[NETFILTER]: Kill lockhelp.h
[linux-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
40 #if 0
41 #define DEBUGP printk
42 #define DEBUGP_VARS
43 #else
44 #define DEBUGP(format, args...)
45 #endif
46
47 /* Protects conntrack->proto.tcp */
48 static DEFINE_RWLOCK(tcp_lock);
49
50 /* "Be conservative in what you do, 
51     be liberal in what you accept from others." 
52     If it's non-zero, we mark only out of window RST segments as INVALID. */
53 int ip_ct_tcp_be_liberal = 0;
54
55 /* When connection is picked up from the middle, how many packets are required
56    to pass in each direction when we assume we are in sync - if any side uses
57    window scaling, we lost the game. 
58    If it is set to zero, we disable picking up already established 
59    connections. */
60 int ip_ct_tcp_loose = 3;
61
62 /* Max number of the retransmitted packets without receiving an (acceptable) 
63    ACK from the destination. If this number is reached, a shorter timer 
64    will be started. */
65 int ip_ct_tcp_max_retrans = 3;
66
67   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
68      closely.  They're more complex. --RR */
69
70 static const char *tcp_conntrack_names[] = {
71         "NONE",
72         "SYN_SENT",
73         "SYN_RECV",
74         "ESTABLISHED",
75         "FIN_WAIT",
76         "CLOSE_WAIT",
77         "LAST_ACK",
78         "TIME_WAIT",
79         "CLOSE",
80         "LISTEN"
81 };
82   
83 #define SECS * HZ
84 #define MINS * 60 SECS
85 #define HOURS * 60 MINS
86 #define DAYS * 24 HOURS
87
88 unsigned long ip_ct_tcp_timeout_syn_sent =      2 MINS;
89 unsigned long ip_ct_tcp_timeout_syn_recv =     60 SECS;
90 unsigned long ip_ct_tcp_timeout_established =   5 DAYS;
91 unsigned long ip_ct_tcp_timeout_fin_wait =      2 MINS;
92 unsigned long ip_ct_tcp_timeout_close_wait =   60 SECS;
93 unsigned long ip_ct_tcp_timeout_last_ack =     30 SECS;
94 unsigned long ip_ct_tcp_timeout_time_wait =     2 MINS;
95 unsigned long ip_ct_tcp_timeout_close =        10 SECS;
96
97 /* RFC1122 says the R2 limit should be at least 100 seconds.
98    Linux uses 15 packets as limit, which corresponds 
99    to ~13-30min depending on RTO. */
100 unsigned long ip_ct_tcp_timeout_max_retrans =     5 MINS;
101  
102 static unsigned long * tcp_timeouts[]
103 = { NULL,                              /*      TCP_CONNTRACK_NONE */
104     &ip_ct_tcp_timeout_syn_sent,       /*      TCP_CONNTRACK_SYN_SENT, */
105     &ip_ct_tcp_timeout_syn_recv,       /*      TCP_CONNTRACK_SYN_RECV, */
106     &ip_ct_tcp_timeout_established,    /*      TCP_CONNTRACK_ESTABLISHED,      */
107     &ip_ct_tcp_timeout_fin_wait,       /*      TCP_CONNTRACK_FIN_WAIT, */
108     &ip_ct_tcp_timeout_close_wait,     /*      TCP_CONNTRACK_CLOSE_WAIT,       */
109     &ip_ct_tcp_timeout_last_ack,       /*      TCP_CONNTRACK_LAST_ACK, */
110     &ip_ct_tcp_timeout_time_wait,      /*      TCP_CONNTRACK_TIME_WAIT,        */
111     &ip_ct_tcp_timeout_close,          /*      TCP_CONNTRACK_CLOSE,    */
112     NULL,                              /*      TCP_CONNTRACK_LISTEN */
113  };
114  
115 #define sNO TCP_CONNTRACK_NONE
116 #define sSS TCP_CONNTRACK_SYN_SENT
117 #define sSR TCP_CONNTRACK_SYN_RECV
118 #define sES TCP_CONNTRACK_ESTABLISHED
119 #define sFW TCP_CONNTRACK_FIN_WAIT
120 #define sCW TCP_CONNTRACK_CLOSE_WAIT
121 #define sLA TCP_CONNTRACK_LAST_ACK
122 #define sTW TCP_CONNTRACK_TIME_WAIT
123 #define sCL TCP_CONNTRACK_CLOSE
124 #define sLI TCP_CONNTRACK_LISTEN
125 #define sIV TCP_CONNTRACK_MAX
126 #define sIG TCP_CONNTRACK_IGNORE
127
128 /* What TCP flags are set from RST/SYN/FIN/ACK. */
129 enum tcp_bit_set {
130         TCP_SYN_SET,
131         TCP_SYNACK_SET,
132         TCP_FIN_SET,
133         TCP_ACK_SET,
134         TCP_RST_SET,
135         TCP_NONE_SET,
136 };
137   
138 /*
139  * The TCP state transition table needs a few words...
