netfilter: ctnetlink: fix expectation mask dump
[linux-2.6.git] / net / netfilter / nf_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
9 #include <linux/types.h>
10 #include <linux/timer.h>
11 #include <linux/module.h>
12 #include <linux/in.h>
13 #include <linux/tcp.h>
14 #include <linux/spinlock.h>
15 #include <linux/skbuff.h>
16 #include <linux/ipv6.h>
17 #include <net/ip6_checksum.h>
18 #include <asm/unaligned.h>
19
20 #include <net/tcp.h>
21
22 #include <linux/netfilter.h>
23 #include <linux/netfilter_ipv4.h>
24 #include <linux/netfilter_ipv6.h>
25 #include <net/netfilter/nf_conntrack.h>
26 #include <net/netfilter/nf_conntrack_l4proto.h>
27 #include <net/netfilter/nf_conntrack_ecache.h>
28 #include <net/netfilter/nf_log.h>
29 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
30 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
31
32 /* "Be conservative in what you do,
33     be liberal in what you accept from others."
34     If it's non-zero, we mark only out of window RST segments as INVALID. */
35 static int nf_ct_tcp_be_liberal __read_mostly = 0;
36
37 /* If it is set to zero, we disable picking up already established
38    connections. */
39 static int nf_ct_tcp_loose __read_mostly = 1;
40
41 /* Max number of the retransmitted packets without receiving an (acceptable)
42    ACK from the destination. If this number is reached, a shorter timer
43    will be started. */
44 static int nf_ct_tcp_max_retrans __read_mostly = 3;
45
46   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
47      closely.  They're more complex. --RR */
48
49 static const char *const tcp_conntrack_names[] = {
50         "NONE",
51         "SYN_SENT",
52         "SYN_RECV",
53         "ESTABLISHED",
54         "FIN_WAIT",
55         "CLOSE_WAIT",
56         "LAST_ACK",
57         "TIME_WAIT",
58         "CLOSE",
59         "SYN_SENT2",
60 };
61
62 #define SECS * HZ
63 #define MINS * 60 SECS
64 #define HOURS * 60 MINS
65 #define DAYS * 24 HOURS
66
67 /* RFC1122 says the R2 limit should be at least 100 seconds.
68    Linux uses 15 packets as limit, which corresponds
69    to ~13-30min depending on RTO. */
70 static unsigned int nf_ct_tcp_timeout_max_retrans __read_mostly    =   5 MINS;
71 static unsigned int nf_ct_tcp_timeout_unacknowledged __read_mostly =   5 MINS;
72
73 static unsigned int tcp_timeouts[TCP_CONNTRACK_MAX] __read_mostly = {
74         [TCP_CONNTRACK_SYN_SENT]        = 2 MINS,
75         [TCP_CONNTRACK_SYN_RECV]        = 60 SECS,
76         [TCP_CONNTRACK_ESTABLISHED]     = 5 DAYS,
77         [TCP_CONNTRACK_FIN_WAIT]        = 2 MINS,
78         [TCP_CONNTRACK_CLOSE_WAIT]      = 60 SECS,
79         [TCP_CONNTRACK_LAST_ACK]        = 30 SECS,
80         [TCP_CONNTRACK_TIME_WAIT]       = 2 MINS,
81         [TCP_CONNTRACK_CLOSE]           = 10 SECS,
82         [TCP_CONNTRACK_SYN_SENT2]       = 2 MINS,
83 };
84
85 #define sNO TCP_CONNTRACK_NONE
86 #define sSS TCP_CONNTRACK_SYN_SENT
87 #define sSR TCP_CONNTRACK_SYN_RECV
88 #define sES TCP_CONNTRACK_ESTABLISHED
89 #define sFW TCP_CONNTRACK_FIN_WAIT
90 #define sCW TCP_CONNTRACK_CLOSE_WAIT
91 #define sLA TCP_CONNTRACK_LAST_ACK
92 #define sTW TCP_CONNTRACK_TIME_WAIT
93 #define sCL TCP_CONNTRACK_CLOSE
94 #define sS2 TCP_CONNTRACK_SYN_SENT2
95 #define sIV TCP_CONNTRACK_MAX
96 #define sIG TCP_CONNTRACK_IGNORE
97
98 /* What TCP flags are set from RST/SYN/FIN/ACK. */
99 enum tcp_bit_set {
100         TCP_SYN_SET,
101         TCP_SYNACK_SET,
102         TCP_FIN_SET,
103         TCP_ACK_SET,
104         TCP_RST_SET,
105         TCP_NONE_SET,
106 };
107
108 /*
109  * The TCP state transition table needs a few words...
