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
19 * 27 Oct 2004: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
20 * - genelized Layer 3 protocol part.
22 * Derived from net/ipv4/netfilter/ip_conntrack_proto_tcp.c
27 #include <linux/types.h>
28 #include <linux/sched.h>
29 #include <linux/timer.h>
30 #include <linux/netfilter.h>
31 #include <linux/module.h>
33 #include <linux/tcp.h>
34 #include <linux/spinlock.h>
35 #include <linux/skbuff.h>
36 #include <linux/ipv6.h>
37 #include <net/ip6_checksum.h>
41 #include <linux/netfilter.h>
42 #include <linux/netfilter_ipv4.h>
43 #include <linux/netfilter_ipv6.h>
44 #include <net/netfilter/nf_conntrack.h>
45 #include <net/netfilter/nf_conntrack_protocol.h>
51 #define DEBUGP(format, args...)
54 /* Protects conntrack->proto.tcp */
55 static DEFINE_RWLOCK(tcp_lock);
57 /* "Be conservative in what you do,
58 be liberal in what you accept from others."
59 If it's non-zero, we mark only out of window RST segments as INVALID. */
60 int nf_ct_tcp_be_liberal = 0;
62 /* When connection is picked up from the middle, how many packets are required
63 to pass in each direction when we assume we are in sync - if any side uses
64 window scaling, we lost the game.
65 If it is set to zero, we disable picking up already established
67 int nf_ct_tcp_loose = 3;
69 /* Max number of the retransmitted packets without receiving an (acceptable)
70 ACK from the destination. If this number is reached, a shorter timer
72 int nf_ct_tcp_max_retrans = 3;
74 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
75 closely. They're more complex. --RR */
77 static const char *tcp_conntrack_names[] = {
91 #define MINS * 60 SECS
92 #define HOURS * 60 MINS
93 #define DAYS * 24 HOURS
95 unsigned int nf_ct_tcp_timeout_syn_sent = 2 MINS;
96 unsigned int nf_ct_tcp_timeout_syn_recv = 60 SECS;
97 unsigned int nf_ct_tcp_timeout_established = 5 DAYS;
98 unsigned int nf_ct_tcp_timeout_fin_wait = 2 MINS;
99 unsigned int nf_ct_tcp_timeout_close_wait = 60 SECS;
100 unsigned int nf_ct_tcp_timeout_last_ack = 30 SECS;
101 unsigned int nf_ct_tcp_timeout_time_wait = 2 MINS;
102 unsigned int nf_ct_tcp_timeout_close = 10 SECS;
104 /* RFC1122 says the R2 limit should be at least 100 seconds.
105 Linux uses 15 packets as limit, which corresponds
106 to ~13-30min depending on RTO. */
107 unsigned int nf_ct_tcp_timeout_max_retrans = 5 MINS;
109 static unsigned int * tcp_timeouts[]
110 = { NULL, /* TCP_CONNTRACK_NONE */
111 &nf_ct_tcp_timeout_syn_sent, /* TCP_CONNTRACK_SYN_SENT, */
112 &nf_ct_tcp_timeout_syn_recv, /* TCP_CONNTRACK_SYN_RECV, */
113 &nf_ct_tcp_timeout_established, /* TCP_CONNTRACK_ESTABLISHED, */
114 &nf_ct_tcp_timeout_fin_wait, /* TCP_CONNTRACK_FIN_WAIT, */
115 &nf_ct_tcp_timeout_close_wait, /* TCP_CONNTRACK_CLOSE_WAIT, */
116 &nf_ct_tcp_timeout_last_ack, /* TCP_CONNTRACK_LAST_ACK, */
117 &nf_ct_tcp_timeout_time_wait, /* TCP_CONNTRACK_TIME_WAIT, */
118 &nf_ct_tcp_timeout_close, /* TCP_CONNTRACK_CLOSE, */
119 NULL, /* TCP_CONNTRACK_LISTEN */
122 #define sNO TCP_CONNTRACK_NONE
123 #define sSS TCP_CONNTRACK_SYN_SENT
124 #define sSR TCP_CONNTRACK_SYN_RECV
125 #define sES TCP_CONNTRACK_ESTABLISHED
126 #define sFW TCP_CONNTRACK_FIN_WAIT
127 #define sCW TCP_CONNTRACK_CLOSE_WAIT
128 #define sLA TCP_CONNTRACK_LAST_ACK
129 #define sTW TCP_CONNTRACK_TIME_WAIT
130 #define sCL TCP_CONNTRACK_CLOSE
131 #define sLI TCP_CONNTRACK_LISTEN
132 #define sIV TCP_CONNTRACK_MAX
133 #define sIG TCP_CONNTRACK_IGNORE
135 /* What TCP flags are set from RST/SYN/FIN/ACK. */
146 * The TCP state transition table needs a few words...
