1 /* (C) 1999-2001 Paul `Rusty' Russell
2 * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>:
9 * - Real stateful connection tracking
10 * - Modified state transitions table
11 * - Window scaling support added
12 * - SACK support added
15 * - State table bugfixes
16 * - More robust state changes
17 * - Tuning timer parameters
22 #include <linux/config.h>
23 #include <linux/types.h>
24 #include <linux/sched.h>
25 #include <linux/timer.h>
26 #include <linux/netfilter.h>
27 #include <linux/module.h>
30 #include <linux/tcp.h>
31 #include <linux/spinlock.h>
35 #include <linux/netfilter.h>
36 #include <linux/netfilter_ipv4.h>
37 #include <linux/netfilter_ipv4/ip_conntrack.h>
38 #include <linux/netfilter_ipv4/ip_conntrack_protocol.h>
44 #define DEBUGP(format, args...)
47 /* Protects conntrack->proto.tcp */
48 static DEFINE_RWLOCK(tcp_lock);
50 /* "Be conservative in what you do,
51 be liberal in what you accept from others."
52 If it's non-zero, we mark only out of window RST segments as INVALID. */
53 int ip_ct_tcp_be_liberal = 0;
55 /* When connection is picked up from the middle, how many packets are required
56 to pass in each direction when we assume we are in sync - if any side uses
57 window scaling, we lost the game.
58 If it is set to zero, we disable picking up already established
60 int ip_ct_tcp_loose = 3;
62 /* Max number of the retransmitted packets without receiving an (acceptable)
63 ACK from the destination. If this number is reached, a shorter timer
65 int ip_ct_tcp_max_retrans = 3;
67 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
68 closely. They're more complex. --RR */
70 static const char *tcp_conntrack_names[] = {
84 #define MINS * 60 SECS
85 #define HOURS * 60 MINS
86 #define DAYS * 24 HOURS
88 unsigned long ip_ct_tcp_timeout_syn_sent = 2 MINS;
89 unsigned long ip_ct_tcp_timeout_syn_recv = 60 SECS;
90 unsigned long ip_ct_tcp_timeout_established = 5 DAYS;
91 unsigned long ip_ct_tcp_timeout_fin_wait = 2 MINS;
92 unsigned long ip_ct_tcp_timeout_close_wait = 60 SECS;
93 unsigned long ip_ct_tcp_timeout_last_ack = 30 SECS;
94 unsigned long ip_ct_tcp_timeout_time_wait = 2 MINS;
95 unsigned long ip_ct_tcp_timeout_close = 10 SECS;
97 /* RFC1122 says the R2 limit should be at least 100 seconds.
98 Linux uses 15 packets as limit, which corresponds
99 to ~13-30min depending on RTO. */
100 unsigned long ip_ct_tcp_timeout_max_retrans = 5 MINS;
102 static unsigned long * tcp_timeouts[]
103 = { NULL, /* TCP_CONNTRACK_NONE */
104 &ip_ct_tcp_timeout_syn_sent, /* TCP_CONNTRACK_SYN_SENT, */
105 &ip_ct_tcp_timeout_syn_recv, /* TCP_CONNTRACK_SYN_RECV, */
106 &ip_ct_tcp_timeout_established, /* TCP_CONNTRACK_ESTABLISHED, */
107 &ip_ct_tcp_timeout_fin_wait, /* TCP_CONNTRACK_FIN_WAIT, */
108 &ip_ct_tcp_timeout_close_wait, /* TCP_CONNTRACK_CLOSE_WAIT, */
109 &ip_ct_tcp_timeout_last_ack, /* TCP_CONNTRACK_LAST_ACK, */
110 &ip_ct_tcp_timeout_time_wait, /* TCP_CONNTRACK_TIME_WAIT, */
111 &ip_ct_tcp_timeout_close, /* TCP_CONNTRACK_CLOSE, */
112 NULL, /* TCP_CONNTRACK_LISTEN */
115 #define sNO TCP_CONNTRACK_NONE
116 #define sSS TCP_CONNTRACK_SYN_SENT
117 #define sSR TCP_CONNTRACK_SYN_RECV
118 #define sES TCP_CONNTRACK_ESTABLISHED
119 #define sFW TCP_CONNTRACK_FIN_WAIT
120 #define sCW TCP_CONNTRACK_CLOSE_WAIT
121 #define sLA TCP_CONNTRACK_LAST_ACK
122 #define sTW TCP_CONNTRACK_TIME_WAIT
123 #define sCL TCP_CONNTRACK_CLOSE
124 #define sLI TCP_CONNTRACK_LISTEN
125 #define sIV TCP_CONNTRACK_MAX
126 #define sIG TCP_CONNTRACK_IGNORE
128 /* What TCP flags are set from RST/SYN/FIN/ACK. */
139 * The TCP state transition table needs a few words...
