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/timer.h>
24 #include <linux/netfilter.h>
25 #include <linux/module.h>
28 #include <linux/tcp.h>
29 #include <linux/spinlock.h>
33 #include <linux/netfilter_ipv4.h>
34 #include <linux/netfilter_ipv4/ip_conntrack.h>
35 #include <linux/netfilter_ipv4/ip_conntrack_protocol.h>
41 #define DEBUGP(format, args...)
44 /* Protects conntrack->proto.tcp */
45 static DEFINE_RWLOCK(tcp_lock
);
47 /* "Be conservative in what you do,
48 be liberal in what you accept from others."
49 If it's non-zero, we mark only out of window RST segments as INVALID. */
50 int ip_ct_tcp_be_liberal __read_mostly
= 0;
52 /* If it is set to zero, we disable picking up already established
54 int ip_ct_tcp_loose __read_mostly
= 1;
56 /* Max number of the retransmitted packets without receiving an (acceptable)
57 ACK from the destination. If this number is reached, a shorter timer
59 int ip_ct_tcp_max_retrans __read_mostly
= 3;
61 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
62 closely. They're more complex. --RR */
64 static const char *tcp_conntrack_names
[] = {
78 #define MINS * 60 SECS
79 #define HOURS * 60 MINS
80 #define DAYS * 24 HOURS
82 unsigned int ip_ct_tcp_timeout_syn_sent __read_mostly
= 2 MINS
;
83 unsigned int ip_ct_tcp_timeout_syn_recv __read_mostly
= 60 SECS
;
84 unsigned int ip_ct_tcp_timeout_established __read_mostly
= 5 DAYS
;
85 unsigned int ip_ct_tcp_timeout_fin_wait __read_mostly
= 2 MINS
;
86 unsigned int ip_ct_tcp_timeout_close_wait __read_mostly
= 60 SECS
;
87 unsigned int ip_ct_tcp_timeout_last_ack __read_mostly
= 30 SECS
;
88 unsigned int ip_ct_tcp_timeout_time_wait __read_mostly
= 2 MINS
;
89 unsigned int ip_ct_tcp_timeout_close __read_mostly
= 10 SECS
;
91 /* RFC1122 says the R2 limit should be at least 100 seconds.
92 Linux uses 15 packets as limit, which corresponds
93 to ~13-30min depending on RTO. */
94 unsigned int ip_ct_tcp_timeout_max_retrans __read_mostly
= 5 MINS
;
96 static const unsigned int * tcp_timeouts
[]
97 = { NULL
, /* TCP_CONNTRACK_NONE */
98 &ip_ct_tcp_timeout_syn_sent
, /* TCP_CONNTRACK_SYN_SENT, */
99 &ip_ct_tcp_timeout_syn_recv
, /* TCP_CONNTRACK_SYN_RECV, */
100 &ip_ct_tcp_timeout_established
, /* TCP_CONNTRACK_ESTABLISHED, */
101 &ip_ct_tcp_timeout_fin_wait
, /* TCP_CONNTRACK_FIN_WAIT, */
102 &ip_ct_tcp_timeout_close_wait
, /* TCP_CONNTRACK_CLOSE_WAIT, */
103 &ip_ct_tcp_timeout_last_ack
, /* TCP_CONNTRACK_LAST_ACK, */
104 &ip_ct_tcp_timeout_time_wait
, /* TCP_CONNTRACK_TIME_WAIT, */
105 &ip_ct_tcp_timeout_close
, /* TCP_CONNTRACK_CLOSE, */
106 NULL
, /* TCP_CONNTRACK_LISTEN */
109 #define sNO TCP_CONNTRACK_NONE
110 #define sSS TCP_CONNTRACK_SYN_SENT
111 #define sSR TCP_CONNTRACK_SYN_RECV
112 #define sES TCP_CONNTRACK_ESTABLISHED
113 #define sFW TCP_CONNTRACK_FIN_WAIT
114 #define sCW TCP_CONNTRACK_CLOSE_WAIT
115 #define sLA TCP_CONNTRACK_LAST_ACK
116 #define sTW TCP_CONNTRACK_TIME_WAIT
117 #define sCL TCP_CONNTRACK_CLOSE
118 #define sLI TCP_CONNTRACK_LISTEN
119 #define sIV TCP_CONNTRACK_MAX
120 #define sIG TCP_CONNTRACK_IGNORE
122 /* What TCP flags are set from RST/SYN/FIN/ACK. */
133 * The TCP state transition table needs a few words...
135 * We are the man in the middle. All the packets go through us
136 * but might get lost in transit to the destination.
137 * It is assumed that the destinations can't receive segments
140 * The checked segment is in window, but our windows are *not*
141 * equivalent with the ones of the sender/receiver. We always
142 * try to guess the state of the current sender.
144 * The meaning of the states are:
146 * NONE: initial state
147 * SYN_SENT: SYN-only packet seen
148 * SYN_RECV: SYN-ACK packet seen
149 * ESTABLISHED: ACK packet seen
150 * FIN_WAIT: FIN packet seen
151 * CLOSE_WAIT: ACK seen (after FIN)
152 * LAST_ACK: FIN seen (after FIN)
153 * TIME_WAIT: last ACK seen
154 * CLOSE: closed connection
156 * LISTEN state is not used.
