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/types.h>
23 #include <linux/sched.h>
24 #include <linux/timer.h>
25 #include <linux/netfilter.h>
26 #include <linux/module.h>
29 #include <linux/tcp.h>
30 #include <linux/spinlock.h>
34 #include <linux/netfilter_ipv4.h>
35 #include <linux/netfilter_ipv4/ip_conntrack.h>
36 #include <linux/netfilter_ipv4/ip_conntrack_protocol.h>
42 #define DEBUGP(format, args...)
45 /* Protects conntrack->proto.tcp */
46 static DEFINE_RWLOCK(tcp_lock);
48 /* "Be conservative in what you do,
49 be liberal in what you accept from others."
50 If it's non-zero, we mark only out of window RST segments as INVALID. */
51 int ip_ct_tcp_be_liberal = 0;
53 /* When connection is picked up from the middle, how many packets are required
54 to pass in each direction when we assume we are in sync - if any side uses
55 window scaling, we lost the game.
56 If it is set to zero, we disable picking up already established
58 int ip_ct_tcp_loose = 3;
60 /* Max number of the retransmitted packets without receiving an (acceptable)
61 ACK from the destination. If this number is reached, a shorter timer
63 int ip_ct_tcp_max_retrans = 3;
65 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
66 closely. They're more complex. --RR */
68 static const char *tcp_conntrack_names[] = {
82 #define MINS * 60 SECS
83 #define HOURS * 60 MINS
84 #define DAYS * 24 HOURS
86 unsigned int ip_ct_tcp_timeout_syn_sent = 2 MINS;
87 unsigned int ip_ct_tcp_timeout_syn_recv = 60 SECS;
88 unsigned int ip_ct_tcp_timeout_established = 5 DAYS;
89 unsigned int ip_ct_tcp_timeout_fin_wait = 2 MINS;
90 unsigned int ip_ct_tcp_timeout_close_wait = 60 SECS;
91 unsigned int ip_ct_tcp_timeout_last_ack = 30 SECS;
92 unsigned int ip_ct_tcp_timeout_time_wait = 2 MINS;
93 unsigned int ip_ct_tcp_timeout_close = 10 SECS;
95 /* RFC1122 says the R2 limit should be at least 100 seconds.
96 Linux uses 15 packets as limit, which corresponds
97 to ~13-30min depending on RTO. */
98 unsigned int ip_ct_tcp_timeout_max_retrans = 5 MINS;
100 static const unsigned int * tcp_timeouts[]
101 = { NULL, /* TCP_CONNTRACK_NONE */
102 &ip_ct_tcp_timeout_syn_sent, /* TCP_CONNTRACK_SYN_SENT, */
103 &ip_ct_tcp_timeout_syn_recv, /* TCP_CONNTRACK_SYN_RECV, */
104 &ip_ct_tcp_timeout_established, /* TCP_CONNTRACK_ESTABLISHED, */
105 &ip_ct_tcp_timeout_fin_wait, /* TCP_CONNTRACK_FIN_WAIT, */
106 &ip_ct_tcp_timeout_close_wait, /* TCP_CONNTRACK_CLOSE_WAIT, */
107 &ip_ct_tcp_timeout_last_ack, /* TCP_CONNTRACK_LAST_ACK, */
108 &ip_ct_tcp_timeout_time_wait, /* TCP_CONNTRACK_TIME_WAIT, */
109 &ip_ct_tcp_timeout_close, /* TCP_CONNTRACK_CLOSE, */
110 NULL, /* TCP_CONNTRACK_LISTEN */
113 #define sNO TCP_CONNTRACK_NONE
114 #define sSS TCP_CONNTRACK_SYN_SENT
115 #define sSR TCP_CONNTRACK_SYN_RECV
116 #define sES TCP_CONNTRACK_ESTABLISHED
117 #define sFW TCP_CONNTRACK_FIN_WAIT
118 #define sCW TCP_CONNTRACK_CLOSE_WAIT
119 #define sLA TCP_CONNTRACK_LAST_ACK
120 #define sTW TCP_CONNTRACK_TIME_WAIT
121 #define sCL TCP_CONNTRACK_CLOSE
122 #define sLI TCP_CONNTRACK_LISTEN
123 #define sIV TCP_CONNTRACK_MAX
124 #define sIG TCP_CONNTRACK_IGNORE
126 /* What TCP flags are set from RST/SYN/FIN/ACK. */
137 * The TCP state transition table needs a few words...