140  *
141  * We are the man in the middle. All the packets go through us
142  * but might get lost in transit to the destination.
143  * It is assumed that the destinations can't receive segments 
144  * we haven't seen.
145  *
146  * The checked segment is in window, but our windows are *not*
147  * equivalent with the ones of the sender/receiver. We always
148  * try to guess the state of the current sender.
149  *
150  * The meaning of the states are:
151  *
152  * NONE:        initial state
153  * SYN_SENT:    SYN-only packet seen 
154  * SYN_RECV:    SYN-ACK packet seen
155  * ESTABLISHED: ACK packet seen
156  * FIN_WAIT:    FIN packet seen
157  * CLOSE_WAIT:  ACK seen (after FIN) 
158  * LAST_ACK:    FIN seen (after FIN)
159  * TIME_WAIT:   last ACK seen
160  * CLOSE:       closed connection
161  *
162  * LISTEN state is not used.
163  *
164  * Packets marked as IGNORED (sIG):
165  *      if they may be either invalid or valid 
166  *      and the receiver may send back a connection 
167  *      closing RST or a SYN/ACK.
168  *
169  * Packets marked as INVALID (sIV):
170  *      if they are invalid
171  *      or we do not support the request (simultaneous open)
172  */
173 static enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
174         {
175 /* ORIGINAL */
176 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
177 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
178 /*
179  *      sNO -> sSS      Initialize a new connection
180  *      sSS -> sSS      Retransmitted SYN
181  *      sSR -> sIG      Late retransmitted SYN?
182  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
183  *                      are errors. Receiver will reply with RST 
184  *                      and close the connection.
185  *                      Or we are not in sync and hold a dead connection.
186  *      sFW -> sIG
187  *      sCW -> sIG
188  *      sLA -> sIG
189  *      sTW -> sSS      Reopened connection (RFC 1122).
190  *      sCL -> sSS
191  */
192 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
193 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
194 /*
195  * A SYN/ACK from the client is always invalid:
196  *      - either it tries to set up a simultaneous open, which is 
197  *        not supported;
198  *      - or the firewall has just been inserted between the two hosts
199  *        during the session set-up. The SYN will be retransmitted 
200  *        by the true client (or it'll time out).
201  */
202 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
203 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
204 /*
205  *      sNO -> sIV      Too late and no reason to do anything...
206  *      sSS -> sIV      Client migth not send FIN in this state:
207  *                      we enforce waiting for a SYN/ACK reply first.
208  *      sSR -> sFW      Close started.
209  *      sES -> sFW      
210  *      sFW -> sLA      FIN seen in both directions, waiting for
211  *                      the last ACK. 
212  *                      Migth be a retransmitted FIN as well...
213  *      sCW -> sLA
214  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
215  *      sTW -> sTW
216  *      sCL -> sCL
217  */
218 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
219 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
220 /*
221  *      sNO -> sES      Assumed.
222  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
223  *      sSR -> sES      Established state is reached.
224  *      sES -> sES      :-)
225  *      sFW -> sCW      Normal close request answered by ACK.
226  *      sCW -> sCW
227  *      sLA -> sTW      Last ACK detected.
228  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
229  *      sCL -> sCL
230  */
231 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
232 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
233 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
234         },
235         {
236 /* REPLY */
237 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
238 /*syn*/    { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
239 /*
240  *      sNO -> sIV      Never reached.
241  *      sSS -> sIV      Simultaneous open, not supported
242  *      sSR -> sIV      Simultaneous open, not supported.
243  *      sES -> sIV      Server may not initiate a connection.
244  *      sFW -> sIV
245  *      sCW -> sIV
246  *      sLA -> sIV
247  *      sTW -> sIV      Reopened connection, but server may not do it.
248  *      sCL -> sIV
249  */
250 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
251 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
252 /*
253  *      sSS -> sSR      Standard open.
254  *      sSR -> sSR      Retransmitted SYN/ACK.
255  *      sES -> sIG      Late retransmitted SYN/ACK?
256  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
257  *      sCW -> sIG
258  *      sLA -> sIG
259  *      sTW -> sIG
260  *      sCL -> sIG
261  */
262 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
263 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
264 /*
265  *      sSS -> sIV      Server might not send FIN in this state.
266  *      sSR -> sFW      Close started.
267  *      sES -> sFW
268  *      sFW -> sLA      FIN seen in both directions.
269  *      sCW -> sLA
270  *      sLA -> sLA      Retransmitted FIN.