141 * We are the man in the middle. All the packets go through us
142 * but might get lost in transit to the destination.
143 * It is assumed that the destinations can't receive segments
146 * The checked segment is in window, but our windows are *not*
147 * equivalent with the ones of the sender/receiver. We always
148 * try to guess the state of the current sender.
150 * The meaning of the states are:
152 * NONE: initial state
153 * SYN_SENT: SYN-only packet seen
154 * SYN_RECV: SYN-ACK packet seen
155 * ESTABLISHED: ACK packet seen
156 * FIN_WAIT: FIN packet seen
157 * CLOSE_WAIT: ACK seen (after FIN)
158 * LAST_ACK: FIN seen (after FIN)
159 * TIME_WAIT: last ACK seen
160 * CLOSE: closed connection
162 * LISTEN state is not used.
164 * Packets marked as IGNORED (sIG):
165 * if they may be either invalid or valid
166 * and the receiver may send back a connection
167 * closing RST or a SYN/ACK.
169 * Packets marked as INVALID (sIV):
170 * if they are invalid
171 * or we do not support the request (simultaneous open)
173 static enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
176 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
177 /*syn*/ { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
179 * sNO -> sSS Initialize a new connection
180 * sSS -> sSS Retransmitted SYN
181 * sSR -> sIG Late retransmitted SYN?
182 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
183 * are errors. Receiver will reply with RST
184 * and close the connection.
185 * Or we are not in sync and hold a dead connection.
189 * sTW -> sSS Reopened connection (RFC 1122).
192 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
193 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
195 * A SYN/ACK from the client is always invalid:
196 * - either it tries to set up a simultaneous open, which is
198 * - or the firewall has just been inserted between the two hosts
199 * during the session set-up. The SYN will be retransmitted
200 * by the true client (or it'll time out).
202 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
203 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
205 * sNO -> sIV Too late and no reason to do anything...
206 * sSS -> sIV Client migth not send FIN in this state:
207 * we enforce waiting for a SYN/ACK reply first.
208 * sSR -> sFW Close started.
210 * sFW -> sLA FIN seen in both directions, waiting for
212 * Migth be a retransmitted FIN as well...
214 * sLA -> sLA Retransmitted FIN. Remain in the same state.
218 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
219 /*ack*/ { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
221 * sNO -> sES Assumed.
222 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
223 * sSR -> sES Established state is reached.
225 * sFW -> sCW Normal close request answered by ACK.
227 * sLA -> sTW Last ACK detected.
228 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
231 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
232 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
233 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
237 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
238 /*syn*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
240 * sNO -> sIV Never reached.
241 * sSS -> sIV Simultaneous open, not supported
242 * sSR -> sIV Simultaneous open, not supported.
243 * sES -> sIV Server may not initiate a connection.
247 * sTW -> sIV Reopened connection, but server may not do it.
250 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
251 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
253 * sSS -> sSR Standard open.
254 * sSR -> sSR Retransmitted SYN/ACK.
255 * sES -> sIG Late retransmitted SYN/ACK?
256 * sFW -> sIG Might be SYN/ACK answering ignored SYN
262 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
263 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
265 * sSS -> sIV Server might not send FIN in this state.
266 * sSR -> sFW Close started.
268 * sFW -> sLA FIN seen in both directions.
270 * sLA -> sLA Retransmitted FIN.
274 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
275 /*ack*/ { sIV, sIV, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
277 * sSS -> sIV Might be a half-open connection.