110  *
111  * We are the man in the middle. All the packets go through us
112  * but might get lost in transit to the destination.
113  * It is assumed that the destinations can't receive segments
114  * we haven't seen.
115  *
116  * The checked segment is in window, but our windows are *not*
117  * equivalent with the ones of the sender/receiver. We always
118  * try to guess the state of the current sender.
119  *
120  * The meaning of the states are:
121  *
122  * NONE:        initial state
123  * SYN_SENT:    SYN-only packet seen
124  * SYN_SENT2:   SYN-only packet seen from reply dir, simultaneous open
125  * SYN_RECV:    SYN-ACK packet seen
126  * ESTABLISHED: ACK packet seen
127  * FIN_WAIT:    FIN packet seen
128  * CLOSE_WAIT:  ACK seen (after FIN)
129  * LAST_ACK:    FIN seen (after FIN)
130  * TIME_WAIT:   last ACK seen
131  * CLOSE:       closed connection (RST)
132  *
133  * Packets marked as IGNORED (sIG):
134  *      if they may be either invalid or valid
135  *      and the receiver may send back a connection
136  *      closing RST or a SYN/ACK.
137  *
138  * Packets marked as INVALID (sIV):
139  *      if we regard them as truly invalid packets
140  */
141 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
142         {
143 /* ORIGINAL */
144 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
145 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sS2 },
146 /*
147  *      sNO -> sSS      Initialize a new connection
148  *      sSS -> sSS      Retransmitted SYN
149  *      sS2 -> sS2      Late retransmitted SYN
150  *      sSR -> sIG
151  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
152  *                      are errors. Receiver will reply with RST
153  *                      and close the connection.
154  *                      Or we are not in sync and hold a dead connection.
155  *      sFW -> sIG
156  *      sCW -> sIG
157  *      sLA -> sIG
158  *      sTW -> sSS      Reopened connection (RFC 1122).
159  *      sCL -> sSS
160  */
161 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
162 /*synack*/ { sIV, sIV, sIG, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
163 /*
164  *      sNO -> sIV      Too late and no reason to do anything
165  *      sSS -> sIV      Client can't send SYN and then SYN/ACK
166  *      sS2 -> sSR      SYN/ACK sent to SYN2 in simultaneous open
167  *      sSR -> sIG
168  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
169  *                      are errors. Receiver will reply with RST
170  *                      and close the connection.
171  *                      Or we are not in sync and hold a dead connection.
172  *      sFW -> sIG
173  *      sCW -> sIG
174  *      sLA -> sIG
175  *      sTW -> sIG
176  *      sCL -> sIG
177  */
178 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
179 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
180 /*
181  *      sNO -> sIV      Too late and no reason to do anything...
182  *      sSS -> sIV      Client migth not send FIN in this state:
183  *                      we enforce waiting for a SYN/ACK reply first.
184  *      sS2 -> sIV
185  *      sSR -> sFW      Close started.
186  *      sES -> sFW
187  *      sFW -> sLA      FIN seen in both directions, waiting for
188  *                      the last ACK.
189  *                      Migth be a retransmitted FIN as well...
190  *      sCW -> sLA
191  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
192  *      sTW -> sTW
193  *      sCL -> sCL
194  */
195 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
196 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
197 /*
198  *      sNO -> sES      Assumed.
199  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
200  *      sS2 -> sIV
201  *      sSR -> sES      Established state is reached.
202  *      sES -> sES      :-)
203  *      sFW -> sCW      Normal close request answered by ACK.
204  *      sCW -> sCW
205  *      sLA -> sTW      Last ACK detected.
206  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
207  *      sCL -> sCL
208  */
209 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
210 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
211 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
212         },
213         {
214 /* REPLY */
215 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
216 /*syn*/    { sIV, sS2, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sS2 },
217 /*
218  *      sNO -> sIV      Never reached.
219  *      sSS -> sS2      Simultaneous open
220  *      sS2 -> sS2      Retransmitted simultaneous SYN
221  *      sSR -> sIV      Invalid SYN packets sent by the server
222  *      sES -> sIV
223  *      sFW -> sIV
224  *      sCW -> sIV
225  *      sLA -> sIV
226  *      sTW -> sIV      Reopened connection, but server may not do it.