148 * We are the man in the middle. All the packets go through us
149 * but might get lost in transit to the destination.
150 * It is assumed that the destinations can't receive segments
153 * The checked segment is in window, but our windows are *not*
154 * equivalent with the ones of the sender/receiver. We always
155 * try to guess the state of the current sender.
157 * The meaning of the states are:
159 * NONE: initial state
160 * SYN_SENT: SYN-only packet seen
161 * SYN_RECV: SYN-ACK packet seen
162 * ESTABLISHED: ACK packet seen
163 * FIN_WAIT: FIN packet seen
164 * CLOSE_WAIT: ACK seen (after FIN)
165 * LAST_ACK: FIN seen (after FIN)
166 * TIME_WAIT: last ACK seen
167 * CLOSE: closed connection
169 * LISTEN state is not used.
171 * Packets marked as IGNORED (sIG):
172 * if they may be either invalid or valid
173 * and the receiver may send back a connection
174 * closing RST or a SYN/ACK.
176 * Packets marked as INVALID (sIV):
177 * if they are invalid
178 * or we do not support the request (simultaneous open)
180 static enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
183 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
184 /*syn*/ { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
186 * sNO -> sSS Initialize a new connection
187 * sSS -> sSS Retransmitted SYN
188 * sSR -> sIG Late retransmitted SYN?
189 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
190 * are errors. Receiver will reply with RST
191 * and close the connection.
192 * Or we are not in sync and hold a dead connection.
196 * sTW -> sSS Reopened connection (RFC 1122).
199 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
200 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
202 * A SYN/ACK from the client is always invalid:
203 * - either it tries to set up a simultaneous open, which is
205 * - or the firewall has just been inserted between the two hosts
206 * during the session set-up. The SYN will be retransmitted
207 * by the true client (or it'll time out).
209 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
210 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
212 * sNO -> sIV Too late and no reason to do anything...
213 * sSS -> sIV Client migth not send FIN in this state:
214 * we enforce waiting for a SYN/ACK reply first.
215 * sSR -> sFW Close started.
217 * sFW -> sLA FIN seen in both directions, waiting for
219 * Migth be a retransmitted FIN as well...
221 * sLA -> sLA Retransmitted FIN. Remain in the same state.
225 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
226 /*ack*/ { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
228 * sNO -> sES Assumed.
229 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
230 * sSR -> sES Established state is reached.
232 * sFW -> sCW Normal close request answered by ACK.
234 * sLA -> sTW Last ACK detected.
235 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
238 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
239 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
240 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
244 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
245 /*syn*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
247 * sNO -> sIV Never reached.
248 * sSS -> sIV Simultaneous open, not supported
249 * sSR -> sIV Simultaneous open, not supported.
250 * sES -> sIV Server may not initiate a connection.
254 * sTW -> sIV Reopened connection, but server may not do it.
257 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
258 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
260 * sSS -> sSR Standard open.
261 * sSR -> sSR Retransmitted SYN/ACK.
262 * sES -> sIG Late retransmitted SYN/ACK?
263 * sFW -> sIG Might be SYN/ACK answering ignored SYN
269 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
270 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
272 * sSS -> sIV Server might not send FIN in this state.
273 * sSR -> sFW Close started.
275 * sFW -> sLA FIN seen in both directions.
277 * sLA -> sLA Retransmitted FIN.
281 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
282 /*ack*/ { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
284 * sSS -> sIG Might be a half-open connection.
285 * sSR -> sSR Might answer late resent SYN.
287 * sFW -> sCW Normal close request answered by ACK.
289 * sLA -> sTW Last ACK detected.
290 * sTW -> sTW Retransmitted last ACK.
