Linux-2.6.12-rc2
[sfrench/cifs-2.6.git] / net / ipv4 / netfilter / ip_conntrack_proto_tcp.c
1 /* (C) 1999-2001 Paul `Rusty' Russell
2  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
3  *
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.
7  *
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
13  *
14  * Willy Tarreau:
15  *      - State table bugfixes
16  *      - More robust state changes
17  *      - Tuning timer parameters
18  *
19  * version 2.2
20  */
21
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>
28 #include <linux/in.h>
29 #include <linux/ip.h>
30 #include <linux/tcp.h>
31 #include <linux/spinlock.h>
32
33 #include <net/tcp.h>
34
35 #include <linux/netfilter.h>
36 #include <linux/netfilter_ipv4.h>
37 #include <linux/netfilter_ipv4/ip_conntrack.h>
38 #include <linux/netfilter_ipv4/ip_conntrack_protocol.h>
39 #include <linux/netfilter_ipv4/lockhelp.h>
40
41 #if 0
42 #define DEBUGP printk
43 #define DEBUGP_VARS
44 #else
45 #define DEBUGP(format, args...)
46 #endif
47
48 /* Protects conntrack->proto.tcp */
49 static DECLARE_RWLOCK(tcp_lock);
50
51 /* "Be conservative in what you do, 
52     be liberal in what you accept from others." 
53     If it's non-zero, we mark only out of window RST segments as INVALID. */
54 int ip_ct_tcp_be_liberal = 0;
55
56 /* When connection is picked up from the middle, how many packets are required
57    to pass in each direction when we assume we are in sync - if any side uses
58    window scaling, we lost the game. 
59    If it is set to zero, we disable picking up already established 
60    connections. */
61 int ip_ct_tcp_loose = 3;
62
63 /* Max number of the retransmitted packets without receiving an (acceptable) 
64    ACK from the destination. If this number is reached, a shorter timer 
65    will be started. */
66 int ip_ct_tcp_max_retrans = 3;
67
68   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
69      closely.  They're more complex. --RR */
70
71 static const char *tcp_conntrack_names[] = {
72         "NONE",
73         "SYN_SENT",
74         "SYN_RECV",
75         "ESTABLISHED",
76         "FIN_WAIT",
77         "CLOSE_WAIT",
78         "LAST_ACK",
79         "TIME_WAIT",
80         "CLOSE",
81         "LISTEN"
82 };
83   
84 #define SECS * HZ
85 #define MINS * 60 SECS
86 #define HOURS * 60 MINS
87 #define DAYS * 24 HOURS
88
89 unsigned long ip_ct_tcp_timeout_syn_sent =      2 MINS;
90 unsigned long ip_ct_tcp_timeout_syn_recv =     60 SECS;
91 unsigned long ip_ct_tcp_timeout_established =   5 DAYS;
92 unsigned long ip_ct_tcp_timeout_fin_wait =      2 MINS;
93 unsigned long ip_ct_tcp_timeout_close_wait =   60 SECS;
94 unsigned long ip_ct_tcp_timeout_last_ack =     30 SECS;
95 unsigned long ip_ct_tcp_timeout_time_wait =     2 MINS;
96 unsigned long ip_ct_tcp_timeout_close =        10 SECS;
97
98 /* RFC1122 says the R2 limit should be at least 100 seconds.
99    Linux uses 15 packets as limit, which corresponds 
100    to ~13-30min depending on RTO. */
101 unsigned long ip_ct_tcp_timeout_max_retrans =     5 MINS;
102  
103 static unsigned long * tcp_timeouts[]
104 = { NULL,                              /*      TCP_CONNTRACK_NONE */
105     &ip_ct_tcp_timeout_syn_sent,       /*      TCP_CONNTRACK_SYN_SENT, */
106     &ip_ct_tcp_timeout_syn_recv,       /*      TCP_CONNTRACK_SYN_RECV, */
107     &ip_ct_tcp_timeout_established,    /*      TCP_CONNTRACK_ESTABLISHED,      */
108     &ip_ct_tcp_timeout_fin_wait,       /*      TCP_CONNTRACK_FIN_WAIT, */
109     &ip_ct_tcp_timeout_close_wait,     /*      TCP_CONNTRACK_CLOSE_WAIT,       */
110     &ip_ct_tcp_timeout_last_ack,       /*      TCP_CONNTRACK_LAST_ACK, */
111     &ip_ct_tcp_timeout_time_wait,      /*      TCP_CONNTRACK_TIME_WAIT,        */
112     &ip_ct_tcp_timeout_close,          /*      TCP_CONNTRACK_CLOSE,    */
113     NULL,                              /*      TCP_CONNTRACK_LISTEN */
114  };
115  
116 #define sNO TCP_CONNTRACK_NONE
117 #define sSS TCP_CONNTRACK_SYN_SENT
118 #define sSR TCP_CONNTRACK_SYN_RECV
119 #define sES TCP_CONNTRACK_ESTABLISHED
120 #define sFW TCP_CONNTRACK_FIN_WAIT
121 #define sCW TCP_CONNTRACK_CLOSE_WAIT
122 #define sLA TCP_CONNTRACK_LAST_ACK
123 #define sTW TCP_CONNTRACK_TIME_WAIT
124 #define sCL TCP_CONNTRACK_CLOSE
125 #define sLI TCP_CONNTRACK_LISTEN
126 #define sIV TCP_CONNTRACK_MAX
127 #define sIG TCP_CONNTRACK_IGNORE
128
129 /* What TCP flags are set from RST/SYN/FIN/ACK. */
130 enum tcp_bit_set {
131         TCP_SYN_SET,
132         TCP_SYNACK_SET,
133         TCP_FIN_SET,
134         TCP_ACK_SET,
135         TCP_RST_SET,
136         TCP_NONE_SET,
137 };
138   
139 /*
140  * The TCP state transition table needs a few words...
