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>
44 #define DEBUGP(format, args...)
47 /* Protects conntrack->proto.tcp */
48 static DEFINE_RWLOCK(tcp_lock);
50 /* "Be conservative in what you do,
51 be liberal in what you accept from others."
52 If it's non-zero, we mark only out of window RST segments as INVALID. */
53 int ip_ct_tcp_be_liberal = 0;
55 /* When connection is picked up from the middle, how many packets are required
56 to pass in each direction when we assume we are in sync - if any side uses
57 window scaling, we lost the game.
58 If it is set to zero, we disable picking up already established
60 int ip_ct_tcp_loose = 3;
62 /* Max number of the retransmitted packets without receiving an (acceptable)
63 ACK from the destination. If this number is reached, a shorter timer
65 int ip_ct_tcp_max_retrans = 3;
67 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
68 closely. They're more complex. --RR */
70 static const char *tcp_conntrack_names[] = {
84 #define MINS * 60 SECS
85 #define HOURS * 60 MINS
86 #define DAYS * 24 HOURS
88 unsigned long ip_ct_tcp_timeout_syn_sent = 2 MINS;
89 unsigned long ip_ct_tcp_timeout_syn_recv = 60 SECS;
90 unsigned long ip_ct_tcp_timeout_established = 5 DAYS;
91 unsigned long ip_ct_tcp_timeout_fin_wait = 2 MINS;
92 unsigned long ip_ct_tcp_timeout_close_wait = 60 SECS;
93 unsigned long ip_ct_tcp_timeout_last_ack = 30 SECS;
94 unsigned long ip_ct_tcp_timeout_time_wait = 2 MINS;
95 unsigned long ip_ct_tcp_timeout_close = 10 SECS;
97 /* RFC1122 says the R2 limit should be at least 100 seconds.
98 Linux uses 15 packets as limit, which corresponds
99 to ~13-30min depending on RTO. */
100 unsigned long ip_ct_tcp_timeout_max_retrans = 5 MINS;
102 static unsigned long * tcp_timeouts[]
103 = { NULL, /* TCP_CONNTRACK_NONE */
104 &ip_ct_tcp_timeout_syn_sent, /* TCP_CONNTRACK_SYN_SENT, */
105 &ip_ct_tcp_timeout_syn_recv, /* TCP_CONNTRACK_SYN_RECV, */
106 &ip_ct_tcp_timeout_established, /* TCP_CONNTRACK_ESTABLISHED, */
107 &ip_ct_tcp_timeout_fin_wait, /* TCP_CONNTRACK_FIN_WAIT, */
108 &ip_ct_tcp_timeout_close_wait, /* TCP_CONNTRACK_CLOSE_WAIT, */
109 &ip_ct_tcp_timeout_last_ack, /* TCP_CONNTRACK_LAST_ACK, */
110 &ip_ct_tcp_timeout_time_wait, /* TCP_CONNTRACK_TIME_WAIT, */
111 &ip_ct_tcp_timeout_close, /* TCP_CONNTRACK_CLOSE, */
112 NULL, /* TCP_CONNTRACK_LISTEN */
115 #define sNO TCP_CONNTRACK_NONE
116 #define sSS TCP_CONNTRACK_SYN_SENT
117 #define sSR TCP_CONNTRACK_SYN_RECV
118 #define sES TCP_CONNTRACK_ESTABLISHED
119 #define sFW TCP_CONNTRACK_FIN_WAIT
120 #define sCW TCP_CONNTRACK_CLOSE_WAIT
121 #define sLA TCP_CONNTRACK_LAST_ACK
122 #define sTW TCP_CONNTRACK_TIME_WAIT
123 #define sCL TCP_CONNTRACK_CLOSE
124 #define sLI TCP_CONNTRACK_LISTEN
125 #define sIV TCP_CONNTRACK_MAX
126 #define sIG TCP_CONNTRACK_IGNORE
128 /* What TCP flags are set from RST/SYN/FIN/ACK. */
139 * The TCP state transition table needs a few words...