278 * sSR -> sSR Might answer late resent SYN.
280 * sFW -> sCW Normal close request answered by ACK.
282 * sLA -> sTW Last ACK detected.
283 * sTW -> sTW Retransmitted last ACK.
286 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
287 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
288 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
292 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
293 unsigned int dataoff,
294 struct ip_conntrack_tuple *tuple)
296 struct tcphdr _hdr, *hp;
298 /* Actually only need first 8 bytes. */
299 hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
303 tuple->src.u.tcp.port = hp->source;
304 tuple->dst.u.tcp.port = hp->dest;
309 static int tcp_invert_tuple(struct ip_conntrack_tuple *tuple,
310 const struct ip_conntrack_tuple *orig)
312 tuple->src.u.tcp.port = orig->dst.u.tcp.port;
313 tuple->dst.u.tcp.port = orig->src.u.tcp.port;
317 /* Print out the per-protocol part of the tuple. */
318 static int tcp_print_tuple(struct seq_file *s,
319 const struct ip_conntrack_tuple *tuple)
321 return seq_printf(s, "sport=%hu dport=%hu ",
322 ntohs(tuple->src.u.tcp.port),
323 ntohs(tuple->dst.u.tcp.port));
326 /* Print out the private part of the conntrack. */
327 static int tcp_print_conntrack(struct seq_file *s,
328 const struct ip_conntrack *conntrack)
330 enum tcp_conntrack state;
332 read_lock_bh(&tcp_lock);
333 state = conntrack->proto.tcp.state;
334 read_unlock_bh(&tcp_lock);
336 return seq_printf(s, "%s ", tcp_conntrack_names[state]);
339 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
340 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
341 static int tcp_to_nfattr(struct sk_buff *skb, struct nfattr *nfa,
342 const struct ip_conntrack *ct)
344 struct nfattr *nest_parms = NFA_NEST(skb, CTA_PROTOINFO_TCP);
346 read_lock_bh(&tcp_lock);
347 NFA_PUT(skb, CTA_PROTOINFO_TCP_STATE, sizeof(u_int8_t),
348 &ct->proto.tcp.state);
349 read_unlock_bh(&tcp_lock);
351 NFA_NEST_END(skb, nest_parms);
356 read_unlock_bh(&tcp_lock);
360 static int nfattr_to_tcp(struct nfattr *cda[], struct ip_conntrack *ct)
362 struct nfattr *attr = cda[CTA_PROTOINFO_TCP-1];
363 struct nfattr *tb[CTA_PROTOINFO_TCP_MAX];
365 if (nfattr_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, attr) < 0)
368 if (!tb[CTA_PROTOINFO_TCP_STATE-1])
371 write_lock_bh(&tcp_lock);
372 ct->proto.tcp.state =
373 *(u_int8_t *)NFA_DATA(tb[CTA_PROTOINFO_TCP_STATE-1]);
374 write_unlock_bh(&tcp_lock);
383 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
385 if (tcph->rst) return TCP_RST_SET;
386 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
387 else if (tcph->fin) return TCP_FIN_SET;
388 else if (tcph->ack) return TCP_ACK_SET;
389 else return TCP_NONE_SET;
392 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
393 in IP Filter' by Guido van Rooij.
395 http://www.nluug.nl/events/sane2000/papers.html
396 http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
398 The boundaries and the conditions are changed according to RFC793:
399 the packet must intersect the window (i.e. segments may be
400 after the right or before the left edge) and thus receivers may ACK
401 segments after the right edge of the window.
403 td_maxend = max(sack + max(win,1)) seen in reply packets
404 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
405 td_maxwin += seq + len - sender.td_maxend
406 if seq + len > sender.td_maxend
407 td_end = max(seq + len) seen in sent packets
409 I. Upper bound for valid data: seq <= sender.td_maxend
410 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
411 III. Upper bound for valid ack: sack <= receiver.td_end
412 IV. Lower bound for valid ack: ack >= receiver.td_end - MAXACKWINDOW
414 where sack is the highest right edge of sack block found in the packet.