141  *
142  * We are the man in the middle. All the packets go through us
143  * but might get lost in transit to the destination.
144  * It is assumed that the destinations can't receive segments 
145  * we haven't seen.
146  *
147  * The checked segment is in window, but our windows are *not*
148  * equivalent with the ones of the sender/receiver. We always
149  * try to guess the state of the current sender.
150  *
151  * The meaning of the states are:
152  *
153  * NONE:        initial state
154  * SYN_SENT:    SYN-only packet seen 
155  * SYN_RECV:    SYN-ACK packet seen
156  * ESTABLISHED: ACK packet seen
157  * FIN_WAIT:    FIN packet seen
158  * CLOSE_WAIT:  ACK seen (after FIN) 
159  * LAST_ACK:    FIN seen (after FIN)
160  * TIME_WAIT:   last ACK seen
161  * CLOSE:       closed connection
162  *
163  * LISTEN state is not used.
164  *
165  * Packets marked as IGNORED (sIG):
166  *      if they may be either invalid or valid 
167  *      and the receiver may send back a connection 
168  *      closing RST or a SYN/ACK.
169  *
170  * Packets marked as INVALID (sIV):
171  *      if they are invalid
172  *      or we do not support the request (simultaneous open)
173  */
174 static enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
175         {
176 /* ORIGINAL */
177 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
178 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
179 /*
180  *      sNO -> sSS      Initialize a new connection
181  *      sSS -> sSS      Retransmitted SYN
182  *      sSR -> sIG      Late retransmitted SYN?
183  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
184  *                      are errors. Receiver will reply with RST 
185  *                      and close the connection.
186  *                      Or we are not in sync and hold a dead connection.
187  *      sFW -> sIG
188  *      sCW -> sIG
189  *      sLA -> sIG
190  *      sTW -> sSS      Reopened connection (RFC 1122).
191  *      sCL -> sSS
192  */
193 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
194 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
195 /*
196  * A SYN/ACK from the client is always invalid:
197  *      - either it tries to set up a simultaneous open, which is 
198  *        not supported;
199  *      - or the firewall has just been inserted between the two hosts
200  *        during the session set-up. The SYN will be retransmitted 
201  *        by the true client (or it'll time out).
202  */
203 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
204 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
205 /*
206  *      sNO -> sIV      Too late and no reason to do anything...
207  *      sSS -> sIV      Client migth not send FIN in this state:
208  *                      we enforce waiting for a SYN/ACK reply first.
209  *      sSR -> sFW      Close started.
210  *      sES -> sFW      
211  *      sFW -> sLA      FIN seen in both directions, waiting for
212  *                      the last ACK. 
213  *                      Migth be a retransmitted FIN as well...
214  *      sCW -> sLA
215  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
216  *      sTW -> sTW
217  *      sCL -> sCL
218  */
219 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
220 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
221 /*
222  *      sNO -> sES      Assumed.
223  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
224  *      sSR -> sES      Established state is reached.
225  *      sES -> sES      :-)
226  *      sFW -> sCW      Normal close request answered by ACK.
227  *      sCW -> sCW
228  *      sLA -> sTW      Last ACK detected.
229  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
230  *      sCL -> sCL
231  */
232 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
233 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
234 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
235         },
236         {
237 /* REPLY */
238 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
239 /*syn*/    { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
240 /*
241  *      sNO -> sIV      Never reached.
242  *      sSS -> sIV      Simultaneous open, not supported
243  *      sSR -> sIV      Simultaneous open, not supported.
244  *      sES -> sIV      Server may not initiate a connection.
245  *      sFW -> sIV
246  *      sCW -> sIV
247  *      sLA -> sIV
248  *      sTW -> sIV      Reopened connection, but server may not do it.
249  *      sCL -> sIV
250  */
251 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
252 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
253 /*
254  *      sSS -> sSR      Standard open.
255  *      sSR -> sSR      Retransmitted SYN/ACK.
256  *      sES -> sIG      Late retransmitted SYN/ACK?
257  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
258  *      sCW -> sIG
259  *      sLA -> sIG
260  *      sTW -> sIG
261  *      sCL -> sIG
262  */
263 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
264 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
265 /*
266  *      sSS -> sIV      Server might not send FIN in this state.
