1 /* (C) 1999-2001 Paul `Rusty' Russell
2 * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>:
9 * - Real stateful connection tracking
10 * - Modified state transitions table
11 * - Window scaling support added
12 * - SACK support added
15 * - State table bugfixes
16 * - More robust state changes
17 * - Tuning timer parameters
22 #include <linux/config.h>
23 #include <linux/types.h>
24 #include <linux/sched.h>
25 #include <linux/timer.h>
26 #include <linux/netfilter.h>
27 #include <linux/module.h>
30 #include <linux/tcp.h>
31 #include <linux/spinlock.h>
35 #include <linux/netfilter_ipv4.h>
36 #include <linux/netfilter_ipv4/ip_conntrack.h>
37 #include <linux/netfilter_ipv4/ip_conntrack_protocol.h>
43 #define DEBUGP(format, args...)
46 /* Protects conntrack->proto.tcp */
47 static DEFINE_RWLOCK(tcp_lock
);
49 /* "Be conservative in what you do,
50 be liberal in what you accept from others."
51 If it's non-zero, we mark only out of window RST segments as INVALID. */
52 int ip_ct_tcp_be_liberal
= 0;
54 /* When connection is picked up from the middle, how many packets are required
55 to pass in each direction when we assume we are in sync - if any side uses
56 window scaling, we lost the game.
57 If it is set to zero, we disable picking up already established
59 int ip_ct_tcp_loose
= 3;
61 /* Max number of the retransmitted packets without receiving an (acceptable)
62 ACK from the destination. If this number is reached, a shorter timer
64 int ip_ct_tcp_max_retrans
= 3;
66 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
67 closely. They're more complex. --RR */
69 static const char *tcp_conntrack_names
[] = {
83 #define MINS * 60 SECS
84 #define HOURS * 60 MINS
85 #define DAYS * 24 HOURS
87 unsigned int ip_ct_tcp_timeout_syn_sent
= 2 MINS
;
88 unsigned int ip_ct_tcp_timeout_syn_recv
= 60 SECS
;
89 unsigned int ip_ct_tcp_timeout_established
= 5 DAYS
;
90 unsigned int ip_ct_tcp_timeout_fin_wait
= 2 MINS
;
91 unsigned int ip_ct_tcp_timeout_close_wait
= 60 SECS
;
92 unsigned int ip_ct_tcp_timeout_last_ack
= 30 SECS
;
93 unsigned int ip_ct_tcp_timeout_time_wait
= 2 MINS
;
94 unsigned int ip_ct_tcp_timeout_close
= 10 SECS
;
96 /* RFC1122 says the R2 limit should be at least 100 seconds.
97 Linux uses 15 packets as limit, which corresponds
98 to ~13-30min depending on RTO. */
99 unsigned int ip_ct_tcp_timeout_max_retrans
= 5 MINS
;
101 static const unsigned int * tcp_timeouts
[]
102 = { NULL
, /* TCP_CONNTRACK_NONE */
103 &ip_ct_tcp_timeout_syn_sent
, /* TCP_CONNTRACK_SYN_SENT, */
104 &ip_ct_tcp_timeout_syn_recv
, /* TCP_CONNTRACK_SYN_RECV, */
105 &ip_ct_tcp_timeout_established
, /* TCP_CONNTRACK_ESTABLISHED, */
106 &ip_ct_tcp_timeout_fin_wait
, /* TCP_CONNTRACK_FIN_WAIT, */
107 &ip_ct_tcp_timeout_close_wait
, /* TCP_CONNTRACK_CLOSE_WAIT, */
108 &ip_ct_tcp_timeout_last_ack
, /* TCP_CONNTRACK_LAST_ACK, */
109 &ip_ct_tcp_timeout_time_wait
, /* TCP_CONNTRACK_TIME_WAIT, */
110 &ip_ct_tcp_timeout_close
, /* TCP_CONNTRACK_CLOSE, */
111 NULL
, /* TCP_CONNTRACK_LISTEN */
114 #define sNO TCP_CONNTRACK_NONE
115 #define sSS TCP_CONNTRACK_SYN_SENT
116 #define sSR TCP_CONNTRACK_SYN_RECV
117 #define sES TCP_CONNTRACK_ESTABLISHED
118 #define sFW TCP_CONNTRACK_FIN_WAIT
119 #define sCW TCP_CONNTRACK_CLOSE_WAIT
120 #define sLA TCP_CONNTRACK_LAST_ACK
121 #define sTW TCP_CONNTRACK_TIME_WAIT
122 #define sCL TCP_CONNTRACK_CLOSE
123 #define sLI TCP_CONNTRACK_LISTEN
124 #define sIV TCP_CONNTRACK_MAX
125 #define sIG TCP_CONNTRACK_IGNORE
127 /* What TCP flags are set from RST/SYN/FIN/ACK. */
138 * The TCP state transition table needs a few words...
140 * We are the man in the middle. All the packets go through us
141 * but might get lost in transit to the destination.
142 * It is assumed that the destinations can't receive segments
145 * The checked segment is in window, but our windows are *not*
146 * equivalent with the ones of the sender/receiver. We always
147 * try to guess the state of the current sender.