227  *      sCL -> sIV
228  */
229 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
230 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
231 /*
232  *      sSS -> sSR      Standard open.
233  *      sS2 -> sSR      Simultaneous open
234  *      sSR -> sSR      Retransmitted SYN/ACK.
235  *      sES -> sIG      Late retransmitted SYN/ACK?
236  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
237  *      sCW -> sIG
238  *      sLA -> sIG
239  *      sTW -> sIG
240  *      sCL -> sIG
241  */
242 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
243 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
244 /*
245  *      sSS -> sIV      Server might not send FIN in this state.
246  *      sS2 -> sIV
247  *      sSR -> sFW      Close started.
248  *      sES -> sFW
249  *      sFW -> sLA      FIN seen in both directions.
250  *      sCW -> sLA
251  *      sLA -> sLA      Retransmitted FIN.
252  *      sTW -> sTW
253  *      sCL -> sCL
254  */
255 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
256 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIG },
257 /*
258  *      sSS -> sIG      Might be a half-open connection.
259  *      sS2 -> sIG
260  *      sSR -> sSR      Might answer late resent SYN.
261  *      sES -> sES      :-)
262  *      sFW -> sCW      Normal close request answered by ACK.
263  *      sCW -> sCW
264  *      sLA -> sTW      Last ACK detected.
265  *      sTW -> sTW      Retransmitted last ACK.
266  *      sCL -> sCL
267  */
268 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
269 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
270 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
271         }
272 };
273
274 static bool tcp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
275                              struct nf_conntrack_tuple *tuple)
276 {
277         const struct tcphdr *hp;
278         struct tcphdr _hdr;
279
280         /* Actually only need first 8 bytes. */
281         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
282         if (hp == NULL)
283                 return false;
284
285         tuple->src.u.tcp.port = hp->source;
286         tuple->dst.u.tcp.port = hp->dest;
287
288         return true;
289 }
290
291 static bool tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
292                              const struct nf_conntrack_tuple *orig)
293 {
294         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
295         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
296         return true;
297 }
298
299 /* Print out the per-protocol part of the tuple. */
300 static int tcp_print_tuple(struct seq_file *s,
301                            const struct nf_conntrack_tuple *tuple)
302 {
303         return seq_printf(s, "sport=%hu dport=%hu ",
304                           ntohs(tuple->src.u.tcp.port),
305                           ntohs(tuple->dst.u.tcp.port));
306 }
307
308 /* Print out the private part of the conntrack. */
309 static int tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
310 {
311         enum tcp_conntrack state;
312
313         spin_lock_bh(&ct->lock);
314         state = ct->proto.tcp.state;
315         spin_unlock_bh(&ct->lock);
316
317         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
318 }
319
320 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
321 {
322         if (tcph->rst) return TCP_RST_SET;
323         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
324         else if (tcph->fin) return TCP_FIN_SET;
325         else if (tcph->ack) return TCP_ACK_SET;
326         else return TCP_NONE_SET;
327 }
328
329 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
330    in IP Filter' by Guido van Rooij.
331
332    http://www.nluug.nl/events/sane2000/papers.html
333    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
334
335    The boundaries and the conditions are changed according to RFC793:
336    the packet must intersect the window (i.e. segments may be
337    after the right or before the left edge) and thus receivers may ACK
338    segments after the right edge of the window.
339
340         td_maxend = max(sack + max(win,1)) seen in reply packets
341         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
342         td_maxwin += seq + len - sender.td_maxend
343                         if seq + len > sender.td_maxend
344         td_end    = max(seq + len) seen in sent packets
345
346    I.   Upper bound for valid data:     seq <= sender.td_maxend
347    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
348    III. Upper bound for valid (s)ack:   sack <= receiver.td_end
349    IV.  Lower bound for valid (s)ack:   sack >= receiver.td_end - MAXACKWINDOW
350
351    where sack is the highest right edge of sack block found in the packet
352    or ack in the case of packet without SACK option.
353
354    The upper bound limit for a valid (s)ack is not ignored -
355    we doesn't have to deal with fragments.
356 */
357
358 static inline __u32 segment_seq_plus_len(__u32 seq,
359                                          size_t len,
360                                          unsigned int dataoff,
361                                          const struct tcphdr *tcph)
362 {
363         /* XXX Should I use payload length field in IP/IPv6 header ?