267  *      sSR -> sFW      Close started.
268  *      sES -> sFW
269  *      sFW -> sLA      FIN seen in both directions.
270  *      sCW -> sLA
271  *      sLA -> sLA      Retransmitted FIN.
272  *      sTW -> sTW
273  *      sCL -> sCL
274  */
275 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
276 /*ack*/    { sIV, sIV, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
277 /*
278  *      sSS -> sIV      Might be a half-open connection.
279  *      sSR -> sSR      Might answer late resent SYN.
280  *      sES -> sES      :-)
281  *      sFW -> sCW      Normal close request answered by ACK.
282  *      sCW -> sCW
283  *      sLA -> sTW      Last ACK detected.
284  *      sTW -> sTW      Retransmitted last ACK.
285  *      sCL -> sCL
286  */
287 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
288 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
289 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
290         }
291 };
292
293 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
294                             unsigned int dataoff,
295                             struct ip_conntrack_tuple *tuple)
296 {
297         struct tcphdr _hdr, *hp;
298
299         /* Actually only need first 8 bytes. */
300         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
301         if (hp == NULL)
302                 return 0;
303
304         tuple->src.u.tcp.port = hp->source;
305         tuple->dst.u.tcp.port = hp->dest;
306
307         return 1;
308 }
309
310 static int tcp_invert_tuple(struct ip_conntrack_tuple *tuple,
311                             const struct ip_conntrack_tuple *orig)
312 {
313         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
314         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
315         return 1;
316 }
317
318 /* Print out the per-protocol part of the tuple. */
319 static int tcp_print_tuple(struct seq_file *s,
320                            const struct ip_conntrack_tuple *tuple)
321 {
322         return seq_printf(s, "sport=%hu dport=%hu ",
323                           ntohs(tuple->src.u.tcp.port),
324                           ntohs(tuple->dst.u.tcp.port));
325 }
326
327 /* Print out the private part of the conntrack. */
328 static int tcp_print_conntrack(struct seq_file *s,
329                                const struct ip_conntrack *conntrack)
330 {
331         enum tcp_conntrack state;
332
333         READ_LOCK(&tcp_lock);
334         state = conntrack->proto.tcp.state;
335         READ_UNLOCK(&tcp_lock);
336
337         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
338 }
339
340 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
341 {
342         if (tcph->rst) return TCP_RST_SET;
343         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
344         else if (tcph->fin) return TCP_FIN_SET;
345         else if (tcph->ack) return TCP_ACK_SET;
346         else return TCP_NONE_SET;
347 }
348
349 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
350    in IP Filter' by Guido van Rooij.
351    
352    http://www.nluug.nl/events/sane2000/papers.html
353    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
354    
355    The boundaries and the conditions are changed according to RFC793:
356    the packet must intersect the window (i.e. segments may be
357    after the right or before the left edge) and thus receivers may ACK
358    segments after the right edge of the window.
359
360         td_maxend = max(sack + max(win,1)) seen in reply packets
361         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
362         td_maxwin += seq + len - sender.td_maxend
363                         if seq + len > sender.td_maxend
364         td_end    = max(seq + len) seen in sent packets
365    
366    I.   Upper bound for valid data:     seq <= sender.td_maxend
367    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
368    III. Upper bound for valid ack:      sack <= receiver.td_end
369    IV.  Lower bound for valid ack:      ack >= receiver.td_end - MAXACKWINDOW
370         
371    where sack is the highest right edge of sack block found in the packet.
372         
373    The upper bound limit for a valid ack is not ignored - 
374    we doesn't have to deal with fragments. 
