct sk_buff *skb,
847 enum ip_conntrack_info ctinfo)
849 enum tcp_conntrack new_state, old_state;
850 enum ip_conntrack_dir dir;
851 struct iphdr *iph = skb->nh.iph;
852 struct tcphdr *th, _tcph;
853 unsigned long timeout;
856 th = skb_header_pointer(skb, iph->ihl * 4,
857 sizeof(_tcph), &_tcph);
860 WRITE_LOCK(&tcp_lock);
861 old_state = conntrack->proto.tcp.state;
862 dir = CTINFO2DIR(ctinfo);
863 index = get_conntrack_index(th);
864 new_state = tcp_conntracks[dir][index][old_state];
867 case TCP_CONNTRACK_IGNORE:
868 /* Either SYN in ORIGINAL
869 * or SYN/ACK in REPLY. */
870 if (index == TCP_SYNACK_SET
871 && conntrack->proto.tcp.last_index == TCP_SYN_SET
872 && conntrack->proto.tcp.last_dir != dir
873 && ntohl(th->ack_seq) ==
874 conntrack->proto.tcp.last_end) {
875 /* This SYN/ACK acknowledges a SYN that we earlier
876 * ignored as invalid. This means that the client and
877 * the server are both in sync, while the firewall is
878 * not. We kill this session and block the SYN/ACK so
879 * that the client cannot but retransmit its SYN and
880 * thus initiate a clean new session.
882 WRITE_UNLOCK(&tcp_lock);
883 if (LOG_INVALID(IPPROTO_TCP))
884 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
885 "ip_ct_tcp: killing out of sync session ");
886 if (del_timer(&conntrack->timeout))
887 conntrack->timeout.function((unsigned long)
891 conntrack->proto.tcp.last_index = index;
892 conntrack->proto.tcp.last_dir = dir;
893 conntrack->proto.tcp.last_seq = ntohl(th->seq);
894 conntrack->proto.tcp.last_end =
895 segment_seq_plus_len(ntohl(th->seq), skb->len, iph, th);
897 WRITE_UNLOCK(&tcp_lock);
898 if (LOG_INVALID(IPPROTO_TCP))
899 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
900 "ip_ct_tcp: invalid packet ignored ");
902 case TCP_CONNTRACK_MAX:
904 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
905 dir, get_conntrack_index(th),
907 WRITE_UNLOCK(&tcp_lock);
908 if (LOG_INVALID(IPPROTO_TCP))
909 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
910 "ip_ct_tcp: invalid state ");
912 case TCP_CONNTRACK_SYN_SENT:
913 if (old_state < TCP_CONNTRACK_TIME_WAIT)
915 if ((conntrack->proto.tcp.seen[dir].flags &
916 IP_CT_TCP_FLAG_CLOSE_INIT)
917 || after(ntohl(th->seq),
918 conntrack->proto.tcp.seen[dir].td_end)) {
919 /* Attempt to reopen a closed connection.
920 * Delete this connection and look up again. */
921 WRITE_UNLOCK(&tcp_lock);
922 if (del_timer(&conntrack->timeout))
923 conntrack->timeout.function((unsigned long)
927 WRITE_UNLOCK(&tcp_lock);
928 if (LOG_INVALID(IPPROTO_TCP))
929 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
930 "ip_ct_tcp: invalid SYN");
933 case TCP_CONNTRACK_CLOSE:
934 if (index == TCP_RST_SET
935 && test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
936 && conntrack->proto.tcp.last_index == TCP_SYN_SET
937 && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
938 /* RST sent to invalid SYN we had let trough
939 * SYN was in window then, tear down connection.