141 * We are the man in the middle. All the packets go through us
142 * but might get lost in transit to the destination.
143 * It is assumed that the destinations can't receive segments
146 * The checked segment is in window, but our windows are *not*
147 * equivalent with the ones of the sender/receiver. We always
148 * try to guess the state of the current sender.
150 * The meaning of the states are:
152 * NONE: initial state
153 * SYN_SENT: SYN-only packet seen
154 * SYN_RECV: SYN-ACK packet seen
155 * ESTABLISHED: ACK packet seen
156 * FIN_WAIT: FIN packet seen
157 * CLOSE_WAIT: ACK seen (after FIN)
158 * LAST_ACK: FIN seen (after FIN)
159 * TIME_WAIT: last ACK seen
160 * CLOSE: closed connection
162 * LISTEN state is not used.
164 * Packets marked as IGNORED (sIG):
165 * if they may be either invalid or valid
166 * and the receiver may send back a connection
167 * closing RST or a SYN/ACK.
169 * Packets marked as INVALID (sIV):
170 * if they are invalid
171 * or we do not support the request (simultaneous open)
173 static enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
176 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
177 /*syn*/ { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
179 * sNO -> sSS Initialize a new connection
180 * sSS -> sSS Retransmitted SYN
181 * sSR -> sIG Late retransmitted SYN?
182 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
183 * are errors. Receiver will reply with RST
184 * and close the connection.
185 * Or we are not in sync and hold a dead connection.
189 * sTW -> sSS Reopened connection (RFC 1122).
192 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
193 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
195 * A SYN/ACK from the client is always invalid:
196 * - either it tries to set up a simultaneous open, which is
198 * - or the firewall has just been inserted between the two hosts
199 * during the session set-up. The SYN will be retransmitted
200 * by the true client (or it'll time out).
202 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
203 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
205 * sNO -> sIV Too late and no reason to do anything...
206 * sSS -> sIV Client migth not send FIN in this state:
207 * we enforce waiting for a SYN/ACK reply first.
208 * sSR -> sFW Close started.
210 * sFW -> sLA FIN seen in both directions, waiting for
212 * Migth be a retransmitted FIN as well...
214 * sLA -> sLA Retransmitted FIN. Remain in the same state.
218 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
219 /*ack*/ { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
221 * sNO -> sES Assumed.
222 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
223 * sSR -> sES Established state is reached.
225 * sFW -> sCW Normal close request answered by ACK.
227 * sLA -> sTW Last ACK detected.
228 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
231 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
232 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
233 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
237 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
238 /*syn*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
240 * sNO -> sIV Never reached.
241 * sSS -> sIV Simultaneous open, not supported
242 * sSR -> sIV Simultaneous open, not supported.
243 * sES -> sIV Server may not initiate a connection.
247 * sTW -> sIV Reopened connection, but server may not do it.
250 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
251 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
253 * sSS -> sSR Standard open.
254 * sSR -> sSR Retransmitted SYN/ACK.
255 * sES -> sIG Late retransmitted SYN/ACK?
256 * sFW -> sIG Might be SYN/ACK answering ignored SYN
262 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
263 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
265 * sSS -> sIV Server might not send FIN in this state.
266 * sSR -> sFW Close started.
268 * sFW -> sLA FIN seen in both directions.
270 * sLA -> sLA Retransmitted FIN.
274 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
275 /*ack*/ { sIV, sIV, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
277 * sSS -> sIV Might be a half-open connection.
278 * sSR -> sSR Might answer late resent SYN.
280 * sFW -> sCW Normal close request answered by ACK.
282 * sLA -> sTW Last ACK detected.
283 * sTW -> sTW Retransmitted last ACK.