271  *      sTW -> sTW
272  *      sCL -> sCL
273  */
274 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
275 /*ack*/    { sIV, sIV, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
276 /*
277  *      sSS -> sIV      Might be a half-open connection.
278  *      sSR -> sSR      Might answer late resent SYN.
279  *      sES -> sES      :-)
280  *      sFW -> sCW      Normal close request answered by ACK.
281  *      sCW -> sCW
282  *      sLA -> sTW      Last ACK detected.
283  *      sTW -> sTW      Retransmitted last ACK.
284  *      sCL -> sCL
285  */
286 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
287 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
288 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
289         }
290 };
291
292 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
293                             unsigned int dataoff,
294                             struct ip_conntrack_tuple *tuple)
295 {
296         struct tcphdr _hdr, *hp;
297
298         /* Actually only need first 8 bytes. */
299         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
300         if (hp == NULL)
301                 return 0;
302
303         tuple->src.u.tcp.port = hp->source;
304         tuple->dst.u.tcp.port = hp->dest;
305
306         return 1;
307 }
308
309 static int tcp_invert_tuple(struct ip_conntrack_tuple *tuple,
310                             const struct ip_conntrack_tuple *orig)
311 {
312         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
313         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
314         return 1;
315 }
316
317 /* Print out the per-protocol part of the tuple. */
318 static int tcp_print_tuple(struct seq_file *s,
319                            const struct ip_conntrack_tuple *tuple)
320 {
321         return seq_printf(s, "sport=%hu dport=%hu ",
322                           ntohs(tuple->src.u.tcp.port),
323                           ntohs(tuple->dst.u.tcp.port));
324 }
325
326 /* Print out the private part of the conntrack. */
327 static int tcp_print_conntrack(struct seq_file *s,
328                                const struct ip_conntrack *conntrack)
329 {
330         enum tcp_conntrack state;
331
332         read_lock_bh(&tcp_lock);
333         state = conntrack->proto.tcp.state;
334         read_unlock_bh(&tcp_lock);
335
336         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
337 }
338
339 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
340 {
341         if (tcph->rst) return TCP_RST_SET;
342         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
343         else if (tcph->fin) return TCP_FIN_SET;
344         else if (tcph->ack) return TCP_ACK_SET;
345         else return TCP_NONE_SET;
346 }
347
348 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
349    in IP Filter' by Guido van Rooij.
350    
351    http://www.nluug.nl/events/sane2000/papers.html
352    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
353    
354    The boundaries and the conditions are changed according to RFC793:
355    the packet must intersect the window (i.e. segments may be
356    after the right or before the left edge) and thus receivers may ACK
357    segments after the right edge of the window.
358
359         td_maxend = max(sack + max(win,1)) seen in reply packets
360         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
361         td_maxwin += seq + len - sender.td_maxend
362                         if seq + len > sender.td_maxend
363         td_end    = max(seq + len) seen in sent packets
364    
365    I.   Upper bound for valid data:     seq <= sender.td_maxend
366    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
367    III. Upper bound for valid ack:      sack <= receiver.td_end
368    IV.  Lower bound for valid ack:      ack >= receiver.td_end - MAXACKWINDOW
369         
370    where sack is the highest right edge of sack block found in the packet.
371         
372    The upper bound limit for a valid ack is not ignored - 
373    we doesn't have to deal with fragments. 
374 */
375
376 static inline __u32 segment_seq_plus_len(__u32 seq,
377                                          size_t len,
378                                          struct iphdr *iph,
379                                          struct tcphdr *tcph)
380 {
381         return (seq + len - (iph->ihl + tcph->doff)*4
382                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
383 }
384   
385 /* Fixme: what about big packets? */
386 #define MAXACKWINCONST                  66000
387 #define MAXACKWINDOW(sender)                                            \
388         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
389                                               : MAXACKWINCONST)
390   
391 /*
392  * Simplified tcp_parse_options routine from tcp_input.c
393  */
394 static void tcp_options(const struct sk_buff *skb,
395                         struct iphdr *iph,
396                         struct tcphdr *tcph, 
397                         struct ip_ct_tcp_state *state)
398 {
399         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
400         unsigned char *ptr;
401         int length = (tcph->doff*4) - sizeof(struct tcphdr);
402         
403         if (!length)
404                 return;
405
406         ptr = skb_header_pointer(skb,
407                                  (iph->ihl * 4) + sizeof(struct tcphdr),
408                                  length, buff);
409         BUG_ON(ptr == NULL);
410
411         state->td_scale = 
412         state->flags = 0;
413         
414         while (length > 0) {
415                 int opcode=*ptr++;
416                 int opsize;
417                 
418                 switch (opcode) {
419                 case TCPOPT_EOL:
420                         return;
421                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
422                         length--;
423                         continue;
424                 default:
425                         opsize=*ptr++;
426                         if (opsize < 2) /* "silly options" */
427                                 return;
428                         if (opsize > length)
429                                 break;  /* don't parse partial options */