416 The upper bound limit for a valid ack is not ignored -
417 we doesn't have to deal with fragments.
420 static inline __u32 segment_seq_plus_len(__u32 seq,
425 return (seq + len - (iph->ihl + tcph->doff)*4
426 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
429 /* Fixme: what about big packets? */
430 #define MAXACKWINCONST 66000
431 #define MAXACKWINDOW(sender) \
432 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
436 * Simplified tcp_parse_options routine from tcp_input.c
438 static void tcp_options(const struct sk_buff *skb,
441 struct ip_ct_tcp_state *state)
443 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
445 int length = (tcph->doff*4) - sizeof(struct tcphdr);
450 ptr = skb_header_pointer(skb,
451 (iph->ihl * 4) + sizeof(struct tcphdr),
465 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
470 if (opsize < 2) /* "silly options" */
473 break; /* don't parse partial options */
475 if (opcode == TCPOPT_SACK_PERM
476 && opsize == TCPOLEN_SACK_PERM)
477 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
478 else if (opcode == TCPOPT_WINDOW
479 && opsize == TCPOLEN_WINDOW) {
480 state->td_scale = *(u_int8_t *)ptr;
482 if (state->td_scale > 14) {
484 state->td_scale = 14;
487 IP_CT_TCP_FLAG_WINDOW_SCALE;
495 static void tcp_sack(const struct sk_buff *skb,
500 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
502 int length = (tcph->doff*4) - sizeof(struct tcphdr);
508 ptr = skb_header_pointer(skb,
509 (iph->ihl * 4) + sizeof(struct tcphdr),
513 /* Fast path for timestamp-only option */
514 if (length == TCPOLEN_TSTAMP_ALIGNED*4
516 __constant_ntohl((TCPOPT_NOP << 24)
518 | (TCPOPT_TIMESTAMP << 8)
519 | TCPOLEN_TIMESTAMP))
529 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
534 if (opsize < 2) /* "silly options" */
537 break; /* don't parse partial options */
539 if (opcode == TCPOPT_SACK
540 && opsize >= (TCPOLEN_SACK_BASE
541 + TCPOLEN_SACK_PERBLOCK)
542 && !((opsize - TCPOLEN_SACK_BASE)
543 % TCPOLEN_SACK_PERBLOCK)) {
545 i < (opsize - TCPOLEN_SACK_BASE);
546 i += TCPOLEN_SACK_PERBLOCK) {
547 tmp = ntohl(*((u_int32_t *)(ptr+i)+1));
549 if (after(tmp, *sack))
560 static int tcp_in_window(struct ip_ct_tcp *state,
561 enum ip_conntrack_dir dir,
563 const struct sk_buff *skb,
567 struct ip_ct_tcp_state *sender = &state->seen[dir];
568 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
569 __u32 seq, ack, sack, end, win, swin;
573 * Get the required data from the packet.
575 seq = ntohl(tcph->seq);
576 ack = sack = ntohl(tcph->ack_seq);
577 win = ntohs(tcph->window);
578 end = segment_seq_plus_len(seq, skb->len, iph, tcph);
580 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
581 tcp_sack(skb, iph, tcph, &sack);
583 DEBUGP("tcp_in_window: START\n");
584 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
585 "seq=%u ack=%u sack=%u win=%u end=%u\n",
586 NIPQUAD(iph->saddr), ntohs(tcph->source),
587 NIPQUAD(iph->daddr), ntohs(tcph->dest),
588 seq, ack, sack, win, end);
589 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
590 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
591 sender->td_end, sender->td_maxend, sender->td_maxwin,
593 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
596 if (sender->td_end == 0) {
598 * Initialize sender data.
600 if (tcph->syn && tcph->ack) {
602 * Outgoing SYN-ACK in reply to a SYN.
605 sender->td_maxend = end;
606 sender->td_maxwin = (win == 0 ? 1 : win);
608 tcp_options(skb, iph, tcph, sender);
611 * Both sides must send the Window Scale option
612 * to enable window scaling in either direction.
614 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
615 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
617 receiver->td_scale = 0;
620 * We are in the middle of a connection,
621 * its history is lost for us.