293 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
294 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
295 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
299 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
300 unsigned int dataoff,
301 struct nf_conntrack_tuple *tuple)
303 struct tcphdr _hdr, *hp;
305 /* Actually only need first 8 bytes. */
306 hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
310 tuple->src.u.tcp.port = hp->source;
311 tuple->dst.u.tcp.port = hp->dest;
316 static int tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
317 const struct nf_conntrack_tuple *orig)
319 tuple->src.u.tcp.port = orig->dst.u.tcp.port;
320 tuple->dst.u.tcp.port = orig->src.u.tcp.port;
324 /* Print out the per-protocol part of the tuple. */
325 static int tcp_print_tuple(struct seq_file *s,
326 const struct nf_conntrack_tuple *tuple)
328 return seq_printf(s, "sport=%hu dport=%hu ",
329 ntohs(tuple->src.u.tcp.port),
330 ntohs(tuple->dst.u.tcp.port));
333 /* Print out the private part of the conntrack. */
334 static int tcp_print_conntrack(struct seq_file *s,
335 const struct nf_conn *conntrack)
337 enum tcp_conntrack state;
339 read_lock_bh(&tcp_lock);
340 state = conntrack->proto.tcp.state;
341 read_unlock_bh(&tcp_lock);
343 return seq_printf(s, "%s ", tcp_conntrack_names[state]);
346 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
348 if (tcph->rst) return TCP_RST_SET;
349 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
350 else if (tcph->fin) return TCP_FIN_SET;
351 else if (tcph->ack) return TCP_ACK_SET;
352 else return TCP_NONE_SET;
355 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
356 in IP Filter' by Guido van Rooij.
358 http://www.nluug.nl/events/sane2000/papers.html
359 http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
361 The boundaries and the conditions are changed according to RFC793:
362 the packet must intersect the window (i.e. segments may be
363 after the right or before the left edge) and thus receivers may ACK
364 segments after the right edge of the window.
366 td_maxend = max(sack + max(win,1)) seen in reply packets
367 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
368 td_maxwin += seq + len - sender.td_maxend
369 if seq + len > sender.td_maxend
370 td_end = max(seq + len) seen in sent packets
372 I. Upper bound for valid data: seq <= sender.td_maxend
373 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
374 III. Upper bound for valid ack: sack <= receiver.td_end
375 IV. Lower bound for valid ack: ack >= receiver.td_end - MAXACKWINDOW
377 where sack is the highest right edge of sack block found in the packet.
379 The upper bound limit for a valid ack is not ignored -
380 we doesn't have to deal with fragments.
383 static inline __u32 segment_seq_plus_len(__u32 seq,
385 unsigned int dataoff,
388 /* XXX Should I use payload length field in IP/IPv6 header ?
390 return (seq + len - dataoff - tcph->doff*4
391 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
394 /* Fixme: what about big packets? */
395 #define MAXACKWINCONST 66000
396 #define MAXACKWINDOW(sender) \
397 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
401 * Simplified tcp_parse_options routine from tcp_input.c
403 static void tcp_options(const struct sk_buff *skb,
404 unsigned int dataoff,
406 struct ip_ct_tcp_state *state)
408 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
410 int length = (tcph->doff*4) - sizeof(struct tcphdr);
415 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
429 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
434 if (opsize < 2) /* "silly options" */
437 break; /* don't parse partial options */
439 if (opcode == TCPOPT_SACK_PERM
440 && opsize == TCPOLEN_SACK_PERM)
441 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
442 else if (opcode == TCPOPT_WINDOW
443 && opsize == TCPOLEN_WINDOW) {
444 state->td_scale = *(u_int8_t *)ptr;
446 if (state->td_scale > 14) {
448 state->td_scale = 14;
451 IP_CT_TCP_FLAG_WINDOW_SCALE;
459 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
460 struct tcphdr *tcph, __u32 *sack)
462 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
464 int length = (tcph->doff*4) - sizeof(struct tcphdr);
470 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
474 /* Fast path for timestamp-only option */
475 if (length == TCPOLEN_TSTAMP_ALIGNED*4
477 __constant_ntohl((TCPOPT_NOP << 24)
479 | (TCPOPT_TIMESTAMP << 8)
480 | TCPOLEN_TIMESTAMP))
490 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
495 if (opsize < 2) /* "silly options" */
498 break; /* don't parse partial options */
500 if (opcode == TCPOPT_SACK
501 && opsize >= (TCPOLEN_SACK_BASE
502 + TCPOLEN_SACK_PERBLOCK)
503 && !((opsize - TCPOLEN_SACK_BASE)
504 % TCPOLEN_SACK_PERBLOCK)) {
506 i < (opsize - TCPOLEN_SACK_BASE);
507 i += TCPOLEN_SACK_PERBLOCK) {
508 memcpy(&tmp, (__u32 *)(ptr + i) + 1,
512 if (after(tmp, *sack))
523 static int tcp_in_window(struct ip_ct_tcp *state,
524 enum ip_conntrack_dir dir,
526 const struct sk_buff *skb,
527 unsigned int dataoff,
531 struct ip_ct_tcp_state *sender = &state->seen[dir];
532 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
533 __u32 seq, ack, sack, end, win, swin;
537 * Get the required data from the packet.