430
431                         if (opcode == TCPOPT_SACK_PERM 
432                             && opsize == TCPOLEN_SACK_PERM)
433                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
434                         else if (opcode == TCPOPT_WINDOW
435                                  && opsize == TCPOLEN_WINDOW) {
436                                 state->td_scale = *(u_int8_t *)ptr;
437                                 
438                                 if (state->td_scale > 14) {
439                                         /* See RFC1323 */
440                                         state->td_scale = 14;
441                                 }
442                                 state->flags |=
443                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
444                         }
445                         ptr += opsize - 2;
446                         length -= opsize;
447                 }
448         }
449 }
450
451 static void tcp_sack(const struct sk_buff *skb,
452                      struct iphdr *iph,
453                      struct tcphdr *tcph,
454                      __u32 *sack)
455 {
456         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
457         unsigned char *ptr;
458         int length = (tcph->doff*4) - sizeof(struct tcphdr);
459         __u32 tmp;
460
461         if (!length)
462                 return;
463
464         ptr = skb_header_pointer(skb,
465                                  (iph->ihl * 4) + sizeof(struct tcphdr),
466                                  length, buff);
467         BUG_ON(ptr == NULL);
468
469         /* Fast path for timestamp-only option */
470         if (length == TCPOLEN_TSTAMP_ALIGNED*4
471             && *(__u32 *)ptr ==
472                 __constant_ntohl((TCPOPT_NOP << 24) 
473                                  | (TCPOPT_NOP << 16)
474                                  | (TCPOPT_TIMESTAMP << 8)
475                                  | TCPOLEN_TIMESTAMP))
476                 return;
477                 
478         while (length > 0) {
479                 int opcode=*ptr++;
480                 int opsize, i;
481                 
482                 switch (opcode) {
483                 case TCPOPT_EOL:
484                         return;
485                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
486                         length--;
487                         continue;
488                 default:
489                         opsize=*ptr++;
490                         if (opsize < 2) /* "silly options" */
491                                 return;
492                         if (opsize > length)
493                                 break;  /* don't parse partial options */
494
495                         if (opcode == TCPOPT_SACK 
496                             && opsize >= (TCPOLEN_SACK_BASE 
497                                           + TCPOLEN_SACK_PERBLOCK)
498                             && !((opsize - TCPOLEN_SACK_BASE) 
499                                  % TCPOLEN_SACK_PERBLOCK)) {
500                                 for (i = 0;
501                                      i < (opsize - TCPOLEN_SACK_BASE);
502                                      i += TCPOLEN_SACK_PERBLOCK) {
503                                         tmp = ntohl(*((u_int32_t *)(ptr+i)+1));
504                                         
505                                         if (after(tmp, *sack))
506                                                 *sack = tmp;
507                                 }
508                                 return;
509                         }
510                         ptr += opsize - 2;
511                         length -= opsize;
512                 }
513         }
514 }
515
516 static int tcp_in_window(struct ip_ct_tcp *state, 
517                          enum ip_conntrack_dir dir,
518                          unsigned int index,
519                          const struct sk_buff *skb,
520                          struct iphdr *iph,
521                          struct tcphdr *tcph)
522 {
523         struct ip_ct_tcp_state *sender = &state->seen[dir];
524         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
525         __u32 seq, ack, sack, end, win, swin;
526         int res;
527         
528         /*
529          * Get the required data from the packet.
530          */
531         seq = ntohl(tcph->seq);
532         ack = sack = ntohl(tcph->ack_seq);
533         win = ntohs(tcph->window);
534         end = segment_seq_plus_len(seq, skb->len, iph, tcph);
535         
536         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
537                 tcp_sack(skb, iph, tcph, &sack);
538                 
539         DEBUGP("tcp_in_window: START\n");
540         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
541                "seq=%u ack=%u sack=%u win=%u end=%u\n",
542                 NIPQUAD(iph->saddr), ntohs(tcph->source), 
543                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
544                 seq, ack, sack, win, end);
545         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
546                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
547                 sender->td_end, sender->td_maxend, sender->td_maxwin,
548                 sender->td_scale, 
549                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin, 
550                 receiver->td_scale);
551                 
552         if (sender->td_end == 0) {
553                 /*
554                  * Initialize sender data.
555                  */
556                 if (tcph->syn && tcph->ack) {
557                         /*
558                          * Outgoing SYN-ACK in reply to a SYN.
559                          */
560                         sender->td_end = 
561                         sender->td_maxend = end;
562                         sender->td_maxwin = (win == 0 ? 1 : win);
563
564                         tcp_options(skb, iph, tcph, sender);
565                         /* 
566                          * RFC 1323:
567                          * Both sides must send the Window Scale option
568                          * to enable window scaling in either direction.