375 */
376
377 static inline __u32 segment_seq_plus_len(__u32 seq,
378                                          size_t len,
379                                          struct iphdr *iph,
380                                          struct tcphdr *tcph)
381 {
382         return (seq + len - (iph->ihl + tcph->doff)*4
383                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
384 }
385   
386 /* Fixme: what about big packets? */
387 #define MAXACKWINCONST                  66000
388 #define MAXACKWINDOW(sender)                                            \
389         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
390                                               : MAXACKWINCONST)
391   
392 /*
393  * Simplified tcp_parse_options routine from tcp_input.c
394  */
395 static void tcp_options(const struct sk_buff *skb,
396                         struct iphdr *iph,
397                         struct tcphdr *tcph, 
398                         struct ip_ct_tcp_state *state)
399 {
400         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
401         unsigned char *ptr;
402         int length = (tcph->doff*4) - sizeof(struct tcphdr);
403         
404         if (!length)
405                 return;
406
407         ptr = skb_header_pointer(skb,
408                                  (iph->ihl * 4) + sizeof(struct tcphdr),
409                                  length, buff);
410         BUG_ON(ptr == NULL);
411
412         state->td_scale = 
413         state->flags = 0;
414         
415         while (length > 0) {
416                 int opcode=*ptr++;
417                 int opsize;
418                 
419                 switch (opcode) {
420                 case TCPOPT_EOL:
421                         return;
422                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
423                         length--;
424                         continue;
425                 default:
426                         opsize=*ptr++;
427                         if (opsize < 2) /* "silly options" */
428                                 return;
429                         if (opsize > length)
430                                 break;  /* don't parse partial options */
431
432                         if (opcode == TCPOPT_SACK_PERM 
433                             && opsize == TCPOLEN_SACK_PERM)
434                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
435                         else if (opcode == TCPOPT_WINDOW
436                                  && opsize == TCPOLEN_WINDOW) {
437                                 state->td_scale = *(u_int8_t *)ptr;
438                                 
439                                 if (state->td_scale > 14) {
440                                         /* See RFC1323 */
441                                         state->td_scale = 14;
442                                 }
443                                 state->flags |=
444                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
445                         }
446                         ptr += opsize - 2;
447                         length -= opsize;
448                 }
449         }
450 }
451
452 static void tcp_sack(const struct sk_buff *skb,
453                      struct iphdr *iph,
454                      struct tcphdr *tcph,
455                      __u32 *sack)
456 {
457         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
458         unsigned char *ptr;
459         int length = (tcph->doff*4) - sizeof(struct tcphdr);
460         __u32 tmp;
461
462         if (!length)
463                 return;
464
465         ptr = skb_header_pointer(skb,
466                                  (iph->ihl * 4) + sizeof(struct tcphdr),
467                                  length, buff);
468         BUG_ON(ptr == NULL);
469
470         /* Fast path for timestamp-only option */
471         if (length == TCPOLEN_TSTAMP_ALIGNED*4
472             && *(__u32 *)ptr ==
473                 __constant_ntohl((TCPOPT_NOP << 24) 
474                                  | (TCPOPT_NOP << 16)
475                                  | (TCPOPT_TIMESTAMP << 8)
476                                  | TCPOLEN_TIMESTAMP))
477                 return;
478                 
479         while (length > 0) {
480                 int opcode=*ptr++;
481                 int opsize, i;
482                 
483                 switch (opcode) {
484                 case TCPOPT_EOL:
485                         return;
486                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
487                         length--;
488                         continue;
489                 default:
490                         opsize=*ptr++;
491                         if (opsize < 2) /* "silly options" */
492                                 return;
493                         if (opsize > length)
494                                 break;  /* don't parse partial options */
495
496                         if (opcode == TCPOPT_SACK 
497                             && opsize >= (TCPOLEN_SACK_BASE 
498                                           + TCPOLEN_SACK_PERBLOCK)
499                             && !((opsize - TCPOLEN_SACK_BASE) 
500                                  % TCPOLEN_SACK_PERBLOCK)) {
501                                 for (i = 0;
502                                      i < (opsize - TCPOLEN_SACK_BASE);
503                                      i += TCPOLEN_SACK_PERBLOCK) {
504                                         tmp = ntohl(*((u_int32_t *)(ptr+i)+1));
505                                         
506                                         if (after(tmp, *sack))
507                                                 *sack = tmp;
508                                 }
509                                 return;
510                         }
511                         ptr += opsize - 2;
512                         length -= opsize;
513                 }
514         }
515 }
516
517 static int tcp_in_window(struct ip_ct_tcp *state, 
518                          enum ip_conntrack_dir dir,
519                          unsigned int index,
520                          const struct sk_buff *skb,
521                          struct iphdr *iph,
522                          struct tcphdr *tcph)
523 {
524         struct ip_ct_tcp_state *sender = &state->seen[dir];
525         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
526         __u32 seq, ack, sack, end, win, swin;
527         int res;
528         
529         /*
530          * Get the required data from the packet.
531          */
532         seq = ntohl(tcph->seq);
533         ack = sack = ntohl(tcph->ack_seq);
534         win = ntohs(tcph->window);
535         end = segment_seq_plus_len(seq, skb->len, iph, tcph);
536         
537         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
538                 tcp_sack(skb, iph, tcph, &sack);
539                 
540         DEBUGP("tcp_in_window: START\n");
541         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
542                "seq=%u ack=%u sack=%u win=%u end=%u\n",
543                 NIPQUAD(iph->saddr), ntohs(tcph->source), 
544                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
545                 seq, ack, sack, win, end);
546         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
547                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
548                 sender->td_end, sender->td_maxend, sender->td_maxwin,
549                 sender->td_scale, 
550                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin, 
551                 receiver->td_scale);
552                 
553         if (sender->td_end == 0) {
554                 /*
555                  * Initialize sender data.
556                  */
557                 if (tcph->syn && tcph->ack) {
558                         /*
559                          * Outgoing SYN-ACK in reply to a SYN.
560                          */
561                         sender->td_end = 
562                         sender->td_maxend = end;
563                         sender->td_maxwin = (win == 0 ? 1 : win);
564
565                         tcp_options(skb, iph, tcph, sender);
566                         /* 
567                          * RFC 1323:
568                          * Both sides must send the Window Scale option
569                          * to enable window scaling in either direction.