539 seq = ntohl(tcph->seq);
540 ack = sack = ntohl(tcph->ack_seq);
541 win = ntohs(tcph->window);
542 end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
544 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
545 tcp_sack(skb, dataoff, tcph, &sack);
547 DEBUGP("tcp_in_window: START\n");
548 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
549 "seq=%u ack=%u sack=%u win=%u end=%u\n",
550 NIPQUAD(iph->saddr), ntohs(tcph->source),
551 NIPQUAD(iph->daddr), ntohs(tcph->dest),
552 seq, ack, sack, win, end);
553 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
554 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
555 sender->td_end, sender->td_maxend, sender->td_maxwin,
557 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
560 if (sender->td_end == 0) {
562 * Initialize sender data.
564 if (tcph->syn && tcph->ack) {
566 * Outgoing SYN-ACK in reply to a SYN.
569 sender->td_maxend = end;
570 sender->td_maxwin = (win == 0 ? 1 : win);
572 tcp_options(skb, dataoff, tcph, sender);
575 * Both sides must send the Window Scale option
576 * to enable window scaling in either direction.
578 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
579 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
581 receiver->td_scale = 0;
584 * We are in the middle of a connection,
585 * its history is lost for us.
586 * Let's try to use the data from the packet.
588 sender->td_end = end;
589 sender->td_maxwin = (win == 0 ? 1 : win);
590 sender->td_maxend = end + sender->td_maxwin;
592 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
593 && dir == IP_CT_DIR_ORIGINAL)
594 || (state->state == TCP_CONNTRACK_SYN_RECV
595 && dir == IP_CT_DIR_REPLY))
596 && after(end, sender->td_end)) {
598 * RFC 793: "if a TCP is reinitialized ... then it need
599 * not wait at all; it must only be sure to use sequence
600 * numbers larger than those recently used."
603 sender->td_maxend = end;
604 sender->td_maxwin = (win == 0 ? 1 : win);
606 tcp_options(skb, dataoff, tcph, sender);
611 * If there is no ACK, just pretend it was set and OK.
613 ack = sack = receiver->td_end;
614 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
615 (TCP_FLAG_ACK|TCP_FLAG_RST))
618 * Broken TCP stacks, that set ACK in RST packets as well
619 * with zero ack value.
621 ack = sack = receiver->td_end;
626 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
628 * Packets contains no data: we assume it is valid
629 * and check the ack value only.
630 * However RST segments are always validated by their
631 * SEQ number, except when seq == 0 (reset sent answering
634 seq = end = sender->td_end;
636 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
637 "seq=%u ack=%u sack =%u win=%u end=%u\n",
638 NIPQUAD(iph->saddr), ntohs(tcph->source),
639 NIPQUAD(iph->daddr), ntohs(tcph->dest),
640 seq, ack, sack, win, end);
641 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
642 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
643 sender->td_end, sender->td_maxend, sender->td_maxwin,
645 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
648 DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
649 before(seq, sender->td_maxend + 1),
650 after(end, sender->td_end - receiver->td_maxwin - 1),
651 before(sack, receiver->td_end + 1),
652 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
654 if (sender->loose || receiver->loose ||
655 (before(seq, sender->td_maxend + 1) &&
656 after(end, sender->td_end - receiver->td_maxwin - 1) &&
657 before(sack, receiver->td_end + 1) &&
658 after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
660 * Take into account window scaling (RFC 1323).
663 win <<= sender->td_scale;
666 * Update sender data.