158 * Packets marked as IGNORED (sIG):
159 * if they may be either invalid or valid
160 * and the receiver may send back a connection
161 * closing RST or a SYN/ACK.
163 * Packets marked as INVALID (sIV):
164 * if they are invalid
165 * or we do not support the request (simultaneous open)
167 static const enum tcp_conntrack tcp_conntracks
[2][6][TCP_CONNTRACK_MAX
] = {
170 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
171 /*syn*/ { sSS
, sSS
, sIG
, sIG
, sIG
, sIG
, sIG
, sSS
, sSS
, sIV
},
173 * sNO -> sSS Initialize a new connection
174 * sSS -> sSS Retransmitted SYN
175 * sSR -> sIG Late retransmitted SYN?
176 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
177 * are errors. Receiver will reply with RST
178 * and close the connection.
179 * Or we are not in sync and hold a dead connection.
183 * sTW -> sSS Reopened connection (RFC 1122).
186 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
187 /*synack*/ { sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
},
189 * A SYN/ACK from the client is always invalid:
190 * - either it tries to set up a simultaneous open, which is
192 * - or the firewall has just been inserted between the two hosts
193 * during the session set-up. The SYN will be retransmitted
194 * by the true client (or it'll time out).
196 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
197 /*fin*/ { sIV
, sIV
, sFW
, sFW
, sLA
, sLA
, sLA
, sTW
, sCL
, sIV
},
199 * sNO -> sIV Too late and no reason to do anything...
200 * sSS -> sIV Client migth not send FIN in this state:
201 * we enforce waiting for a SYN/ACK reply first.
202 * sSR -> sFW Close started.
204 * sFW -> sLA FIN seen in both directions, waiting for
206 * Migth be a retransmitted FIN as well...
208 * sLA -> sLA Retransmitted FIN. Remain in the same state.
212 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
213 /*ack*/ { sES
, sIV
, sES
, sES
, sCW
, sCW
, sTW
, sTW
, sCL
, sIV
},
215 * sNO -> sES Assumed.
216 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
217 * sSR -> sES Established state is reached.
219 * sFW -> sCW Normal close request answered by ACK.
221 * sLA -> sTW Last ACK detected.
222 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
225 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
226 /*rst*/ { sIV
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sIV
},
227 /*none*/ { sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
}
231 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
232 /*syn*/ { sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
},
234 * sNO -> sIV Never reached.
235 * sSS -> sIV Simultaneous open, not supported
236 * sSR -> sIV Simultaneous open, not supported.
237 * sES -> sIV Server may not initiate a connection.
241 * sTW -> sIV Reopened connection, but server may not do it.
244 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
245 /*synack*/ { sIV
, sSR
, sSR
, sIG
, sIG
, sIG
, sIG
, sIG
, sIG
, sIV
},
247 * sSS -> sSR Standard open.
248 * sSR -> sSR Retransmitted SYN/ACK.
249 * sES -> sIG Late retransmitted SYN/ACK?
250 * sFW -> sIG Might be SYN/ACK answering ignored SYN
256 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
257 /*fin*/ { sIV
, sIV
, sFW
, sFW
, sLA
, sLA
, sLA
, sTW
, sCL
, sIV
},
259 * sSS -> sIV Server might not send FIN in this state.
260 * sSR -> sFW Close started.
262 * sFW -> sLA FIN seen in both directions.
264 * sLA -> sLA Retransmitted FIN.
268 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
269 /*ack*/ { sIV
, sIG
, sSR
, sES
, sCW
, sCW
, sTW
, sTW
, sCL
, sIV
},
271 * sSS -> sIG Might be a half-open connection.
272 * sSR -> sSR Might answer late resent SYN.
274 * sFW -> sCW Normal close request answered by ACK.
276 * sLA -> sTW Last ACK detected.
277 * sTW -> sTW Retransmitted last ACK.