149 * The meaning of the states are:
151 * NONE: initial state
152 * SYN_SENT: SYN-only packet seen
153 * SYN_RECV: SYN-ACK packet seen
154 * ESTABLISHED: ACK packet seen
155 * FIN_WAIT: FIN packet seen
156 * CLOSE_WAIT: ACK seen (after FIN)
157 * LAST_ACK: FIN seen (after FIN)
158 * TIME_WAIT: last ACK seen
159 * CLOSE: closed connection
161 * LISTEN state is not used.
163 * Packets marked as IGNORED (sIG):
164 * if they may be either invalid or valid
165 * and the receiver may send back a connection
166 * closing RST or a SYN/ACK.
168 * Packets marked as INVALID (sIV):
169 * if they are invalid
170 * or we do not support the request (simultaneous open)
172 static const enum tcp_conntrack tcp_conntracks
[2][6][TCP_CONNTRACK_MAX
] = {
175 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
176 /*syn*/ { sSS
, sSS
, sIG
, sIG
, sIG
, sIG
, sIG
, sSS
, sSS
, sIV
},
178 * sNO -> sSS Initialize a new connection
179 * sSS -> sSS Retransmitted SYN
180 * sSR -> sIG Late retransmitted SYN?
181 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
182 * are errors. Receiver will reply with RST
183 * and close the connection.
184 * Or we are not in sync and hold a dead connection.
188 * sTW -> sSS Reopened connection (RFC 1122).
191 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
192 /*synack*/ { sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
},
194 * A SYN/ACK from the client is always invalid:
195 * - either it tries to set up a simultaneous open, which is
197 * - or the firewall has just been inserted between the two hosts
198 * during the session set-up. The SYN will be retransmitted
199 * by the true client (or it'll time out).
201 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
202 /*fin*/ { sIV
, sIV
, sFW
, sFW
, sLA
, sLA
, sLA
, sTW
, sCL
, sIV
},
204 * sNO -> sIV Too late and no reason to do anything...
205 * sSS -> sIV Client migth not send FIN in this state:
206 * we enforce waiting for a SYN/ACK reply first.
207 * sSR -> sFW Close started.
209 * sFW -> sLA FIN seen in both directions, waiting for
211 * Migth be a retransmitted FIN as well...
213 * sLA -> sLA Retransmitted FIN. Remain in the same state.
217 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
218 /*ack*/ { sES
, sIV
, sES
, sES
, sCW
, sCW
, sTW
, sTW
, sCL
, sIV
},
220 * sNO -> sES Assumed.
221 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
222 * sSR -> sES Established state is reached.
224 * sFW -> sCW Normal close request answered by ACK.
226 * sLA -> sTW Last ACK detected.
227 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
230 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
231 /*rst*/ { sIV
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sIV
},
232 /*none*/ { sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
}
236 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
237 /*syn*/ { sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
},
239 * sNO -> sIV Never reached.
240 * sSS -> sIV Simultaneous open, not supported
241 * sSR -> sIV Simultaneous open, not supported.
242 * sES -> sIV Server may not initiate a connection.
246 * sTW -> sIV Reopened connection, but server may not do it.
249 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
250 /*synack*/ { sIV
, sSR
, sSR
, sIG
, sIG
, sIG
, sIG
, sIG
, sIG
, sIV
},
252 * sSS -> sSR Standard open.
253 * sSR -> sSR Retransmitted SYN/ACK.
254 * sES -> sIG Late retransmitted SYN/ACK?
255 * sFW -> sIG Might be SYN/ACK answering ignored SYN
261 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
262 /*fin*/ { sIV
, sIV
, sFW
, sFW
, sLA
, sLA
, sLA
, sTW
, sCL
, sIV
},
264 * sSS -> sIV Server might not send FIN in this state.
265 * sSR -> sFW Close started.
267 * sFW -> sLA FIN seen in both directions.
269 * sLA -> sLA Retransmitted FIN.
273 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
274 /*ack*/ { sIV
, sIG
, sSR
, sES
, sCW
, sCW
, sTW
, sTW
, sCL
, sIV
},
276 * sSS -> sIG Might be a half-open connection.
277 * sSR -> sSR Might answer late resent SYN.
279 * sFW -> sCW Normal close request answered by ACK.
281 * sLA -> sTW Last ACK detected.
282 * sTW -> sTW Retransmitted last ACK.