668 swin = win + (sack - ack);
669 if (sender->td_maxwin < swin)
670 sender->td_maxwin = swin;
671 if (after(end, sender->td_end))
672 sender->td_end = end;
674 * Update receiver data.
676 if (after(end, sender->td_maxend))
677 receiver->td_maxwin += end - sender->td_maxend;
678 if (after(sack + win, receiver->td_maxend - 1)) {
679 receiver->td_maxend = sack + win;
681 receiver->td_maxend++;
685 * Check retransmissions.
687 if (index == TCP_ACK_SET) {
688 if (state->last_dir == dir
689 && state->last_seq == seq
690 && state->last_ack == ack
691 && state->last_end == end)
694 state->last_dir = dir;
695 state->last_seq = seq;
696 state->last_ack = ack;
697 state->last_end = end;
702 * Close the window of disabled window tracking :-)
709 if (LOG_INVALID(IPPROTO_TCP))
710 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
712 before(seq, sender->td_maxend + 1) ?
713 after(end, sender->td_end - receiver->td_maxwin - 1) ?
714 before(sack, receiver->td_end + 1) ?
715 after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
716 : "ACK is under the lower bound (possible overly delayed ACK)"
717 : "ACK is over the upper bound (ACKed data not seen yet)"
718 : "SEQ is under the lower bound (already ACKed data retransmitted)"
719 : "SEQ is over the upper bound (over the window of the receiver)");
721 res = nf_ct_tcp_be_liberal;
724 DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
725 "receiver end=%u maxend=%u maxwin=%u\n",
726 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
727 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
732 #ifdef CONFIG_IP_NF_NAT_NEEDED
733 /* Update sender->td_end after NAT successfully mangled the packet */
734 /* Caller must linearize skb at tcp header. */
735 void nf_conntrack_tcp_update(struct sk_buff *skb,
736 unsigned int dataoff,
737 struct nf_conn *conntrack,
740 struct tcphdr *tcph = (void *)skb->data + dataoff;
743 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
744 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
747 end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
749 write_lock_bh(&tcp_lock);
751 * We have to worry for the ack in the reply packet only...
753 if (after(end, conntrack->proto.tcp.seen[dir].td_end))
754 conntrack->proto.tcp.seen[dir].td_end = end;
755 conntrack->proto.tcp.last_end = end;
756 write_unlock_bh(&tcp_lock);
757 DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
758 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
759 sender->td_end, sender->td_maxend, sender->td_maxwin,
761 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
776 /* table of valid flag combinations - ECE and CWR are always valid */
777 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
781 [TH_SYN|TH_PUSH] = 1,
782 [TH_SYN|TH_ACK|TH_PUSH] = 1,
785 [TH_RST|TH_ACK|TH_PUSH] = 1,
788 [TH_ACK|TH_PUSH] = 1,
790 [TH_ACK|TH_URG|TH_PUSH] = 1,
791 [TH_FIN|TH_ACK|TH_PUSH] = 1,
792 [TH_FIN|TH_ACK|TH_URG] = 1,
793 [TH_FIN|TH_ACK|TH_URG|TH_PUSH] = 1,
796 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
797 static int tcp_error(struct sk_buff *skb,
798 unsigned int dataoff,
799 enum ip_conntrack_info *ctinfo,
801 unsigned int hooknum)
803 struct tcphdr _tcph, *th;
804 unsigned int tcplen = skb->len - dataoff;
807 /* Smaller that minimal TCP header? */
808 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
810 if (LOG_INVALID(IPPROTO_TCP))
811 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
812 "nf_ct_tcp: short packet ");
816 /* Not whole TCP header or malformed packet */
817 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
818 if (LOG_INVALID(IPPROTO_TCP))
819 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
820 "nf_ct_tcp: truncated/malformed packet ");
824 /* Checksum invalid? Ignore.
825 * We skip checking packets on the outgoing path
826 * because the semantic of CHECKSUM_HW is different there
827 * and moreover root might send raw packets.
