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 const 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 const 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, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
277 * sSS -> sIG 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;
346 read_lock_bh(&tcp_lock);
347 nest_parms = NFA_NEST(skb, CTA_PROTOINFO_TCP);
348 NFA_PUT(skb, CTA_PROTOINFO_TCP_STATE, sizeof(u_int8_t),
349 &ct->proto.tcp.state);
350 read_unlock_bh(&tcp_lock);
352 NFA_NEST_END(skb, nest_parms);
357 read_unlock_bh(&tcp_lock);
361 static const size_t cta_min_tcp[CTA_PROTOINFO_TCP_MAX] = {
362 [CTA_PROTOINFO_TCP_STATE-1] = sizeof(u_int8_t),
365 static int nfattr_to_tcp(struct nfattr *cda[], struct ip_conntrack *ct)
367 struct nfattr *attr = cda[CTA_PROTOINFO_TCP-1];
368 struct nfattr *tb[CTA_PROTOINFO_TCP_MAX];
370 /* updates could not contain anything about the private
371 * protocol info, in that case skip the parsing */
375 nfattr_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, attr);
377 if (nfattr_bad_size(tb, CTA_PROTOINFO_TCP_MAX, cta_min_tcp))
380 if (!tb[CTA_PROTOINFO_TCP_STATE-1])
383 write_lock_bh(&tcp_lock);
384 ct->proto.tcp.state =
385 *(u_int8_t *)NFA_DATA(tb[CTA_PROTOINFO_TCP_STATE-1]);
386 write_unlock_bh(&tcp_lock);
392 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
394 if (tcph->rst) return TCP_RST_SET;
395 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
396 else if (tcph->fin) return TCP_FIN_SET;
397 else if (tcph->ack) return TCP_ACK_SET;
398 else return TCP_NONE_SET;
401 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
402 in IP Filter' by Guido van Rooij.
404 http://www.nluug.nl/events/sane2000/papers.html
405 http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
407 The boundaries and the conditions are changed according to RFC793:
408 the packet must intersect the window (i.e. segments may be
409 after the right or before the left edge) and thus receivers may ACK
410 segments after the right edge of the window.
412 td_maxend = max(sack + max(win,1)) seen in reply packets
413 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
414 td_maxwin += seq + len - sender.td_maxend
415 if seq + len > sender.td_maxend
416 td_end = max(seq + len) seen in sent packets
418 I. Upper bound for valid data: seq <= sender.td_maxend
419 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
420 III. Upper bound for valid ack: sack <= receiver.td_end
421 IV. Lower bound for valid ack: ack >= receiver.td_end - MAXACKWINDOW
423 where sack is the highest right edge of sack block found in the packet.
425 The upper bound limit for a valid ack is not ignored -
426 we doesn't have to deal with fragments.