139 * We are the man in the middle. All the packets go through us
140 * but might get lost in transit to the destination.
141 * It is assumed that the destinations can't receive segments
144 * The checked segment is in window, but our windows are *not*
145 * equivalent with the ones of the sender/receiver. We always
146 * try to guess the state of the current sender.
148 * The meaning of the states are:
150 * NONE: initial state
151 * SYN_SENT: SYN-only packet seen
152 * SYN_RECV: SYN-ACK packet seen
153 * ESTABLISHED: ACK packet seen
154 * FIN_WAIT: FIN packet seen
155 * CLOSE_WAIT: ACK seen (after FIN)
156 * LAST_ACK: FIN seen (after FIN)
157 * TIME_WAIT: last ACK seen
158 * CLOSE: closed connection
160 * LISTEN state is not used.
162 * Packets marked as IGNORED (sIG):
163 * if they may be either invalid or valid
164 * and the receiver may send back a connection
165 * closing RST or a SYN/ACK.
167 * Packets marked as INVALID (sIV):
168 * if they are invalid
169 * or we do not support the request (simultaneous open)
171 static const enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
174 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
175 /*syn*/ { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
177 * sNO -> sSS Initialize a new connection
178 * sSS -> sSS Retransmitted SYN
179 * sSR -> sIG Late retransmitted SYN?
180 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
181 * are errors. Receiver will reply with RST
182 * and close the connection.
183 * Or we are not in sync and hold a dead connection.
187 * sTW -> sSS Reopened connection (RFC 1122).
190 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
191 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
193 * A SYN/ACK from the client is always invalid:
194 * - either it tries to set up a simultaneous open, which is
196 * - or the firewall has just been inserted between the two hosts
197 * during the session set-up. The SYN will be retransmitted
198 * by the true client (or it'll time out).
200 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
201 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
203 * sNO -> sIV Too late and no reason to do anything...
204 * sSS -> sIV Client migth not send FIN in this state:
205 * we enforce waiting for a SYN/ACK reply first.
206 * sSR -> sFW Close started.
208 * sFW -> sLA FIN seen in both directions, waiting for
210 * Migth be a retransmitted FIN as well...
212 * sLA -> sLA Retransmitted FIN. Remain in the same state.
216 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
217 /*ack*/ { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
219 * sNO -> sES Assumed.
220 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
221 * sSR -> sES Established state is reached.
223 * sFW -> sCW Normal close request answered by ACK.
225 * sLA -> sTW Last ACK detected.
226 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
229 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
230 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
231 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
235 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
236 /*syn*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
238 * sNO -> sIV Never reached.
239 * sSS -> sIV Simultaneous open, not supported
240 * sSR -> sIV Simultaneous open, not supported.
241 * sES -> sIV Server may not initiate a connection.
245 * sTW -> sIV Reopened connection, but server may not do it.
248 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
249 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
251 * sSS -> sSR Standard open.
252 * sSR -> sSR Retransmitted SYN/ACK.
253 * sES -> sIG Late retransmitted SYN/ACK?
254 * sFW -> sIG Might be SYN/ACK answering ignored SYN
260 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
261 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
263 * sSS -> sIV Server might not send FIN in this state.
264 * sSR -> sFW Close started.
266 * sFW -> sLA FIN seen in both directions.
268 * sLA -> sLA Retransmitted FIN.
272 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
273 /*ack*/ { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
275 * sSS -> sIG Might be a half-open connection.
276 * sSR -> sSR Might answer late resent SYN.
278 * sFW -> sCW Normal close request answered by ACK.
280 * sLA -> sTW Last ACK detected.
281 * sTW -> sTW Retransmitted last ACK.