364          * - YK */
365         return (seq + len - dataoff - tcph->doff*4
366                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
367 }
368
369 /* Fixme: what about big packets? */
370 #define MAXACKWINCONST                  66000
371 #define MAXACKWINDOW(sender)                                            \
372         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
373                                               : MAXACKWINCONST)
374
375 /*
376  * Simplified tcp_parse_options routine from tcp_input.c
377  */
378 static void tcp_options(const struct sk_buff *skb,
379                         unsigned int dataoff,
380                         const struct tcphdr *tcph,
381                         struct ip_ct_tcp_state *state)
382 {
383         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
384         const unsigned char *ptr;
385         int length = (tcph->doff*4) - sizeof(struct tcphdr);
386
387         if (!length)
388                 return;
389
390         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
391                                  length, buff);
392         BUG_ON(ptr == NULL);
393
394         state->td_scale =
395         state->flags = 0;
396
397         while (length > 0) {
398                 int opcode=*ptr++;
399                 int opsize;
400
401                 switch (opcode) {
402                 case TCPOPT_EOL:
403                         return;
404                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
405                         length--;
406                         continue;
407                 default:
408                         opsize=*ptr++;
409                         if (opsize < 2) /* "silly options" */
410                                 return;
411                         if (opsize > length)
412                                 break;  /* don't parse partial options */
413
414                         if (opcode == TCPOPT_SACK_PERM
415                             && opsize == TCPOLEN_SACK_PERM)
416                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
417                         else if (opcode == TCPOPT_WINDOW
418                                  && opsize == TCPOLEN_WINDOW) {
419                                 state->td_scale = *(u_int8_t *)ptr;
420
421                                 if (state->td_scale > 14) {
422                                         /* See RFC1323 */
423                                         state->td_scale = 14;
424                                 }
425                                 state->flags |=
426                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
427                         }
428                         ptr += opsize - 2;
429                         length -= opsize;
430                 }
431         }
432 }
433
434 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
435                      const struct tcphdr *tcph, __u32 *sack)
436 {
437         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
438         const unsigned char *ptr;
439         int length = (tcph->doff*4) - sizeof(struct tcphdr);
440         __u32 tmp;
441
442         if (!length)
443                 return;
444
445         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
446                                  length, buff);
447         BUG_ON(ptr == NULL);
448
449         /* Fast path for timestamp-only option */
450         if (length == TCPOLEN_TSTAMP_ALIGNED*4
451             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
452                                        | (TCPOPT_NOP << 16)
453                                        | (TCPOPT_TIMESTAMP << 8)
454                                        | TCPOLEN_TIMESTAMP))
455                 return;
456
457         while (length > 0) {
458                 int opcode = *ptr++;
459                 int opsize, i;
460
461                 switch (opcode) {
462                 case TCPOPT_EOL:
463                         return;
464                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
465                         length--;
466                         continue;
467                 default:
468                         opsize = *ptr++;
469                         if (opsize < 2) /* "silly options" */
470                                 return;
471                         if (opsize > length)
472                                 break;  /* don't parse partial options */
473
474                         if (opcode == TCPOPT_SACK
475                             && opsize >= (TCPOLEN_SACK_BASE
476                                           + TCPOLEN_SACK_PERBLOCK)
477                             && !((opsize - TCPOLEN_SACK_BASE)
478                                  % TCPOLEN_SACK_PERBLOCK)) {
479                                 for (i = 0;
480                                      i < (opsize - TCPOLEN_SACK_BASE);
481                                      i += TCPOLEN_SACK_PERBLOCK) {
482                                         tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
483
484                                         if (after(tmp, *sack))
485                                                 *sack = tmp;
486                                 }
487                                 return;
488                         }
489                         ptr += opsize - 2;
490                         length -= opsize;
491                 }
492         }
493 }
494
495 #ifdef CONFIG_NF_NAT_NEEDED
496 static inline s16 nat_offset(const struct nf_conn *ct,
497                              enum ip_conntrack_dir dir,
498                              u32 seq)
499 {
500         typeof(nf_ct_nat_offset) get_offset = rcu_dereference(nf_ct_nat_offset);
501
502         return get_offset != NULL ? get_offset(ct, dir, seq) : 0;
503 }
504 #define NAT_OFFSET(pf, ct, dir, seq) \
505         (pf == NFPROTO_IPV4 ? nat_offset(ct, dir, seq) : 0)
506 #else
507 #define NAT_OFFSET(pf, ct, dir, seq)    0
508 #endif
509
510 static bool tcp_in_window(const struct nf_conn *ct,
511                           struct ip_ct_tcp *state,
512                           enum ip_conntrack_dir dir,
513                           unsigned int index,
514                           const struct sk_buff *skb,
515                           unsigned int dataoff,
516                           const struct tcphdr *tcph,
517                           u_int8_t pf)
518 {
519         struct net *net = nf_ct_net(ct);
520         struct ip_ct_tcp_state *sender = &state->seen[dir];
521         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
522         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
523         __u32 seq, ack, sack, end, win, swin;
524         s16 receiver_offset;
525         bool res;
526
527         /*
528          * Get the required data from the packet.