829 /* FIXME: Source route IP option packets --RR */
830 if (nf_conntrack_checksum &&
831 ((pf == PF_INET && hooknum == NF_IP_PRE_ROUTING) ||
832 (pf == PF_INET6 && hooknum == NF_IP6_PRE_ROUTING)) &&
833 nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
834 if (LOG_INVALID(IPPROTO_TCP))
835 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
836 "nf_ct_tcp: bad TCP checksum ");
840 /* Check TCP flags. */
841 tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
842 if (!tcp_valid_flags[tcpflags]) {
843 if (LOG_INVALID(IPPROTO_TCP))
844 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
845 "nf_ct_tcp: invalid TCP flag combination ");
852 /* Returns verdict for packet, or -1 for invalid. */
853 static int tcp_packet(struct nf_conn *conntrack,
854 const struct sk_buff *skb,
855 unsigned int dataoff,
856 enum ip_conntrack_info ctinfo,
858 unsigned int hooknum)
860 enum tcp_conntrack new_state, old_state;
861 enum ip_conntrack_dir dir;
862 struct tcphdr *th, _tcph;
863 unsigned long timeout;
866 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
869 write_lock_bh(&tcp_lock);
870 old_state = conntrack->proto.tcp.state;
871 dir = CTINFO2DIR(ctinfo);
872 index = get_conntrack_index(th);
873 new_state = tcp_conntracks[dir][index][old_state];
876 case TCP_CONNTRACK_IGNORE:
880 * b) SYN/ACK in REPLY
881 * c) ACK in reply direction after initial SYN in original.
883 if (index == TCP_SYNACK_SET
884 && conntrack->proto.tcp.last_index == TCP_SYN_SET
885 && conntrack->proto.tcp.last_dir != dir
886 && ntohl(th->ack_seq) ==
887 conntrack->proto.tcp.last_end) {
888 /* This SYN/ACK acknowledges a SYN that we earlier
889 * ignored as invalid. This means that the client and
890 * the server are both in sync, while the firewall is
891 * not. We kill this session and block the SYN/ACK so
892 * that the client cannot but retransmit its SYN and
893 * thus initiate a clean new session.
895 write_unlock_bh(&tcp_lock);
896 if (LOG_INVALID(IPPROTO_TCP))
897 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
898 "nf_ct_tcp: killing out of sync session ");
899 if (del_timer(&conntrack->timeout))
900 conntrack->timeout.function((unsigned long)
904 conntrack->proto.tcp.last_index = index;
905 conntrack->proto.tcp.last_dir = dir;
906 conntrack->proto.tcp.last_seq = ntohl(th->seq);
907 conntrack->proto.tcp.last_end =
908 segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
910 write_unlock_bh(&tcp_lock);
911 if (LOG_INVALID(IPPROTO_TCP))
912 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
913 "nf_ct_tcp: invalid packed ignored ");
915 case TCP_CONNTRACK_MAX:
917 DEBUGP("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
918 dir, get_conntrack_index(th),
920 write_unlock_bh(&tcp_lock);
921 if (LOG_INVALID(IPPROTO_TCP))
922 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
923 "nf_ct_tcp: invalid state ");
925 case TCP_CONNTRACK_SYN_SENT:
926 if (old_state < TCP_CONNTRACK_TIME_WAIT)
928 if ((conntrack->proto.tcp.seen[dir].flags &
929 IP_CT_TCP_FLAG_CLOSE_INIT)
930 || after(ntohl(th->seq),
931 conntrack->proto.tcp.seen[dir].td_end)) {
932 /* Attempt to reopen a closed connection.
933 * Delete this connection and look up again. */
934 write_unlock_bh(&tcp_lock);
935 if (del_timer(&conntrack->timeout))
936 conntrack->timeout.function((unsigned long)
940 write_unlock_bh(&tcp_lock);
941 if (LOG_INVALID(IPPROTO_TCP))
942 nf_log_packet(pf, 0, skb, NULL, NULL,
943 NULL, "nf_ct_tcp: invalid SYN");
946 case TCP_CONNTRACK_CLOSE:
947 if (index == TCP_RST_SET
948 && ((test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
949 && conntrack->proto.tcp.last_index == TCP_SYN_SET)
950 || (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
951 && conntrack->proto.tcp.last_index == TCP_ACK_SET))
952 && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
953 /* RST sent to invalid SYN or ACK we had let through
954 * at a) and c) above:
956 * a) SYN was in window then
957 * c) we hold a half-open connection.
959 * Delete our connection entry.