286 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
287 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
288 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
292 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
293 unsigned int dataoff,
294 struct ip_conntrack_tuple *tuple)
296 struct tcphdr _hdr, *hp;
298 /* Actually only need first 8 bytes. */
299 hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
303 tuple->src.u.tcp.port = hp->source;
304 tuple->dst.u.tcp.port = hp->dest;
309 static int tcp_invert_tuple(struct ip_conntrack_tuple *tuple,
310 const struct ip_conntrack_tuple *orig)
312 tuple->src.u.tcp.port = orig->dst.u.tcp.port;
313 tuple->dst.u.tcp.port = orig->src.u.tcp.port;
317 /* Print out the per-protocol part of the tuple. */
318 static int tcp_print_tuple(struct seq_file *s,
319 const struct ip_conntrack_tuple *tuple)
321 return seq_printf(s, "sport=%hu dport=%hu ",
322 ntohs(tuple->src.u.tcp.port),
323 ntohs(tuple->dst.u.tcp.port));
326 /* Print out the private part of the conntrack. */
327 static int tcp_print_conntrack(struct seq_file *s,
328 const struct ip_conntrack *conntrack)
330 enum tcp_conntrack state;
332 read_lock_bh(&tcp_lock);
333 state = conntrack->proto.tcp.state;
334 read_unlock_bh(&tcp_lock);
336 return seq_printf(s, "%s ", tcp_conntrack_names[state]);
339 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
340 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
341 static int tcp_to_nfattr(struct sk_buff *skb, struct nfattr *nfa,
342 const struct ip_conntrack *ct)
344 struct nfattr *nest_parms = NFA_NEST(skb, CTA_PROTOINFO_TCP);
346 read_lock_bh(&tcp_lock);
347 NFA_PUT(skb, CTA_PROTOINFO_TCP_STATE, sizeof(u_int8_t),
348 &ct->proto.tcp.state);
349 read_unlock_bh(&tcp_lock);
351 NFA_NEST_END(skb, nest_parms);
356 read_unlock_bh(&tcp_lock);
360 static int nfattr_to_tcp(struct nfattr *cda[], struct ip_conntrack *ct)
362 struct nfattr *attr = cda[CTA_PROTOINFO_TCP-1];
363 struct nfattr *tb[CTA_PROTOINFO_TCP_MAX];
365 /* updates could not contain anything about the private
366 * protocol info, in that case skip the parsing */
370 nfattr_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, attr);
372 if (!tb[CTA_PROTOINFO_TCP_STATE-1])
375 write_lock_bh(&tcp_lock);
376 ct->proto.tcp.state =
377 *(u_int8_t *)NFA_DATA(tb[CTA_PROTOINFO_TCP_STATE-1]);
378 write_unlock_bh(&tcp_lock);
384 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
386 if (tcph->rst) return TCP_RST_SET;
387 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
388 else if (tcph->fin) return TCP_FIN_SET;
389 else if (tcph->ack) return TCP_ACK_SET;
390 else return TCP_NONE_SET;
393 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
394 in IP Filter' by Guido van Rooij.
396 http://www.nluug.nl/events/sane2000/papers.html
397 http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
399 The boundaries and the conditions are changed according to RFC793:
400 the packet must intersect the window (i.e. segments may be
401 after the right or before the left edge) and thus receivers may ACK
402 segments after the right edge of the window.
404 td_maxend = max(sack + max(win,1)) seen in reply packets
405 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
406 td_maxwin += seq + len - sender.td_maxend
407 if seq + len > sender.td_maxend
408 td_end = max(seq + len) seen in sent packets
410 I. Upper bound for valid data: seq <= sender.td_maxend
411 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
412 III. Upper bound for valid ack: sack <= receiver.td_end
413 IV. Lower bound for valid ack: ack >= receiver.td_end - MAXACKWINDOW
415 where sack is the highest right edge of sack block found in the packet.
417 The upper bound limit for a valid ack is not ignored -
418 we doesn't have to deal with fragments.