622 * Let's try to use the data from the packet.
624 sender->td_end = end;
625 sender->td_maxwin = (win == 0 ? 1 : win);
626 sender->td_maxend = end + sender->td_maxwin;
628 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
629 && dir == IP_CT_DIR_ORIGINAL)
630 || (state->state == TCP_CONNTRACK_SYN_RECV
631 && dir == IP_CT_DIR_REPLY))
632 && after(end, sender->td_end)) {
634 * RFC 793: "if a TCP is reinitialized ... then it need
635 * not wait at all; it must only be sure to use sequence
636 * numbers larger than those recently used."
639 sender->td_maxend = end;
640 sender->td_maxwin = (win == 0 ? 1 : win);
642 tcp_options(skb, iph, tcph, sender);
647 * If there is no ACK, just pretend it was set and OK.
649 ack = sack = receiver->td_end;
650 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
651 (TCP_FLAG_ACK|TCP_FLAG_RST))
654 * Broken TCP stacks, that set ACK in RST packets as well
655 * with zero ack value.
657 ack = sack = receiver->td_end;
662 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
664 * Packets contains no data: we assume it is valid
665 * and check the ack value only.
666 * However RST segments are always validated by their
667 * SEQ number, except when seq == 0 (reset sent answering
670 seq = end = sender->td_end;
672 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
673 "seq=%u ack=%u sack =%u win=%u end=%u\n",
674 NIPQUAD(iph->saddr), ntohs(tcph->source),
675 NIPQUAD(iph->daddr), ntohs(tcph->dest),
676 seq, ack, sack, win, end);
677 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
678 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
679 sender->td_end, sender->td_maxend, sender->td_maxwin,
681 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
684 DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
685 before(seq, sender->td_maxend + 1),
686 after(end, sender->td_end - receiver->td_maxwin - 1),
687 before(sack, receiver->td_end + 1),
688 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
690 if (sender->loose || receiver->loose ||
691 (before(seq, sender->td_maxend + 1) &&
692 after(end, sender->td_end - receiver->td_maxwin - 1) &&
693 before(sack, receiver->td_end + 1) &&
694 after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
696 * Take into account window scaling (RFC 1323).
699 win <<= sender->td_scale;
702 * Update sender data.
704 swin = win + (sack - ack);
705 if (sender->td_maxwin < swin)
706 sender->td_maxwin = swin;
707 if (after(end, sender->td_end))
708 sender->td_end = end;
710 * Update receiver data.
712 if (after(end, sender->td_maxend))
713 receiver->td_maxwin += end - sender->td_maxend;
714 if (after(sack + win, receiver->td_maxend - 1)) {
715 receiver->td_maxend = sack + win;
717 receiver->td_maxend++;
721 * Check retransmissions.
723 if (index == TCP_ACK_SET) {
724 if (state->last_dir == dir
725 && state->last_seq == seq
726 && state->last_ack == ack
727 && state->last_end == end)
730 state->last_dir = dir;
731 state->last_seq = seq;
732 state->last_ack = ack;
733 state->last_end = end;
738 * Close the window of disabled window tracking :-)
745 if (LOG_INVALID(IPPROTO_TCP))
746 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
748 before(seq, sender->td_maxend + 1) ?
749 after(end, sender->td_end - receiver->td_maxwin - 1) ?
750 before(sack, receiver->td_end + 1) ?
751 after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
752 : "ACK is under the lower bound (possible overly delayed ACK)"
753 : "ACK is over the upper bound (ACKed data not seen yet)"
754 : "SEQ is under the lower bound (already ACKed data retransmitted)"
755 : "SEQ is over the upper bound (over the window of the receiver)");
757 res = ip_ct_tcp_be_liberal;
760 DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
761 "receiver end=%u maxend=%u maxwin=%u\n",
762 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
763 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
768 #ifdef CONFIG_IP_NF_NAT_NEEDED
769 /* Update sender->td_end after NAT successfully mangled the packet */
770 void ip_conntrack_tcp_update(struct sk_buff *skb,
771 struct ip_conntrack *conntrack,
772 enum ip_conntrack_dir dir)
774 struct iphdr *iph = skb->nh.iph;
775 struct tcphdr *tcph = (void *)skb->nh.iph + skb->nh.iph->ihl*4;
778 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
779 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
782 end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, iph, tcph);
784 write_lock_bh(&tcp_lock);
786 * We have to worry for the ack in the reply packet only...