429 static inline __u32 segment_seq_plus_len(__u32 seq,
434 return (seq + len - (iph->ihl + tcph->doff)*4
435 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
438 /* Fixme: what about big packets? */
439 #define MAXACKWINCONST 66000
440 #define MAXACKWINDOW(sender) \
441 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
445 * Simplified tcp_parse_options routine from tcp_input.c
447 static void tcp_options(const struct sk_buff *skb,
450 struct ip_ct_tcp_state *state)
452 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
454 int length = (tcph->doff*4) - sizeof(struct tcphdr);
459 ptr = skb_header_pointer(skb,
460 (iph->ihl * 4) + sizeof(struct tcphdr),
474 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
479 if (opsize < 2) /* "silly options" */
482 break; /* don't parse partial options */
484 if (opcode == TCPOPT_SACK_PERM
485 && opsize == TCPOLEN_SACK_PERM)
486 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
487 else if (opcode == TCPOPT_WINDOW
488 && opsize == TCPOLEN_WINDOW) {
489 state->td_scale = *(u_int8_t *)ptr;
491 if (state->td_scale > 14) {
493 state->td_scale = 14;
496 IP_CT_TCP_FLAG_WINDOW_SCALE;
504 static void tcp_sack(const struct sk_buff *skb,
509 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
511 int length = (tcph->doff*4) - sizeof(struct tcphdr);
517 ptr = skb_header_pointer(skb,
518 (iph->ihl * 4) + sizeof(struct tcphdr),
522 /* Fast path for timestamp-only option */
523 if (length == TCPOLEN_TSTAMP_ALIGNED*4
525 __constant_ntohl((TCPOPT_NOP << 24)
527 | (TCPOPT_TIMESTAMP << 8)
528 | TCPOLEN_TIMESTAMP))
538 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
543 if (opsize < 2) /* "silly options" */
546 break; /* don't parse partial options */
548 if (opcode == TCPOPT_SACK
549 && opsize >= (TCPOLEN_SACK_BASE
550 + TCPOLEN_SACK_PERBLOCK)
551 && !((opsize - TCPOLEN_SACK_BASE)
552 % TCPOLEN_SACK_PERBLOCK)) {
554 i < (opsize - TCPOLEN_SACK_BASE);
555 i += TCPOLEN_SACK_PERBLOCK) {
556 tmp = ntohl(*((u_int32_t *)(ptr+i)+1));
558 if (after(tmp, *sack))
569 static int tcp_in_window(struct ip_ct_tcp *state,
570 enum ip_conntrack_dir dir,
572 const struct sk_buff *skb,
576 struct ip_ct_tcp_state *sender = &state->seen[dir];
577 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
578 __u32 seq, ack, sack, end, win, swin;
582 * Get the required data from the packet.
584 seq = ntohl(tcph->seq);
585 ack = sack = ntohl(tcph->ack_seq);
586 win = ntohs(tcph->window);
587 end = segment_seq_plus_len(seq, skb->len, iph, tcph);
589 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
590 tcp_sack(skb, iph, tcph, &sack);
592 DEBUGP("tcp_in_window: START\n");
593 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
594 "seq=%u ack=%u sack=%u win=%u end=%u\n",
595 NIPQUAD(iph->saddr), ntohs(tcph->source),
596 NIPQUAD(iph->daddr), ntohs(tcph->dest),
597 seq, ack, sack, win, end);
598 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
599 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
600 sender->td_end, sender->td_maxend, sender->td_maxwin,
602 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
605 if (sender->td_end == 0) {
607 * Initialize sender data.
609 if (tcph->syn && tcph->ack) {
611 * Outgoing SYN-ACK in reply to a SYN.
614 sender->td_maxend = end;
615 sender->td_maxwin = (win == 0 ? 1 : win);
617 tcp_options(skb, iph, tcph, sender);
620 * Both sides must send the Window Scale option
621 * to enable window scaling in either direction.
623 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
624 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
626 receiver->td_scale = 0;
629 * We are in the middle of a connection,
630 * its history is lost for us.
631 * Let's try to use the data from the packet.
633 sender->td_end = end;
634 sender->td_maxwin = (win == 0 ? 1 : win);
635 sender->td_maxend = end + sender->td_maxwin;
637 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
638 && dir == IP_CT_DIR_ORIGINAL)
639 || (state->state == TCP_CONNTRACK_SYN_RECV
640 && dir == IP_CT_DIR_REPLY))
641 && after(end, sender->td_end)) {
643 * RFC 793: "if a TCP is reinitialized ... then it need
644 * not wait at all; it must only be sure to use sequence
645 * numbers larger than those recently used."
648 sender->td_maxend = end;
649 sender->td_maxwin = (win == 0 ? 1 : win);
651 tcp_options(skb, iph, tcph, sender);
656 * If there is no ACK, just pretend it was set and OK.
658 ack = sack = receiver->td_end;
659 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
660 (TCP_FLAG_ACK|TCP_FLAG_RST))
663 * Broken TCP stacks, that set ACK in RST packets as well
664 * with zero ack value.