421 static inline __u32 segment_seq_plus_len(__u32 seq,
426 return (seq + len - (iph->ihl + tcph->doff)*4
427 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
430 /* Fixme: what about big packets? */
431 #define MAXACKWINCONST 66000
432 #define MAXACKWINDOW(sender) \
433 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
437 * Simplified tcp_parse_options routine from tcp_input.c
439 static void tcp_options(const struct sk_buff *skb,
442 struct ip_ct_tcp_state *state)
444 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
446 int length = (tcph->doff*4) - sizeof(struct tcphdr);
451 ptr = skb_header_pointer(skb,
452 (iph->ihl * 4) + sizeof(struct tcphdr),
466 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
471 if (opsize < 2) /* "silly options" */
474 break; /* don't parse partial options */
476 if (opcode == TCPOPT_SACK_PERM
477 && opsize == TCPOLEN_SACK_PERM)
478 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
479 else if (opcode == TCPOPT_WINDOW
480 && opsize == TCPOLEN_WINDOW) {
481 state->td_scale = *(u_int8_t *)ptr;
483 if (state->td_scale > 14) {
485 state->td_scale = 14;
488 IP_CT_TCP_FLAG_WINDOW_SCALE;
496 static void tcp_sack(const struct sk_buff *skb,
501 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
503 int length = (tcph->doff*4) - sizeof(struct tcphdr);
509 ptr = skb_header_pointer(skb,
510 (iph->ihl * 4) + sizeof(struct tcphdr),
514 /* Fast path for timestamp-only option */
515 if (length == TCPOLEN_TSTAMP_ALIGNED*4
517 __constant_ntohl((TCPOPT_NOP << 24)
519 | (TCPOPT_TIMESTAMP << 8)
520 | TCPOLEN_TIMESTAMP))
530 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
535 if (opsize < 2) /* "silly options" */
538 break; /* don't parse partial options */
540 if (opcode == TCPOPT_SACK
541 && opsize >= (TCPOLEN_SACK_BASE
542 + TCPOLEN_SACK_PERBLOCK)
543 && !((opsize - TCPOLEN_SACK_BASE)
544 % TCPOLEN_SACK_PERBLOCK)) {
546 i < (opsize - TCPOLEN_SACK_BASE);
547 i += TCPOLEN_SACK_PERBLOCK) {
548 tmp = ntohl(*((u_int32_t *)(ptr+i)+1));
550 if (after(tmp, *sack))
561 static int tcp_in_window(struct ip_ct_tcp *state,
562 enum ip_conntrack_dir dir,
564 const struct sk_buff *skb,
568 struct ip_ct_tcp_state *sender = &state->seen[dir];
569 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
570 __u32 seq, ack, sack, end, win, swin;
574 * Get the required data from the packet.
576 seq = ntohl(tcph->seq);
577 ack = sack = ntohl(tcph->ack_seq);
578 win = ntohs(tcph->window);
579 end = segment_seq_plus_len(seq, skb->len, iph, tcph);
581 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
582 tcp_sack(skb, iph, tcph, &sack);
584 DEBUGP("tcp_in_window: START\n");
585 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
586 "seq=%u ack=%u sack=%u win=%u end=%u\n",
587 NIPQUAD(iph->saddr), ntohs(tcph->source),
588 NIPQUAD(iph->daddr), ntohs(tcph->dest),
589 seq, ack, sack, win, end);
590 DEBUGP("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,
594 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
597 if (sender->td_end == 0) {
599 * Initialize sender data.
601 if (tcph->syn && tcph->ack) {
603 * Outgoing SYN-ACK in reply to a SYN.
606 sender->td_maxend = end;
607 sender->td_maxwin = (win == 0 ? 1 : win);
609 tcp_options(skb, iph, tcph, sender);
612 * Both sides must send the Window Scale option
613 * to enable window scaling in either direction.
615 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
616 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
618 receiver->td_scale = 0;
621 * We are in the middle of a connection,
622 * its history is lost for us.
623 * Let's try to use the data from the packet.
625 sender->td_end = end;
626 sender->td_maxwin = (win == 0 ? 1 : win);
627 sender->td_maxend = end + sender->td_maxwin;
629 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
630 && dir == IP_CT_DIR_ORIGINAL)
631 || (state->state == TCP_CONNTRACK_SYN_RECV
632 && dir == IP_CT_DIR_REPLY))
633 && after(end, sender->td_end)) {
635 * RFC 793: "if a TCP is reinitialized ... then it need
636 * not wait at all; it must only be sure to use sequence
637 * numbers larger than those recently used."