284 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
285 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
286 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
290 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
291 unsigned int dataoff,
292 struct ip_conntrack_tuple *tuple)
294 struct tcphdr _hdr, *hp;
296 /* Actually only need first 8 bytes. */
297 hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
301 tuple->src.u.tcp.port = hp->source;
302 tuple->dst.u.tcp.port = hp->dest;
307 static int tcp_invert_tuple(struct ip_conntrack_tuple *tuple,
308 const struct ip_conntrack_tuple *orig)
310 tuple->src.u.tcp.port = orig->dst.u.tcp.port;
311 tuple->dst.u.tcp.port = orig->src.u.tcp.port;
315 /* Print out the per-protocol part of the tuple. */
316 static int tcp_print_tuple(struct seq_file *s,
317 const struct ip_conntrack_tuple *tuple)
319 return seq_printf(s, "sport=%hu dport=%hu ",
320 ntohs(tuple->src.u.tcp.port),
321 ntohs(tuple->dst.u.tcp.port));
324 /* Print out the private part of the conntrack. */
325 static int tcp_print_conntrack(struct seq_file *s,
326 const struct ip_conntrack *conntrack)
328 enum tcp_conntrack state;
330 read_lock_bh(&tcp_lock);
331 state = conntrack->proto.tcp.state;
332 read_unlock_bh(&tcp_lock);
334 return seq_printf(s, "%s ", tcp_conntrack_names[state]);
337 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
338 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
339 static int tcp_to_nfattr(struct sk_buff *skb, struct nfattr *nfa,
340 const struct ip_conntrack *ct)
342 struct nfattr *nest_parms;
344 read_lock_bh(&tcp_lock);
345 nest_parms = NFA_NEST(skb, CTA_PROTOINFO_TCP);
346 NFA_PUT(skb, CTA_PROTOINFO_TCP_STATE, sizeof(u_int8_t),
347 &ct->proto.tcp.state);
348 read_unlock_bh(&tcp_lock);
350 NFA_NEST_END(skb, nest_parms);
355 read_unlock_bh(&tcp_lock);
359 static const size_t cta_min_tcp[CTA_PROTOINFO_TCP_MAX] = {
360 [CTA_PROTOINFO_TCP_STATE-1] = sizeof(u_int8_t),
363 static int nfattr_to_tcp(struct nfattr *cda[], struct ip_conntrack *ct)
365 struct nfattr *attr = cda[CTA_PROTOINFO_TCP-1];
366 struct nfattr *tb[CTA_PROTOINFO_TCP_MAX];
368 /* updates could not contain anything about the private
369 * protocol info, in that case skip the parsing */
373 nfattr_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, attr);
375 if (nfattr_bad_size(tb, CTA_PROTOINFO_TCP_MAX, cta_min_tcp))
378 if (!tb[CTA_PROTOINFO_TCP_STATE-1])
381 write_lock_bh(&tcp_lock);
382 ct->proto.tcp.state =
383 *(u_int8_t *)NFA_DATA(tb[CTA_PROTOINFO_TCP_STATE-1]);
384 write_unlock_bh(&tcp_lock);
390 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
392 if (tcph->rst) return TCP_RST_SET;
393 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
394 else if (tcph->fin) return TCP_FIN_SET;
395 else if (tcph->ack) return TCP_ACK_SET;
396 else return TCP_NONE_SET;
399 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
400 in IP Filter' by Guido van Rooij.
402 http://www.nluug.nl/events/sane2000/papers.html
403 http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
405 The boundaries and the conditions are changed according to RFC793:
406 the packet must intersect the window (i.e. segments may be
407 after the right or before the left edge) and thus receivers may ACK
408 segments after the right edge of the window.
410 td_maxend = max(sack + max(win,1)) seen in reply packets
411 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
412 td_maxwin += seq + len - sender.td_maxend
413 if seq + len > sender.td_maxend
414 td_end = max(seq + len) seen in sent packets
416 I. Upper bound for valid data: seq <= sender.td_maxend
417 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
418 III. Upper bound for valid ack: sack <= receiver.td_end
419 IV. Lower bound for valid ack: ack >= receiver.td_end - MAXACKWINDOW
421 where sack is the highest right edge of sack block found in the packet.
423 The upper bound limit for a valid ack is not ignored -
424 we doesn't have to deal with fragments.