529          */
530         seq = ntohl(tcph->seq);
531         ack = sack = ntohl(tcph->ack_seq);
532         win = ntohs(tcph->window);
533         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
534
535         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
536                 tcp_sack(skb, dataoff, tcph, &sack);
537
538         /* Take into account NAT sequence number mangling */
539         receiver_offset = NAT_OFFSET(pf, ct, !dir, ack - 1);
540         ack -= receiver_offset;
541         sack -= receiver_offset;
542
543         pr_debug("tcp_in_window: START\n");
544         pr_debug("tcp_in_window: ");
545         nf_ct_dump_tuple(tuple);
546         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
547                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
548         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
549                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
550                  sender->td_end, sender->td_maxend, sender->td_maxwin,
551                  sender->td_scale,
552                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
553                  receiver->td_scale);
554
555         if (sender->td_maxwin == 0) {
556                 /*
557                  * Initialize sender data.
558                  */
559                 if (tcph->syn) {
560                         /*
561                          * SYN-ACK in reply to a SYN
562                          * or SYN from reply direction in simultaneous open.
563                          */
564                         sender->td_end =
565                         sender->td_maxend = end;
566                         sender->td_maxwin = (win == 0 ? 1 : win);
567
568                         tcp_options(skb, dataoff, tcph, sender);
569                         /*
570                          * RFC 1323:
571                          * Both sides must send the Window Scale option
572                          * to enable window scaling in either direction.
573                          */
574                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
575                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
576                                 sender->td_scale =
577                                 receiver->td_scale = 0;
578                         if (!tcph->ack)
579                                 /* Simultaneous open */
580                                 return true;
581                 } else {
582                         /*
583                          * We are in the middle of a connection,
584                          * its history is lost for us.
585                          * Let's try to use the data from the packet.
586                          */
587                         sender->td_end = end;
588                         sender->td_maxwin = (win == 0 ? 1 : win);
589                         sender->td_maxend = end + sender->td_maxwin;
590                 }
591         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
592                      && dir == IP_CT_DIR_ORIGINAL)
593                    || (state->state == TCP_CONNTRACK_SYN_RECV
594                      && dir == IP_CT_DIR_REPLY))
595                    && after(end, sender->td_end)) {
596                 /*
597                  * RFC 793: "if a TCP is reinitialized ... then it need
598                  * not wait at all; it must only be sure to use sequence
599                  * numbers larger than those recently used."
600                  */
601                 sender->td_end =
602                 sender->td_maxend = end;
603                 sender->td_maxwin = (win == 0 ? 1 : win);
604
605                 tcp_options(skb, dataoff, tcph, sender);
606         }
607
608         if (!(tcph->ack)) {
609                 /*
610                  * If there is no ACK, just pretend it was set and OK.
611                  */
612                 ack = sack = receiver->td_end;
613         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
614                     (TCP_FLAG_ACK|TCP_FLAG_RST))
615                    && (ack == 0)) {
616                 /*
617                  * Broken TCP stacks, that set ACK in RST packets as well
618                  * with zero ack value.
619                  */
620                 ack = sack = receiver->td_end;
621         }
622
623         if (seq == end
624             && (!tcph->rst
625                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
626                 /*
627                  * Packets contains no data: we assume it is valid
628                  * and check the ack value only.
629                  * However RST segments are always validated by their
630                  * SEQ number, except when seq == 0 (reset sent answering
631                  * SYN.
632                  */
633                 seq = end = sender->td_end;
634
635         pr_debug("tcp_in_window: ");
636         nf_ct_dump_tuple(tuple);
637         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
638                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
639         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
640                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
641                  sender->td_end, sender->td_maxend, sender->td_maxwin,
642                  sender->td_scale,
643                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
644                  receiver->td_scale);
645
646         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
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(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
651
652         if (before(seq, sender->td_maxend + 1) &&
653             after(end, sender->td_end - receiver->td_maxwin - 1) &&
654             before(sack, receiver->td_end + 1) &&
655             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
656                 /*
657                  * Take into account window scaling (RFC 1323).