666 ack = sack = receiver->td_end;
671 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
673 * Packets contains no data: we assume it is valid
674 * and check the ack value only.
675 * However RST segments are always validated by their
676 * SEQ number, except when seq == 0 (reset sent answering
679 seq = end = sender->td_end;
681 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
682 "seq=%u ack=%u sack =%u win=%u end=%u\n",
683 NIPQUAD(iph->saddr), ntohs(tcph->source),
684 NIPQUAD(iph->daddr), ntohs(tcph->dest),
685 seq, ack, sack, win, end);
686 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
687 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
688 sender->td_end, sender->td_maxend, sender->td_maxwin,
690 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
693 DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
694 before(seq, sender->td_maxend + 1),
695 after(end, sender->td_end - receiver->td_maxwin - 1),
696 before(sack, receiver->td_end + 1),
697 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
699 if (sender->loose || receiver->loose ||
700 (before(seq, sender->td_maxend + 1) &&
701 after(end, sender->td_end - receiver->td_maxwin - 1) &&
702 before(sack, receiver->td_end + 1) &&
703 after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
705 * Take into account window scaling (RFC 1323).
708 win <<= sender->td_scale;
711 * Update sender data.
713 swin = win + (sack - ack);
714 if (sender->td_maxwin < swin)
715 sender->td_maxwin = swin;
716 if (after(end, sender->td_end))
717 sender->td_end = end;
719 * Update receiver data.
721 if (after(end, sender->td_maxend))
722 receiver->td_maxwin += end - sender->td_maxend;
723 if (after(sack + win, receiver->td_maxend - 1)) {
724 receiver->td_maxend = sack + win;
726 receiver->td_maxend++;
730 * Check retransmissions.
732 if (index == TCP_ACK_SET) {
733 if (state->last_dir == dir
734 && state->last_seq == seq
735 && state->last_ack == ack
736 && state->last_end == end)
739 state->last_dir = dir;
740 state->last_seq = seq;
741 state->last_ack = ack;
742 state->last_end = end;
747 * Close the window of disabled window tracking :-)
754 if (LOG_INVALID(IPPROTO_TCP))
755 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
757 before(seq, sender->td_maxend + 1) ?
758 after(end, sender->td_end - receiver->td_maxwin - 1) ?
759 before(sack, receiver->td_end + 1) ?
760 after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
761 : "ACK is under the lower bound (possible overly delayed ACK)"
762 : "ACK is over the upper bound (ACKed data not seen yet)"
763 : "SEQ is under the lower bound (already ACKed data retransmitted)"
764 : "SEQ is over the upper bound (over the window of the receiver)");
766 res = ip_ct_tcp_be_liberal;
769 DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
770 "receiver end=%u maxend=%u maxwin=%u\n",
771 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
772 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
777 #ifdef CONFIG_IP_NF_NAT_NEEDED
778 /* Update sender->td_end after NAT successfully mangled the packet */
779 void ip_conntrack_tcp_update(struct sk_buff *skb,
780 struct ip_conntrack *conntrack,
781 enum ip_conntrack_dir dir)
783 struct iphdr *iph = skb->nh.iph;
784 struct tcphdr *tcph = (void *)skb->nh.iph + skb->nh.iph->ihl*4;
787 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
788 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
791 end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, iph, tcph);
793 write_lock_bh(&tcp_lock);
795 * We have to worry for the ack in the reply packet only...
