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);
352 read_unlock_bh(&tcp_lock);
357 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
359 if (tcph->rst) return TCP_RST_SET;
360 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
361 else if (tcph->fin) return TCP_FIN_SET;
362 else if (tcph->ack) return TCP_ACK_SET;
363 else return TCP_NONE_SET;
366 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
367 in IP Filter' by Guido van Rooij.
369 http://www.nluug.nl/events/sane2000/papers.html
370 http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
372 The boundaries and the conditions are changed according to RFC793:
373 the packet must intersect the window (i.e. segments may be
374 after the right or before the left edge) and thus receivers may ACK
375 segments after the right edge of the window.
377 td_maxend = max(sack + max(win,1)) seen in reply packets
378 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
379 td_maxwin += seq + len - sender.td_maxend
380 if seq + len > sender.td_maxend
381 td_end = max(seq + len) seen in sent packets
383 I. Upper bound for valid data: seq <= sender.td_maxend
384 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
385 III. Upper bound for valid ack: sack <= receiver.td_end
386 IV. Lower bound for valid ack: ack >= receiver.td_end - MAXACKWINDOW
388 where sack is the highest right edge of sack block found in the packet.
390 The upper bound limit for a valid ack is not ignored -
391 we doesn't have to deal with fragments.
394 static inline __u32 segment_seq_plus_len(__u32 seq,
399 return (seq + len - (iph->ihl + tcph->doff)*4
400 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
403 /* Fixme: what about big packets? */
404 #define MAXACKWINCONST 66000
405 #define MAXACKWINDOW(sender) \
406 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
410 * Simplified tcp_parse_options routine from tcp_input.c
412 static void tcp_options(const struct sk_buff *skb,
415 struct ip_ct_tcp_state *state)
417 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
419 int length = (tcph->doff*4) - sizeof(struct tcphdr);
424 ptr = skb_header_pointer(skb,
425 (iph->ihl * 4) + sizeof(struct tcphdr),
439 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
444 if (opsize < 2) /* "silly options" */
447 break; /* don't parse partial options */
449 if (opcode == TCPOPT_SACK_PERM
450 && opsize == TCPOLEN_SACK_PERM)
451 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
452 else if (opcode == TCPOPT_WINDOW
453 && opsize == TCPOLEN_WINDOW) {
454 state->td_scale = *(u_int8_t *)ptr;
456 if (state->td_scale > 14) {
458 state->td_scale = 14;
461 IP_CT_TCP_FLAG_WINDOW_SCALE;
469 static void tcp_sack(const struct sk_buff *skb,
474 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
476 int length = (tcph->doff*4) - sizeof(struct tcphdr);
482 ptr = skb_header_pointer(skb,
483 (iph->ihl * 4) + sizeof(struct tcphdr),
487 /* Fast path for timestamp-only option */
488 if (length == TCPOLEN_TSTAMP_ALIGNED*4
490 __constant_ntohl((TCPOPT_NOP << 24)
492 | (TCPOPT_TIMESTAMP << 8)
493 | TCPOLEN_TIMESTAMP))
503 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
508 if (opsize < 2) /* "silly options" */
511 break; /* don't parse partial options */
513 if (opcode == TCPOPT_SACK
514 && opsize >= (TCPOLEN_SACK_BASE
515 + TCPOLEN_SACK_PERBLOCK)
516 && !((opsize - TCPOLEN_SACK_BASE)
517 % TCPOLEN_SACK_PERBLOCK)) {
519 i < (opsize - TCPOLEN_SACK_BASE);
520 i += TCPOLEN_SACK_PERBLOCK) {
521 tmp = ntohl(*((u_int32_t *)(ptr+i)+1));
523 if (after(tmp, *sack))
534 static int tcp_in_window(struct ip_ct_tcp *state,
535 enum ip_conntrack_dir dir,
537 const struct sk_buff *skb,
541 struct ip_ct_tcp_state *sender = &state->seen[dir];
542 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
543 __u32 seq, ack, sack, end, win, swin;
547 * Get the required data from the packet.