280 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
281 /*rst*/ { sIV
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sIV
},
282 /*none*/ { sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
}
286 static int tcp_pkt_to_tuple(const struct sk_buff
*skb
,
287 unsigned int dataoff
,
288 struct ip_conntrack_tuple
*tuple
)
290 struct tcphdr _hdr
, *hp
;
292 /* Actually only need first 8 bytes. */
293 hp
= skb_header_pointer(skb
, dataoff
, 8, &_hdr
);
297 tuple
->src
.u
.tcp
.port
= hp
->source
;
298 tuple
->dst
.u
.tcp
.port
= hp
->dest
;
303 static int tcp_invert_tuple(struct ip_conntrack_tuple
*tuple
,
304 const struct ip_conntrack_tuple
*orig
)
306 tuple
->src
.u
.tcp
.port
= orig
->dst
.u
.tcp
.port
;
307 tuple
->dst
.u
.tcp
.port
= orig
->src
.u
.tcp
.port
;
311 /* Print out the per-protocol part of the tuple. */
312 static int tcp_print_tuple(struct seq_file
*s
,
313 const struct ip_conntrack_tuple
*tuple
)
315 return seq_printf(s
, "sport=%hu dport=%hu ",
316 ntohs(tuple
->src
.u
.tcp
.port
),
317 ntohs(tuple
->dst
.u
.tcp
.port
));
320 /* Print out the private part of the conntrack. */
321 static int tcp_print_conntrack(struct seq_file
*s
,
322 const struct ip_conntrack
*conntrack
)
324 enum tcp_conntrack state
;
326 read_lock_bh(&tcp_lock
);
327 state
= conntrack
->proto
.tcp
.state
;
328 read_unlock_bh(&tcp_lock
);
330 return seq_printf(s
, "%s ", tcp_conntrack_names
[state
]);
333 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
334 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
335 static int tcp_to_nfattr(struct sk_buff
*skb
, struct nfattr
*nfa
,
336 const struct ip_conntrack
*ct
)
338 struct nfattr
*nest_parms
;
340 read_lock_bh(&tcp_lock
);
341 nest_parms
= NFA_NEST(skb
, CTA_PROTOINFO_TCP
);
342 NFA_PUT(skb
, CTA_PROTOINFO_TCP_STATE
, sizeof(u_int8_t
),
343 &ct
->proto
.tcp
.state
);
344 read_unlock_bh(&tcp_lock
);
346 NFA_NEST_END(skb
, nest_parms
);
351 read_unlock_bh(&tcp_lock
);
355 static const size_t cta_min_tcp
[CTA_PROTOINFO_TCP_MAX
] = {
356 [CTA_PROTOINFO_TCP_STATE
-1] = sizeof(u_int8_t
),
359 static int nfattr_to_tcp(struct nfattr
*cda
[], struct ip_conntrack
*ct
)
361 struct nfattr
*attr
= cda
[CTA_PROTOINFO_TCP
-1];
362 struct nfattr
*tb
[CTA_PROTOINFO_TCP_MAX
];
364 /* updates could not contain anything about the private
365 * protocol info, in that case skip the parsing */
369 nfattr_parse_nested(tb
, CTA_PROTOINFO_TCP_MAX
, attr
);
371 if (nfattr_bad_size(tb
, CTA_PROTOINFO_TCP_MAX
, cta_min_tcp
))
374 if (!tb
[CTA_PROTOINFO_TCP_STATE
-1])
377 write_lock_bh(&tcp_lock
);
378 ct
->proto
.tcp
.state
=
379 *(u_int8_t
*)NFA_DATA(tb
[CTA_PROTOINFO_TCP_STATE
-1]);
380 write_unlock_bh(&tcp_lock
);
386 static unsigned int get_conntrack_index(const struct tcphdr
*tcph
)
388 if (tcph
->rst
) return TCP_RST_SET
;
389 else if (tcph
->syn
) return (tcph
->ack
? TCP_SYNACK_SET
: TCP_SYN_SET
);
390 else if (tcph
->fin
) return TCP_FIN_SET
;
391 else if (tcph
->ack
) return TCP_ACK_SET
;
392 else return TCP_NONE_SET
;
395 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
396 in IP Filter' by Guido van Rooij.
398 http://www.nluug.nl/events/sane2000/papers.html
399 http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
401 The boundaries and the conditions are changed according to RFC793:
402 the packet must intersect the window (i.e. segments may be
403 after the right or before the left edge) and thus receivers may ACK
404 segments after the right edge of the window.
406 td_maxend = max(sack + max(win,1)) seen in reply packets
407 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
408 td_maxwin += seq + len - sender.td_maxend
409 if seq + len > sender.td_maxend
410 td_end = max(seq + len) seen in sent packets
412 I. Upper bound for valid data: seq <= sender.td_maxend
413 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
414 III. Upper bound for valid ack: sack <= receiver.td_end
415 IV. Lower bound for valid ack: ack >= receiver.td_end - MAXACKWINDOW
417 where sack is the highest right edge of sack block found in the packet.
419 The upper bound limit for a valid ack is not ignored -
420 we doesn't have to deal with fragments.