285 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI */
286 /*rst*/ { sIV
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sCL
, sIV
},
287 /*none*/ { sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
, sIV
}
291 static int tcp_pkt_to_tuple(const struct sk_buff
*skb
,
292 unsigned int dataoff
,
293 struct ip_conntrack_tuple
*tuple
)
295 struct tcphdr _hdr
, *hp
;
297 /* Actually only need first 8 bytes. */
298 hp
= skb_header_pointer(skb
, dataoff
, 8, &_hdr
);
302 tuple
->src
.u
.tcp
.port
= hp
->source
;
303 tuple
->dst
.u
.tcp
.port
= hp
->dest
;
308 static int tcp_invert_tuple(struct ip_conntrack_tuple
*tuple
,
309 const struct ip_conntrack_tuple
*orig
)
311 tuple
->src
.u
.tcp
.port
= orig
->dst
.u
.tcp
.port
;
312 tuple
->dst
.u
.tcp
.port
= orig
->src
.u
.tcp
.port
;
316 /* Print out the per-protocol part of the tuple. */
317 static int tcp_print_tuple(struct seq_file
*s
,
318 const struct ip_conntrack_tuple
*tuple
)
320 return seq_printf(s
, "sport=%hu dport=%hu ",
321 ntohs(tuple
->src
.u
.tcp
.port
),
322 ntohs(tuple
->dst
.u
.tcp
.port
));
325 /* Print out the private part of the conntrack. */
326 static int tcp_print_conntrack(struct seq_file
*s
,
327 const struct ip_conntrack
*conntrack
)
329 enum tcp_conntrack state
;
331 read_lock_bh(&tcp_lock
);
332 state
= conntrack
->proto
.tcp
.state
;
333 read_unlock_bh(&tcp_lock
);
335 return seq_printf(s
, "%s ", tcp_conntrack_names
[state
]);
338 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
339 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
340 static int tcp_to_nfattr(struct sk_buff
*skb
, struct nfattr
*nfa
,
341 const struct ip_conntrack
*ct
)
343 struct nfattr
*nest_parms
;
345 read_lock_bh(&tcp_lock
);
346 nest_parms
= NFA_NEST(skb
, CTA_PROTOINFO_TCP
);
347 NFA_PUT(skb
, CTA_PROTOINFO_TCP_STATE
, sizeof(u_int8_t
),
348 &ct
->proto
.tcp
.state
);
349 read_unlock_bh(&tcp_lock
);
351 NFA_NEST_END(skb
, nest_parms
);
356 read_unlock_bh(&tcp_lock
);
360 static const size_t cta_min_tcp
[CTA_PROTOINFO_TCP_MAX
] = {
361 [CTA_PROTOINFO_TCP_STATE
-1] = sizeof(u_int8_t
),
364 static int nfattr_to_tcp(struct nfattr
*cda
[], struct ip_conntrack
*ct
)
366 struct nfattr
*attr
= cda
[CTA_PROTOINFO_TCP
-1];
367 struct nfattr
*tb
[CTA_PROTOINFO_TCP_MAX
];
369 /* updates could not contain anything about the private
370 * protocol info, in that case skip the parsing */
374 nfattr_parse_nested(tb
, CTA_PROTOINFO_TCP_MAX
, attr
);
376 if (nfattr_bad_size(tb
, CTA_PROTOINFO_TCP_MAX
, cta_min_tcp
))
379 if (!tb
[CTA_PROTOINFO_TCP_STATE
-1])
382 write_lock_bh(&tcp_lock
);
383 ct
->proto
.tcp
.state
=
384 *(u_int8_t
*)NFA_DATA(tb
[CTA_PROTOINFO_TCP_STATE
-1]);
385 write_unlock_bh(&tcp_lock
);
391 static unsigned int get_conntrack_index(const struct tcphdr
*tcph
)
393 if (tcph
->rst
) return TCP_RST_SET
;
394 else if (tcph
->syn
) return (tcph
->ack
? TCP_SYNACK_SET
: TCP_SYN_SET
);
395 else if (tcph
->fin
) return TCP_FIN_SET
;
396 else if (tcph
->ack
) return TCP_ACK_SET
;
397 else return TCP_NONE_SET
;
400 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
401 in IP Filter' by Guido van Rooij.
403 http://www.nluug.nl/events/sane2000/papers.html
404 http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
406 The boundaries and the conditions are changed according to RFC793:
407 the packet must intersect the window (i.e. segments may be
408 after the right or before the left edge) and thus receivers may ACK
409 segments after the right edge of the window.
411 td_maxend = max(sack + max(win,1)) seen in reply packets
412 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
413 td_maxwin += seq + len - sender.td_maxend
414 if seq + len > sender.td_maxend
415 td_end = max(seq + len) seen in sent packets
417 I. Upper bound for valid data: seq <= sender.td_maxend
418 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
419 III. Upper bound for valid ack: sack <= receiver.td_end
420 IV. Lower bound for valid ack: ack >= receiver.td_end - MAXACKWINDOW
422 where sack is the highest right edge of sack block found in the packet.
424 The upper bound limit for a valid ack is not ignored -
425 we doesn't have to deal with fragments.