640 sender->td_maxend = end;
641 sender->td_maxwin = (win == 0 ? 1 : win);
643 tcp_options(skb, iph, tcph, sender);
648 * If there is no ACK, just pretend it was set and OK.
650 ack = sack = receiver->td_end;
651 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
652 (TCP_FLAG_ACK|TCP_FLAG_RST))
655 * Broken TCP stacks, that set ACK in RST packets as well
656 * with zero ack value.
658 ack = sack = receiver->td_end;
663 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
665 * Packets contains no data: we assume it is valid
666 * and check the ack value only.
667 * However RST segments are always validated by their
668 * SEQ number, except when seq == 0 (reset sent answering
671 seq = end = sender->td_end;
673 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
674 "seq=%u ack=%u sack =%u win=%u end=%u\n",
675 NIPQUAD(iph->saddr), ntohs(tcph->source),
676 NIPQUAD(iph->daddr), ntohs(tcph->dest),
677 seq, ack, sack, win, end);
678 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
679 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
680 sender->td_end, sender->td_maxend, sender->td_maxwin,
682 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
685 DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
686 before(seq, sender->td_maxend + 1),
687 after(end, sender->td_end - receiver->td_maxwin - 1),
688 before(sack, receiver->td_end + 1),
689 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
691 if (sender->loose || receiver->loose ||
692 (before(seq, sender->td_maxend + 1) &&
693 after(end, sender->td_end - receiver->td_maxwin - 1) &&
694 before(sack, receiver->td_end + 1) &&
695 after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
697 * Take into account window scaling (RFC 1323).
700 win <<= sender->td_scale;
703 * Update sender data.
705 swin = win + (sack - ack);
706 if (sender->td_maxwin < swin)
707 sender->td_maxwin = swin;
708 if (after(end, sender->td_end))
709 sender->td_end = end;
711 * Update receiver data.
713 if (after(end, sender->td_maxend))
714 receiver->td_maxwin += end - sender->td_maxend;
715 if (after(sack + win, receiver->td_maxend - 1)) {
716 receiver->td_maxend = sack + win;
718 receiver->td_maxend++;
722 * Check retransmissions.
724 if (index == TCP_ACK_SET) {
725 if (state->last_dir == dir
726 && state->last_seq == seq
727 && state->last_ack == ack
728 && state->last_end == end)
731 state->last_dir = dir;
732 state->last_seq = seq;
733 state->last_ack = ack;
734 state->last_end = end;
739 * Close the window of disabled window tracking :-)
746 if (LOG_INVALID(IPPROTO_TCP))
747 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
749 before(seq, sender->td_maxend + 1) ?
750 after(end, sender->td_end - receiver->td_maxwin - 1) ?
751 before(sack, receiver->td_end + 1) ?
752 after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
753 : "ACK is under the lower bound (possible overly delayed ACK)"
754 : "ACK is over the upper bound (ACKed data not seen yet)"
755 : "SEQ is under the lower bound (already ACKed data retransmitted)"
756 : "SEQ is over the upper bound (over the window of the receiver)");
758 res = ip_ct_tcp_be_liberal;
761 DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
762 "receiver end=%u maxend=%u maxwin=%u\n",
763 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
764 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
769 #ifdef CONFIG_IP_NF_NAT_NEEDED
770 /* Update sender->td_end after NAT successfully mangled the packet */
771 void ip_conntrack_tcp_update(struct sk_buff *skb,
772 struct ip_conntrack *conntrack,
773 enum ip_conntrack_dir dir)
775 struct iphdr *iph = skb->nh.iph;
776 struct tcphdr *tcph = (void *)skb->nh.iph + skb->nh.iph->ihl*4;
779 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
780 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
783 end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, iph, tcph);
785 write_lock_bh(&tcp_lock);
787 * We have to worry for the ack in the reply packet only...