570                          */
571                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
572                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
573                                 sender->td_scale = 
574                                 receiver->td_scale = 0;
575                 } else {
576                         /*
577                          * We are in the middle of a connection,
578                          * its history is lost for us.
579                          * Let's try to use the data from the packet.
580                          */
581                         sender->td_end = end;
582                         sender->td_maxwin = (win == 0 ? 1 : win);
583                         sender->td_maxend = end + sender->td_maxwin;
584                 }
585         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
586                      && dir == IP_CT_DIR_ORIGINAL)
587                     || (state->state == TCP_CONNTRACK_SYN_RECV
588                         && dir == IP_CT_DIR_REPLY))
589                     && after(end, sender->td_end)) {
590                 /*
591                  * RFC 793: "if a TCP is reinitialized ... then it need
592                  * not wait at all; it must only be sure to use sequence 
593                  * numbers larger than those recently used."
594                  */
595                 sender->td_end =
596                 sender->td_maxend = end;
597                 sender->td_maxwin = (win == 0 ? 1 : win);
598
599                 tcp_options(skb, iph, tcph, sender);
600         }
601         
602         if (!(tcph->ack)) {
603                 /*
604                  * If there is no ACK, just pretend it was set and OK.
605                  */
606                 ack = sack = receiver->td_end;
607         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) == 
608                     (TCP_FLAG_ACK|TCP_FLAG_RST)) 
609                    && (ack == 0)) {
610                 /*
611                  * Broken TCP stacks, that set ACK in RST packets as well
612                  * with zero ack value.
613                  */
614                 ack = sack = receiver->td_end;
615         }
616
617         if (seq == end
618             && (!tcph->rst 
619                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
620                 /*
621                  * Packets contains no data: we assume it is valid
622                  * and check the ack value only.
623                  * However RST segments are always validated by their
624                  * SEQ number, except when seq == 0 (reset sent answering
625                  * SYN.
626                  */
627                 seq = end = sender->td_end;
628                 
629         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
630                "seq=%u ack=%u sack =%u win=%u end=%u\n",
631                 NIPQUAD(iph->saddr), ntohs(tcph->source),
632                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
633                 seq, ack, sack, win, end);
634         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
635                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
636                 sender->td_end, sender->td_maxend, sender->td_maxwin,
637                 sender->td_scale, 
638                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
639                 receiver->td_scale);
640         
641         DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
642                 before(seq, sender->td_maxend + 1),
643                 after(end, sender->td_end - receiver->td_maxwin - 1),
644                 before(sack, receiver->td_end + 1),
645                 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
646         
647         if (sender->loose || receiver->loose ||
648             (before(seq, sender->td_maxend + 1) &&
649              after(end, sender->td_end - receiver->td_maxwin - 1) &&
650              before(sack, receiver->td_end + 1) &&
651              after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
652                 /*
653                  * Take into account window scaling (RFC 1323).
654                  */
655                 if (!tcph->syn)
656                         win <<= sender->td_scale;
657                 
658                 /*
659                  * Update sender data.
660                  */
661                 swin = win + (sack - ack);
662                 if (sender->td_maxwin < swin)
663                         sender->td_maxwin = swin;
664                 if (after(end, sender->td_end))
665                         sender->td_end = end;
666                 /*
667                  * Update receiver data.
668                  */
669                 if (after(end, sender->td_maxend))
670                         receiver->td_maxwin += end - sender->td_maxend;
671                 if (after(sack + win, receiver->td_maxend - 1)) {
672                         receiver->td_maxend = sack + win;
673                         if (win == 0)
674                                 receiver->td_maxend++;
675                 }
676
677                 /* 
678                  * Check retransmissions.
679                  */
680                 if (index == TCP_ACK_SET) {
681                         if (state->last_dir == dir
682                             && state->last_seq == seq
683                             && state->last_ack == ack
684                             && state->last_end == end)
685                                 state->retrans++;
686                         else {
687                                 state->last_dir = dir;
688                                 state->last_seq = seq;
689                                 state->last_ack = ack;
690                                 state->last_end = end;
691                                 state->retrans = 0;
692                         }
693                 }
694                 /*
695                  * Close the window of disabled window tracking :-)
696                  */
697                 if (sender->loose)
698                         sender->loose--;
699                 
700                 res = 1;
701         } else {
702                 if (LOG_INVALID(IPPROTO_TCP))
703                         nf_log_packet(PF_INET, 0, skb, NULL, NULL,
704                         "ip_ct_tcp: %s ",
705                         before(seq, sender->td_maxend + 1) ?
706                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
707                         before(sack, receiver->td_end + 1) ?