427 static inline __u32 segment_seq_plus_len(__u32 seq,
432 return (seq + len - (iph->ihl + tcph->doff)*4
433 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
436 /* Fixme: what about big packets? */
437 #define MAXACKWINCONST 66000
438 #define MAXACKWINDOW(sender) \
439 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
443 * Simplified tcp_parse_options routine from tcp_input.c
445 static void tcp_options(const struct sk_buff *skb,
448 struct ip_ct_tcp_state *state)
450 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
452 int length = (tcph->doff*4) - sizeof(struct tcphdr);
457 ptr = skb_header_pointer(skb,
458 (iph->ihl * 4) + sizeof(struct tcphdr),
472 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
477 if (opsize < 2) /* "silly options" */
480 break; /* don't parse partial options */
482 if (opcode == TCPOPT_SACK_PERM
483 && opsize == TCPOLEN_SACK_PERM)
484 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
485 else if (opcode == TCPOPT_WINDOW
486 && opsize == TCPOLEN_WINDOW) {
487 state->td_scale = *(u_int8_t *)ptr;
489 if (state->td_scale > 14) {
491 state->td_scale = 14;
494 IP_CT_TCP_FLAG_WINDOW_SCALE;
502 static void tcp_sack(const struct sk_buff *skb,
507 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
509 int length = (tcph->doff*4) - sizeof(struct tcphdr);
515 ptr = skb_header_pointer(skb,
516 (iph->ihl * 4) + sizeof(struct tcphdr),
520 /* Fast path for timestamp-only option */
521 if (length == TCPOLEN_TSTAMP_ALIGNED*4
523 __constant_ntohl((TCPOPT_NOP << 24)
525 | (TCPOPT_TIMESTAMP << 8)
526 | TCPOLEN_TIMESTAMP))
536 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
541 if (opsize < 2) /* "silly options" */
544 break; /* don't parse partial options */
546 if (opcode == TCPOPT_SACK
547 && opsize >= (TCPOLEN_SACK_BASE
548 + TCPOLEN_SACK_PERBLOCK)
549 && !((opsize - TCPOLEN_SACK_BASE)
550 % TCPOLEN_SACK_PERBLOCK)) {
552 i < (opsize - TCPOLEN_SACK_BASE);
553 i += TCPOLEN_SACK_PERBLOCK) {
554 tmp = ntohl(*((u_int32_t *)(ptr+i)+1));
556 if (after(tmp, *sack))
567 static int tcp_in_window(struct ip_ct_tcp *state,
568 enum ip_conntrack_dir dir,
570 const struct sk_buff *skb,
574 struct ip_ct_tcp_state *sender = &state->seen[dir];
575 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
576 __u32 seq, ack, sack, end, win, swin;
580 * Get the required data from the packet.
582 seq = ntohl(tcph->seq);
583 ack = sack = ntohl(tcph->ack_seq);
584 win = ntohs(tcph->window);
585 end = segment_seq_plus_len(seq, skb->len, iph, tcph);
587 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
588 tcp_sack(skb, iph, tcph, &sack);
590 DEBUGP("tcp_in_window: START\n");
591 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
592 "seq=%u ack=%u sack=%u win=%u end=%u\n",
593 NIPQUAD(iph->saddr), ntohs(tcph->source),
594 NIPQUAD(iph->daddr), ntohs(tcph->dest),
595 seq, ack, sack, win, end);
596 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
597 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
598 sender->td_end, sender->td_maxend, sender->td_maxwin,
600 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
603 if (sender->td_end == 0) {
605 * Initialize sender data.
607 if (tcph->syn && tcph->ack) {
609 * Outgoing SYN-ACK in reply to a SYN.
612 sender->td_maxend = end;
613 sender->td_maxwin = (win == 0 ? 1 : win);
615 tcp_options(skb, iph, tcph, sender);
618 * Both sides must send the Window Scale option
619 * to enable window scaling in either direction.
621 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
622 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
624 receiver->td_scale = 0;
627 * We are in the middle of a connection,
628 * its history is lost for us.
629 * Let's try to use the data from the packet.