428 static inline __u32
segment_seq_plus_len(__u32 seq
,
433 return (seq
+ len
- (iph
->ihl
+ tcph
->doff
)*4
434 + (tcph
->syn
? 1 : 0) + (tcph
->fin
? 1 : 0));
437 /* Fixme: what about big packets? */
438 #define MAXACKWINCONST 66000
439 #define MAXACKWINDOW(sender) \
440 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
444 * Simplified tcp_parse_options routine from tcp_input.c
446 static void tcp_options(const struct sk_buff
*skb
,
449 struct ip_ct_tcp_state
*state
)
451 unsigned char buff
[(15 * 4) - sizeof(struct tcphdr
)];
453 int length
= (tcph
->doff
*4) - sizeof(struct tcphdr
);
458 ptr
= skb_header_pointer(skb
,
459 (iph
->ihl
* 4) + sizeof(struct tcphdr
),
473 case TCPOPT_NOP
: /* Ref: RFC 793 section 3.1 */
478 if (opsize
< 2) /* "silly options" */
481 break; /* don't parse partial options */
483 if (opcode
== TCPOPT_SACK_PERM
484 && opsize
== TCPOLEN_SACK_PERM
)
485 state
->flags
|= IP_CT_TCP_FLAG_SACK_PERM
;
486 else if (opcode
== TCPOPT_WINDOW
487 && opsize
== TCPOLEN_WINDOW
) {
488 state
->td_scale
= *(u_int8_t
*)ptr
;
490 if (state
->td_scale
> 14) {
492 state
->td_scale
= 14;
495 IP_CT_TCP_FLAG_WINDOW_SCALE
;
503 static void tcp_sack(const struct sk_buff
*skb
,
508 unsigned char buff
[(15 * 4) - sizeof(struct tcphdr
)];
510 int length
= (tcph
->doff
*4) - sizeof(struct tcphdr
);
516 ptr
= skb_header_pointer(skb
,
517 (iph
->ihl
* 4) + sizeof(struct tcphdr
),
521 /* Fast path for timestamp-only option */
522 if (length
== TCPOLEN_TSTAMP_ALIGNED
*4
524 __constant_ntohl((TCPOPT_NOP
<< 24)
526 | (TCPOPT_TIMESTAMP
<< 8)
527 | TCPOLEN_TIMESTAMP
))
537 case TCPOPT_NOP
: /* Ref: RFC 793 section 3.1 */
542 if (opsize
< 2) /* "silly options" */
545 break; /* don't parse partial options */
547 if (opcode
== TCPOPT_SACK
548 && opsize
>= (TCPOLEN_SACK_BASE
549 + TCPOLEN_SACK_PERBLOCK
)
550 && !((opsize
- TCPOLEN_SACK_BASE
)
551 % TCPOLEN_SACK_PERBLOCK
)) {
553 i
< (opsize
- TCPOLEN_SACK_BASE
);
554 i
+= TCPOLEN_SACK_PERBLOCK
) {
555 tmp
= ntohl(*((u_int32_t
*)(ptr
+i
)+1));
557 if (after(tmp
, *sack
))
568 static int tcp_in_window(struct ip_ct_tcp
*state
,
569 enum ip_conntrack_dir dir
,
571 const struct sk_buff
*skb
,
575 struct ip_ct_tcp_state
*sender
= &state
->seen
[dir
];
576 struct ip_ct_tcp_state
*receiver
= &state
->seen
[!dir
];
577 __u32 seq
, ack
, sack
, end
, win
, swin
;
581 * Get the required data from the packet.
583 seq
= ntohl(tcph
->seq
);
584 ack
= sack
= ntohl(tcph
->ack_seq
);
585 win
= ntohs(tcph
->window
);
586 end
= segment_seq_plus_len(seq
, skb
->len
, iph
, tcph
);
588 if (receiver
->flags
& IP_CT_TCP_FLAG_SACK_PERM
)
589 tcp_sack(skb
, iph
, tcph
, &sack
);
591 DEBUGP("tcp_in_window: START\n");
592 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
593 "seq=%u ack=%u sack=%u win=%u end=%u\n",
594 NIPQUAD(iph
->saddr
), ntohs(tcph
->source
),
595 NIPQUAD(iph
->daddr
), ntohs(tcph
->dest
),
596 seq
, ack
, sack
, win
, end
);
597 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
598 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
599 sender
->td_end
, sender
->td_maxend
, sender
->td_maxwin
,
601 receiver
->td_end
, receiver
->td_maxend
, receiver
->td_maxwin
,
604 if (sender
->td_end
== 0) {
606 * Initialize sender data.
608 if (tcph
->syn
&& tcph
->ack
) {
610 * Outgoing SYN-ACK in reply to a SYN.
613 sender
->td_maxend
= end
;
614 sender
->td_maxwin
= (win
== 0 ? 1 : win
);
616 tcp_options(skb
, iph
, tcph
, sender
);
619 * Both sides must send the Window Scale option
620 * to enable window scaling in either direction.
622 if (!(sender
->flags
& IP_CT_TCP_FLAG_WINDOW_SCALE
623 && receiver
->flags
& IP_CT_TCP_FLAG_WINDOW_SCALE
))
625 receiver
->td_scale
= 0;
628 * We are in the middle of a connection,
629 * its history is lost for us.
630 * Let's try to use the data from the packet.
632 sender
->td_end
= end
;
633 sender
->td_maxwin
= (win
== 0 ? 1 : win
);
634 sender
->td_maxend
= end
+ sender
->td_maxwin
;
636 } else if (((state
->state
== TCP_CONNTRACK_SYN_SENT
637 && dir
== IP_CT_DIR_ORIGINAL
)
638 || (state
->state
== TCP_CONNTRACK_SYN_RECV
639 && dir
== IP_CT_DIR_REPLY
))
640 && after(end
, sender
->td_end
)) {
642 * RFC 793: "if a TCP is reinitialized ... then it need
643 * not wait at all; it must only be sure to use sequence
644 * numbers larger than those recently used."
647 sender
->td_maxend
= end
;
648 sender
->td_maxwin
= (win
== 0 ? 1 : win
);
650 tcp_options(skb
, iph
, tcph
, sender
);
655 * If there is no ACK, just pretend it was set and OK.