708                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
709                         : "ACK is under the lower bound (possible overly delayed ACK)"
710                         : "ACK is over the upper bound (ACKed data not seen yet)"
711                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
712                         : "SEQ is over the upper bound (over the window of the receiver)");
713
714                 res = ip_ct_tcp_be_liberal;
715         }
716   
717         DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
718                "receiver end=%u maxend=%u maxwin=%u\n",
719                 res, sender->td_end, sender->td_maxend, sender->td_maxwin, 
720                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
721
722         return res;
723 }
724
725 #ifdef CONFIG_IP_NF_NAT_NEEDED
726 /* Update sender->td_end after NAT successfully mangled the packet */
727 void ip_conntrack_tcp_update(struct sk_buff *skb,
728                              struct ip_conntrack *conntrack, 
729                              enum ip_conntrack_dir dir)
730 {
731         struct iphdr *iph = skb->nh.iph;
732         struct tcphdr *tcph = (void *)skb->nh.iph + skb->nh.iph->ihl*4;
733         __u32 end;
734 #ifdef DEBUGP_VARS
735         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
736         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
737 #endif
738
739         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, iph, tcph);
740         
741         WRITE_LOCK(&tcp_lock);
742         /*
743          * We have to worry for the ack in the reply packet only...
744          */
745         if (after(end, conntrack->proto.tcp.seen[dir].td_end))
746                 conntrack->proto.tcp.seen[dir].td_end = end;
747         conntrack->proto.tcp.last_end = end;
748         WRITE_UNLOCK(&tcp_lock);
749         DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
750                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
751                 sender->td_end, sender->td_maxend, sender->td_maxwin,
752                 sender->td_scale, 
753                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
754                 receiver->td_scale);
755 }
756  
757 #endif
758
759 #define TH_FIN  0x01
760 #define TH_SYN  0x02
761 #define TH_RST  0x04
762 #define TH_PUSH 0x08
763 #define TH_ACK  0x10
764 #define TH_URG  0x20
765 #define TH_ECE  0x40
766 #define TH_CWR  0x80
767
768 /* table of valid flag combinations - ECE and CWR are always valid */
769 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
770 {
771         [TH_SYN]                        = 1,
772         [TH_SYN|TH_ACK]                 = 1,
773         [TH_RST]                        = 1,
774         [TH_RST|TH_ACK]                 = 1,
775         [TH_RST|TH_ACK|TH_PUSH]         = 1,
776         [TH_FIN|TH_ACK]                 = 1,
777         [TH_ACK]                        = 1,
778         [TH_ACK|TH_PUSH]                = 1,
779         [TH_ACK|TH_URG]                 = 1,
780         [TH_ACK|TH_URG|TH_PUSH]         = 1,
781         [TH_FIN|TH_ACK|TH_PUSH]         = 1,
782         [TH_FIN|TH_ACK|TH_URG]          = 1,
783         [TH_FIN|TH_ACK|TH_URG|TH_PUSH]  = 1,
784 };
785
786 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
787 static int tcp_error(struct sk_buff *skb,
788                      enum ip_conntrack_info *ctinfo,
789                      unsigned int hooknum)
790 {
791         struct iphdr *iph = skb->nh.iph;
792         struct tcphdr _tcph, *th;
793         unsigned int tcplen = skb->len - iph->ihl * 4;
794         u_int8_t tcpflags;
795
796         /* Smaller that minimal TCP header? */
797         th = skb_header_pointer(skb, iph->ihl * 4,
798                                 sizeof(_tcph), &_tcph);
799         if (th == NULL) {
800                 if (LOG_INVALID(IPPROTO_TCP))
801                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
802                                 "ip_ct_tcp: short packet ");
803                 return -NF_ACCEPT;
804         }
805   
806         /* Not whole TCP header or malformed packet */
807         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
808                 if (LOG_INVALID(IPPROTO_TCP))
809                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
810                                 "ip_ct_tcp: truncated/malformed packet ");
811                 return -NF_ACCEPT;
812         }
813   
814         /* Checksum invalid? Ignore.
815          * We skip checking packets on the outgoing path
816          * because the semantic of CHECKSUM_HW is different there 
817          * and moreover root might send raw packets.
818          */
819         /* FIXME: Source route IP option packets --RR */
820         if (hooknum == NF_IP_PRE_ROUTING
821             && csum_tcpudp_magic(iph->saddr, iph->daddr, tcplen, IPPROTO_TCP,
822                                  skb->ip_summed == CHECKSUM_HW ? skb->csum
823                                  : skb_checksum(skb, iph->ihl*4, tcplen, 0))) {
824                 if (LOG_INVALID(IPPROTO_TCP))
825                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
826                                   "ip_ct_tcp: bad TCP checksum ");
827                 return -NF_ACCEPT;
828         }
829
830         /* Check TCP flags. */
831         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
832         if (!tcp_valid_flags[tcpflags]) {
833                 if (LOG_INVALID(IPPROTO_TCP))
834                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
835                                   "ip_ct_tcp: invalid TCP flag combination ");
836                 return -NF_ACCEPT;
837         }
838
839         return NF_ACCEPT;
840 }
841
842 /* Returns verdict for packet, or -1 for invalid. */
843 static int tcp_packet(struct ip_conntrack *conntrack,
844                       const struct sk_buff *skb,
845                       enum ip_conntrack_info ctinfo)
846 {
847         enum tcp_conntrack new_state, old_state;
848         enum ip_conntrack_dir dir;
849         struct iphdr *iph = skb->nh.iph;
850         struct tcphdr *th, _tcph;
851         unsigned long timeout;
852         unsigned int index;
853         
854         th = skb_header_pointer(skb, iph->ihl * 4,
855                                 sizeof(_tcph), &_tcph);
856         BUG_ON(th == NULL);
857         
858         WRITE_LOCK(&tcp_lock);
859         old_state = conntrack->proto.tcp.state;
860         dir = CTINFO2DIR(ctinfo);
861         index = get_conntrack_index(th);
862         new_state = tcp_conntracks[dir][index][old_state];
863
864         switch (new_state) {
865         case TCP_CONNTRACK_IGNORE:
866                 /* Either SYN in ORIGINAL
867                  * or SYN/ACK in REPLY. */
868                 if (index == TCP_SYNACK_SET
869                     && conntrack->proto.tcp.last_index == TCP_SYN_SET
870                     && conntrack->proto.tcp.last_dir != dir
871                     && ntohl(th->ack_seq) ==
872                              conntrack->proto.tcp.last_end) {
873                         /* This SYN/ACK acknowledges a SYN that we earlier 
874                          * ignored as invalid. This means that the client and
875                          * the server are both in sync, while the firewall is
876                          * not. We kill this session and block the SYN/ACK so
877                          * that the client cannot but retransmit its SYN and 
878                          * thus initiate a clean new session.