423 static inline __u32
segment_seq_plus_len(__u32 seq
,
428 return (seq
+ len
- (iph
->ihl
+ tcph
->doff
)*4
429 + (tcph
->syn
? 1 : 0) + (tcph
->fin
? 1 : 0));
432 /* Fixme: what about big packets? */
433 #define MAXACKWINCONST 66000
434 #define MAXACKWINDOW(sender) \
435 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
439 * Simplified tcp_parse_options routine from tcp_input.c
441 static void tcp_options(const struct sk_buff
*skb
,
444 struct ip_ct_tcp_state
*state
)
446 unsigned char buff
[(15 * 4) - sizeof(struct tcphdr
)];
448 int length
= (tcph
->doff
*4) - sizeof(struct tcphdr
);
453 ptr
= skb_header_pointer(skb
,
454 (iph
->ihl
* 4) + sizeof(struct tcphdr
),
468 case TCPOPT_NOP
: /* Ref: RFC 793 section 3.1 */
473 if (opsize
< 2) /* "silly options" */
476 break; /* don't parse partial options */
478 if (opcode
== TCPOPT_SACK_PERM
479 && opsize
== TCPOLEN_SACK_PERM
)
480 state
->flags
|= IP_CT_TCP_FLAG_SACK_PERM
;
481 else if (opcode
== TCPOPT_WINDOW
482 && opsize
== TCPOLEN_WINDOW
) {
483 state
->td_scale
= *(u_int8_t
*)ptr
;
485 if (state
->td_scale
> 14) {
487 state
->td_scale
= 14;
490 IP_CT_TCP_FLAG_WINDOW_SCALE
;
498 static void tcp_sack(const struct sk_buff
*skb
,
503 unsigned char buff
[(15 * 4) - sizeof(struct tcphdr
)];
505 int length
= (tcph
->doff
*4) - sizeof(struct tcphdr
);
511 ptr
= skb_header_pointer(skb
,
512 (iph
->ihl
* 4) + sizeof(struct tcphdr
),
516 /* Fast path for timestamp-only option */
517 if (length
== TCPOLEN_TSTAMP_ALIGNED
*4
518 && *(__be32
*)ptr
== htonl((TCPOPT_NOP
<< 24)
520 | (TCPOPT_TIMESTAMP
<< 8)
521 | TCPOLEN_TIMESTAMP
))
531 case TCPOPT_NOP
: /* Ref: RFC 793 section 3.1 */
536 if (opsize
< 2) /* "silly options" */
539 break; /* don't parse partial options */
541 if (opcode
== TCPOPT_SACK
542 && opsize
>= (TCPOLEN_SACK_BASE
543 + TCPOLEN_SACK_PERBLOCK
)
544 && !((opsize
- TCPOLEN_SACK_BASE
)
545 % TCPOLEN_SACK_PERBLOCK
)) {
547 i
< (opsize
- TCPOLEN_SACK_BASE
);
548 i
+= TCPOLEN_SACK_PERBLOCK
) {
549 tmp
= ntohl(*((__be32
*)(ptr
+i
)+1));
551 if (after(tmp
, *sack
))
562 static int tcp_in_window(struct ip_ct_tcp
*state
,
563 enum ip_conntrack_dir dir
,
565 const struct sk_buff
*skb
,
569 struct ip_ct_tcp_state
*sender
= &state
->seen
[dir
];
570 struct ip_ct_tcp_state
*receiver
= &state
->seen
[!dir
];
571 __u32 seq
, ack
, sack
, end
, win
, swin
;
575 * Get the required data from the packet.
577 seq
= ntohl(tcph
->seq
);
578 ack
= sack
= ntohl(tcph
->ack_seq
);
579 win
= ntohs(tcph
->window
);
580 end
= segment_seq_plus_len(seq
, skb
->len
, iph
, tcph
);
582 if (receiver
->flags
& IP_CT_TCP_FLAG_SACK_PERM
)
583 tcp_sack(skb
, iph
, tcph
, &sack
);
585 DEBUGP("tcp_in_window: START\n");
586 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
587 "seq=%u ack=%u sack=%u win=%u end=%u\n",
588 NIPQUAD(iph
->saddr
), ntohs(tcph
->source
),
589 NIPQUAD(iph
->daddr
), ntohs(tcph
->dest
),
590 seq
, ack
, sack
, win
, end
);
591 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
592 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
593 sender
->td_end
, sender
->td_maxend
, sender
->td_maxwin
,
595 receiver
->td_end
, receiver
->td_maxend
, receiver
->td_maxwin
,
598 if (sender
->td_end
== 0) {
600 * Initialize sender data.
602 if (tcph
->syn
&& tcph
->ack
) {
604 * Outgoing SYN-ACK in reply to a SYN.
607 sender
->td_maxend
= end
;
608 sender
->td_maxwin
= (win
== 0 ? 1 : win
);
610 tcp_options(skb
, iph
, tcph
, sender
);
613 * Both sides must send the Window Scale option
614 * to enable window scaling in either direction.
616 if (!(sender
->flags
& IP_CT_TCP_FLAG_WINDOW_SCALE
617 && receiver
->flags
& IP_CT_TCP_FLAG_WINDOW_SCALE
))
619 receiver
->td_scale
= 0;
622 * We are in the middle of a connection,
623 * its history is lost for us.
624 * Let's try to use the data from the packet.
626 sender
->td_end
= end
;
627 sender
->td_maxwin
= (win
== 0 ? 1 : win
);
628 sender
->td_maxend
= end
+ sender
->td_maxwin
;
630 } else if (((state
->state
== TCP_CONNTRACK_SYN_SENT
631 && dir
== IP_CT_DIR_ORIGINAL
)
632 || (state
->state
== TCP_CONNTRACK_SYN_RECV
633 && dir
== IP_CT_DIR_REPLY
))
634 && after(end
, sender
->td_end
)) {
636 * RFC 793: "if a TCP is reinitialized ... then it need
637 * not wait at all; it must only be sure to use sequence
638 * numbers larger than those recently used."
641 sender
->td_maxend
= end
;
642 sender
->td_maxwin
= (win
== 0 ? 1 : win
);
644 tcp_options(skb
, iph
, tcph
, sender
);
649 * If there is no ACK, just pretend it was set and OK.