960 * We skip window checking, because packet might ACK
961 * segments we ignored. */
964 /* Just fall through */
966 /* Keep compilers happy. */
970 if (!tcp_in_window(&conntrack->proto.tcp, dir, index,
971 skb, dataoff, th, pf)) {
972 write_unlock_bh(&tcp_lock);
976 /* From now on we have got in-window packets */
977 conntrack->proto.tcp.last_index = index;
979 DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
980 "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
981 NIPQUAD(iph->saddr), ntohs(th->source),
982 NIPQUAD(iph->daddr), ntohs(th->dest),
983 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
984 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
985 old_state, new_state);
987 conntrack->proto.tcp.state = new_state;
988 if (old_state != new_state
989 && (new_state == TCP_CONNTRACK_FIN_WAIT
990 || new_state == TCP_CONNTRACK_CLOSE))
991 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
992 timeout = conntrack->proto.tcp.retrans >= nf_ct_tcp_max_retrans
993 && *tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans
994 ? nf_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
995 write_unlock_bh(&tcp_lock);
997 nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
998 if (new_state != old_state)
999 nf_conntrack_event_cache(IPCT_PROTOINFO, skb);
1001 if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
1002 /* If only reply is a RST, we can consider ourselves not to
1003 have an established connection: this is a fairly common
1004 problem case, so we can delete the conntrack
1005 immediately. --RR */
1007 if (del_timer(&conntrack->timeout))
1008 conntrack->timeout.function((unsigned long)
1012 } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
1013 && (old_state == TCP_CONNTRACK_SYN_RECV
1014 || old_state == TCP_CONNTRACK_ESTABLISHED)
1015 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1016 /* Set ASSURED if we see see valid ack in ESTABLISHED
1017 after SYN_RECV or a valid answer for a picked up
1019 set_bit(IPS_ASSURED_BIT, &conntrack->status);
1020 nf_conntrack_event_cache(IPCT_STATUS, skb);
1022 nf_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
1027 /* Called when a new connection for this protocol found. */
1028 static int tcp_new(struct nf_conn *conntrack,
1029 const struct sk_buff *skb,
1030 unsigned int dataoff)
1032 enum tcp_conntrack new_state;
1033 struct tcphdr *th, _tcph;
1035 struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1036 struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1039 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1042 /* Don't need lock here: this conntrack not in circulation yet */
1044 = tcp_conntracks[0][get_conntrack_index(th)]
1045 [TCP_CONNTRACK_NONE];
1047 /* Invalid: delete conntrack */
1048 if (new_state >= TCP_CONNTRACK_MAX) {
1049 DEBUGP("nf_ct_tcp: invalid new deleting.\n");
1053 if (new_state == TCP_CONNTRACK_SYN_SENT) {
1055 conntrack->proto.tcp.seen[0].td_end =
1056 segment_seq_plus_len(ntohl(th->seq), skb->len,
1058 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1059 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1060 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1061 conntrack->proto.tcp.seen[0].td_maxend =
1062 conntrack->proto.tcp.seen[0].td_end;
1064 tcp_options(skb, dataoff, th, &conntrack->proto.tcp.seen[0]);
1065 conntrack->proto.tcp.seen[1].flags = 0;
1066 conntrack->proto.tcp.seen[0].loose =
1067 conntrack->proto.tcp.seen[1].loose = 0;
1068 } else if (nf_ct_tcp_loose == 0) {
1069 /* Don't try to pick up connections. */
1073 * We are in the middle of a connection,
1074 * its history is lost for us.
1075 * Let's try to use the data from the packet.