657 ack
= sack
= receiver
->td_end
;
658 } else if (((tcp_flag_word(tcph
) & (TCP_FLAG_ACK
|TCP_FLAG_RST
)) ==
659 (TCP_FLAG_ACK
|TCP_FLAG_RST
))
662 * Broken TCP stacks, that set ACK in RST packets as well
663 * with zero ack value.
665 ack
= sack
= receiver
->td_end
;
670 || (seq
== 0 && state
->state
== TCP_CONNTRACK_SYN_SENT
)))
672 * Packets contains no data: we assume it is valid
673 * and check the ack value only.
674 * However RST segments are always validated by their
675 * SEQ number, except when seq == 0 (reset sent answering
678 seq
= end
= sender
->td_end
;
680 DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
681 "seq=%u ack=%u sack =%u win=%u end=%u\n",
682 NIPQUAD(iph
->saddr
), ntohs(tcph
->source
),
683 NIPQUAD(iph
->daddr
), ntohs(tcph
->dest
),
684 seq
, ack
, sack
, win
, end
);
685 DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
686 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
687 sender
->td_end
, sender
->td_maxend
, sender
->td_maxwin
,
689 receiver
->td_end
, receiver
->td_maxend
, receiver
->td_maxwin
,
692 DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
693 before(seq
, sender
->td_maxend
+ 1),
694 after(end
, sender
->td_end
- receiver
->td_maxwin
- 1),
695 before(sack
, receiver
->td_end
+ 1),
696 after(ack
, receiver
->td_end
- MAXACKWINDOW(sender
)));
698 if (sender
->loose
|| receiver
->loose
||
699 (before(seq
, sender
->td_maxend
+ 1) &&
700 after(end
, sender
->td_end
- receiver
->td_maxwin
- 1) &&
701 before(sack
, receiver
->td_end
+ 1) &&
702 after(ack
, receiver
->td_end
- MAXACKWINDOW(sender
)))) {
704 * Take into account window scaling (RFC 1323).
707 win
<<= sender
->td_scale
;
710 * Update sender data.
712 swin
= win
+ (sack
- ack
);
713 if (sender
->td_maxwin
< swin
)
714 sender
->td_maxwin
= swin
;
715 if (after(end
, sender
->td_end
))
716 sender
->td_end
= end
;
718 * Update receiver data.
720 if (after(end
, sender
->td_maxend
))
721 receiver
->td_maxwin
+= end
- sender
->td_maxend
;
722 if (after(sack
+ win
, receiver
->td_maxend
- 1)) {
723 receiver
->td_maxend
= sack
+ win
;
725 receiver
->td_maxend
++;
729 * Check retransmissions.
731 if (index
== TCP_ACK_SET
) {
732 if (state
->last_dir
== dir
733 && state
->last_seq
== seq
734 && state
->last_ack
== ack
735 && state
->last_end
== end
)
738 state
->last_dir
= dir
;
739 state
->last_seq
= seq
;
740 state
->last_ack
= ack
;
741 state
->last_end
= end
;
746 * Close the window of disabled window tracking :-)
753 if (LOG_INVALID(IPPROTO_TCP
))
754 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
756 before(seq
, sender
->td_maxend
+ 1) ?
757 after(end
, sender
->td_end
- receiver
->td_maxwin
- 1) ?
758 before(sack
, receiver
->td_end
+ 1) ?
759 after(ack
, receiver
->td_end
- MAXACKWINDOW(sender
)) ? "BUG"
760 : "ACK is under the lower bound (possible overly delayed ACK)"
761 : "ACK is over the upper bound (ACKed data not seen yet)"
762 : "SEQ is under the lower bound (already ACKed data retransmitted)"
763 : "SEQ is over the upper bound (over the window of the receiver)");
765 res
= ip_ct_tcp_be_liberal
;
768 DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
769 "receiver end=%u maxend=%u maxwin=%u\n",
770 res
, sender
->td_end
, sender
->td_maxend
, sender
->td_maxwin
,
771 receiver
->td_end
, receiver
->td_maxend
, receiver
->td_maxwin
);
776 #ifdef CONFIG_IP_NF_NAT_NEEDED
777 /* Update sender->td_end after NAT successfully mangled the packet */
778 void ip_conntrack_tcp_update(struct sk_buff
*skb
,
779 struct ip_conntrack
*conntrack
,
780 enum ip_conntrack_dir dir
)
782 struct iphdr
*iph
= skb
->nh
.iph
;
783 struct tcphdr
*tcph
= (void *)skb
->nh
.iph
+ skb
->nh
.iph
->ihl
*4;
786 struct ip_ct_tcp_state
*sender
= &conntrack
->proto
.tcp
.seen
[dir
];
787 struct ip_ct_tcp_state
*receiver
= &conntrack
->proto
.tcp
.seen
[!dir
];
790 end
= segment_seq_plus_len(ntohl(tcph
->seq
), skb
->len
, iph
, tcph
);
792 write_lock_bh(&tcp_lock
);
794 * We have to worry for the ack in the reply packet only...
