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 read_lock_bh(&tcp_lock);
345 NFA_PUT(skb, CTA_PROTOINFO_TCP_STATE, sizeof(u_int8_t),
346 &ct->proto.tcp.state);
347 read_unlock_bh(&tcp_lock);
356 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
358 if (tcph->rst) return TCP_RST_SET;
359 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
360 else if (tcph->fin) return TCP_FIN_SET;
361 else if (tcph->ack) return TCP_ACK_SET;
362 else return TCP_NONE_SET;
365 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
366 in IP Filter' by Guido van Rooij.
368 http://www.nluug.nl/events/sane2000/papers.html
369 http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
371 The boundaries and the conditions are changed according to RFC793:
372 the packet must intersect the window (i.e. segments may be
373 after the right or before the left edge) and thus receivers may ACK
374 segments after the right edge of the window.
376 td_maxend = max(sack + max(win,1)) seen in reply packets
377 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
378 td_maxwin += seq + len - sender.td_maxend
379 if seq + len > sender.td_maxend
380 td_end = max(seq + len) seen in sent packets
382 I. Upper bound for valid data: seq <= sender.td_maxend
383 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
384 III. Upper bound for valid ack: sack <= receiver.td_end
385 IV. Lower bound for valid ack: ack >= receiver.td_end - MAXACKWINDOW
387 where sack is the highest right edge of sack block found in the packet.
389 The upper bound limit for a valid ack is not ignored -
390 we doesn't have to deal with fragments.
393 static inline __u32 segment_seq_plus_len(__u32 seq,
398 return (seq + len - (iph->ihl + tcph->doff)*4
399 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
402 /* Fixme: what about big packets? */
403 #define MAXACKWINCONST 66000
404 #define MAXACKWINDOW(sender) \
405 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
409 * Simplified tcp_parse_options routine from tcp_input.c
411 static void tcp_options(const struct sk_buff *skb,
414 struct ip_ct_tcp_state *state)
416 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
418 int length = (tcph->doff*4) - sizeof(struct tcphdr);
423 ptr = skb_header_pointer(skb,
424 (iph->ihl * 4) + sizeof(struct tcphdr),
438 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
443 if (opsize < 2) /* "silly options" */
446 break; /* don't parse partial options */
448 if (opcode == TCPOPT_SACK_PERM
449 && opsize == TCPOLEN_SACK_PERM)
450 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
451 else if (opcode == TCPOPT_WINDOW
452 && opsize == TCPOLEN_WINDOW) {
453 state->td_scale = *(u_int8_t *)ptr;
455 if (state->td_scale > 14) {
457 state->td_scale = 14;
460 IP_CT_TCP_FLAG_WINDOW_SCALE;
468 static void tcp_sack(const struct sk_buff *skb,
473 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
475 int length = (tcph->doff*4) - sizeof(struct tcphdr);
481 ptr = skb_header_pointer(skb,
482 (iph->ihl * 4) + sizeof(struct tcphdr),
486 /* Fast path for timestamp-only option */
487 if (length == TCPOLEN_TSTAMP_ALIGNED*4
489 __constant_ntohl((TCPOPT_NOP << 24)
491 | (TCPOPT_TIMESTAMP << 8)
492 | TCPOLEN_TIMESTAMP))
502 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
507 if (opsize < 2) /* "silly options" */
510 break; /* don't parse partial options */
512 if (opcode == TCPOPT_SACK
513 && opsize >= (TCPOLEN_SACK_BASE
514 + TCPOLEN_SACK_PERBLOCK)
515 && !((opsize - TCPOLEN_SACK_BASE)
516 % TCPOLEN_SACK_PERBLOCK)) {
518 i < (opsize - TCPOLEN_SACK_BASE);
519 i += TCPOLEN_SACK_PERBLOCK) {
520 tmp = ntohl(*((u_int32_t *)(ptr+i)+1));
522 if (after(tmp, *sack))
533 static int tcp_in_window(struct ip_ct_tcp *state,
534 enum ip_conntrack_dir dir,
536 const struct sk_buff *skb,
540 struct ip_ct_tcp_state *sender = &state->seen[dir];
541 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
542 __u32 seq, ack, sack, end, win, swin;
546 * Get the required data from the packet.