569                          */
570                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
571                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
572                                 sender->td_scale = 
573                                 receiver->td_scale = 0;
574                 } else {
575                         /*
576                          * We are in the middle of a connection,
577                          * its history is lost for us.
578                          * Let's try to use the data from the packet.
579                          */
580                         sender->td_end = end;
581                         sender->td_maxwin = (win == 0 ? 1 : win);
582                         sender->td_maxend = end + sender->td_maxwin;
583                 }
584         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
585                      && dir == IP_CT_DIR_ORIGINAL)
586                     || (state->state == TCP_CONNTRACK_SYN_RECV
587                         && dir == IP_CT_DIR_REPLY))
588                     && after(end, sender->td_end)) {
589                 /*
590                  * RFC 793: "if a TCP is reinitialized ... then it need
591                  * not wait at all; it must only be sure to use sequence 
592                  * numbers larger than those recently used."
593                  */
594                 sender->td_end =
595                 sender->td_maxend = end;
596                 sender->td_maxwin = (win == 0 ? 1 : win);
597
598                 tcp_options(skb, iph, tcph, sender);
599         }
600         
601         if (!(tcph->ack)) {
602                 /*
603                  * If there is no ACK, just pretend it was set and OK.
604                  */
605                 ack = sack = receiver->td_end;
606         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) == 
607                     (TCP_FLAG_ACK|TCP_FLAG_RST)) 
608                    && (ack == 0)) {
609                 /*
610                  * Broken TCP stacks, that set ACK in RST packets as well
611                  * with zero ack value.
612                  */
613                 ack = sack = receiver->td_end;
614         }
615
616         if (seq == end
617             && (!tcph->rst 
618                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
619                 /*
620                  * Packets contains no data: we assume it is valid
621                  * and check the ack value only.
622                  * However RST segments are always validated by their
623                  * SEQ number, except when seq == 0 (reset sent answering
624                  * SYN.
625                  */
626                 seq = end = sender->td_end;
627                 
628         DEBUGP("tcp_in_window: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
629                "seq=%u ack=%u sack =%u win=%u end=%u\n",
630                 NIPQUAD(iph->saddr), ntohs(tcph->source),
631                 NIPQUAD(iph->daddr), ntohs(tcph->dest),
632                 seq, ack, sack, win, end);
633         DEBUGP("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
634                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
635                 sender->td_end, sender->td_maxend, sender->td_maxwin,
636                 sender->td_scale, 
637                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
638                 receiver->td_scale);
639         
640         DEBUGP("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
641                 before(seq, sender->td_maxend + 1),
642                 after(end, sender->td_end - receiver->td_maxwin - 1),
643                 before(sack, receiver->td_end + 1),
644                 after(ack, receiver->td_end - MAXACKWINDOW(sender)));
645         
646         if (sender->loose || receiver->loose ||
647             (before(seq, sender->td_maxend + 1) &&
648              after(end, sender->td_end - receiver->td_maxwin - 1) &&
649              before(sack, receiver->td_end + 1) &&
650              after(ack, receiver->td_end - MAXACKWINDOW(sender)))) {
651                 /*
652                  * Take into account window scaling (RFC 1323).
653                  */
654                 if (!tcph->syn)
655                         win <<= sender->td_scale;
656                 
657                 /*
658                  * Update sender data.
659                  */
660                 swin = win + (sack - ack);
661                 if (sender->td_maxwin < swin)
662                         sender->td_maxwin = swin;
663                 if (after(end, sender->td_end))
664                         sender->td_end = end;
665                 /*
666                  * Update receiver data.
667                  */
668                 if (after(end, sender->td_maxend))
669                         receiver->td_maxwin += end - sender->td_maxend;
670                 if (after(sack + win, receiver->td_maxend - 1)) {
671                         receiver->td_maxend = sack + win;
672                         if (win == 0)
673                                 receiver->td_maxend++;
674                 }
675
676                 /* 
677                  * Check retransmissions.
678                  */
679                 if (index == TCP_ACK_SET) {
680                         if (state->last_dir == dir
681                             && state->last_seq == seq
682                             && state->last_ack == ack
683                             && state->last_end == end)
684                                 state->retrans++;
685                         else {
686                                 state->last_dir = dir;
687                                 state->last_seq = seq;
688                                 state->last_ack = ack;
689                                 state->last_end = end;
690                                 state->retrans = 0;
691                         }
692                 }
693                 /*
694                  * Close the window of disabled window tracking :-)
695                  */
696                 if (sender->loose)
697                         sender->loose--;
698                 
699                 res = 1;
700         } else {
701                 if (LOG_INVALID(IPPROTO_TCP))
702                         nf_log_packet(PF_INET, 0, skb, NULL, NULL,
703                         "ip_ct_tcp: %s ",
704                         before(seq, sender->td_maxend + 1) ?