651 ack
= sack
= receiver
->td_end
;
652 } else if (((tcp_flag_word(tcph
) & (TCP_FLAG_ACK
|TCP_FLAG_RST
)) ==
653 (TCP_FLAG_ACK
|TCP_FLAG_RST
))
656 * Broken TCP stacks, that set ACK in RST packets as well
657 * with zero ack value.
659 ack
= sack
= receiver
->td_end
;
664 || (seq
== 0 && state
->state
== TCP_CONNTRACK_SYN_SENT
)))
666 * Packets contains no data: we assume it is valid
667 * and check the ack value only.
668 * However RST segments are always validated by their
669 * SEQ number, except when seq == 0 (reset sent answering
672 seq
= end
= sender
->td_end
;
674 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
675 "seq=%u ack=%u sack =%u win=%u end=%u\n",
676 NIPQUAD(iph
->saddr
), ntohs(tcph
->source
),
677 NIPQUAD(iph
->daddr
), ntohs(tcph
->dest
),
678 seq
, ack
, sack
, win
, end
);
679 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
680 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
681 sender
->td_end
, sender
->td_maxend
, sender
->td_maxwin
,
683 receiver
->td_end
, receiver
->td_maxend
, receiver
->td_maxwin
,
686 DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
687 before(seq
, sender
->td_maxend
+ 1),
688 after(end
, sender
->td_end
- receiver
->td_maxwin
- 1),
689 before(sack
, receiver
->td_end
+ 1),
690 after(ack
, receiver
->td_end
- MAXACKWINDOW(sender
)));
692 if (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 * Take into account window scaling (RFC 1323).
700 win
<<= sender
->td_scale
;
703 * Update sender data.
705 swin
= win
+ (sack
- ack
);
706 if (sender
->td_maxwin
< swin
)
707 sender
->td_maxwin
= swin
;
708 if (after(end
, sender
->td_end
))
709 sender
->td_end
= end
;
711 * Update receiver data.
713 if (after(end
, sender
->td_maxend
))
714 receiver
->td_maxwin
+= end
- sender
->td_maxend
;
715 if (after(sack
+ win
, receiver
->td_maxend
- 1)) {
716 receiver
->td_maxend
= sack
+ win
;
718 receiver
->td_maxend
++;
722 * Check retransmissions.
724 if (index
== TCP_ACK_SET
) {
725 if (state
->last_dir
== dir
726 && state
->last_seq
== seq
727 && state
->last_ack
== ack
728 && state
->last_end
== end
729 && state
->last_win
== win
)
732 state
->last_dir
= dir
;
733 state
->last_seq
= seq
;
734 state
->last_ack
= ack
;
735 state
->last_end
= end
;
736 state
->last_win
= win
;
743 if (sender
->flags
& IP_CT_TCP_FLAG_BE_LIBERAL
||
744 ip_ct_tcp_be_liberal
)
746 if (!res
&& LOG_INVALID(IPPROTO_TCP
))
747 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
749 before(seq
, sender
->td_maxend
+ 1) ?
750 after(end
, sender
->td_end
- receiver
->td_maxwin
- 1) ?
751 before(sack
, receiver
->td_end
+ 1) ?
752 after(ack
, receiver
->td_end
- MAXACKWINDOW(sender
)) ? "BUG"
753 : "ACK is under the lower bound (possible overly delayed ACK)"
754 : "ACK is over the upper bound (ACKed data not seen yet)"
755 : "SEQ is under the lower bound (already ACKed data retransmitted)"
756 : "SEQ is over the upper bound (over the window of the receiver)");
759 DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
760 "receiver end=%u maxend=%u maxwin=%u\n",
761 res
, sender
->td_end
, sender
->td_maxend
, sender
->td_maxwin
,
762 receiver
->td_end
, receiver
->td_maxend
, receiver
->td_maxwin
);
767 #ifdef CONFIG_IP_NF_NAT_NEEDED
768 /* Update sender->td_end after NAT successfully mangled the packet */
769 void ip_conntrack_tcp_update(struct sk_buff
*skb
,
770 struct ip_conntrack
*conntrack
,
771 enum ip_conntrack_dir dir
)
773 struct iphdr
*iph
= skb
->nh
.iph
;
774 struct tcphdr
*tcph
= (void *)skb
->nh
.iph
+ ip_hdrlen(skb
);
777 struct ip_ct_tcp_state
*sender
= &conntrack
->proto
.tcp
.seen
[dir
];
778 struct ip_ct_tcp_state
*receiver
= &conntrack
->proto
.tcp
.seen
[!dir
];
781 end
= segment_seq_plus_len(ntohl(tcph
->seq
), skb
->len
, iph
, tcph
);
783 write_lock_bh(&tcp_lock
);
785 * We have to worry for the ack in the reply packet only...