549 seq = ntohl(tcph->seq);
550 ack = sack = ntohl(tcph->ack_seq);
551 win = ntohs(tcph->window);
552 end = segment_seq_plus_len(seq, skb->len, iph, tcph);
554 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
555 tcp_sack(skb, iph, tcph, &sack);
557 DEBUGP("tcp_in_window: START\n");
558 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
559 "seq=%u ack=%u sack=%u win=%u end=%u\n",
560 NIPQUAD(iph->saddr), ntohs(tcph->source),
561 NIPQUAD(iph->daddr), ntohs(tcph->dest),
562 seq, ack, sack, win, end);
563 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
564 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
565 sender->td_end, sender->td_maxend, sender->td_maxwin,
567 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
570 if (sender->td_end == 0) {
572 * Initialize sender data.
574 if (tcph->syn && tcph->ack) {
576 * Outgoing SYN-ACK in reply to a SYN.
579 sender->td_maxend = end;
580 sender->td_maxwin = (win == 0 ? 1 : win);
582 tcp_options(skb, iph, tcph, sender);
585 * Both sides must send the Window Scale option
586 * to enable window scaling in either direction.
588 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
589 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
591 receiver->td_scale = 0;
594 * We are in the middle of a connection,
595 * its history is lost for us.
596 * Let's try to use the data from the packet.
598 sender->td_end = end;
599 sender->td_maxwin = (win == 0 ? 1 : win);
600 sender->td_maxend = end + sender->td_maxwin;
602 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
603 && dir == IP_CT_DIR_ORIGINAL)
604 || (state->state == TCP_CONNTRACK_SYN_RECV
605 && dir == IP_CT_DIR_REPLY))
606 && after(end, sender->td_end)) {
608 * RFC 793: "if a TCP is reinitialized ... then it need
609 * not wait at all; it must only be sure to use sequence
610 * numbers larger than those recently used."
613 sender->td_maxend = end;
614 sender->td_maxwin = (win == 0 ? 1 : win);
616 tcp_options(skb, iph, tcph, sender);
621 * If there is no ACK, just pretend it was set and OK.
623 ack = sack = receiver->td_end;
624 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
625 (TCP_FLAG_ACK|TCP_FLAG_RST))
628 * Broken TCP stacks, that set ACK in RST packets as well
629 * with zero ack value.
631 ack = sack = receiver->td_end;
636 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
638 * Packets contains no data: we assume it is valid
639 * and check the ack value only.
640 * However RST segments are always validated by their
641 * SEQ number, except when seq == 0 (reset sent answering
644 seq = end = sender->td_end;
646 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
647 "seq=%u ack=%u sack =%u win=%u end=%u\n",
648 NIPQUAD(iph->saddr), ntohs(tcph->source),
649 NIPQUAD(iph->daddr), ntohs(tcph->dest),
650 seq, ack, sack, win, end);
651 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
652 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
653 sender->td_end, sender->td_maxend, sender->td_maxwin,
655 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
658 DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
659 before(seq, sender->td_maxend + 1),
660 after(end, sender->td_end - receiver->td_maxwin - 1),
661 before(sack, receiver->td_end + 1),
662 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
664 if (sender->loose || receiver->loose ||
665 (before(seq, sender->td_maxend + 1) &&
666 after(end, sender->td_end - receiver->td_maxwin - 1) &&
667 before(sack, receiver->td_end + 1) &&
668 after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
670 * Take into account window scaling (RFC 1323).
673 win <<= sender->td_scale;
676 * Update sender data.
678 swin = win + (sack - ack);
679 if (sender->td_maxwin < swin)
680 sender->td_maxwin = swin;
681 if (after(end, sender->td_end))
682 sender->td_end = end;
684 * Update receiver data.
686 if (after(end, sender->td_maxend))
687 receiver->td_maxwin += end - sender->td_maxend;
688 if (after(sack + win, receiver->td_maxend - 1)) {
689 receiver->td_maxend = sack + win;
691 receiver->td_maxend++;
695 * Check retransmissions.
697 if (index == TCP_ACK_SET) {
698 if (state->last_dir == dir
699 && state->last_seq == seq
700 && state->last_ack == ack
701 && state->last_end == end)
704 state->last_dir = dir;
705 state->last_seq = seq;
706 state->last_ack = ack;
707 state->last_end = end;
712 * Close the window of disabled window tracking :-)
719 if (LOG_INVALID(IPPROTO_TCP))
720 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
722 before(seq, sender->td_maxend + 1) ?
723 after(end, sender->td_end - receiver->td_maxwin - 1) ?
724 before(sack, receiver->td_end + 1) ?