658                  */
659                 if (!tcph->syn)
660                         win <<= sender->td_scale;
661
662                 /*
663                  * Update sender data.
664                  */
665                 swin = win + (sack - ack);
666                 if (sender->td_maxwin < swin)
667                         sender->td_maxwin = swin;
668                 if (after(end, sender->td_end)) {
669                         sender->td_end = end;
670                         sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
671                 }
672                 if (tcph->ack) {
673                         if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
674                                 sender->td_maxack = ack;
675                                 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
676                         } else if (after(ack, sender->td_maxack))
677                                 sender->td_maxack = ack;
678                 }
679
680                 /*
681                  * Update receiver data.
682                  */
683                 if (after(end, sender->td_maxend))
684                         receiver->td_maxwin += end - sender->td_maxend;
685                 if (after(sack + win, receiver->td_maxend - 1)) {
686                         receiver->td_maxend = sack + win;
687                         if (win == 0)
688                                 receiver->td_maxend++;
689                 }
690                 if (ack == receiver->td_end)
691                         receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
692
693                 /*
694                  * Check retransmissions.
695                  */
696                 if (index == TCP_ACK_SET) {
697                         if (state->last_dir == dir
698                             && state->last_seq == seq
699                             && state->last_ack == ack
700                             && state->last_end == end
701                             && state->last_win == win)
702                                 state->retrans++;
703                         else {
704                                 state->last_dir = dir;
705                                 state->last_seq = seq;
706                                 state->last_ack = ack;
707                                 state->last_end = end;
708                                 state->last_win = win;
709                                 state->retrans = 0;
710                         }
711                 }
712                 res = true;
713         } else {
714                 res = false;
715                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
716                     nf_ct_tcp_be_liberal)
717                         res = true;
718                 if (!res && LOG_INVALID(net, IPPROTO_TCP))
719                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
720                         "nf_ct_tcp: %s ",
721                         before(seq, sender->td_maxend + 1) ?
722                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
723                         before(sack, receiver->td_end + 1) ?
724                         after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
725                         : "ACK is under the lower bound (possible overly delayed ACK)"
726                         : "ACK is over the upper bound (ACKed data not seen yet)"
727                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
728                         : "SEQ is over the upper bound (over the window of the receiver)");
729         }
730
731         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
732                  "receiver end=%u maxend=%u maxwin=%u\n",
733                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
734                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
735
736         return res;
737 }
738
739 #define TH_FIN  0x01
740 #define TH_SYN  0x02
741 #define TH_RST  0x04
742 #define TH_PUSH 0x08
743 #define TH_ACK  0x10
744 #define TH_URG  0x20
745 #define TH_ECE  0x40
746 #define TH_CWR  0x80
747
748 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
749 static const u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_ACK|TH_URG) + 1] =
750 {
751         [TH_SYN]                        = 1,
752         [TH_SYN|TH_URG]                 = 1,
753         [TH_SYN|TH_ACK]                 = 1,
754         [TH_RST]                        = 1,
755         [TH_RST|TH_ACK]                 = 1,
756         [TH_FIN|TH_ACK]                 = 1,
757         [TH_FIN|TH_ACK|TH_URG]          = 1,
758         [TH_ACK]                        = 1,
759         [TH_ACK|TH_URG]                 = 1,
760 };
761
762 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
763 static int tcp_error(struct net *net,
764                      struct sk_buff *skb,
765                      unsigned int dataoff,
766                      enum ip_conntrack_info *ctinfo,
767                      u_int8_t pf,
768                      unsigned int hooknum)
769 {
770         const struct tcphdr *th;
771         struct tcphdr _tcph;
772         unsigned int tcplen = skb->len - dataoff;
773         u_int8_t tcpflags;
774
775         /* Smaller that minimal TCP header? */
776         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
777         if (th == NULL) {
778                 if (LOG_INVALID(net, IPPROTO_TCP))
779                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
780                                 "nf_ct_tcp: short packet ");
781                 return -NF_ACCEPT;
782         }
783
784         /* Not whole TCP header or malformed packet */
785         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
786                 if (LOG_INVALID(net, IPPROTO_TCP))
787                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
788                                 "nf_ct_tcp: truncated/malformed packet ");
789                 return -NF_ACCEPT;
790         }
791
792         /* Checksum invalid? Ignore.