797 if (after(end, conntrack->proto.tcp.seen[dir].td_end))
798 conntrack->proto.tcp.seen[dir].td_end = end;
799 conntrack->proto.tcp.last_end = end;
800 write_unlock_bh(&tcp_lock);
801 DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
802 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
803 sender->td_end, sender->td_maxend, sender->td_maxwin,
805 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
820 /* table of valid flag combinations - ECE and CWR are always valid */
821 static const u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
825 [TH_SYN|TH_PUSH] = 1,
826 [TH_SYN|TH_ACK|TH_PUSH] = 1,
829 [TH_RST|TH_ACK|TH_PUSH] = 1,
832 [TH_ACK|TH_PUSH] = 1,
834 [TH_ACK|TH_URG|TH_PUSH] = 1,
835 [TH_FIN|TH_ACK|TH_PUSH] = 1,
836 [TH_FIN|TH_ACK|TH_URG] = 1,
837 [TH_FIN|TH_ACK|TH_URG|TH_PUSH] = 1,
840 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
841 static int tcp_error(struct sk_buff *skb,
842 enum ip_conntrack_info *ctinfo,
843 unsigned int hooknum)
845 struct iphdr *iph = skb->nh.iph;
846 struct tcphdr _tcph, *th;
847 unsigned int tcplen = skb->len - iph->ihl * 4;
850 /* Smaller that minimal TCP header? */
851 th = skb_header_pointer(skb, iph->ihl * 4,
852 sizeof(_tcph), &_tcph);
854 if (LOG_INVALID(IPPROTO_TCP))
855 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
856 "ip_ct_tcp: short packet ");
860 /* Not whole TCP header or malformed packet */
861 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
862 if (LOG_INVALID(IPPROTO_TCP))
863 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
864 "ip_ct_tcp: truncated/malformed packet ");
868 /* Checksum invalid? Ignore.
869 * We skip checking packets on the outgoing path
870 * because the semantic of CHECKSUM_HW is different there
871 * and moreover root might send raw packets.
873 /* FIXME: Source route IP option packets --RR */
874 if (hooknum == NF_IP_PRE_ROUTING
875 && skb->ip_summed != CHECKSUM_UNNECESSARY
876 && csum_tcpudp_magic(iph->saddr, iph->daddr, tcplen, IPPROTO_TCP,
877 skb->ip_summed == CHECKSUM_HW ? skb->csum
878 : skb_checksum(skb, iph->ihl*4, tcplen, 0))) {
879 if (LOG_INVALID(IPPROTO_TCP))
880 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
881 "ip_ct_tcp: bad TCP checksum ");
885 /* Check TCP flags. */
886 tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
887 if (!tcp_valid_flags[tcpflags]) {
888 if (LOG_INVALID(IPPROTO_TCP))
889 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
890 "ip_ct_tcp: invalid TCP flag combination ");
897 /* Returns verdict for packet, or -1 for invalid. */
898 static int tcp_packet(struct ip_conntrack *conntrack,
899 const struct sk_buff *skb,
900 enum ip_conntrack_info ctinfo)
902 enum tcp_conntrack new_state, old_state;
903 enum ip_conntrack_dir dir;
904 struct iphdr *iph = skb->nh.iph;
905 struct tcphdr *th, _tcph;
906 unsigned long timeout;
909 th = skb_header_pointer(skb, iph->ihl * 4,
910 sizeof(_tcph), &_tcph);
913 write_lock_bh(&tcp_lock);
914 old_state = conntrack->proto.tcp.state;
915 dir = CTINFO2DIR(ctinfo);
916 index = get_conntrack_index(th);
917 new_state = tcp_conntracks[dir][index][old_state];
920 case TCP_CONNTRACK_IGNORE:
924 * b) SYN/ACK in REPLY
925 * c) ACK in reply direction after initial SYN in original.
927 if (index == TCP_SYNACK_SET
928 && conntrack->proto.tcp.last_index == TCP_SYN_SET
929 && conntrack->proto.tcp.last_dir != dir
930 && ntohl(th->ack_seq) ==
931 conntrack->proto.tcp.last_end) {
932 /* This SYN/ACK acknowledges a SYN that we earlier
933 * ignored as invalid. This means that the client and
934 * the server are both in sync, while the firewall is
935 * not. We kill this session and block the SYN/ACK so
936 * that the client cannot but retransmit its SYN and
937 * thus initiate a clean new session.