548 seq = ntohl(tcph->seq);
549 ack = sack = ntohl(tcph->ack_seq);
550 win = ntohs(tcph->window);
551 end = segment_seq_plus_len(seq, skb->len, iph, tcph);
553 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
554 tcp_sack(skb, iph, tcph, &sack);
556 DEBUGP("tcp_in_window: START\n");
557 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
558 "seq=%u ack=%u sack=%u win=%u end=%u\n",
559 NIPQUAD(iph->saddr), ntohs(tcph->source),
560 NIPQUAD(iph->daddr), ntohs(tcph->dest),
561 seq, ack, sack, win, end);
562 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
563 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
564 sender->td_end, sender->td_maxend, sender->td_maxwin,
566 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
569 if (sender->td_end == 0) {
571 * Initialize sender data.
573 if (tcph->syn && tcph->ack) {
575 * Outgoing SYN-ACK in reply to a SYN.
578 sender->td_maxend = end;
579 sender->td_maxwin = (win == 0 ? 1 : win);
581 tcp_options(skb, iph, tcph, sender);
584 * Both sides must send the Window Scale option
585 * to enable window scaling in either direction.
587 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
588 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
590 receiver->td_scale = 0;
593 * We are in the middle of a connection,
594 * its history is lost for us.
595 * Let's try to use the data from the packet.
597 sender->td_end = end;
598 sender->td_maxwin = (win == 0 ? 1 : win);
599 sender->td_maxend = end + sender->td_maxwin;
601 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
602 && dir == IP_CT_DIR_ORIGINAL)
603 || (state->state == TCP_CONNTRACK_SYN_RECV
604 && dir == IP_CT_DIR_REPLY))
605 && after(end, sender->td_end)) {
607 * RFC 793: "if a TCP is reinitialized ... then it need
608 * not wait at all; it must only be sure to use sequence
609 * numbers larger than those recently used."
612 sender->td_maxend = end;
613 sender->td_maxwin = (win == 0 ? 1 : win);
615 tcp_options(skb, iph, tcph, sender);
620 * If there is no ACK, just pretend it was set and OK.
622 ack = sack = receiver->td_end;
623 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
624 (TCP_FLAG_ACK|TCP_FLAG_RST))
627 * Broken TCP stacks, that set ACK in RST packets as well
628 * with zero ack value.
630 ack = sack = receiver->td_end;
635 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
637 * Packets contains no data: we assume it is valid
638 * and check the ack value only.
639 * However RST segments are always validated by their
640 * SEQ number, except when seq == 0 (reset sent answering
643 seq = end = sender->td_end;
645 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
646 "seq=%u ack=%u sack =%u win=%u end=%u\n",
647 NIPQUAD(iph->saddr), ntohs(tcph->source),
648 NIPQUAD(iph->daddr), ntohs(tcph->dest),
649 seq, ack, sack, win, end);
650 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
651 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
652 sender->td_end, sender->td_maxend, sender->td_maxwin,
654 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
657 DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
658 before(seq, sender->td_maxend + 1),
659 after(end, sender->td_end - receiver->td_maxwin - 1),
660 before(sack, receiver->td_end + 1),
661 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
663 if (sender->loose || receiver->loose ||
664 (before(seq, sender->td_maxend + 1) &&
665 after(end, sender->td_end - receiver->td_maxwin - 1) &&
666 before(sack, receiver->td_end + 1) &&
667 after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
669 * Take into account window scaling (RFC 1323).
672 win <<= sender->td_scale;
675 * Update sender data.
677 swin = win + (sack - ack);
678 if (sender->td_maxwin < swin)
679 sender->td_maxwin = swin;
680 if (after(end, sender->td_end))
681 sender->td_end = end;
683 * Update receiver data.
685 if (after(end, sender->td_maxend))
686 receiver->td_maxwin += end - sender->td_maxend;
687 if (after(sack + win, receiver->td_maxend - 1)) {
688 receiver->td_maxend = sack + win;
690 receiver->td_maxend++;
694 * Check retransmissions.
696 if (index == TCP_ACK_SET) {
697 if (state->last_dir == dir
698 && state->last_seq == seq
699 && state->last_ack == ack
700 && state->last_end == end)
703 state->last_dir = dir;
704 state->last_seq = seq;
705 state->last_ack = ack;
706 state->last_end = end;
711 * Close the window of disabled window tracking :-)
718 if (LOG_INVALID(IPPROTO_TCP))
719 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
721 before(seq, sender->td_maxend + 1) ?