789 if (after(end, conntrack->proto.tcp.seen[dir].td_end))
790 conntrack->proto.tcp.seen[dir].td_end = end;
791 conntrack->proto.tcp.last_end = end;
792 write_unlock_bh(&tcp_lock);
793 DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
794 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
795 sender->td_end, sender->td_maxend, sender->td_maxwin,
797 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
812 /* table of valid flag combinations - ECE and CWR are always valid */
813 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
817 [TH_SYN|TH_ACK|TH_PUSH] = 1,
820 [TH_RST|TH_ACK|TH_PUSH] = 1,
823 [TH_ACK|TH_PUSH] = 1,
825 [TH_ACK|TH_URG|TH_PUSH] = 1,
826 [TH_FIN|TH_ACK|TH_PUSH] = 1,
827 [TH_FIN|TH_ACK|TH_URG] = 1,
828 [TH_FIN|TH_ACK|TH_URG|TH_PUSH] = 1,
831 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
832 static int tcp_error(struct sk_buff *skb,
833 enum ip_conntrack_info *ctinfo,
834 unsigned int hooknum)
836 struct iphdr *iph = skb->nh.iph;
837 struct tcphdr _tcph, *th;
838 unsigned int tcplen = skb->len - iph->ihl * 4;
841 /* Smaller that minimal TCP header? */
842 th = skb_header_pointer(skb, iph->ihl * 4,
843 sizeof(_tcph), &_tcph);
845 if (LOG_INVALID(IPPROTO_TCP))
846 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
847 "ip_ct_tcp: short packet ");
851 /* Not whole TCP header or malformed packet */
852 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
853 if (LOG_INVALID(IPPROTO_TCP))
854 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
855 "ip_ct_tcp: truncated/malformed packet ");
859 /* Checksum invalid? Ignore.
860 * We skip checking packets on the outgoing path
861 * because the semantic of CHECKSUM_HW is different there
862 * and moreover root might send raw packets.
864 /* FIXME: Source route IP option packets --RR */
865 if (hooknum == NF_IP_PRE_ROUTING
866 && skb->ip_summed != CHECKSUM_UNNECESSARY
867 && csum_tcpudp_magic(iph->saddr, iph->daddr, tcplen, IPPROTO_TCP,
868 skb->ip_summed == CHECKSUM_HW ? skb->csum
869 : skb_checksum(skb, iph->ihl*4, tcplen, 0))) {
870 if (LOG_INVALID(IPPROTO_TCP))
871 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
872 "ip_ct_tcp: bad TCP checksum ");
876 /* Check TCP flags. */
877 tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
878 if (!tcp_valid_flags[tcpflags]) {
879 if (LOG_INVALID(IPPROTO_TCP))
880 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
881 "ip_ct_tcp: invalid TCP flag combination ");
888 /* Returns verdict for packet, or -1 for invalid. */
889 static int tcp_packet(struct ip_conntrack *conntrack,
890 const struct sk_buff *skb,
891 enum ip_conntrack_info ctinfo)
893 enum tcp_conntrack new_state, old_state;
894 enum ip_conntrack_dir dir;
895 struct iphdr *iph = skb->nh.iph;
896 struct tcphdr *th, _tcph;
897 unsigned long timeout;
900 th = skb_header_pointer(skb, iph->ihl * 4,
901 sizeof(_tcph), &_tcph);
904 write_lock_bh(&tcp_lock);
905 old_state = conntrack->proto.tcp.state;
906 dir = CTINFO2DIR(ctinfo);
907 index = get_conntrack_index(th);
908 new_state = tcp_conntracks[dir][index][old_state];
911 case TCP_CONNTRACK_IGNORE:
912 /* Either SYN in ORIGINAL
913 * or SYN/ACK in REPLY. */
914 if (index == TCP_SYNACK_SET
915 && conntrack->proto.tcp.last_index == TCP_SYN_SET
916 && conntrack->proto.tcp.last_dir != dir
917 && ntohl(th->ack_seq) ==
918 conntrack->proto.tcp.last_end) {
919 /* This SYN/ACK acknowledges a SYN that we earlier
920 * ignored as invalid. This means that the client and
921 * the server are both in sync, while the firewall is
922 * not. We kill this session and block the SYN/ACK so
923 * that the client cannot but retransmit its SYN and
924 * thus initiate a clean new session.