879                          */
880                         WRITE_UNLOCK(&tcp_lock);
881                         if (LOG_INVALID(IPPROTO_TCP))
882                                 nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
883                                           "ip_ct_tcp: killing out of sync session ");
884                         if (del_timer(&conntrack->timeout))
885                                 conntrack->timeout.function((unsigned long)
886                                                             conntrack);
887                         return -NF_DROP;
888                 }
889                 conntrack->proto.tcp.last_index = index;
890                 conntrack->proto.tcp.last_dir = dir;
891                 conntrack->proto.tcp.last_seq = ntohl(th->seq);
892                 conntrack->proto.tcp.last_end = 
893                     segment_seq_plus_len(ntohl(th->seq), skb->len, iph, th);
894                 
895                 WRITE_UNLOCK(&tcp_lock);
896                 if (LOG_INVALID(IPPROTO_TCP))
897                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
898                                   "ip_ct_tcp: invalid packet ignored ");
899                 return NF_ACCEPT;
900         case TCP_CONNTRACK_MAX:
901                 /* Invalid packet */
902                 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
903                        dir, get_conntrack_index(th),
904                        old_state);
905                 WRITE_UNLOCK(&tcp_lock);
906                 if (LOG_INVALID(IPPROTO_TCP))
907                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
908                                   "ip_ct_tcp: invalid state ");
909                 return -NF_ACCEPT;
910         case TCP_CONNTRACK_SYN_SENT:
911                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
912                         break;
913                 if ((conntrack->proto.tcp.seen[dir].flags &
914                          IP_CT_TCP_FLAG_CLOSE_INIT)
915                     || after(ntohl(th->seq),
916                              conntrack->proto.tcp.seen[dir].td_end)) {  
917                         /* Attempt to reopen a closed connection.
918                         * Delete this connection and look up again. */
919                         WRITE_UNLOCK(&tcp_lock);
920                         if (del_timer(&conntrack->timeout))
921                                 conntrack->timeout.function((unsigned long)
922                                                             conntrack);
923                         return -NF_REPEAT;
924                 } else {
925                         WRITE_UNLOCK(&tcp_lock);
926                         if (LOG_INVALID(IPPROTO_TCP))
927                                 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
928                                               "ip_ct_tcp: invalid SYN");
929                         return -NF_ACCEPT;
930                 }
931         case TCP_CONNTRACK_CLOSE:
932                 if (index == TCP_RST_SET
933                     && test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
934                     && conntrack->proto.tcp.last_index == TCP_SYN_SET
935                     && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
936                         /* RST sent to invalid SYN we had let trough
937                          * SYN was in window then, tear down connection.