793          * We skip checking packets on the outgoing path
794          * because the checksum is assumed to be correct.
795          */
796         /* FIXME: Source route IP option packets --RR */
797         if (net->ct.sysctl_checksum && hooknum == NF_INET_PRE_ROUTING &&
798             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
799                 if (LOG_INVALID(net, IPPROTO_TCP))
800                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
801                                   "nf_ct_tcp: bad TCP checksum ");
802                 return -NF_ACCEPT;
803         }
804
805         /* Check TCP flags. */
806         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR|TH_PUSH));
807         if (!tcp_valid_flags[tcpflags]) {
808                 if (LOG_INVALID(net, IPPROTO_TCP))
809                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
810                                   "nf_ct_tcp: invalid TCP flag combination ");
811                 return -NF_ACCEPT;
812         }
813
814         return NF_ACCEPT;
815 }
816
817 /* Returns verdict for packet, or -1 for invalid. */
818 static int tcp_packet(struct nf_conn *ct,
819                       const struct sk_buff *skb,
820                       unsigned int dataoff,
821                       enum ip_conntrack_info ctinfo,
822                       u_int8_t pf,
823                       unsigned int hooknum)
824 {
825         struct net *net = nf_ct_net(ct);
826         struct nf_conntrack_tuple *tuple;
827         enum tcp_conntrack new_state, old_state;
828         enum ip_conntrack_dir dir;
829         const struct tcphdr *th;
830         struct tcphdr _tcph;
831         unsigned long timeout;
832         unsigned int index;
833
834         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
835         BUG_ON(th == NULL);
836
837         spin_lock_bh(&ct->lock);
838         old_state = ct->proto.tcp.state;
839         dir = CTINFO2DIR(ctinfo);
840         index = get_conntrack_index(th);
841         new_state = tcp_conntracks[dir][index][old_state];
842         tuple = &ct->tuplehash[dir].tuple;
843
844         switch (new_state) {
845         case TCP_CONNTRACK_SYN_SENT:
846                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
847                         break;
848                 /* RFC 1122: "When a connection is closed actively,
849                  * it MUST linger in TIME-WAIT state for a time 2xMSL
850                  * (Maximum Segment Lifetime). However, it MAY accept
851                  * a new SYN from the remote TCP to reopen the connection
852                  * directly from TIME-WAIT state, if..."
853                  * We ignore the conditions because we are in the
854                  * TIME-WAIT state anyway.
855                  *
856                  * Handle aborted connections: we and the server
857                  * think there is an existing connection but the client
858                  * aborts it and starts a new one.
859                  */
860                 if (((ct->proto.tcp.seen[dir].flags
861                       | ct->proto.tcp.seen[!dir].flags)
862                      & IP_CT_TCP_FLAG_CLOSE_INIT)
863                     || (ct->proto.tcp.last_dir == dir
864                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
865                         /* Attempt to reopen a closed/aborted connection.
866                          * Delete this connection and look up again. */
867                         spin_unlock_bh(&ct->lock);
868
869                         /* Only repeat if we can actually remove the timer.
870                          * Destruction may already be in progress in process
871                          * context and we must give it a chance to terminate.
872                          */
873                         if (nf_ct_kill(ct))
874                                 return -NF_REPEAT;
875                         return NF_DROP;
876                 }
877                 /* Fall through */
878         case TCP_CONNTRACK_IGNORE:
879                 /* Ignored packets:
880                  *
881                  * Our connection entry may be out of sync, so ignore
882                  * packets which may signal the real connection between
883                  * the client and the server.
884                  *
885                  * a) SYN in ORIGINAL
886                  * b) SYN/ACK in REPLY
887                  * c) ACK in reply direction after initial SYN in original.
888                  *
889                  * If the ignored packet is invalid, the receiver will send
890                  * a RST we'll catch below.
891                  */
892                 if (index == TCP_SYNACK_SET
893                     && ct->proto.tcp.last_index == TCP_SYN_SET
894                     && ct->proto.tcp.last_dir != dir
895                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
896                         /* b) This SYN/ACK acknowledges a SYN that we earlier
897                          * ignored as invalid. This means that the client and
898                          * the server are both in sync, while the firewall is
899                          * not. We get in sync from the previously annotated
900                          * values.