705                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
706                         before(sack, receiver->td_end + 1) ?
707                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
708                         : "ACK is under the lower bound (possible overly delayed ACK)"
709                         : "ACK is over the upper bound (ACKed data not seen yet)"
710                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
711                         : "SEQ is over the upper bound (over the window of the receiver)");
712
713                 res = ip_ct_tcp_be_liberal;
714         }
715   
716         DEBUGP("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
717                "receiver end=%u maxend=%u maxwin=%u\n",
718                 res, sender->td_end, sender->td_maxend, sender->td_maxwin, 
719                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
720
721         return res;
722 }
723
724 #ifdef CONFIG_IP_NF_NAT_NEEDED
725 /* Update sender->td_end after NAT successfully mangled the packet */
726 void ip_conntrack_tcp_update(struct sk_buff *skb,
727                              struct ip_conntrack *conntrack, 
728                              enum ip_conntrack_dir dir)
729 {
730         struct iphdr *iph = skb->nh.iph;
731         struct tcphdr *tcph = (void *)skb->nh.iph + skb->nh.iph->ihl*4;
732         __u32 end;
733 #ifdef DEBUGP_VARS
734         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
735         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
736 #endif
737
738         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, iph, tcph);
739         
740         write_lock_bh(&tcp_lock);
741         /*
742          * We have to worry for the ack in the reply packet only...
743          */
744         if (after(end, conntrack->proto.tcp.seen[dir].td_end))
745                 conntrack->proto.tcp.seen[dir].td_end = end;
746         conntrack->proto.tcp.last_end = end;
747         write_unlock_bh(&tcp_lock);
748         DEBUGP("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
749                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
750                 sender->td_end, sender->td_maxend, sender->td_maxwin,
751                 sender->td_scale, 
752                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
753                 receiver->td_scale);
754 }
755  
756 #endif
757
758 #define TH_FIN  0x01
759 #define TH_SYN  0x02
760 #define TH_RST  0x04
761 #define TH_PUSH 0x08
762 #define TH_ACK  0x10
763 #define TH_URG  0x20
764 #define TH_ECE  0x40
765 #define TH_CWR  0x80
766
767 /* table of valid flag combinations - ECE and CWR are always valid */
768 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_PUSH|TH_ACK|TH_URG) + 1] =
769 {
770         [TH_SYN]                        = 1,
771         [TH_SYN|TH_ACK]                 = 1,
772         [TH_SYN|TH_ACK|TH_PUSH]         = 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             && skb->ip_summed != CHECKSUM_UNNECESSARY
822             && csum_tcpudp_magic(iph->saddr, iph->daddr, tcplen, IPPROTO_TCP,
823                                  skb->ip_summed == CHECKSUM_HW ? skb->csum
824                                  : skb_checksum(skb, iph->ihl*4, tcplen, 0))) {
825                 if (LOG_INVALID(IPPROTO_TCP))
826                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
827                                   "ip_ct_tcp: bad TCP checksum ");
828                 return -NF_ACCEPT;
829         }
830
831         /* Check TCP flags. */
832         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR));
833         if (!tcp_valid_flags[tcpflags]) {
834                 if (LOG_INVALID(IPPROTO_TCP))
835                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
836                                   "ip_ct_tcp: invalid TCP flag combination ");
837                 return -NF_ACCEPT;
838         }
839
840         return NF_ACCEPT;
841 }
842
843 /* Returns verdict for packet, or -1 for invalid. */
844 static int tcp_packet(struct ip_conntrack *conntrack,
845                       const struct sk_buff *skb,
846                       enum ip_conntrack_info ctinfo)
847 {
848         enum tcp_conntrack new_state, old_state;
849         enum ip_conntrack_dir dir;
850         struct iphdr *iph = skb->nh.iph;
851         struct tcphdr *th, _tcph;
852         unsigned long timeout;
853         unsigned int index;
854         
855         th = skb_header_pointer(skb, iph->ihl * 4,
856                                 sizeof(_tcph), &_tcph);
857         BUG_ON(th == NULL);
858         
859         write_lock_bh(&tcp_lock);
860         old_state = conntrack->proto.tcp.state;
861         dir = CTINFO2DIR(ctinfo);
862         index = get_conntrack_index(th);
863         new_state = tcp_conntracks[dir][index][old_state];
864
865         switch (new_state) {
866         case TCP_CONNTRACK_IGNORE:
867                 /* Either SYN in ORIGINAL
868                  * or SYN/ACK in REPLY. */
869                 if (index == TCP_SYNACK_SET
870                     && conntrack->proto.tcp.last_index == TCP_SYN_SET
871                     && conntrack->proto.tcp.last_dir != dir
872                     && ntohl(th->ack_seq) ==
873                              conntrack->proto.tcp.last_end) {
874                         /* This SYN/ACK acknowledges a SYN that we earlier 
875                          * ignored as invalid. This means that the client and
876                          * the server are both in sync, while the firewall is
877                          * not. We kill this session and block the SYN/ACK so
878                          * that the client cannot but retransmit its SYN and 
879                          * thus initiate a clean new session.