788 if (after(end, conntrack->proto.tcp.seen[dir].td_end))
789 conntrack->proto.tcp.seen[dir].td_end = end;
790 conntrack->proto.tcp.last_end = end;
791 write_unlock_bh(&tcp_lock);
792 DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
793 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
794 sender->td_end, sender->td_maxend, sender->td_maxwin,
796 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
811 /* table of valid flag combinations - ECE and CWR are always valid */
812 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
816 [TH_SYN|TH_ACK|TH_PUSH] = 1,
819 [TH_RST|TH_ACK|TH_PUSH] = 1,
822 [TH_ACK|TH_PUSH] = 1,
824 [TH_ACK|TH_URG|TH_PUSH] = 1,
825 [TH_FIN|TH_ACK|TH_PUSH] = 1,
826 [TH_FIN|TH_ACK|TH_URG] = 1,
827 [TH_FIN|TH_ACK|TH_URG|TH_PUSH] = 1,
830 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
831 static int tcp_error(struct sk_buff *skb,
832 enum ip_conntrack_info *ctinfo,
833 unsigned int hooknum)
835 struct iphdr *iph = skb->nh.iph;
836 struct tcphdr _tcph, *th;
837 unsigned int tcplen = skb->len - iph->ihl * 4;
840 /* Smaller that minimal TCP header? */
841 th = skb_header_pointer(skb, iph->ihl * 4,
842 sizeof(_tcph), &_tcph);
844 if (LOG_INVALID(IPPROTO_TCP))
845 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
846 "ip_ct_tcp: short packet ");
850 /* Not whole TCP header or malformed packet */
851 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
852 if (LOG_INVALID(IPPROTO_TCP))
853 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
854 "ip_ct_tcp: truncated/malformed packet ");
858 /* Checksum invalid? Ignore.
859 * We skip checking packets on the outgoing path
860 * because the semantic of CHECKSUM_HW is different there
861 * and moreover root might send raw packets.
863 /* FIXME: Source route IP option packets --RR */
864 if (hooknum == NF_IP_PRE_ROUTING
865 && skb->ip_summed != CHECKSUM_UNNECESSARY
866 && csum_tcpudp_magic(iph->saddr, iph->daddr, tcplen, IPPROTO_TCP,
867 skb->ip_summed == CHECKSUM_HW ? skb->csum
868 : skb_checksum(skb, iph->ihl*4, tcplen, 0))) {
869 if (LOG_INVALID(IPPROTO_TCP))
870 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
871 "ip_ct_tcp: bad TCP checksum ");
875 /* Check TCP flags. */
876 tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
877 if (!tcp_valid_flags[tcpflags]) {
878 if (LOG_INVALID(IPPROTO_TCP))
879 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
880 "ip_ct_tcp: invalid TCP flag combination ");
887 /* Returns verdict for packet, or -1 for invalid. */
888 static int tcp_packet(struct ip_conntrack *conntrack,
889 const struct sk_buff *skb,
890 enum ip_conntrack_info ctinfo)
892 enum tcp_conntrack new_state, old_state;
893 enum ip_conntrack_dir dir;
894 struct iphdr *iph = skb->nh.iph;
895 struct tcphdr *th, _tcph;
896 unsigned long timeout;
899 th = skb_header_pointer(skb, iph->ihl * 4,
900 sizeof(_tcph), &_tcph);
903 write_lock_bh(&tcp_lock);
904 old_state = conntrack->proto.tcp.state;
905 dir = CTINFO2DIR(ctinfo);
906 index = get_conntrack_index(th);
907 new_state = tcp_conntracks[dir][index][old_state];
910 case TCP_CONNTRACK_IGNORE:
911 /* Either SYN in ORIGINAL
912 * or SYN/ACK in REPLY. */
913 if (index == TCP_SYNACK_SET
914 && conntrack->proto.tcp.last_index == TCP_SYN_SET
915 && conntrack->proto.tcp.last_dir != dir
916 && ntohl(th->ack_seq) ==
917 conntrack->proto.tcp.last_end) {
918 /* This SYN/ACK acknowledges a SYN that we earlier
919 * ignored as invalid. This means that the client and
920 * the server are both in sync, while the firewall is
921 * not. We kill this session and block the SYN/ACK so
922 * that the client cannot but retransmit its SYN and
923 * thus initiate a clean new session.