631 sender->td_end = end;
632 sender->td_maxwin = (win == 0 ? 1 : win);
633 sender->td_maxend = end + sender->td_maxwin;
635 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
636 && dir == IP_CT_DIR_ORIGINAL)
637 || (state->state == TCP_CONNTRACK_SYN_RECV
638 && dir == IP_CT_DIR_REPLY))
639 && after(end, sender->td_end)) {
641 * RFC 793: "if a TCP is reinitialized ... then it need
642 * not wait at all; it must only be sure to use sequence
643 * numbers larger than those recently used."
646 sender->td_maxend = end;
647 sender->td_maxwin = (win == 0 ? 1 : win);
649 tcp_options(skb, iph, tcph, sender);
654 * If there is no ACK, just pretend it was set and OK.
656 ack = sack = receiver->td_end;
657 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
658 (TCP_FLAG_ACK|TCP_FLAG_RST))
661 * Broken TCP stacks, that set ACK in RST packets as well
662 * with zero ack value.
664 ack = sack = receiver->td_end;
669 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
671 * Packets contains no data: we assume it is valid
672 * and check the ack value only.
673 * However RST segments are always validated by their
674 * SEQ number, except when seq == 0 (reset sent answering
677 seq = end = sender->td_end;
679 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
680 "seq=%u ack=%u sack =%u win=%u end=%u\n",
681 NIPQUAD(iph->saddr), ntohs(tcph->source),
682 NIPQUAD(iph->daddr), ntohs(tcph->dest),
683 seq, ack, sack, win, end);
684 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
685 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
686 sender->td_end, sender->td_maxend, sender->td_maxwin,
688 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
691 DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
692 before(seq, sender->td_maxend + 1),
693 after(end, sender->td_end - receiver->td_maxwin - 1),
694 before(sack, receiver->td_end + 1),
695 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
697 if (sender->loose || receiver->loose ||
698 (before(seq, sender->td_maxend + 1) &&
699 after(end, sender->td_end - receiver->td_maxwin - 1) &&
700 before(sack, receiver->td_end + 1) &&
701 after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
703 * Take into account window scaling (RFC 1323).
706 win <<= sender->td_scale;
709 * Update sender data.
711 swin = win + (sack - ack);
712 if (sender->td_maxwin < swin)
713 sender->td_maxwin = swin;
714 if (after(end, sender->td_end))
715 sender->td_end = end;
717 * Update receiver data.
719 if (after(end, sender->td_maxend))
720 receiver->td_maxwin += end - sender->td_maxend;
721 if (after(sack + win, receiver->td_maxend - 1)) {
722 receiver->td_maxend = sack + win;
724 receiver->td_maxend++;
728 * Check retransmissions.
730 if (index == TCP_ACK_SET) {
731 if (state->last_dir == dir
732 && state->last_seq == seq
733 && state->last_ack == ack
734 && state->last_end == end)
737 state->last_dir = dir;
738 state->last_seq = seq;
739 state->last_ack = ack;
740 state->last_end = end;
745 * Close the window of disabled window tracking :-)
752 if (LOG_INVALID(IPPROTO_TCP))
753 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
755 before(seq, sender->td_maxend + 1) ?
756 after(end, sender->td_end - receiver->td_maxwin - 1) ?
757 before(sack, receiver->td_end + 1) ?
758 after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
759 : "ACK is under the lower bound (possible overly delayed ACK)"
760 : "ACK is over the upper bound (ACKed data not seen yet)"
761 : "SEQ is under the lower bound (already ACKed data retransmitted)"
762 : "SEQ is over the upper bound (over the window of the receiver)");
764 res = ip_ct_tcp_be_liberal;
767 DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
768 "receiver end=%u maxend=%u maxwin=%u\n",
769 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
770 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
775 #ifdef CONFIG_IP_NF_NAT_NEEDED
776 /* Update sender->td_end after NAT successfully mangled the packet */
777 void ip_conntrack_tcp_update(struct sk_buff *skb,
778 struct ip_conntrack *conntrack,
779 enum ip_conntrack_dir dir)
781 struct iphdr *iph = skb->nh.iph;
782 struct tcphdr *tcph = (void *)skb->nh.iph + skb->nh.iph->ihl*4;
785 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
786 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
789 end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, iph, tcph);
791 write_lock_bh(&tcp_lock);
793 * We have to worry for the ack in the reply packet only...