787 if (after(end
, conntrack
->proto
.tcp
.seen
[dir
].td_end
))
788 conntrack
->proto
.tcp
.seen
[dir
].td_end
= end
;
789 conntrack
->proto
.tcp
.last_end
= end
;
790 write_unlock_bh(&tcp_lock
);
791 DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
792 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
793 sender
->td_end
, sender
->td_maxend
, sender
->td_maxwin
,
795 receiver
->td_end
, receiver
->td_maxend
, receiver
->td_maxwin
,
810 /* table of valid flag combinations - ECE and CWR are always valid */
811 static const u8 tcp_valid_flags
[(TH_FIN
|TH_SYN
|TH_RST
|TH_PUSH
|TH_ACK
|TH_URG
) + 1] =
814 [TH_SYN
|TH_PUSH
] = 1,
816 [TH_SYN
|TH_PUSH
|TH_URG
] = 1,
818 [TH_SYN
|TH_ACK
|TH_PUSH
] = 1,
821 [TH_RST
|TH_ACK
|TH_PUSH
] = 1,
824 [TH_ACK
|TH_PUSH
] = 1,
826 [TH_ACK
|TH_URG
|TH_PUSH
] = 1,
827 [TH_FIN
|TH_ACK
|TH_PUSH
] = 1,
828 [TH_FIN
|TH_ACK
|TH_URG
] = 1,
829 [TH_FIN
|TH_ACK
|TH_URG
|TH_PUSH
] = 1,
832 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
833 static int tcp_error(struct sk_buff
*skb
,
834 enum ip_conntrack_info
*ctinfo
,
835 unsigned int hooknum
)
837 struct iphdr
*iph
= skb
->nh
.iph
;
838 struct tcphdr _tcph
, *th
;
839 unsigned int tcplen
= skb
->len
- iph
->ihl
* 4;
842 /* Smaller that minimal TCP header? */
843 th
= skb_header_pointer(skb
, iph
->ihl
* 4,
844 sizeof(_tcph
), &_tcph
);
846 if (LOG_INVALID(IPPROTO_TCP
))
847 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
848 "ip_ct_tcp: short packet ");
852 /* Not whole TCP header or malformed packet */
853 if (th
->doff
*4 < sizeof(struct tcphdr
) || tcplen
< th
->doff
*4) {
854 if (LOG_INVALID(IPPROTO_TCP
))
855 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
856 "ip_ct_tcp: truncated/malformed packet ");
860 /* Checksum invalid? Ignore.
861 * We skip checking packets on the outgoing path
862 * because it is assumed to be correct.
864 /* FIXME: Source route IP option packets --RR */
865 if (ip_conntrack_checksum
&& hooknum
== NF_IP_PRE_ROUTING
&&
866 nf_ip_checksum(skb
, hooknum
, iph
->ihl
* 4, IPPROTO_TCP
)) {
867 if (LOG_INVALID(IPPROTO_TCP
))
868 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
869 "ip_ct_tcp: bad TCP checksum ");
873 /* Check TCP flags. */
874 tcpflags
= (((u_int8_t
*)th
)[13] & ~(TH_ECE
|TH_CWR
));
875 if (!tcp_valid_flags
[tcpflags
]) {
876 if (LOG_INVALID(IPPROTO_TCP
))
877 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
878 "ip_ct_tcp: invalid TCP flag combination ");
885 /* Returns verdict for packet, or -1 for invalid. */
886 static int tcp_packet(struct ip_conntrack
*conntrack
,
887 const struct sk_buff
*skb
,
888 enum ip_conntrack_info ctinfo
)
890 enum tcp_conntrack new_state
, old_state
;
891 enum ip_conntrack_dir dir
;
892 struct iphdr
*iph
= skb
->nh
.iph
;
893 struct tcphdr
*th
, _tcph
;
894 unsigned long timeout
;
897 th
= skb_header_pointer(skb
, iph
->ihl
* 4,
898 sizeof(_tcph
), &_tcph
);
901 write_lock_bh(&tcp_lock
);
902 old_state
= conntrack
->proto
.tcp
.state
;
903 dir
= CTINFO2DIR(ctinfo
);
904 index
= get_conntrack_index(th
);
905 new_state
= tcp_conntracks
[dir
][index
][old_state
];
908 case TCP_CONNTRACK_IGNORE
:
912 * b) SYN/ACK in REPLY
913 * c) ACK in reply direction after initial SYN in original.
915 if (index
== TCP_SYNACK_SET
916 && conntrack
->proto
.tcp
.last_index
== TCP_SYN_SET
917 && conntrack
->proto
.tcp
.last_dir
!= dir
918 && ntohl(th
->ack_seq
) ==
919 conntrack
->proto
.tcp
.last_end
) {
920 /* This SYN/ACK acknowledges a SYN that we earlier
921 * ignored as invalid. This means that the client and
922 * the server are both in sync, while the firewall is
923 * not. We kill this session and block the SYN/ACK so
924 * that the client cannot but retransmit its SYN and
925 * thus initiate a clean new session.