880                          */
881                         write_unlock_bh(&tcp_lock);
882                         if (LOG_INVALID(IPPROTO_TCP))
883                                 nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
884                                           "ip_ct_tcp: killing out of sync session ");
885                         if (del_timer(&conntrack->timeout))
886                                 conntrack->timeout.function((unsigned long)
887                                                             conntrack);
888                         return -NF_DROP;
889                 }
890                 conntrack->proto.tcp.last_index = index;
891                 conntrack->proto.tcp.last_dir = dir;
892                 conntrack->proto.tcp.last_seq = ntohl(th->seq);
893                 conntrack->proto.tcp.last_end = 
894                     segment_seq_plus_len(ntohl(th->seq), skb->len, iph, th);
895                 
896                 write_unlock_bh(&tcp_lock);
897                 if (LOG_INVALID(IPPROTO_TCP))
898                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
899                                   "ip_ct_tcp: invalid packet ignored ");
900                 return NF_ACCEPT;
901         case TCP_CONNTRACK_MAX:
902                 /* Invalid packet */
903                 DEBUGP("ip_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
904                        dir, get_conntrack_index(th),
905                        old_state);
906                 write_unlock_bh(&tcp_lock);
907                 if (LOG_INVALID(IPPROTO_TCP))
908                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, 
909                                   "ip_ct_tcp: invalid state ");
910                 return -NF_ACCEPT;
911         case TCP_CONNTRACK_SYN_SENT:
912                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
913                         break;
914                 if ((conntrack->proto.tcp.seen[dir].flags &
915                          IP_CT_TCP_FLAG_CLOSE_INIT)
916                     || after(ntohl(th->seq),
917                              conntrack->proto.tcp.seen[dir].td_end)) {  
918                         /* Attempt to reopen a closed connection.
919                         * Delete this connection and look up again. */
920                         write_unlock_bh(&tcp_lock);
921                         if (del_timer(&conntrack->timeout))
922                                 conntrack->timeout.function((unsigned long)
923                                                             conntrack);
924                         return -NF_REPEAT;
925                 } else {
926                         write_unlock_bh(&tcp_lock);
927                         if (LOG_INVALID(IPPROTO_TCP))
928                                 nf_log_packet(PF_INET, 0, skb, NULL, NULL,
929                                               "ip_ct_tcp: invalid SYN");
930                         return -NF_ACCEPT;
931                 }
932         case TCP_CONNTRACK_CLOSE:
933                 if (index == TCP_RST_SET
934                     && test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
935                     && conntrack->proto.tcp.last_index == TCP_SYN_SET
936                     && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
937                         /* RST sent to invalid SYN we had let trough
938                          * SYN was in window then, tear down connection.
939                          * We skip window checking, because packet might ACK
940                          * segments we ignored in the SYN. */
941                         goto in_window;
942                 }
943                 /* Just fall trough */
944         default:
945                 /* Keep compilers happy. */
946                 break;
947         }
948
949         if (!tcp_in_window(&conntrack->proto.tcp, dir, index, 
950                            skb, iph, th)) {
951                 write_unlock_bh(&tcp_lock);
952                 return -NF_ACCEPT;
953         }
954     in_window:
955         /* From now on we have got in-window packets */ 
956         conntrack->proto.tcp.last_index = index;
957
958         DEBUGP("tcp_conntracks: src=%u.%u.%u.%u:%hu dst=%u.%u.%u.%u:%hu "
959                "syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
960                 NIPQUAD(iph->saddr), ntohs(th->source),
961                 NIPQUAD(iph->daddr), ntohs(th->dest),
962                 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
963                 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
964                 old_state, new_state);
965
966         conntrack->proto.tcp.state = new_state;
967         if (old_state != new_state 
968             && (new_state == TCP_CONNTRACK_FIN_WAIT
969                 || new_state == TCP_CONNTRACK_CLOSE))
970                 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
971         timeout = conntrack->proto.tcp.retrans >= ip_ct_tcp_max_retrans
972                   && *tcp_timeouts[new_state] > ip_ct_tcp_timeout_max_retrans
973                   ? ip_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
974         write_unlock_bh(&tcp_lock);
975
976         if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
977                 /* If only reply is a RST, we can consider ourselves not to
978                    have an established connection: this is a fairly common
979                    problem case, so we can delete the conntrack
980                    immediately.  --RR */
981                 if (th->rst) {
982                         if (del_timer(&conntrack->timeout))