939 write_unlock_bh(&tcp_lock);
940 if (LOG_INVALID(IPPROTO_TCP))
941 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
943 "killing out of sync session ");
944 if (del_timer(&conntrack->timeout))
945 conntrack->timeout.function((unsigned long)
949 conntrack->proto.tcp.last_index = index;
950 conntrack->proto.tcp.last_dir = dir;
951 conntrack->proto.tcp.last_seq = ntohl(th->seq);
952 conntrack->proto.tcp.last_end =
953 segment_seq_plus_len(ntohl(th->seq), skb->len, iph, th);
955 write_unlock_bh(&tcp_lock);
956 if (LOG_INVALID(IPPROTO_TCP))
957 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
958 "ip_ct_tcp: invalid packet ignored ");
960 case TCP_CONNTRACK_MAX:
962 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
963 dir, get_conntrack_index(th),
965 write_unlock_bh(&tcp_lock);
966 if (LOG_INVALID(IPPROTO_TCP))
967 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
968 "ip_ct_tcp: invalid state ");
970 case TCP_CONNTRACK_SYN_SENT:
971 if (old_state < TCP_CONNTRACK_TIME_WAIT)
973 if ((conntrack->proto.tcp.seen[dir].flags &
974 IP_CT_TCP_FLAG_CLOSE_INIT)
975 || after(ntohl(th->seq),
976 conntrack->proto.tcp.seen[dir].td_end)) {
977 /* Attempt to reopen a closed connection.
978 * Delete this connection and look up again. */
979 write_unlock_bh(&tcp_lock);
980 if (del_timer(&conntrack->timeout))
981 conntrack->timeout.function((unsigned long)
985 write_unlock_bh(&tcp_lock);
986 if (LOG_INVALID(IPPROTO_TCP))
987 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
988 NULL, "ip_ct_tcp: invalid SYN");
991 case TCP_CONNTRACK_CLOSE:
992 if (index == TCP_RST_SET
993 && ((test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
994 && conntrack->proto.tcp.last_index == TCP_SYN_SET)
995 || (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
996 && conntrack->proto.tcp.last_index == TCP_ACK_SET))
997 && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
998 /* RST sent to invalid SYN or ACK we had let trough
999 * at a) and c) above:
1001 * a) SYN was in window then
1002 * c) we hold a half-open connection.
1004 * Delete our connection entry.
1005 * We skip window checking, because packet might ACK
1006 * segments we ignored. */
1009 /* Just fall trough */
1011 /* Keep compilers happy. */
1015 if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
1017 write_unlock_bh(&tcp_lock);
1021 /* From now on we have got in-window packets */
1022 conntrack->proto.tcp.last_index = index;
1024 DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
1025 "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1026 NIPQUAD(iph->saddr), ntohs(th->source),
1027 NIPQUAD(iph->daddr), ntohs(th->dest),
1028 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1029 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1030 old_state, new_state);
1032 conntrack->proto.tcp.state = new_state;
1033 if (old_state != new_state
1034 && (new_state == TCP_CONNTRACK_FIN_WAIT
1035 || new_state == TCP_CONNTRACK_CLOSE))
1036 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1037 timeout = conntrack->proto.tcp.retrans >= ip_ct_tcp_max_retrans
1038 && *tcp_timeouts[new_state] > ip_ct_tcp_timeout_max_retrans
1039 ? ip_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
1040 write_unlock_bh(&tcp_lock);
1042 ip_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
1043 if (new_state != old_state)
1044 ip_conntrack_event_cache(IPCT_PROTOINFO, skb);
1046 if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
1047 /* If only reply is a RST, we can consider ourselves not to
1048 have an established connection: this is a fairly common
1049 problem case, so we can delete the conntrack
1050 immediately. --RR */
1052 if (del_timer(&conntrack->timeout))
1053 conntrack->timeout.function((unsigned long)
1057 } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
1058 && (old_state == TCP_CONNTRACK_SYN_RECV