725 after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
726 : "ACK is under the lower bound (possible overly delayed ACK)"
727 : "ACK is over the upper bound (ACKed data not seen yet)"
728 : "SEQ is under the lower bound (already ACKed data retransmitted)"
729 : "SEQ is over the upper bound (over the window of the receiver)");
731 res = ip_ct_tcp_be_liberal;
734 DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
735 "receiver end=%u maxend=%u maxwin=%u\n",
736 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
737 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
742 #ifdef CONFIG_IP_NF_NAT_NEEDED
743 /* Update sender->td_end after NAT successfully mangled the packet */
744 void ip_conntrack_tcp_update(struct sk_buff *skb,
745 struct ip_conntrack *conntrack,
746 enum ip_conntrack_dir dir)
748 struct iphdr *iph = skb->nh.iph;
749 struct tcphdr *tcph = (void *)skb->nh.iph + skb->nh.iph->ihl*4;
752 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
753 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
756 end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, iph, tcph);
758 write_lock_bh(&tcp_lock);
760 * We have to worry for the ack in the reply packet only...
762 if (after(end, conntrack->proto.tcp.seen[dir].td_end))
763 conntrack->proto.tcp.seen[dir].td_end = end;
764 conntrack->proto.tcp.last_end = end;
765 write_unlock_bh(&tcp_lock);
766 DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
767 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
768 sender->td_end, sender->td_maxend, sender->td_maxwin,
770 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
785 /* table of valid flag combinations - ECE and CWR are always valid */
786 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
790 [TH_SYN|TH_ACK|TH_PUSH] = 1,
793 [TH_RST|TH_ACK|TH_PUSH] = 1,
796 [TH_ACK|TH_PUSH] = 1,
798 [TH_ACK|TH_URG|TH_PUSH] = 1,
799 [TH_FIN|TH_ACK|TH_PUSH] = 1,
800 [TH_FIN|TH_ACK|TH_URG] = 1,
801 [TH_FIN|TH_ACK|TH_URG|TH_PUSH] = 1,
804 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
805 static int tcp_error(struct sk_buff *skb,
806 enum ip_conntrack_info *ctinfo,
807 unsigned int hooknum)
809 struct iphdr *iph = skb->nh.iph;
810 struct tcphdr _tcph, *th;
811 unsigned int tcplen = skb->len - iph->ihl * 4;
814 /* Smaller that minimal TCP header? */
815 th = skb_header_pointer(skb, iph->ihl * 4,
816 sizeof(_tcph), &_tcph);
818 if (LOG_INVALID(IPPROTO_TCP))
819 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
820 "ip_ct_tcp: short packet ");
824 /* Not whole TCP header or malformed packet */
825 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
826 if (LOG_INVALID(IPPROTO_TCP))
827 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
828 "ip_ct_tcp: truncated/malformed packet ");
832 /* Checksum invalid? Ignore.
833 * We skip checking packets on the outgoing path
834 * because the semantic of CHECKSUM_HW is different there
835 * and moreover root might send raw packets.
837 /* FIXME: Source route IP option packets --RR */
838 if (hooknum == NF_IP_PRE_ROUTING
839 && skb->ip_summed != CHECKSUM_UNNECESSARY
840 && csum_tcpudp_magic(iph->saddr, iph->daddr, tcplen, IPPROTO_TCP,
841 skb->ip_summed == CHECKSUM_HW ? skb->csum
842 : skb_checksum(skb, iph->ihl*4, tcplen, 0))) {
843 if (LOG_INVALID(IPPROTO_TCP))
844 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
845 "ip_ct_tcp: bad TCP checksum ");
849 /* Check TCP flags. */
850 tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
851 if (!tcp_valid_flags[tcpflags]) {
852 if (LOG_INVALID(IPPROTO_TCP))
853 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
854 "ip_ct_tcp: invalid TCP flag combination ");
861 /* Returns verdict for packet, or -1 for invalid. */
862 static int tcp_packet(struct ip_conntrack *conntrack,
863 const struct sk_buff *skb,
864 enum ip_conntrack_info ctinfo)
866 enum tcp_conntrack new_state, old_state;
867 enum ip_conntrack_dir dir;
868 struct iphdr *iph = skb->nh.iph;
869 struct tcphdr *th, _tcph;
870 unsigned long timeout;
873 th = skb_header_pointer(skb, iph->ihl * 4,
874 sizeof(_tcph), &_tcph);
877 write_lock_bh(&tcp_lock);
878 old_state = conntrack->proto.tcp.state;
879 dir = CTINFO2DIR(ctinfo);
880 index = get_conntrack_index(th);
881 new_state = tcp_conntracks[dir][index][old_state];
884 case TCP_CONNTRACK_IGNORE:
885 /* Either SYN in ORIGINAL
886 * or SYN/ACK in REPLY. */
887 if (index == TCP_SYNACK_SET
888 && conntrack->proto.tcp.last_index == TCP_SYN_SET
889 && conntrack->proto.tcp.last_dir != dir
890 && ntohl(th->ack_seq) ==
891 conntrack->proto.tcp.last_end) {
892 /* This SYN/ACK acknowledges a SYN that we earlier
893 * ignored as invalid. This means that the client and
894 * the server are both in sync, while the firewall is
895 * not. We kill this session and block the SYN/ACK so
896 * that the client cannot but retransmit its SYN and
897 * thus initiate a clean new session.