925 write_unlock_bh(&tcp_lock);
926 if (LOG_INVALID(IPPROTO_TCP))
927 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
929 "killing out of sync session ");
930 if (del_timer(&conntrack->timeout))
931 conntrack->timeout.function((unsigned long)
935 conntrack->proto.tcp.last_index = index;
936 conntrack->proto.tcp.last_dir = dir;
937 conntrack->proto.tcp.last_seq = ntohl(th->seq);
938 conntrack->proto.tcp.last_end =
939 segment_seq_plus_len(ntohl(th->seq), skb->len, iph, th);
941 write_unlock_bh(&tcp_lock);
942 if (LOG_INVALID(IPPROTO_TCP))
943 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
944 "ip_ct_tcp: invalid packet ignored ");
946 case TCP_CONNTRACK_MAX:
948 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
949 dir, get_conntrack_index(th),
951 write_unlock_bh(&tcp_lock);
952 if (LOG_INVALID(IPPROTO_TCP))
953 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
954 "ip_ct_tcp: invalid state ");
956 case TCP_CONNTRACK_SYN_SENT:
957 if (old_state < TCP_CONNTRACK_TIME_WAIT)
959 if ((conntrack->proto.tcp.seen[dir].flags &
960 IP_CT_TCP_FLAG_CLOSE_INIT)
961 || after(ntohl(th->seq),
962 conntrack->proto.tcp.seen[dir].td_end)) {
963 /* Attempt to reopen a closed connection.
964 * Delete this connection and look up again. */
965 write_unlock_bh(&tcp_lock);
966 if (del_timer(&conntrack->timeout))
967 conntrack->timeout.function((unsigned long)
971 write_unlock_bh(&tcp_lock);
972 if (LOG_INVALID(IPPROTO_TCP))
973 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
974 NULL, "ip_ct_tcp: invalid SYN");
977 case TCP_CONNTRACK_CLOSE:
978 if (index == TCP_RST_SET
979 && test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
980 && conntrack->proto.tcp.last_index == TCP_SYN_SET
981 && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
982 /* RST sent to invalid SYN we had let trough
983 * SYN was in window then, tear down connection.
984 * We skip window checking, because packet might ACK
985 * segments we ignored in the SYN. */
988 /* Just fall trough */
990 /* Keep compilers happy. */
994 if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
996 write_unlock_bh(&tcp_lock);
1000 /* From now on we have got in-window packets */
1001 conntrack->proto.tcp.last_index = index;
1003 DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
1004 "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1005 NIPQUAD(iph->saddr), ntohs(th->source),
1006 NIPQUAD(iph->daddr), ntohs(th->dest),
1007 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1008 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1009 old_state, new_state);
1011 conntrack->proto.tcp.state = new_state;
1012 if (old_state != new_state
1013 && (new_state == TCP_CONNTRACK_FIN_WAIT
1014 || new_state == TCP_CONNTRACK_CLOSE))
1015 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1016 timeout = conntrack->proto.tcp.retrans >= ip_ct_tcp_max_retrans
1017 && *tcp_timeouts[new_state] > ip_ct_tcp_timeout_max_retrans
1018 ? ip_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
1019 write_unlock_bh(&tcp_lock);
1021 ip_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
1022 if (new_state != old_state)
1023 ip_conntrack_event_cache(IPCT_PROTOINFO, skb);
1025 if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
1026 /* If only reply is a RST, we can consider ourselves not to
1027 have an established connection: this is a fairly common
1028 problem case, so we can delete the conntrack
1029 immediately. --RR */
1031 if (del_timer(&conntrack->timeout))
1032 conntrack->timeout.function((unsigned long)
1036 } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
1037 && (old_state == TCP_CONNTRACK_SYN_RECV
1038 || old_state == TCP_CONNTRACK_ESTABLISHED)