796 if (after(end
, conntrack
->proto
.tcp
.seen
[dir
].td_end
))
797 conntrack
->proto
.tcp
.seen
[dir
].td_end
= end
;
798 conntrack
->proto
.tcp
.last_end
= end
;
799 write_unlock_bh(&tcp_lock
);
800 DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
801 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
802 sender
->td_end
, sender
->td_maxend
, sender
->td_maxwin
,
804 receiver
->td_end
, receiver
->td_maxend
, receiver
->td_maxwin
,
819 /* table of valid flag combinations - ECE and CWR are always valid */
820 static const u8 tcp_valid_flags
[(TH_FIN
|TH_SYN
|TH_RST
|TH_PUSH
|TH_ACK
|TH_URG
) + 1] =
824 [TH_SYN
|TH_PUSH
] = 1,
825 [TH_SYN
|TH_ACK
|TH_PUSH
] = 1,
828 [TH_RST
|TH_ACK
|TH_PUSH
] = 1,
831 [TH_ACK
|TH_PUSH
] = 1,
833 [TH_ACK
|TH_URG
|TH_PUSH
] = 1,
834 [TH_FIN
|TH_ACK
|TH_PUSH
] = 1,
835 [TH_FIN
|TH_ACK
|TH_URG
] = 1,
836 [TH_FIN
|TH_ACK
|TH_URG
|TH_PUSH
] = 1,
839 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
840 static int tcp_error(struct sk_buff
*skb
,
841 enum ip_conntrack_info
*ctinfo
,
842 unsigned int hooknum
)
844 struct iphdr
*iph
= skb
->nh
.iph
;
845 struct tcphdr _tcph
, *th
;
846 unsigned int tcplen
= skb
->len
- iph
->ihl
* 4;
849 /* Smaller that minimal TCP header? */
850 th
= skb_header_pointer(skb
, iph
->ihl
* 4,
851 sizeof(_tcph
), &_tcph
);
853 if (LOG_INVALID(IPPROTO_TCP
))
854 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
855 "ip_ct_tcp: short packet ");
859 /* Not whole TCP header or malformed packet */
860 if (th
->doff
*4 < sizeof(struct tcphdr
) || tcplen
< th
->doff
*4) {
861 if (LOG_INVALID(IPPROTO_TCP
))
862 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
863 "ip_ct_tcp: truncated/malformed packet ");
867 /* Checksum invalid? Ignore.
868 * We skip checking packets on the outgoing path
869 * because the semantic of CHECKSUM_HW is different there
870 * and moreover root might send raw packets.
872 /* FIXME: Source route IP option packets --RR */
873 if (ip_conntrack_checksum
&& hooknum
== NF_IP_PRE_ROUTING
&&
874 nf_ip_checksum(skb
, hooknum
, iph
->ihl
* 4, IPPROTO_TCP
)) {
875 if (LOG_INVALID(IPPROTO_TCP
))
876 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
877 "ip_ct_tcp: bad TCP checksum ");
881 /* Check TCP flags. */
882 tcpflags
= (((u_int8_t
*)th
)[13] & ~(TH_ECE
|TH_CWR
));
883 if (!tcp_valid_flags
[tcpflags
]) {
884 if (LOG_INVALID(IPPROTO_TCP
))
885 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
886 "ip_ct_tcp: invalid TCP flag combination ");
893 /* Returns verdict for packet, or -1 for invalid. */
894 static int tcp_packet(struct ip_conntrack
*conntrack
,
895 const struct sk_buff
*skb
,
896 enum ip_conntrack_info ctinfo
)
898 enum tcp_conntrack new_state
, old_state
;
899 enum ip_conntrack_dir dir
;
900 struct iphdr
*iph
= skb
->nh
.iph
;
901 struct tcphdr
*th
, _tcph
;
902 unsigned long timeout
;
905 th
= skb_header_pointer(skb
, iph
->ihl
* 4,
906 sizeof(_tcph
), &_tcph
);
909 write_lock_bh(&tcp_lock
);
910 old_state
= conntrack
->proto
.tcp
.state
;
911 dir
= CTINFO2DIR(ctinfo
);
912 index
= get_conntrack_index(th
);
913 new_state
= tcp_conntracks
[dir
][index
][old_state
];
916 case TCP_CONNTRACK_IGNORE
:
920 * b) SYN/ACK in REPLY
921 * c) ACK in reply direction after initial SYN in original.
923 if (index
== TCP_SYNACK_SET
924 && conntrack
->proto
.tcp
.last_index
== TCP_SYN_SET
925 && conntrack
->proto
.tcp
.last_dir
!= dir
926 && ntohl(th
->ack_seq
) ==
927 conntrack
->proto
.tcp
.last_end
) {
928 /* This SYN/ACK acknowledges a SYN that we earlier
929 * ignored as invalid. This means that the client and
930 * the server are both in sync, while the firewall is
931 * not. We kill this session and block the SYN/ACK so
932 * that the client cannot but retransmit its SYN and
933 * thus initiate a clean new session.