795 if (after(end, conntrack->proto.tcp.seen[dir].td_end))
796 conntrack->proto.tcp.seen[dir].td_end = end;
797 conntrack->proto.tcp.last_end = end;
798 write_unlock_bh(&tcp_lock);
799 DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
800 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
801 sender->td_end, sender->td_maxend, sender->td_maxwin,
803 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
818 /* table of valid flag combinations - ECE and CWR are always valid */
819 static const u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
823 [TH_SYN|TH_PUSH] = 1,
824 [TH_SYN|TH_ACK|TH_PUSH] = 1,
827 [TH_RST|TH_ACK|TH_PUSH] = 1,
830 [TH_ACK|TH_PUSH] = 1,
832 [TH_ACK|TH_URG|TH_PUSH] = 1,
833 [TH_FIN|TH_ACK|TH_PUSH] = 1,
834 [TH_FIN|TH_ACK|TH_URG] = 1,
835 [TH_FIN|TH_ACK|TH_URG|TH_PUSH] = 1,
838 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
839 static int tcp_error(struct sk_buff *skb,
840 enum ip_conntrack_info *ctinfo,
841 unsigned int hooknum)
843 struct iphdr *iph = skb->nh.iph;
844 struct tcphdr _tcph, *th;
845 unsigned int tcplen = skb->len - iph->ihl * 4;
848 /* Smaller that minimal TCP header? */
849 th = skb_header_pointer(skb, iph->ihl * 4,
850 sizeof(_tcph), &_tcph);
852 if (LOG_INVALID(IPPROTO_TCP))
853 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
854 "ip_ct_tcp: short packet ");
858 /* Not whole TCP header or malformed packet */
859 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
860 if (LOG_INVALID(IPPROTO_TCP))
861 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
862 "ip_ct_tcp: truncated/malformed packet ");
866 /* Checksum invalid? Ignore.
867 * We skip checking packets on the outgoing path
868 * because the semantic of CHECKSUM_HW is different there
869 * and moreover root might send raw packets.
871 /* FIXME: Source route IP option packets --RR */
872 if (ip_conntrack_checksum && hooknum == NF_IP_PRE_ROUTING &&
873 nf_ip_checksum(skb, hooknum, iph->ihl * 4, IPPROTO_TCP)) {
874 if (LOG_INVALID(IPPROTO_TCP))
875 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
876 "ip_ct_tcp: bad TCP checksum ");
880 /* Check TCP flags. */
881 tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
882 if (!tcp_valid_flags[tcpflags]) {
883 if (LOG_INVALID(IPPROTO_TCP))
884 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
885 "ip_ct_tcp: invalid TCP flag combination ");
892 /* Returns verdict for packet, or -1 for invalid. */
893 static int tcp_packet(struct ip_conntrack *conntrack,
894 const struct sk_buff *skb,
895 enum ip_conntrack_info ctinfo)
897 enum tcp_conntrack new_state, old_state;
898 enum ip_conntrack_dir dir;
899 struct iphdr *iph = skb->nh.iph;
900 struct tcphdr *th, _tcph;
901 unsigned long timeout;
904 th = skb_header_pointer(skb, iph->ihl * 4,
905 sizeof(_tcph), &_tcph);
908 write_lock_bh(&tcp_lock);
909 old_state = conntrack->proto.tcp.state;
910 dir = CTINFO2DIR(ctinfo);
911 index = get_conntrack_index(th);
912 new_state = tcp_conntracks[dir][index][old_state];
915 case TCP_CONNTRACK_IGNORE:
919 * b) SYN/ACK in REPLY
920 * c) ACK in reply direction after initial SYN in original.