940 * We skip window checking, because packet might ACK
941 * segments we ignored in the SYN. */
944 /* Just fall trough */
946 /* Keep compilers happy. */
950 if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
952 WRITE_UNLOCK(&tcp_lock);
956 /* From now on we have got in-window packets */
957 conntrack->proto.tcp.last_index = index;
959 DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
960 "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
961 NIPQUAD(iph->saddr), ntohs(th->source),
962 NIPQUAD(iph->daddr), ntohs(th->dest),
963 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
964 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
965 old_state, new_state);
967 conntrack->proto.tcp.state = new_state;
968 if (old_state != new_state
969 && (new_state == TCP_CONNTRACK_FIN_WAIT
970 || new_state == TCP_CONNTRACK_CLOSE))
971 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
972 timeout = conntrack->proto.tcp.retrans >= ip_ct_tcp_max_retrans
973 && *tcp_timeouts[new_state] > ip_ct_tcp_timeout_max_retrans
974 ? ip_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
975 WRITE_UNLOCK(&tcp_lock);
977 if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
978 /* If only reply is a RST, we can consider ourselves not to
979 have an established connection: this is a fairly common
980 problem case, so we can delete the conntrack
983 if (del_timer(&conntrack->timeout))
984 conntrack->timeout.function((unsigned long)
988 } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
989 && (old_state == TCP_CONNTRACK_SYN_RECV
990 || old_state == TCP_CONNTRACK_ESTABLISHED)
991 && new_state == TCP_CONNTRACK_ESTABLISHED) {
992 /* Set ASSURED if we see see valid ack in ESTABLISHED
993 after SYN_RECV or a valid answer for a picked up
995 set_bit(IPS_ASSURED_BIT, &conntrack->status);
997 ip_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1002 /* Called when a new connection for this protocol found. */
1003 static int tcp_new(struct ip_conntrack *conntrack,
1004 const struct sk_buff *skb)
1006 enum tcp_conntrack new_state;
1007 struct iphdr *iph = skb->nh.iph;
1008 struct tcphdr *th, _tcph;
1010 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1011 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1014 th = skb_header_pointer(skb, iph->ihl * 4,
1015 sizeof(_tcph), &_tcph);
1018 /* Don't need lock here: this conntrack not in circulation yet */
1020 = tcp_conntracks[0][get_conntrack_index(th)]
1021 [TCP_CONNTRACK_NONE];
1023 /* Invalid: delete conntrack */
1024 if (new_state >= TCP_CONNTRACK_MAX) {
1025 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1029 if (new_state == TCP_CONNTRACK_SYN_SENT) {
1031 conntrack->proto.tcp.seen[0].td_end =
1032 segment_seq_plus_len(ntohl(th->seq), skb->len,
1034 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1035 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1036 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1037 conntrack->proto.tcp.seen[0].td_maxend =
1038 conntrack->proto.tcp.seen[0].td_end;
1040 tcp_options(skb, iph, th, &conntrack->proto.tcp.seen[0]);
1041 conntrack->proto.tcp.seen[1].flags = 0;
1042 conntrack->proto.tcp.seen[0].loose =
1043 conntrack->proto.tcp.seen[1].loose = 0;
1044 } else if (ip_ct_tcp_loose == 0) {
1045 /* Don't try to pick up connections. */
1049 * We are in the middle of a connection,
1050 * its history is lost for us.
1051 * Let's try to use the data from the packet.
1053 conntrack->proto.tcp.seen[0].td_end =
1054 segment_seq_plus_len(ntohl(th->seq), skb->len,
1056 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1057 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1058 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1059 conntrack->proto.tcp.seen[0].td_maxend =
1060 conntrack->proto.tcp.seen[0].td_end +
1061 conntrack->proto.tcp.seen[0].td_maxwin;
1062 conntrack->proto.tcp.seen[0].td_scale = 0;
1064 /* We assume SACK. Should we assume window scaling too? */
1065 conntrack->proto.tcp.seen[0].flags =
1066 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1067 conntrack->proto.tcp.seen[0].loose =
1068 conntrack->proto.tcp.seen[1].loose = ip_ct_tcp_loose;
1071 conntrack->proto.tcp.seen[1].td_end = 0;
1072 conntrack->proto.tcp.seen[1].td_maxend = 0;
1073 conntrack->proto.tcp.seen[1].td_maxwin = 1;
1074 conntrack->proto.tcp.seen[1].td_scale = 0;
1076 /* tcp_packet will set them */
1077 conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1078 conntrack->proto.tcp.last_index = TCP_NONE_SET;
1080 DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1081 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1082 sender->td_end, sender->td_maxend, sender->td_maxwin,
1084 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1085 receiver->td_scale);
1089 struct ip_conntrack_protocol ip_conntrack_protocol_tcp =
1091 .proto = IPPROTO_TCP,
1093 .pkt_to_tuple = tcp_pkt_to_tuple,
1094 .invert_tuple = tcp_invert_tuple,
1095 .print_tuple = tcp_print_tuple,
1096 .print_conntrack = tcp_print_conntrack,
1097 .packet = tcp_packet,