926 write_unlock_bh(&tcp_lock);
927 if (LOG_INVALID(IPPROTO_TCP))
928 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
930 "killing out of sync session ");
931 if (del_timer(&conntrack->timeout))
932 conntrack->timeout.function((unsigned long)
936 conntrack->proto.tcp.last_index = index;
937 conntrack->proto.tcp.last_dir = dir;
938 conntrack->proto.tcp.last_seq = ntohl(th->seq);
939 conntrack->proto.tcp.last_end =
940 segment_seq_plus_len(ntohl(th->seq), skb->len, iph, th);
942 write_unlock_bh(&tcp_lock);
943 if (LOG_INVALID(IPPROTO_TCP))
944 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
945 "ip_ct_tcp: invalid packet ignored ");
947 case TCP_CONNTRACK_MAX:
949 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
950 dir, get_conntrack_index(th),
952 write_unlock_bh(&tcp_lock);
953 if (LOG_INVALID(IPPROTO_TCP))
954 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
955 "ip_ct_tcp: invalid state ");
957 case TCP_CONNTRACK_SYN_SENT:
958 if (old_state < TCP_CONNTRACK_TIME_WAIT)
960 if ((conntrack->proto.tcp.seen[dir].flags &
961 IP_CT_TCP_FLAG_CLOSE_INIT)
962 || after(ntohl(th->seq),
963 conntrack->proto.tcp.seen[dir].td_end)) {
964 /* Attempt to reopen a closed connection.
965 * Delete this connection and look up again. */
966 write_unlock_bh(&tcp_lock);
967 if (del_timer(&conntrack->timeout))
968 conntrack->timeout.function((unsigned long)
972 write_unlock_bh(&tcp_lock);
973 if (LOG_INVALID(IPPROTO_TCP))
974 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
975 NULL, "ip_ct_tcp: invalid SYN");
978 case TCP_CONNTRACK_CLOSE:
979 if (index == TCP_RST_SET
980 && test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
981 && conntrack->proto.tcp.last_index == TCP_SYN_SET
982 && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
983 /* RST sent to invalid SYN we had let trough
984 * SYN was in window then, tear down connection.
985 * We skip window checking, because packet might ACK
986 * segments we ignored in the SYN. */
989 /* Just fall trough */
991 /* Keep compilers happy. */
995 if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
997 write_unlock_bh(&tcp_lock);
1001 /* From now on we have got in-window packets */
1002 conntrack->proto.tcp.last_index = index;
1004 DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
1005 "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1006 NIPQUAD(iph->saddr), ntohs(th->source),
1007 NIPQUAD(iph->daddr), ntohs(th->dest),
1008 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1009 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1010 old_state, new_state);
1012 conntrack->proto.tcp.state = new_state;
1013 if (old_state != new_state
1014 && (new_state == TCP_CONNTRACK_FIN_WAIT
1015 || new_state == TCP_CONNTRACK_CLOSE))
1016 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1017 timeout = conntrack->proto.tcp.retrans >= ip_ct_tcp_max_retrans
1018 && *tcp_timeouts[new_state] > ip_ct_tcp_timeout_max_retrans
1019 ? ip_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
1020 write_unlock_bh(&tcp_lock);
1022 ip_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
1023 if (new_state != old_state)
1024 ip_conntrack_event_cache(IPCT_PROTOINFO, skb);
1026 if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
1027 /* If only reply is a RST, we can consider ourselves not to
1028 have an established connection: this is a fairly common
1029 problem case, so we can delete the conntrack
1030 immediately. --RR */
1032 if (del_timer(&conntrack->timeout))
1033 conntrack->timeout.function((unsigned long)
1037 } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
1038 && (old_state == TCP_CONNTRACK_SYN_RECV
1039 || old_state == TCP_CONNTRACK_ESTABLISHED)