922 if (index == TCP_SYNACK_SET
923 && conntrack->proto.tcp.last_index == TCP_SYN_SET
924 && conntrack->proto.tcp.last_dir != dir
925 && ntohl(th->ack_seq) ==
926 conntrack->proto.tcp.last_end) {
927 /* This SYN/ACK acknowledges a SYN that we earlier
928 * ignored as invalid. This means that the client and
929 * the server are both in sync, while the firewall is
930 * not. We kill this session and block the SYN/ACK so
931 * that the client cannot but retransmit its SYN and
932 * thus initiate a clean new session.
934 write_unlock_bh(&tcp_lock);
935 if (LOG_INVALID(IPPROTO_TCP))
936 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
938 "killing out of sync session ");
939 if (del_timer(&conntrack->timeout))
940 conntrack->timeout.function((unsigned long)
944 conntrack->proto.tcp.last_index = index;
945 conntrack->proto.tcp.last_dir = dir;
946 conntrack->proto.tcp.last_seq = ntohl(th->seq);
947 conntrack->proto.tcp.last_end =
948 segment_seq_plus_len(ntohl(th->seq), skb->len, iph, th);
950 write_unlock_bh(&tcp_lock);
951 if (LOG_INVALID(IPPROTO_TCP))
952 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
953 "ip_ct_tcp: invalid packet ignored ");
955 case TCP_CONNTRACK_MAX:
957 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
958 dir, get_conntrack_index(th),
960 write_unlock_bh(&tcp_lock);
961 if (LOG_INVALID(IPPROTO_TCP))
962 nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
963 "ip_ct_tcp: invalid state ");
965 case TCP_CONNTRACK_SYN_SENT:
966 if (old_state < TCP_CONNTRACK_TIME_WAIT)
968 if ((conntrack->proto.tcp.seen[dir].flags &
969 IP_CT_TCP_FLAG_CLOSE_INIT)
970 || after(ntohl(th->seq),
971 conntrack->proto.tcp.seen[dir].td_end)) {
972 /* Attempt to reopen a closed connection.
973 * Delete this connection and look up again. */
974 write_unlock_bh(&tcp_lock);
975 if (del_timer(&conntrack->timeout))
976 conntrack->timeout.function((unsigned long)
980 write_unlock_bh(&tcp_lock);
981 if (LOG_INVALID(IPPROTO_TCP))
982 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
983 NULL, "ip_ct_tcp: invalid SYN");
986 case TCP_CONNTRACK_CLOSE:
987 if (index == TCP_RST_SET
988 && ((test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
989 && conntrack->proto.tcp.last_index == TCP_SYN_SET)
990 || (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
991 && conntrack->proto.tcp.last_index == TCP_ACK_SET))
992 && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
993 /* RST sent to invalid SYN or ACK we had let through
994 * at a) and c) above:
996 * a) SYN was in window then
997 * c) we hold a half-open connection.
999 * Delete our connection entry.
1000 * We skip window checking, because packet might ACK
1001 * segments we ignored. */
1004 /* Just fall through */
1006 /* Keep compilers happy. */
1010 if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
1012 write_unlock_bh(&tcp_lock);
1016 /* From now on we have got in-window packets */
1017 conntrack->proto.tcp.last_index = index;
1019 DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
1020 "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1021 NIPQUAD(iph->saddr), ntohs(th->source),
1022 NIPQUAD(iph->daddr), ntohs(th->dest),
1023 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1024 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1025 old_state, new_state);
1027 conntrack->proto.tcp.state = new_state;
1028 if (old_state != new_state
1029 && (new_state == TCP_CONNTRACK_FIN_WAIT
1030 || new_state == TCP_CONNTRACK_CLOSE))
1031 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1032 timeout = conntrack->proto.tcp.retrans >= ip_ct_tcp_max_retrans
1033 && *tcp_timeouts[new_state] > ip_ct_tcp_timeout_max_retrans
1034 ? ip_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
1035 write_unlock_bh(&tcp_lock);
1037 ip_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
1038 if (new_state != old_state)
1039 ip_conntrack_event_cache(IPCT_PROTOINFO, skb);
1041 if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
1042 /* If only reply is a RST, we can consider ourselves not to
1043 have an established connection: this is a fairly common