722 after(end, sender->td_end - receiver->td_maxwin - 1) ?
723 before(sack, receiver->td_end + 1) ?
724 after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
725 : "ACK is under the lower bound (possible overly delayed ACK)"
726 : "ACK is over the upper bound (ACKed data not seen yet)"
727 : "SEQ is under the lower bound (already ACKed data retransmitted)"
728 : "SEQ is over the upper bound (over the window of the receiver)");
730 res = ip_ct_tcp_be_liberal;
733 DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
734 "receiver end=%u maxend=%u maxwin=%u\n",
735 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
736 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
741 #ifdef CONFIG_IP_NF_NAT_NEEDED
742 /* Update sender->td_end after NAT successfully mangled the packet */
743 void ip_conntrack_tcp_update(struct sk_buff *skb,
744 struct ip_conntrack *conntrack,
745 enum ip_conntrack_dir dir)
747 struct iphdr *iph = skb->nh.iph;
748 struct tcphdr *tcph = (void *)skb->nh.iph + skb->nh.iph->ihl*4;
751 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
752 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
755 end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, iph, tcph);
757 write_lock_bh(&tcp_lock);
759 * We have to worry for the ack in the reply packet only...
761 if (after(end, conntrack->proto.tcp.seen[dir].td_end))
762 conntrack->proto.tcp.seen[dir].td_end = end;
763 conntrack->proto.tcp.last_end = end;
764 write_unlock_bh(&tcp_lock);
765 DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
766 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
767 sender->td_end, sender->td_maxend, sender->td_maxwin,
769 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
784 /* table of valid flag combinations - ECE and CWR are always valid */
785 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
789 [TH_SYN|TH_ACK|TH_PUSH] = 1,
792 [TH_RST|TH_ACK|TH_PUSH] = 1,
795 [TH_ACK|TH_PUSH] = 1,
797 [TH_ACK|TH_URG|TH_PUSH] = 1,
798 [TH_FIN|TH_ACK|TH_PUSH] = 1,
799 [TH_FIN|TH_ACK|TH_URG] = 1,
800 [TH_FIN|TH_ACK|TH_URG|TH_PUSH] = 1,
803 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
804 static int tcp_error(struct sk_buff *skb,
805 enum ip_conntrack_info *ctinfo,
806 unsigned int hooknum)
808 struct iphdr *iph = skb->nh.iph;
809 struct tcphdr _tcph, *th;
810 unsigned int tcplen = skb->len - iph->ihl * 4;
813 /* Smaller that minimal TCP header? */
814 th = skb_header_pointer(skb, iph->ihl * 4,
815 sizeof(_tcph), &_tcph);
817 if (LOG_INVALID(IPPROTO_TCP))
818 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
819 "ip_ct_tcp: short packet ");
823 /* Not whole TCP header or malformed packet */
824 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
825 if (LOG_INVALID(IPPROTO_TCP))
826 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
827 "ip_ct_tcp: truncated/malformed packet ");
831 /* Checksum invalid? Ignore.
832 * We skip checking packets on the outgoing path
833 * because the semantic of CHECKSUM_HW is different there
834 * and moreover root might send raw packets.
836 /* FIXME: Source route IP option packets --RR */
837 if (hooknum == NF_IP_PRE_ROUTING
838 && skb->ip_summed != CHECKSUM_UNNECESSARY
839 && csum_tcpudp_magic(iph->saddr, iph->daddr, tcplen, IPPROTO_TCP,
840 skb->ip_summed == CHECKSUM_HW ? skb->csum
841 : skb_checksum(skb, iph->ihl*4, tcplen, 0))) {
842 if (LOG_INVALID(IPPROTO_TCP))
843 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
844 "ip_ct_tcp: bad TCP checksum ");
848 /* Check TCP flags. */
849 tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
850 if (!tcp_valid_flags[tcpflags]) {
851 if (LOG_INVALID(IPPROTO_TCP))
852 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
853 "ip_ct_tcp: invalid TCP flag combination ");
860 /* Returns verdict for packet, or -1 for invalid. */
861 static int tcp_packet(struct ip_conntrack *conntrack,
862 const struct sk_buff *skb,
863 enum ip_conntrack_info ctinfo)
865 enum tcp_conntrack new_state, old_state;
866 enum ip_conntrack_dir dir;
867 struct iphdr *iph = skb->nh.iph;
868 struct tcphdr *th, _tcph;
869 unsigned long timeout;
872 th = skb_header_pointer(skb, iph->ihl * 4,
873 sizeof(_tcph), &_tcph);
876 write_lock_bh(&tcp_lock);
877 old_state = conntrack->proto.tcp.state;
878 dir = CTINFO2DIR(ctinfo);
879 index = get_conntrack_index(th);
880 new_state = tcp_conntracks[dir][index][old_state];
883 case TCP_CONNTRACK_IGNORE:
884 /* Either SYN in ORIGINAL
885 * or SYN/ACK in REPLY. */
886 if (index == TCP_SYNACK_SET
887 && conntrack->proto.tcp.last_index == TCP_SYN_SET
888 && conntrack->proto.tcp.last_dir != dir
889 && ntohl(th->ack_seq) ==
890 conntrack->proto.tcp.last_end) {
891 /* This SYN/ACK acknowledges a SYN that we earlier
892 * ignored as invalid. This means that the client and
893 * the server are both in sync, while the firewall is
894 * not. We kill this session and block the SYN/ACK so
895 * that the client cannot but retransmit its SYN and
896 * thus initiate a clean new session.