938                          * We skip window checking, because packet might ACK
939                          * segments we ignored in the SYN. */
940                         goto in_window;
941                 }
942                 /* Just fall trough */
943         default:
944                 /* Keep compilers happy. */
945                 break;
946         }
947
948         if (!tcp_in_window(&conntrack->proto.tcp, dir, index, 
949                            skb, iph, th)) {
950                 WRITE_UNLOCK(&tcp_lock);
951                 return -NF_ACCEPT;
952         }
953     in_window:
954         /* From now on we have got in-window packets */ 
955         conntrack->proto.tcp.last_index = index;
956
957         DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
958                "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
959                 NIPQUAD(iph->saddr), ntohs(th->source),
960                 NIPQUAD(iph->daddr), ntohs(th->dest),
961                 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
962                 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
963                 old_state, new_state);
964
965         conntrack->proto.tcp.state = new_state;
966         if (old_state != new_state 
967             && (new_state == TCP_CONNTRACK_FIN_WAIT
968                 || new_state == TCP_CONNTRACK_CLOSE))
969                 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
970         timeout = conntrack->proto.tcp.retrans >= ip_ct_tcp_max_retrans
971                   && *tcp_timeouts[new_state] > ip_ct_tcp_timeout_max_retrans
972                   ? ip_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
973         WRITE_UNLOCK(&tcp_lock);
974
975         if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
976                 /* If only reply is a RST, we can consider ourselves not to
977                    have an established connection: this is a fairly common
978                    problem case, so we can delete the conntrack
979                    immediately.  --RR */
980                 if (th->rst) {
981                         if (del_timer(&conntrack->timeout))
982                                 conntrack->timeout.function((unsigned long)
983                                                             conntrack);
984                         return NF_ACCEPT;
985                 }
986         } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
987                    && (old_state == TCP_CONNTRACK_SYN_RECV
988                        || old_state == TCP_CONNTRACK_ESTABLISHED)
989                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
990                 /* Set ASSURED if we see see valid ack in ESTABLISHED 
991                    after SYN_RECV or a valid answer for a picked up 
992                    connection. */
993                         set_bit(IPS_ASSURED_BIT, &conntrack->status);
994         }
995         ip_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
996
997         return NF_ACCEPT;
998 }
999  
1000 /* Called when a new connection for this protocol found. */
1001 static int tcp_new(struct ip_conntrack *conntrack,
1002                    const struct sk_buff *skb)
1003 {
1004         enum tcp_conntrack new_state;
1005         struct iphdr *iph = skb->nh.iph;
1006         struct tcphdr *th, _tcph;
1007 #ifdef DEBUGP_VARS
1008         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1009         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1010 #endif
1011
1012         th = skb_header_pointer(skb, iph->ihl * 4,
1013                                 sizeof(_tcph), &_tcph);
1014         BUG_ON(th == NULL);
1015         
1016         /* Don't need lock here: this conntrack not in circulation yet */
1017         new_state
1018                 = tcp_conntracks[0][get_conntrack_index(th)]
1019                 [TCP_CONNTRACK_NONE];
1020
1021         /* Invalid: delete conntrack */
1022         if (new_state >= TCP_CONNTRACK_MAX) {
1023                 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1024                 return 0;
1025         }
1026
1027         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1028                 /* SYN packet */
1029                 conntrack->proto.tcp.seen[0].td_end =
1030                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1031                                              iph, th);
1032                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1033                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1034                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1035                 conntrack->proto.tcp.seen[0].td_maxend =
1036                         conntrack->proto.tcp.seen[0].td_end;
1037
1038                 tcp_options(skb, iph, th, &conntrack->proto.tcp.seen[0]);
1039                 conntrack->proto.tcp.seen[1].flags = 0;
1040                 conntrack->proto.tcp.seen[0].loose = 
1041                 conntrack->proto.tcp.seen[1].loose = 0;
1042         } else if (ip_ct_tcp_loose == 0) {
1043                 /* Don't try to pick up connections. */
1044                 return 0;
1045         } else {
1046                 /*
1047                  * We are in the middle of a connection,
1048                  * its history is lost for us.
1049                  * Let's try to use the data from the packet.
1050                  */
1051                 conntrack->proto.tcp.seen[0].td_end =
1052                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1053                                              iph, th);
1054                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1055                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1056                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1057                 conntrack->proto.tcp.seen[0].td_maxend =
1058                         conntrack->proto.tcp.seen[0].td_end + 
1059                         conntrack->proto.tcp.seen[0].td_maxwin;
1060                 conntrack->proto.tcp.seen[0].td_scale = 0;
1061
1062                 /* We assume SACK. Should we assume window scaling too? */
1063                 conntrack->proto.tcp.seen[0].flags =
1064                 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1065                 conntrack->proto.tcp.seen[0].loose = 
1066                 conntrack->proto.tcp.seen[1].loose = ip_ct_tcp_loose;
1067         }
1068     
1069         conntrack->proto.tcp.seen[1].td_end = 0;
1070         conntrack->proto.tcp.seen[1].td_maxend = 0;
1071         conntrack->proto.tcp.seen[1].td_maxwin = 1;
1072         conntrack->proto.tcp.seen[1].td_scale = 0;      
1073
1074         /* tcp_packet will set them */
1075         conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1076         conntrack->proto.tcp.last_index = TCP_NONE_SET;
1077          
1078         DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1079                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1080                 sender->td_end, sender->td_maxend, sender->td_maxwin,
1081                 sender->td_scale, 
1082                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1083                 receiver->td_scale);
1084         return 1;
1085 }
1086   
1087 struct ip_conntrack_protocol ip_conntrack_protocol_tcp =
1088 {
1089         .proto                  = IPPROTO_TCP,
1090         .name                   = "tcp",
1091         .pkt_to_tuple           = tcp_pkt_to_tuple,
1092         .invert_tuple           = tcp_invert_tuple,
1093         .print_tuple            = tcp_print_tuple,
1094         .print_conntrack        = tcp_print_conntrack,
1095         .packet                 = tcp_packet,
1096         .new                    = tcp_new,
1097         .error                  = tcp_error,
1098 };