899 write_unlock_bh(&tcp_lock);
900 if (LOG_INVALID(IPPROTO_TCP))
901 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
903 "killing out of sync session ");
904 if (del_timer(&conntrack->timeout))
905 conntrack->timeout.function((unsigned long)
909 conntrack->proto.tcp.last_index = index;
910 conntrack->proto.tcp.last_dir = dir;
911 conntrack->proto.tcp.last_seq = ntohl(th->seq);
912 conntrack->proto.tcp.last_end =
913 segment_seq_plus_len(ntohl(th->seq), skb->len, iph, th);
915 write_unlock_bh(&tcp_lock);
916 if (LOG_INVALID(IPPROTO_TCP))
917 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
918 "ip_ct_tcp: invalid packet ignored ");
920 case TCP_CONNTRACK_MAX:
922 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
923 dir, get_conntrack_index(th),
925 write_unlock_bh(&tcp_lock);
926 if (LOG_INVALID(IPPROTO_TCP))
927 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
928 "ip_ct_tcp: invalid state ");
930 case TCP_CONNTRACK_SYN_SENT:
931 if (old_state < TCP_CONNTRACK_TIME_WAIT)
933 if ((conntrack->proto.tcp.seen[dir].flags &
934 IP_CT_TCP_FLAG_CLOSE_INIT)
935 || after(ntohl(th->seq),
936 conntrack->proto.tcp.seen[dir].td_end)) {
937 /* Attempt to reopen a closed connection.
938 * Delete this connection and look up again. */
939 write_unlock_bh(&tcp_lock);
940 if (del_timer(&conntrack->timeout))
941 conntrack->timeout.function((unsigned long)
945 write_unlock_bh(&tcp_lock);
946 if (LOG_INVALID(IPPROTO_TCP))
947 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
948 NULL, "ip_ct_tcp: invalid SYN");
951 case TCP_CONNTRACK_CLOSE:
952 if (index == TCP_RST_SET
953 && test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
954 && conntrack->proto.tcp.last_index == TCP_SYN_SET
955 && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
956 /* RST sent to invalid SYN we had let trough
957 * SYN was in window then, tear down connection.