1077 conntrack->proto.tcp.seen[0].td_end =
1078 segment_seq_plus_len(ntohl(th->seq), skb->len,
1080 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1081 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1082 conntrack->proto.tcp.seen[0].td_maxwin = 1;
1083 conntrack->proto.tcp.seen[0].td_maxend =
1084 conntrack->proto.tcp.seen[0].td_end +
1085 conntrack->proto.tcp.seen[0].td_maxwin;
1086 conntrack->proto.tcp.seen[0].td_scale = 0;
1088 /* We assume SACK. Should we assume window scaling too? */
1089 conntrack->proto.tcp.seen[0].flags =
1090 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1091 conntrack->proto.tcp.seen[0].loose =
1092 conntrack->proto.tcp.seen[1].loose = nf_ct_tcp_loose;
1095 conntrack->proto.tcp.seen[1].td_end = 0;
1096 conntrack->proto.tcp.seen[1].td_maxend = 0;
1097 conntrack->proto.tcp.seen[1].td_maxwin = 1;
1098 conntrack->proto.tcp.seen[1].td_scale = 0;
1100 /* tcp_packet will set them */
1101 conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1102 conntrack->proto.tcp.last_index = TCP_NONE_SET;
1104 DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1105 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1106 sender->td_end, sender->td_maxend, sender->td_maxwin,
1108 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1109 receiver->td_scale);
1113 #if defined(CONFIG_NF_CT_NETLINK) || \
1114 defined(CONFIG_NF_CT_NETLINK_MODULE)
1116 #include <linux/netfilter/nfnetlink.h>
1117 #include <linux/netfilter/nfnetlink_conntrack.h>
1119 static int tcp_to_nfattr(struct sk_buff *skb, struct nfattr *nfa,
1120 const struct nf_conn *ct)
1122 struct nfattr *nest_parms;
1124 read_lock_bh(&tcp_lock);
1125 nest_parms = NFA_NEST(skb, CTA_PROTOINFO_TCP);
1126 NFA_PUT(skb, CTA_PROTOINFO_TCP_STATE, sizeof(u_int8_t),
1127 &ct->proto.tcp.state);
1128 read_unlock_bh(&tcp_lock);
1130 NFA_NEST_END(skb, nest_parms);
1135 read_unlock_bh(&tcp_lock);
1139 static const size_t cta_min_tcp[CTA_PROTOINFO_TCP_MAX] = {
1140 [CTA_PROTOINFO_TCP_STATE-1] = sizeof(u_int8_t),
1143 static int nfattr_to_tcp(struct nfattr *cda[], struct nf_conn *ct)
1145 struct nfattr *attr = cda[CTA_PROTOINFO_TCP-1];
1146 struct nfattr *tb[CTA_PROTOINFO_TCP_MAX];
1148 /* updates could not contain anything about the private
1149 * protocol info, in that case skip the parsing */
1153 nfattr_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, attr);
1155 if (nfattr_bad_size(tb, CTA_PROTOINFO_TCP_MAX, cta_min_tcp))
1158 if (!tb[CTA_PROTOINFO_TCP_STATE-1])
1161 write_lock_bh(&tcp_lock);
1162 ct->proto.tcp.state =
1163 *(u_int8_t *)NFA_DATA(tb[CTA_PROTOINFO_TCP_STATE-1]);
1164 write_unlock_bh(&tcp_lock);
1170 struct nf_conntrack_protocol nf_conntrack_protocol_tcp4 =
1173 .proto = IPPROTO_TCP,
1175 .pkt_to_tuple = tcp_pkt_to_tuple,
1176 .invert_tuple = tcp_invert_tuple,
1177 .print_tuple = tcp_print_tuple,
1178 .print_conntrack = tcp_print_conntrack,
1179 .packet = tcp_packet,
1182 #if defined(CONFIG_NF_CT_NETLINK) || \
1183 defined(CONFIG_NF_CT_NETLINK_MODULE)
1184 .to_nfattr = tcp_to_nfattr,
1185 .from_nfattr = nfattr_to_tcp,
1186 .tuple_to_nfattr = nf_ct_port_tuple_to_nfattr,
1187 .nfattr_to_tuple = nf_ct_port_nfattr_to_tuple,
1191 struct nf_conntrack_protocol nf_conntrack_protocol_tcp6 =
1193 .l3proto = PF_INET6,
1194 .proto = IPPROTO_TCP,
1196 .pkt_to_tuple = tcp_pkt_to_tuple,
1197 .invert_tuple = tcp_invert_tuple,
1198 .print_tuple = tcp_print_tuple,
1199 .print_conntrack = tcp_print_conntrack,
1200 .packet = tcp_packet,
1203 #if defined(CONFIG_NF_CT_NETLINK) || \
1204 defined(CONFIG_NF_CT_NETLINK_MODULE)
1205 .to_nfattr = tcp_to_nfattr,
1206 .from_nfattr = nfattr_to_tcp,
1207 .tuple_to_nfattr = nf_ct_port_tuple_to_nfattr,
1208 .nfattr_to_tuple = nf_ct_port_nfattr_to_tuple,
1212 EXPORT_SYMBOL(nf_conntrack_protocol_tcp4);
1213 EXPORT_SYMBOL(nf_conntrack_protocol_tcp6);