898 write_unlock_bh(&tcp_lock);
899 if (LOG_INVALID(IPPROTO_TCP))
900 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
901 "ip_ct_tcp: killing out of sync session ");
902 if (del_timer(&conntrack->timeout))
903 conntrack->timeout.function((unsigned long)
907 conntrack->proto.tcp.last_index = index;
908 conntrack->proto.tcp.last_dir = dir;
909 conntrack->proto.tcp.last_seq = ntohl(th->seq);
910 conntrack->proto.tcp.last_end =
911 segment_seq_plus_len(ntohl(th->seq), skb->len, iph, th);
913 write_unlock_bh(&tcp_lock);
914 if (LOG_INVALID(IPPROTO_TCP))
915 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
916 "ip_ct_tcp: invalid packet ignored ");
918 case TCP_CONNTRACK_MAX:
920 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
921 dir, get_conntrack_index(th),
923 write_unlock_bh(&tcp_lock);
924 if (LOG_INVALID(IPPROTO_TCP))
925 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
926 "ip_ct_tcp: invalid state ");
928 case TCP_CONNTRACK_SYN_SENT:
929 if (old_state < TCP_CONNTRACK_TIME_WAIT)
931 if ((conntrack->proto.tcp.seen[dir].flags &
932 IP_CT_TCP_FLAG_CLOSE_INIT)
933 || after(ntohl(th->seq),
934 conntrack->proto.tcp.seen[dir].td_end)) {
935 /* Attempt to reopen a closed connection.
936 * Delete this connection and look up again. */
937 write_unlock_bh(&tcp_lock);
938 if (del_timer(&conntrack->timeout))
939 conntrack->timeout.function((unsigned long)
943 write_unlock_bh(&tcp_lock);
944 if (LOG_INVALID(IPPROTO_TCP))
945 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
946 "ip_ct_tcp: invalid SYN");
949 case TCP_CONNTRACK_CLOSE:
950 if (index == TCP_RST_SET
951 && test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
952 && conntrack->proto.tcp.last_index == TCP_SYN_SET
953 && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
954 /* RST sent to invalid SYN we had let trough
955 * SYN was in window then, tear down connection.