927 write_unlock_bh(&tcp_lock
);
928 if (LOG_INVALID(IPPROTO_TCP
))
929 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
,
931 "killing out of sync session ");
932 if (del_timer(&conntrack
->timeout
))
933 conntrack
->timeout
.function((unsigned long)
937 conntrack
->proto
.tcp
.last_index
= index
;
938 conntrack
->proto
.tcp
.last_dir
= dir
;
939 conntrack
->proto
.tcp
.last_seq
= ntohl(th
->seq
);
940 conntrack
->proto
.tcp
.last_end
=
941 segment_seq_plus_len(ntohl(th
->seq
), skb
->len
, iph
, th
);
943 write_unlock_bh(&tcp_lock
);
944 if (LOG_INVALID(IPPROTO_TCP
))
945 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
946 "ip_ct_tcp: invalid packet ignored ");
948 case TCP_CONNTRACK_MAX
:
950 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
951 dir
, get_conntrack_index(th
),
953 write_unlock_bh(&tcp_lock
);
954 if (LOG_INVALID(IPPROTO_TCP
))
955 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
956 "ip_ct_tcp: invalid state ");
958 case TCP_CONNTRACK_SYN_SENT
:
959 if (old_state
< TCP_CONNTRACK_TIME_WAIT
)
961 if ((conntrack
->proto
.tcp
.seen
[dir
].flags
&
962 IP_CT_TCP_FLAG_CLOSE_INIT
)
963 || after(ntohl(th
->seq
),
964 conntrack
->proto
.tcp
.seen
[dir
].td_end
)) {
965 /* Attempt to reopen a closed connection.
966 * Delete this connection and look up again. */
967 write_unlock_bh(&tcp_lock
);
968 if (del_timer(&conntrack
->timeout
))
969 conntrack
->timeout
.function((unsigned long)
973 write_unlock_bh(&tcp_lock
);
974 if (LOG_INVALID(IPPROTO_TCP
))
975 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
,
976 NULL
, "ip_ct_tcp: invalid SYN");
979 case TCP_CONNTRACK_CLOSE
:
980 if (index
== TCP_RST_SET
981 && ((test_bit(IPS_SEEN_REPLY_BIT
, &conntrack
->status
)
982 && conntrack
->proto
.tcp
.last_index
== TCP_SYN_SET
)
983 || (!test_bit(IPS_ASSURED_BIT
, &conntrack
->status
)
984 && conntrack
->proto
.tcp
.last_index
== TCP_ACK_SET
))
985 && ntohl(th
->ack_seq
) == conntrack
->proto
.tcp
.last_end
) {
986 /* RST sent to invalid SYN or ACK we had let through
987 * at a) and c) above:
989 * a) SYN was in window then
990 * c) we hold a half-open connection.
992 * Delete our connection entry.
993 * We skip window checking, because packet might ACK
994 * segments we ignored. */
997 /* Just fall through */
999 /* Keep compilers happy. */
1003 if (!tcp_in_window(&conntrack
->proto
.tcp
, dir
, index
,
1005 write_unlock_bh(&tcp_lock
);
1009 /* From now on we have got in-window packets */
1010 conntrack
->proto
.tcp
.last_index
= index
;
1012 DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
1013 "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1014 NIPQUAD(iph
->saddr
), ntohs(th
->source
),
1015 NIPQUAD(iph
->daddr
), ntohs(th
->dest
),
1016 (th
->syn
? 1 : 0), (th
->ack
? 1 : 0),
1017 (th
->fin
? 1 : 0), (th
->rst
? 1 : 0),
1018 old_state
, new_state
);
1020 conntrack
->proto
.tcp
.state
= new_state
;
1021 if (old_state
!= new_state
1022 && (new_state
== TCP_CONNTRACK_FIN_WAIT
1023 || new_state
== TCP_CONNTRACK_CLOSE
))
1024 conntrack
->proto
.tcp
.seen
[dir
].flags
|= IP_CT_TCP_FLAG_CLOSE_INIT
;
1025 timeout
= conntrack
->proto
.tcp
.retrans
>= ip_ct_tcp_max_retrans
1026 && *tcp_timeouts
[new_state
] > ip_ct_tcp_timeout_max_retrans
1027 ? ip_ct_tcp_timeout_max_retrans
: *tcp_timeouts
[new_state
];
1028 write_unlock_bh(&tcp_lock
);
1030 ip_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE
, skb
);
1031 if (new_state
!= old_state
)
1032 ip_conntrack_event_cache(IPCT_PROTOINFO
, skb
);
1034 if (!test_bit(IPS_SEEN_REPLY_BIT
, &conntrack
->status
)) {
1035 /* If only reply is a RST, we can consider ourselves not to
1036 have an established connection: this is a fairly common
1037 problem case, so we can delete the conntrack
1038 immediately. --RR */
1040 if (del_timer(&conntrack
->timeout
))
1041 conntrack
->timeout
.function((unsigned long)
1045 } else if (!test_bit(IPS_ASSURED_BIT
, &conntrack