983                                 conntrack->timeout.function((unsigned long)
984                                                             conntrack);
985                         return NF_ACCEPT;
986                 }
987         } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
988                    && (old_state == TCP_CONNTRACK_SYN_RECV
989                        || old_state == TCP_CONNTRACK_ESTABLISHED)
990                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
991                 /* Set ASSURED if we see see valid ack in ESTABLISHED 
992                    after SYN_RECV or a valid answer for a picked up 
993                    connection. */
994                         set_bit(IPS_ASSURED_BIT, &conntrack->status);
995         }
996         ip_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
997
998         return NF_ACCEPT;
999 }
1000  
1001 /* Called when a new connection for this protocol found. */
1002 static int tcp_new(struct ip_conntrack *conntrack,
1003                    const struct sk_buff *skb)
1004 {
1005         enum tcp_conntrack new_state;
1006         struct iphdr *iph = skb->nh.iph;
1007         struct tcphdr *th, _tcph;
1008 #ifdef DEBUGP_VARS
1009         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
1010         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
1011 #endif
1012
1013         th = skb_header_pointer(skb, iph->ihl * 4,
1014                                 sizeof(_tcph), &_tcph);
1015         BUG_ON(th == NULL);
1016         
1017         /* Don't need lock here: this conntrack not in circulation yet */
1018         new_state
1019                 = tcp_conntracks[0][get_conntrack_index(th)]
1020                 [TCP_CONNTRACK_NONE];
1021
1022         /* Invalid: delete conntrack */
1023         if (new_state >= TCP_CONNTRACK_MAX) {
1024                 DEBUGP("ip_ct_tcp: invalid new deleting.\n");
1025                 return 0;
1026         }
1027
1028         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1029                 /* SYN packet */
1030                 conntrack->proto.tcp.seen[0].td_end =
1031                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1032                                              iph, th);
1033                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1034                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1035                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1036                 conntrack->proto.tcp.seen[0].td_maxend =
1037                         conntrack->proto.tcp.seen[0].td_end;
1038
1039                 tcp_options(skb, iph, th, &conntrack->proto.tcp.seen[0]);
1040                 conntrack->proto.tcp.seen[1].flags = 0;
1041                 conntrack->proto.tcp.seen[0].loose = 
1042                 conntrack->proto.tcp.seen[1].loose = 0;
1043         } else if (ip_ct_tcp_loose == 0) {
1044                 /* Don't try to pick up connections. */
1045                 return 0;
1046         } else {
1047                 /*
1048                  * We are in the middle of a connection,
1049                  * its history is lost for us.
1050                  * Let's try to use the data from the packet.
1051                  */
1052                 conntrack->proto.tcp.seen[0].td_end =
1053                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1054                                              iph, th);
1055                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1056                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1057                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1058                 conntrack->proto.tcp.seen[0].td_maxend =
1059                         conntrack->proto.tcp.seen[0].td_end + 
1060                         conntrack->proto.tcp.seen[0].td_maxwin;
1061                 conntrack->proto.tcp.seen[0].td_scale = 0;
1062
1063                 /* We assume SACK. Should we assume window scaling too? */
1064                 conntrack->proto.tcp.seen[0].flags =
1065                 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM;
1066                 conntrack->proto.tcp.seen[0].loose = 
1067                 conntrack->proto.tcp.seen[1].loose = ip_ct_tcp_loose;
1068         }
1069     
1070         conntrack->proto.tcp.seen[1].td_end = 0;
1071         conntrack->proto.tcp.seen[1].td_maxend = 0;
1072         conntrack->proto.tcp.seen[1].td_maxwin = 1;
1073         conntrack->proto.tcp.seen[1].td_scale = 0;      
1074
1075         /* tcp_packet will set them */
1076         conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1077         conntrack->proto.tcp.last_index = TCP_NONE_SET;
1078          
1079         DEBUGP("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1080                "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1081                 sender->td_end, sender->td_maxend, sender->td_maxwin,
1082                 sender->td_scale, 
1083                 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1084                 receiver->td_scale);
1085         return 1;
1086 }
1087   
1088 struct ip_conntrack_protocol ip_conntrack_protocol_tcp =
1089 {
1090         .proto                  = IPPROTO_TCP,
1091         .name                   = "tcp",
1092         .pkt_to_tuple           = tcp_pkt_to_tuple,
1093         .invert_tuple           = tcp_invert_tuple,
1094         .print_tuple            = tcp_print_tuple,
1095         .print_conntrack        = tcp_print_conntrack,
1096         .packet                 = tcp_packet,
1097         .new                    = tcp_new,
1098         .error                  = tcp_error,
1099 };