958 * We skip window checking, because packet might ACK
959 * segments we ignored in the SYN. */
962 /* Just fall trough */
964 /* Keep compilers happy. */
968 if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
970 write_unlock_bh(&tcp_lock);
974 /* From now on we have got in-window packets */
975 conntrack->proto.tcp.last_index = index;
977 DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
978 "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
979 NIPQUAD(iph->saddr), ntohs(th->source),
980 NIPQUAD(iph->daddr), ntohs(th->dest),
981 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
982 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
983 old_state, new_state);
985 conntrack->proto.tcp.state = new_state;
986 if (old_state != new_state
987 && (new_state == TCP_CONNTRACK_FIN_WAIT
988 || new_state == TCP_CONNTRACK_CLOSE))
989 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
990 timeout = conntrack->proto.tcp.retrans >= ip_ct_tcp_max_retrans
991 && *tcp_timeouts[new_state] > ip_ct_tcp_timeout_max_retrans
992 ? ip_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
993 write_unlock_bh(&tcp_lock);
995 ip_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
996 if (new_state != old_state)
997 ip_conntrack_event_cache(IPCT_PROTOINFO, skb);
999 if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
1000 /* If only reply is a RST, we can consider ourselves not to
1001 have an established connection: this is a fairly common
1002 problem case, so we can delete the conntrack
1003 immediately. --RR */
1005 if (del_timer(&conntrack->timeout))
1006 conntrack->timeout.function((unsigned long)
1010 } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
1011 && (old_state == TCP_CONNTRACK_SYN_RECV
1012 || old_state == TCP_CONNTRACK_ESTABLISHED)
1013 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1014 /* Set ASSURED if we see see valid ack in ESTABLISHED
1015 after SYN_RECV or a valid answer for a picked up
1017 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1018 ip_conntrack_event_cache(IPCT_STATUS, skb);
1020 ip_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1025 /* Called when a new connection for this protocol found. */
1026 static int tcp_new(struct ip_conntrack *conntrack,
1027 const struct sk_buff *skb)
1029 enum tcp_conntrack new_state;
1030 struct iphdr *iph = skb->nh.iph;
1031 struct tcphdr *th, _tcph;
1033 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1034 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1037 th = skb_header_pointer(skb, iph->ihl * 4,
1038 sizeof(_tcph), &_tcph);
1041 /* Don't need lock here: this conntrack not in circulation yet */
1043 = tcp_conntracks[0][get_conntrack_index(th)]
1044 [TCP_CONNTRACK_NONE];
1046 /* Invalid: delete conntrack */
1047 if (new_state >= TCP_CONNTRACK_MAX) {
1048 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1052 if (new_state == TCP_CONNTRACK_SYN_SENT) {
1054 conntrack->proto.tcp.seen[0].td_end =
1055 segment_seq_plus_len(ntohl(th->seq), skb->len,
1057 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1058 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1059 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1060 conntrack->proto.tcp.seen[0].td_maxend =
1061 conntrack->proto.tcp.seen[0].td_end;
1063 tcp_options(skb, iph, th, &conntrack->proto.tcp.seen[0]);
1064 conntrack->proto.tcp.seen[1].flags = 0;
1065 conntrack->proto.tcp.seen[0].loose =
1066 conntrack->proto.tcp.seen[1].loose = 0;
1067 } else if (ip_ct_tcp_loose == 0) {
1068 /* Don't try to pick up connections. */
1072 * We are in the middle of a connection,
1073 * its history is lost for us.
1074 * Let's try to use the data from the packet.
1076 conntrack->proto.tcp.seen[0].td_end =
1077 segment_seq_plus_len(ntohl(th->seq), skb->len,
1079 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1080 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1081 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1082 conntrack->proto.tcp.seen[0].td_maxend =
1083 conntrack->proto.tcp.seen[0].td_end +
1084 conntrack->proto.tcp.seen[0].td_maxwin;
1085 conntrack->proto.tcp.seen[0].td_scale = 0;
1087 /* We assume SACK. Should we assume window scaling too? */
1088 conntrack->proto.tcp.seen[0].flags =
1089 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1090 conntrack->proto.tcp.seen[0].loose =
1091 conntrack->proto.tcp.seen[1].loose = ip_ct_tcp_loose;
1094 conntrack->proto.tcp.seen[1].td_end = 0;
1095 conntrack->proto.tcp.seen[1].td_maxend = 0;
1096 conntrack->proto.tcp.seen[1].td_maxwin = 1;
1097 conntrack->proto.tcp.seen[1].td_scale = 0;
1099 /* tcp_packet will set them */
1100 conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1101 conntrack->proto.tcp.last_index = TCP_NONE_SET;
1103 DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1104 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1105 sender->td_end, sender->td_maxend, sender->td_maxwin,
1107 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1108 receiver->td_scale);
1112 struct ip_conntrack_protocol ip_conntrack_protocol_tcp =
1114 .proto = IPPROTO_TCP,
1116 .pkt_to_tuple = tcp_pkt_to_tuple,
1117 .invert_tuple = tcp_invert_tuple,
1118 .print_tuple = tcp_print_tuple,
1119 .print_conntrack = tcp_print_conntrack,
1120 .packet = tcp_packet,
1123 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
1124 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
1125 .to_nfattr = tcp_to_nfattr,
1126 .tuple_to_nfattr = ip_ct_port_tuple_to_nfattr,
1127 .nfattr_to_tuple = ip_ct_port_nfattr_to_tuple,