956 * We skip window checking, because packet might ACK
957 * segments we ignored in the SYN. */
960 /* Just fall trough */
962 /* Keep compilers happy. */
966 if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
968 write_unlock_bh(&tcp_lock);
972 /* From now on we have got in-window packets */
973 conntrack->proto.tcp.last_index = index;
975 DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
976 "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
977 NIPQUAD(iph->saddr), ntohs(th->source),
978 NIPQUAD(iph->daddr), ntohs(th->dest),
979 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
980 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
981 old_state, new_state);
983 conntrack->proto.tcp.state = new_state;
984 if (old_state != new_state
985 && (new_state == TCP_CONNTRACK_FIN_WAIT
986 || new_state == TCP_CONNTRACK_CLOSE))
987 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
988 timeout = conntrack->proto.tcp.retrans >= ip_ct_tcp_max_retrans
989 && *tcp_timeouts[new_state] > ip_ct_tcp_timeout_max_retrans
990 ? ip_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
991 write_unlock_bh(&tcp_lock);
993 ip_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
994 if (new_state != old_state)
995 ip_conntrack_event_cache(IPCT_PROTOINFO, skb);
997 if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
998 /* If only reply is a RST, we can consider ourselves not to
999 have an established connection: this is a fairly common
1000 problem case, so we can delete the conntrack
1001 immediately. --RR */
1003 if (del_timer(&conntrack->timeout))
1004 conntrack->timeout.function((unsigned long)
1008 } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
1009 && (old_state == TCP_CONNTRACK_SYN_RECV
1010 || old_state == TCP_CONNTRACK_ESTABLISHED)
1011 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1012 /* Set ASSURED if we see see valid ack in ESTABLISHED
1013 after SYN_RECV or a valid answer for a picked up
1015 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1017 ip_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1022 /* Called when a new connection for this protocol found. */
1023 static int tcp_new(struct ip_conntrack *conntrack,
1024 const struct sk_buff *skb)
1026 enum tcp_conntrack new_state;
1027 struct iphdr *iph = skb->nh.iph;
1028 struct tcphdr *th, _tcph;
1030 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1031 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1034 th = skb_header_pointer(skb, iph->ihl * 4,
1035 sizeof(_tcph), &_tcph);
1038 /* Don't need lock here: this conntrack not in circulation yet */
1040 = tcp_conntracks[0][get_conntrack_index(th)]
1041 [TCP_CONNTRACK_NONE];
1043 /* Invalid: delete conntrack */
1044 if (new_state >= TCP_CONNTRACK_MAX) {
1045 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1049 if (new_state == TCP_CONNTRACK_SYN_SENT) {
1051 conntrack->proto.tcp.seen[0].td_end =
1052 segment_seq_plus_len(ntohl(th->seq), skb->len,
1054 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1055 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1056 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1057 conntrack->proto.tcp.seen[0].td_maxend =
1058 conntrack->proto.tcp.seen[0].td_end;
1060 tcp_options(skb, iph, th, &conntrack->proto.tcp.seen[0]);
1061 conntrack->proto.tcp.seen[1].flags = 0;
1062 conntrack->proto.tcp.seen[0].loose =
1063 conntrack->proto.tcp.seen[1].loose = 0;
1064 } else if (ip_ct_tcp_loose == 0) {
1065 /* Don't try to pick up connections. */
1069 * We are in the middle of a connection,
1070 * its history is lost for us.
1071 * Let's try to use the data from the packet.
1073 conntrack->proto.tcp.seen[0].td_end =
1074 segment_seq_plus_len(ntohl(th->seq), skb->len,
1076 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1077 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1078 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1079 conntrack->proto.tcp.seen[0].td_maxend =
1080 conntrack->proto.tcp.seen[0].td_end +
1081 conntrack->proto.tcp.seen[0].td_maxwin;
1082 conntrack->proto.tcp.seen[0].td_scale = 0;
1084 /* We assume SACK. Should we assume window scaling too? */
1085 conntrack->proto.tcp.seen[0].flags =
1086 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1087 conntrack->proto.tcp.seen[0].loose =
1088 conntrack->proto.tcp.seen[1].loose = ip_ct_tcp_loose;
1091 conntrack->proto.tcp.seen[1].td_end = 0;
1092 conntrack->proto.tcp.seen[1].td_maxend = 0;
1093 conntrack->proto.tcp.seen[1].td_maxwin = 1;
1094 conntrack->proto.tcp.seen[1].td_scale = 0;
1096 /* tcp_packet will set them */
1097 conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1098 conntrack->proto.tcp.last_index = TCP_NONE_SET;
1100 DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1101 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1102 sender->td_end, sender->td_maxend, sender->td_maxwin,
1104 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1105 receiver->td_scale);
1109 struct ip_conntrack_protocol ip_conntrack_protocol_tcp =
1111 .proto = IPPROTO_TCP,
1113 .pkt_to_tuple = tcp_pkt_to_tuple,
1114 .invert_tuple = tcp_invert_tuple,
1115 .print_tuple = tcp_print_tuple,
1116 .print_conntrack = tcp_print_conntrack,
1117 .packet = tcp_packet,
1120 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
1121 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
1122 .to_nfattr = tcp_to_nfattr,
1123 .tuple_to_nfattr = ip_ct_port_tuple_to_nfattr,
1124 .nfattr_to_tuple = ip_ct_port_nfattr_to_tuple,