1044 problem case, so we can delete the conntrack
1045 immediately. --RR */
1047 if (del_timer(&conntrack->timeout))
1048 conntrack->timeout.function((unsigned long)
1052 } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
1053 && (old_state == TCP_CONNTRACK_SYN_RECV
1054 || old_state == TCP_CONNTRACK_ESTABLISHED)
1055 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1056 /* Set ASSURED if we see see valid ack in ESTABLISHED
1057 after SYN_RECV or a valid answer for a picked up
1059 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1060 ip_conntrack_event_cache(IPCT_STATUS, skb);
1062 ip_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1067 /* Called when a new connection for this protocol found. */
1068 static int tcp_new(struct ip_conntrack *conntrack,
1069 const struct sk_buff *skb)
1071 enum tcp_conntrack new_state;
1072 struct iphdr *iph = skb->nh.iph;
1073 struct tcphdr *th, _tcph;
1075 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1076 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1079 th = skb_header_pointer(skb, iph->ihl * 4,
1080 sizeof(_tcph), &_tcph);
1083 /* Don't need lock here: this conntrack not in circulation yet */
1085 = tcp_conntracks[0][get_conntrack_index(th)]
1086 [TCP_CONNTRACK_NONE];
1088 /* Invalid: delete conntrack */
1089 if (new_state >= TCP_CONNTRACK_MAX) {
1090 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1094 if (new_state == TCP_CONNTRACK_SYN_SENT) {
1096 conntrack->proto.tcp.seen[0].td_end =
1097 segment_seq_plus_len(ntohl(th->seq), skb->len,
1099 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1100 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1101 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1102 conntrack->proto.tcp.seen[0].td_maxend =
1103 conntrack->proto.tcp.seen[0].td_end;
1105 tcp_options(skb, iph, th, &conntrack->proto.tcp.seen[0]);
1106 conntrack->proto.tcp.seen[1].flags = 0;
1107 conntrack->proto.tcp.seen[0].loose =
1108 conntrack->proto.tcp.seen[1].loose = 0;
1109 } else if (ip_ct_tcp_loose == 0) {
1110 /* Don't try to pick up connections. */
1114 * We are in the middle of a connection,
1115 * its history is lost for us.
1116 * Let's try to use the data from the packet.
1118 conntrack->proto.tcp.seen[0].td_end =
1119 segment_seq_plus_len(ntohl(th->seq), skb->len,
1121 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1122 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1123 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1124 conntrack->proto.tcp.seen[0].td_maxend =
1125 conntrack->proto.tcp.seen[0].td_end +
1126 conntrack->proto.tcp.seen[0].td_maxwin;
1127 conntrack->proto.tcp.seen[0].td_scale = 0;
1129 /* We assume SACK. Should we assume window scaling too? */
1130 conntrack->proto.tcp.seen[0].flags =
1131 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1132 conntrack->proto.tcp.seen[0].loose =
1133 conntrack->proto.tcp.seen[1].loose = ip_ct_tcp_loose;
1136 conntrack->proto.tcp.seen[1].td_end = 0;
1137 conntrack->proto.tcp.seen[1].td_maxend = 0;
1138 conntrack->proto.tcp.seen[1].td_maxwin = 1;
1139 conntrack->proto.tcp.seen[1].td_scale = 0;
1141 /* tcp_packet will set them */
1142 conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1143 conntrack->proto.tcp.last_index = TCP_NONE_SET;
1145 DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1146 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1147 sender->td_end, sender->td_maxend, sender->td_maxwin,
1149 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1150 receiver->td_scale);
1154 struct ip_conntrack_protocol ip_conntrack_protocol_tcp =
1156 .proto = IPPROTO_TCP,
1158 .pkt_to_tuple = tcp_pkt_to_tuple,
1159 .invert_tuple = tcp_invert_tuple,
1160 .print_tuple = tcp_print_tuple,
1161 .print_conntrack = tcp_print_conntrack,
1162 .packet = tcp_packet,
1165 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
1166 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
1167 .to_nfattr = tcp_to_nfattr,
1168 .from_nfattr = nfattr_to_tcp,
1169 .tuple_to_nfattr = ip_ct_port_tuple_to_nfattr,
1170 .nfattr_to_tuple = ip_ct_port_nfattr_to_tuple,