->status
)
1046 && (old_state
== TCP_CONNTRACK_SYN_RECV
1047 || old_state
== TCP_CONNTRACK_ESTABLISHED
)
1048 && new_state
== TCP_CONNTRACK_ESTABLISHED
) {
1049 /* Set ASSURED if we see see valid ack in ESTABLISHED
1050 after SYN_RECV or a valid answer for a picked up
1052 set_bit(IPS_ASSURED_BIT
, &conntrack
->status
);
1053 ip_conntrack_event_cache(IPCT_STATUS
, skb
);
1055 ip_ct_refresh_acct(conntrack
, ctinfo
, skb
, timeout
);
1060 /* Called when a new connection for this protocol found. */
1061 static int tcp_new(struct ip_conntrack
*conntrack
,
1062 const struct sk_buff
*skb
)
1064 enum tcp_conntrack new_state
;
1065 struct iphdr
*iph
= skb
->nh
.iph
;
1066 struct tcphdr
*th
, _tcph
;
1068 struct ip_ct_tcp_state
*sender
= &conntrack
->proto
.tcp
.seen
[0];
1069 struct ip_ct_tcp_state
*receiver
= &conntrack
->proto
.tcp
.seen
[1];
1072 th
= skb_header_pointer(skb
, iph
->ihl
* 4,
1073 sizeof(_tcph
), &_tcph
);
1076 /* Don't need lock here: this conntrack not in circulation yet */
1078 = tcp_conntracks
[0][get_conntrack_index(th
)]
1079 [TCP_CONNTRACK_NONE
];
1081 /* Invalid: delete conntrack */
1082 if (new_state
>= TCP_CONNTRACK_MAX
) {
1083 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1087 if (new_state
== TCP_CONNTRACK_SYN_SENT
) {
1089 conntrack
->proto
.tcp
.seen
[0].td_end
=
1090 segment_seq_plus_len(ntohl(th
->seq
), skb
->len
,
1092 conntrack
->proto
.tcp
.seen
[0].td_maxwin
= ntohs(th
->window
);
1093 if (conntrack
->proto
.tcp
.seen
[0].td_maxwin
== 0)
1094 conntrack
->proto
.tcp
.seen
[0].td_maxwin
= 1;
1095 conntrack
->proto
.tcp
.seen
[0].td_maxend
=
1096 conntrack
->proto
.tcp
.seen
[0].td_end
;
1098 tcp_options(skb
, iph
, th
, &conntrack
->proto
.tcp
.seen
[0]);
1099 conntrack
->proto
.tcp
.seen
[1].flags
= 0;
1100 } else if (ip_ct_tcp_loose
== 0) {
1101 /* Don't try to pick up connections. */
1105 * We are in the middle of a connection,
1106 * its history is lost for us.
1107 * Let's try to use the data from the packet.
1109 conntrack
->proto
.tcp
.seen
[0].td_end
=
1110 segment_seq_plus_len(ntohl(th
->seq
), skb
->len
,
1112 conntrack
->proto
.tcp
.seen
[0].td_maxwin
= ntohs(th
->window
);
1113 if (conntrack
->proto
.tcp
.seen
[0].td_maxwin
== 0)
1114 conntrack
->proto
.tcp
.seen
[0].td_maxwin
= 1;
1115 conntrack
->proto
.tcp
.seen
[0].td_maxend
=
1116 conntrack
->proto
.tcp
.seen
[0].td_end
+
1117 conntrack
->proto
.tcp
.seen
[0].td_maxwin
;
1118 conntrack
->proto
.tcp
.seen
[0].td_scale
= 0;
1120 /* We assume SACK and liberal window checking to handle
1122 conntrack
->proto
.tcp
.seen
[0].flags
=
1123 conntrack
->proto
.tcp
.seen
[1].flags
= IP_CT_TCP_FLAG_SACK_PERM
|
1124 IP_CT_TCP_FLAG_BE_LIBERAL
;
1127 conntrack
->proto
.tcp
.seen
[1].td_end
= 0;
1128 conntrack
->proto
.tcp
.seen
[1].td_maxend
= 0;
1129 conntrack
->proto
.tcp
.seen
[1].td_maxwin
= 1;
1130 conntrack
->proto
.tcp
.seen
[1].td_scale
= 0;
1132 /* tcp_packet will set them */
1133 conntrack
->proto
.tcp
.state
= TCP_CONNTRACK_NONE
;
1134 conntrack
->proto
.tcp
.last_index
= TCP_NONE_SET
;
1136 DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1137 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1138 sender
->td_end
, sender
->td_maxend
, sender
->td_maxwin
,
1140 receiver
->td_end
, receiver
->td_maxend
, receiver
->td_maxwin
,
1141 receiver
->td_scale
);
1145 struct ip_conntrack_protocol ip_conntrack_protocol_tcp
=
1147 .proto
= IPPROTO_TCP
,
1149 .pkt_to_tuple
= tcp_pkt_to_tuple
,
1150 .invert_tuple
= tcp_invert_tuple
,
1151 .print_tuple
= tcp_print_tuple
,
1152 .print_conntrack
= tcp_print_conntrack
,
1153 .packet
= tcp_packet
,
1156 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
1157 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
1158 .to_nfattr
= tcp_to_nfattr
,
1159 .from_nfattr
= nfattr_to_tcp
,
1160 .tuple_to_nfattr
= ip_ct_port_tuple_to_nfattr
,
1161 .nfattr_to_tuple
= ip_ct_port_nfattr_to_tuple
,