935 write_unlock_bh(&tcp_lock
);
936 if (LOG_INVALID(IPPROTO_TCP
))
937 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
,
939 "killing out of sync session ");
940 if (del_timer(&conntrack
->timeout
))
941 conntrack
->timeout
.function((unsigned long)
945 conntrack
->proto
.tcp
.last_index
= index
;
946 conntrack
->proto
.tcp
.last_dir
= dir
;
947 conntrack
->proto
.tcp
.last_seq
= ntohl(th
->seq
);
948 conntrack
->proto
.tcp
.last_end
=
949 segment_seq_plus_len(ntohl(th
->seq
), skb
->len
, iph
, th
);
951 write_unlock_bh(&tcp_lock
);
952 if (LOG_INVALID(IPPROTO_TCP
))
953 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
954 "ip_ct_tcp: invalid packet ignored ");
956 case TCP_CONNTRACK_MAX
:
958 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
959 dir
, get_conntrack_index(th
),
961 write_unlock_bh(&tcp_lock
);
962 if (LOG_INVALID(IPPROTO_TCP
))
963 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
, NULL
,
964 "ip_ct_tcp: invalid state ");
966 case TCP_CONNTRACK_SYN_SENT
:
967 if (old_state
< TCP_CONNTRACK_TIME_WAIT
)
969 if ((conntrack
->proto
.tcp
.seen
[dir
].flags
&
970 IP_CT_TCP_FLAG_CLOSE_INIT
)
971 || after(ntohl(th
->seq
),
972 conntrack
->proto
.tcp
.seen
[dir
].td_end
)) {
973 /* Attempt to reopen a closed connection.
974 * Delete this connection and look up again. */
975 write_unlock_bh(&tcp_lock
);
976 if (del_timer(&conntrack
->timeout
))
977 conntrack
->timeout
.function((unsigned long)
981 write_unlock_bh(&tcp_lock
);
982 if (LOG_INVALID(IPPROTO_TCP
))
983 nf_log_packet(PF_INET
, 0, skb
, NULL
, NULL
,
984 NULL
, "ip_ct_tcp: invalid SYN");
987 case TCP_CONNTRACK_CLOSE
:
988 if (index
== TCP_RST_SET
989 && ((test_bit(IPS_SEEN_REPLY_BIT
, &conntrack
->status
)
990 && conntrack
->proto
.tcp
.last_index
== TCP_SYN_SET
)
991 || (!test_bit(IPS_ASSURED_BIT
, &conntrack
->status
)
992 && conntrack
->proto
.tcp
.last_index
== TCP_ACK_SET
))
993 && ntohl(th
->ack_seq
) == conntrack
->proto
.tcp
.last_end
) {
994 /* RST sent to invalid SYN or ACK we had let through
995 * at a) and c) above:
997 * a) SYN was in window then
998 * c) we hold a half-open connection.
1000 * Delete our connection entry.
1001 * We skip window checking, because packet might ACK
1002 * segments we ignored. */
1005 /* Just fall through */
1007 /* Keep compilers happy. */
1011 if (!tcp_in_window(&conntrack
->proto
.tcp
, dir
, index
,
1013 write_unlock_bh(&tcp_lock
);
1017 /* From now on we have got in-window packets */
1018 conntrack
->proto
.tcp
.last_index
= index
;
1020 DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
1021 "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1022 NIPQUAD(iph
->saddr
), ntohs(th
->source
),
1023 NIPQUAD(iph
->daddr
), ntohs(th
->dest
),
1024 (th
->syn
? 1 : 0), (th
->ack
? 1 : 0),
1025 (th
->fin
? 1 : 0), (th
->rst
? 1 : 0),
1026 old_state
, new_state
);
1028 conntrack
->proto
.tcp
.state
= new_state
;
1029 if (old_state
!= new_state
1030 && (new_state
== TCP_CONNTRACK_FIN_WAIT
1031 || new_state
== TCP_CONNTRACK_CLOSE
))
1032 conntrack
->proto
.tcp
.seen
[dir
].flags
|= IP_CT_TCP_FLAG_CLOSE_INIT
;
1033 timeout
= conntrack
->proto
.tcp
.retrans
>= ip_ct_tcp_max_retrans
1034 && *tcp_timeouts
[new_state
] > ip_ct_tcp_timeout_max_retrans
1035 ? ip_ct_tcp_timeout_max_retrans
: *tcp_timeouts
[new_state
];
1036 write_unlock_bh(&tcp_lock
);
1038 ip_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE
, skb
);
1039 if (new_state
!= old_state
)
1040 ip_conntrack_event_cache(IPCT_PROTOINFO
, skb
);
1042 if (!test_bit(IPS_SEEN_REPLY_BIT
, &conntrack
->status
)) {
1043 /* If only reply is a RST, we can consider ourselves not to
1044 have an established connection: this is a fairly common
1045 problem case, so we can delete the conntrack
1046 immediately. --RR */
1048 if (del_timer(&conntrack
->timeout
))
1049 conntrack
->timeout
.function((unsigned long)
1053 } else if (!test_bit(IPS_ASSURED_BIT
, &conntrack
->status
)
1054 && (old_state
== TCP_CONNTRACK_SYN_RECV
1055 || old_state