1040 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1041 /* Set ASSURED if we see see valid ack in ESTABLISHED
1042 after SYN_RECV or a valid answer for a picked up
1044 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1045 ip_conntrack_event_cache(IPCT_STATUS, skb);
1047 ip_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1052 /* Called when a new connection for this protocol found. */
1053 static int tcp_new(struct ip_conntrack *conntrack,
1054 const struct sk_buff *skb)
1056 enum tcp_conntrack new_state;
1057 struct iphdr *iph = skb->nh.iph;
1058 struct tcphdr *th, _tcph;
1060 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1061 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1064 th = skb_header_pointer(skb, iph->ihl * 4,
1065 sizeof(_tcph), &_tcph);
1068 /* Don't need lock here: this conntrack not in circulation yet */
1070 = tcp_conntracks[0][get_conntrack_index(th)]
1071 [TCP_CONNTRACK_NONE];
1073 /* Invalid: delete conntrack */
1074 if (new_state >= TCP_CONNTRACK_MAX) {
1075 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1079 if (new_state == TCP_CONNTRACK_SYN_SENT) {
1081 conntrack->proto.tcp.seen[0].td_end =
1082 segment_seq_plus_len(ntohl(th->seq), skb->len,
1084 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1085 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1086 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1087 conntrack->proto.tcp.seen[0].td_maxend =
1088 conntrack->proto.tcp.seen[0].td_end;
1090 tcp_options(skb, iph, th, &conntrack->proto.tcp.seen[0]);
1091 conntrack->proto.tcp.seen[1].flags = 0;
1092 conntrack->proto.tcp.seen[0].loose =
1093 conntrack->proto.tcp.seen[1].loose = 0;
1094 } else if (ip_ct_tcp_loose == 0) {
1095 /* Don't try to pick up connections. */
1099 * We are in the middle of a connection,
1100 * its history is lost for us.
1101 * Let's try to use the data from the packet.
1103 conntrack->proto.tcp.seen[0].td_end =
1104 segment_seq_plus_len(ntohl(th->seq), skb->len,
1106 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1107 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1108 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1109 conntrack->proto.tcp.seen[0].td_maxend =
1110 conntrack->proto.tcp.seen[0].td_end +
1111 conntrack->proto.tcp.seen[0].td_maxwin;
1112 conntrack->proto.tcp.seen[0].td_scale = 0;
1114 /* We assume SACK. Should we assume window scaling too? */
1115 conntrack->proto.tcp.seen[0].flags =
1116 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1117 conntrack->proto.tcp.seen[0].loose =
1118 conntrack->proto.tcp.seen[1].loose = ip_ct_tcp_loose;
1121 conntrack->proto.tcp.seen[1].td_end = 0;
1122 conntrack->proto.tcp.seen[1].td_maxend = 0;
1123 conntrack->proto.tcp.seen[1].td_maxwin = 1;
1124 conntrack->proto.tcp.seen[1].td_scale = 0;
1126 /* tcp_packet will set them */
1127 conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1128 conntrack->proto.tcp.last_index = TCP_NONE_SET;
1130 DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1131 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1132 sender->td_end, sender->td_maxend, sender->td_maxwin,
1134 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1135 receiver->td_scale);
1139 struct ip_conntrack_protocol ip_conntrack_protocol_tcp =
1141 .proto = IPPROTO_TCP,
1143 .pkt_to_tuple = tcp_pkt_to_tuple,
1144 .invert_tuple = tcp_invert_tuple,
1145 .print_tuple = tcp_print_tuple,
1146 .print_conntrack = tcp_print_conntrack,
1147 .packet = tcp_packet,
1150 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
1151 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
1152 .to_nfattr = tcp_to_nfattr,
1153 .from_nfattr = nfattr_to_tcp,
1154 .tuple_to_nfattr = ip_ct_port_tuple_to_nfattr,
1155 .nfattr_to_tuple = ip_ct_port_nfattr_to_tuple,