1059 || old_state == TCP_CONNTRACK_ESTABLISHED)
1060 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1061 /* Set ASSURED if we see see valid ack in ESTABLISHED
1062 after SYN_RECV or a valid answer for a picked up
1064 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1065 ip_conntrack_event_cache(IPCT_STATUS, skb);
1067 ip_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1072 /* Called when a new connection for this protocol found. */
1073 static int tcp_new(struct ip_conntrack *conntrack,
1074 const struct sk_buff *skb)
1076 enum tcp_conntrack new_state;
1077 struct iphdr *iph = skb->nh.iph;
1078 struct tcphdr *th, _tcph;
1080 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1081 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1084 th = skb_header_pointer(skb, iph->ihl * 4,
1085 sizeof(_tcph), &_tcph);
1088 /* Don't need lock here: this conntrack not in circulation yet */
1090 = tcp_conntracks[0][get_conntrack_index(th)]
1091 [TCP_CONNTRACK_NONE];
1093 /* Invalid: delete conntrack */
1094 if (new_state >= TCP_CONNTRACK_MAX) {
1095 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1099 if (new_state == TCP_CONNTRACK_SYN_SENT) {
1101 conntrack->proto.tcp.seen[0].td_end =
1102 segment_seq_plus_len(ntohl(th->seq), skb->len,
1104 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1105 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1106 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1107 conntrack->proto.tcp.seen[0].td_maxend =
1108 conntrack->proto.tcp.seen[0].td_end;
1110 tcp_options(skb, iph, th, &conntrack->proto.tcp.seen[0]);
1111 conntrack->proto.tcp.seen[1].flags = 0;
1112 conntrack->proto.tcp.seen[0].loose =
1113 conntrack->proto.tcp.seen[1].loose = 0;
1114 } else if (ip_ct_tcp_loose == 0) {
1115 /* Don't try to pick up connections. */
1119 * We are in the middle of a connection,
1120 * its history is lost for us.
1121 * Let's try to use the data from the packet.
1123 conntrack->proto.tcp.seen[0].td_end =
1124 segment_seq_plus_len(ntohl(th->seq), skb->len,
1126 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1127 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1128 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1129 conntrack->proto.tcp.seen[0].td_maxend =
1130 conntrack->proto.tcp.seen[0].td_end +
1131 conntrack->proto.tcp.seen[0].td_maxwin;
1132 conntrack->proto.tcp.seen[0].td_scale = 0;
1134 /* We assume SACK. Should we assume window scaling too? */
1135 conntrack->proto.tcp.seen[0].flags =
1136 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1137 conntrack->proto.tcp.seen[0].loose =
1138 conntrack->proto.tcp.seen[1].loose = ip_ct_tcp_loose;
1141 conntrack->proto.tcp.seen[1].td_end = 0;
1142 conntrack->proto.tcp.seen[1].td_maxend = 0;
1143 conntrack->proto.tcp.seen[1].td_maxwin = 1;
1144 conntrack->proto.tcp.seen[1].td_scale = 0;
1146 /* tcp_packet will set them */
1147 conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1148 conntrack->proto.tcp.last_index = TCP_NONE_SET;
1150 DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1151 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1152 sender->td_end, sender->td_maxend, sender->td_maxwin,
1154 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1155 receiver->td_scale);
1159 struct ip_conntrack_protocol ip_conntrack_protocol_tcp =
1161 .proto = IPPROTO_TCP,
1163 .pkt_to_tuple = tcp_pkt_to_tuple,
1164 .invert_tuple = tcp_invert_tuple,
1165 .print_tuple = tcp_print_tuple,
1166 .print_conntrack = tcp_print_conntrack,
1167 .packet = tcp_packet,
1170 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
1171 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
1172 .to_nfattr = tcp_to_nfattr,
1173 .from_nfattr = nfattr_to_tcp,
1174 .tuple_to_nfattr = ip_ct_port_tuple_to_nfattr,
1175 .nfattr_to_tuple = ip_ct_port_nfattr_to_tuple,