== TCP_CONNTRACK_ESTABLISHED
)
1056 && new_state
== TCP_CONNTRACK_ESTABLISHED
) {
1057 /* Set ASSURED if we see see valid ack in ESTABLISHED
1058 after SYN_RECV or a valid answer for a picked up
1060 set_bit(IPS_ASSURED_BIT
, &conntrack
->status
);
1061 ip_conntrack_event_cache(IPCT_STATUS
, skb
);
1063 ip_ct_refresh_acct(conntrack
, ctinfo
, skb
, timeout
);
1068 /* Called when a new connection for this protocol found. */
1069 static int tcp_new(struct ip_conntrack
*conntrack
,
1070 const struct sk_buff
*skb
)
1072 enum tcp_conntrack new_state
;
1073 struct iphdr
*iph
= skb
->nh
.iph
;
1074 struct tcphdr
*th
, _tcph
;
1076 struct ip_ct_tcp_state
*sender
= &conntrack
->proto
.tcp
.seen
[0];
1077 struct ip_ct_tcp_state
*receiver
= &conntrack
->proto
.tcp
.seen
[1];
1080 th
= skb_header_pointer(skb
, iph
->ihl
* 4,
1081 sizeof(_tcph
), &_tcph
);
1084 /* Don't need lock here: this conntrack not in circulation yet */
1086 = tcp_conntracks
[0][get_conntrack_index(th
)]
1087 [TCP_CONNTRACK_NONE
];
1089 /* Invalid: delete conntrack */
1090 if (new_state
>= TCP_CONNTRACK_MAX
) {
1091 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1095 if (new_state
== TCP_CONNTRACK_SYN_SENT
) {
1097 conntrack
->proto
.tcp
.seen
[0].td_end
=
1098 segment_seq_plus_len(ntohl(th
->seq
), skb
->len
,
1100 conntrack
->proto
.tcp
.seen
[0].td_maxwin
= ntohs(th
->window
);
1101 if (conntrack
->proto
.tcp
.seen
[0].td_maxwin
== 0)
1102 conntrack
->proto
.tcp
.seen
[0].td_maxwin
= 1;
1103 conntrack
->proto
.tcp
.seen
[0].td_maxend
=
1104 conntrack
->proto
.tcp
.seen
[0].td_end
;
1106 tcp_options(skb
, iph
, th
, &conntrack
->proto
.tcp
.seen
[0]);
1107 conntrack
->proto
.tcp
.seen
[1].flags
= 0;
1108 conntrack
->proto
.tcp
.seen
[0].loose
=
1109 conntrack
->proto
.tcp
.seen
[1].loose
= 0;
1110 } else if (ip_ct_tcp_loose
== 0) {
1111 /* Don't try to pick up connections. */
1115 * We are in the middle of a connection,
1116 * its history is lost for us.
1117 * Let's try to use the data from the packet.
1119 conntrack
->proto
.tcp
.seen
[0].td_end
=
1120 segment_seq_plus_len(ntohl(th
->seq
), skb
->len
,
1122 conntrack
->proto
.tcp
.seen
[0].td_maxwin
= ntohs(th
->window
);
1123 if (conntrack
->proto
.tcp
.seen
[0].td_maxwin
== 0)
1124 conntrack
->proto
.tcp
.seen
[0].td_maxwin
= 1;
1125 conntrack
->proto
.tcp
.seen
[0].td_maxend
=
1126 conntrack
->proto
.tcp
.seen
[0].td_end
+
1127 conntrack
->proto
.tcp
.seen
[0].td_maxwin
;
1128 conntrack
->proto
.tcp
.seen
[0].td_scale
= 0;
1130 /* We assume SACK. Should we assume window scaling too? */
1131 conntrack
->proto
.tcp
.seen
[0].flags
=
1132 conntrack
->proto
.tcp
.seen
[1].flags
= IP_CT_TCP_FLAG_SACK_PERM
;
1133 conntrack
->proto
.tcp
.seen
[0].loose
=
1134 conntrack
->proto
.tcp
.seen
[1].loose
= ip_ct_tcp_loose
;
1137 conntrack
->proto
.tcp
.seen
[1].td_end
= 0;
1138 conntrack
->proto
.tcp
.seen
[1].td_maxend
= 0;
1139 conntrack
->proto
.tcp
.seen
[1].td_maxwin
= 1;
1140 conntrack
->proto
.tcp
.seen
[1].td_scale
= 0;
1142 /* tcp_packet will set them */
1143 conntrack
->proto
.tcp
.state
= TCP_CONNTRACK_NONE
;
1144 conntrack
->proto
.tcp
.last_index
= TCP_NONE_SET
;
1146 DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1147 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1148 sender
->td_end
, sender
->td_maxend
, sender
->td_maxwin
,
1150 receiver
->td_end
, receiver
->td_maxend
, receiver
->td_maxwin
,
1151 receiver
->td_scale
);
1155 struct ip_conntrack_protocol ip_conntrack_protocol_tcp
=
1157 .proto
= IPPROTO_TCP
,
1159 .pkt_to_tuple
= tcp_pkt_to_tuple
,
1160 .invert_tuple
= tcp_invert_tuple
,
1161 .print_tuple
= tcp_print_tuple
,
1162 .print_conntrack
= tcp_print_conntrack
,
1163 .packet
= tcp_packet
,
1166 #if defined(CONFIG_IP_NF_CONNTRACK_NETLINK) || \
1167 defined(CONFIG_IP_NF_CONNTRACK_NETLINK_MODULE)
1168 .to_nfattr
= tcp_to_nfattr
,
1169 .from_nfattr
= nfattr_to_tcp
,
1170 .tuple_to_nfattr
= ip_ct_port_tuple_to_nfattr
,
1171 .nfattr_to_tuple
= ip_ct_port_nfattr_to_tuple
,