1039 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1040 /* Set ASSURED if we see see valid ack in ESTABLISHED
1041 after SYN_RECV or a valid answer for a picked up
1043 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1044 ip_conntrack_event_cache(IPCT_STATUS, skb);
1046 ip_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1051 /* Called when a new connection for this protocol found. */
1052 static int tcp_new(struct ip_conntrack *conntrack,
1053 const struct sk_buff *skb)
1055 enum tcp_conntrack new_state;
1056 struct iphdr *iph = skb->nh.iph;
1057 struct tcphdr *th, _tcph;
1059 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1060 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1063 th = skb_header_pointer(skb, iph->ihl * 4,
1064 sizeof(_tcph), &_tcph);
1067 /* Don't need lock here: this conntrack not in circulation yet */
1069 = tcp_conntracks[0][get_conntrack_index(th)]
1070 [TCP_CONNTRACK_NONE];
1072 /* Invalid: delete conntrack */
1073 if (new_state >= TCP_CONNTRACK_MAX) {
1074 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1078 if (new_state == TCP_CONNTRACK_SYN_SENT) {
1080 conntrack->proto.tcp.seen[0].td_end =
1081 segment_seq_plus_len(ntohl(th->seq), skb->len,
1083 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1084 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1085 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1086 conntrack->proto.tcp.seen[0].td_maxend =
1087 conntrack->proto.tcp.seen[0].td_end;
1089 tcp_options(skb, iph, th, &conntrack->proto.tcp.seen[0]);
1090 conntrack->proto.tcp.seen[1].flags = 0;
1091 conntrack->proto.tcp.seen[0].loose =
1092 conntrack->proto.tcp.seen[1].loose = 0;
1093 } else if (ip_ct_tcp_loose == 0) {
1094 /* Don't try to pick up connections. */
1098 * We are in the middle of a connection,
1099 * its history is lost for us.
1100 * Let's try to use the data from the packet.
1102 conntrack->proto.tcp.seen[0].td_end =
1103 segment_seq_plus_len(ntohl(th->seq), skb->len,
1105 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1106 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1107 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1108 conntrack->proto.tcp.seen[0].td_maxend =
1109 conntrack->proto.tcp.seen[0].td_end +
1110 conntrack->proto.tcp.seen[0].td_maxwin;
1111 conntrack->proto.tcp.seen[0].td_scale = 0;
1113 /* We assume SACK. Should we assume window scaling too? */
1114 conntrack->proto.tcp.seen[0].flags =
1115 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1116 conntrack->proto.tcp.seen[0].loose =
1117 conntrack->proto.tcp.seen[1].loose = ip_ct_tcp_loose;
1120 conntrack->proto.tcp.seen[1].td_end = 0;
1121 conntrack->proto.tcp.seen[1].td_maxend = 0;
1122 conntrack->proto.tcp.seen[1].td_maxwin = 1;
1123 conntrack->proto.tcp.seen[1].td_scale = 0;
1125 /* tcp_packet will set them */
1126 conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1127 conntrack->proto.tcp.last_index = TCP_NONE_SET;
1129 DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1130 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1131 sender->td_end, sender->td_maxend, sender->td_maxwin,
1133 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1134 receiver->td_scale);
1138 struct ip_conntrack_protocol ip_conntrack_protocol_tcp =
1140 .proto = IPPROTO_TCP,
1142 .pkt_to_tuple = tcp_pkt_to_tuple,
1143 .invert_tuple = tcp_invert_tuple,
1144 .print_tuple = tcp_print_tuple,
1145 .print_conntrack = tcp_print_conntrack,
1146 .packet = tcp_packet,
1149 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
1150 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
1151 .to_nfattr = tcp_to_nfattr,
1152 .from_nfattr = nfattr_to_tcp,
1153 .tuple_to_nfattr = ip_ct_port_tuple_to_nfattr,
1154 .nfattr_to_tuple = ip_ct_port_nfattr_to_tuple,