901                          */
902                         old_state = TCP_CONNTRACK_SYN_SENT;
903                         new_state = TCP_CONNTRACK_SYN_RECV;
904                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
905                                 ct->proto.tcp.last_end;
906                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
907                                 ct->proto.tcp.last_end;
908                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
909                                 ct->proto.tcp.last_win == 0 ?
910                                         1 : ct->proto.tcp.last_win;
911                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
912                                 ct->proto.tcp.last_wscale;
913                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
914                                 ct->proto.tcp.last_flags;
915                         memset(&ct->proto.tcp.seen[dir], 0,
916                                sizeof(struct ip_ct_tcp_state));
917                         break;
918                 }
919                 ct->proto.tcp.last_index = index;
920                 ct->proto.tcp.last_dir = dir;
921                 ct->proto.tcp.last_seq = ntohl(th->seq);
922                 ct->proto.tcp.last_end =
923                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
924                 ct->proto.tcp.last_win = ntohs(th->window);
925
926                 /* a) This is a SYN in ORIGINAL. The client and the server
927                  * may be in sync but we are not. In that case, we annotate
928                  * the TCP options and let the packet go through. If it is a
929                  * valid SYN packet, the server will reply with a SYN/ACK, and
930                  * then we'll get in sync. Otherwise, the server ignores it. */
931                 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
932                         struct ip_ct_tcp_state seen = {};
933
934                         ct->proto.tcp.last_flags =
935                         ct->proto.tcp.last_wscale = 0;
936                         tcp_options(skb, dataoff, th, &seen);
937                         if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
938                                 ct->proto.tcp.last_flags |=
939                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
940                                 ct->proto.tcp.last_wscale = seen.td_scale;
941                         }
942                         if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
943                                 ct->proto.tcp.last_flags |=
944                                         IP_CT_TCP_FLAG_SACK_PERM;
945                         }
946                 }
947                 spin_unlock_bh(&ct->lock);
948                 if (LOG_INVALID(net, IPPROTO_TCP))
949                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
950                                   "nf_ct_tcp: invalid packet ignored ");
951                 return NF_ACCEPT;
952         case TCP_CONNTRACK_MAX:
953                 /* Invalid packet */
954                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
955                          dir, get_conntrack_index(th), old_state);
956                 spin_unlock_bh(&ct->lock);
957                 if (LOG_INVALID(net, IPPROTO_TCP))
958                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
959                                   "nf_ct_tcp: invalid state ");
960                 return -NF_ACCEPT;
961         case TCP_CONNTRACK_CLOSE:
962                 if (index == TCP_RST_SET
963                     && (ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET)
964                     && before(ntohl(th->seq), ct->proto.tcp.seen[!dir].td_maxack)) {
965                         /* Invalid RST  */
966                         spin_unlock_bh(&ct->lock);
967                         if (LOG_INVALID(net, IPPROTO_TCP))
968                                 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
969                                           "nf_ct_tcp: invalid RST ");
970                         return -NF_ACCEPT;
971                 }
972                 if (index == TCP_RST_SET
973                     && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
974                          && ct->proto.tcp.last_index == TCP_SYN_SET)
975                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
976                             && ct->proto.tcp.last_index == TCP_ACK_SET))
977                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
978                         /* RST sent to invalid SYN or ACK we had let through
979                          * at a) and c) above:
980                          *
981                          * a) SYN was in window then
982                          * c) we hold a half-open connection.
983                          *
984                          * Delete our connection entry.
985                          * We skip window checking, because packet might ACK
986                          * segments we ignored. */
987                         goto in_window;
988                 }
989                 /* Just fall through */
990         default:
991                 /* Keep compilers happy. */
992                 break;
993         }
994
995         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
996                            skb, dataoff, th, pf)) {
997                 spin_unlock_bh(&ct->lock);
998                 return -NF_ACCEPT;
999         }
1000      in_window:
1001         /* From now on we have got in-window packets */
1002         ct->proto.tcp.last_index = index;
1003         ct->proto.tcp.last_dir = dir;
1004
1005         pr_debug("tcp_conntracks: ");
1006         nf_ct_dump_tuple(tuple);
1007         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1008                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1009                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1010                  old_state, new_state);
1011
1012         ct->proto.tcp.state = new_state;
1013         if (old_state != new_state
1014             && new_state == TCP_CONNTRACK_FIN_WAIT)
1015                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1016
1017         if (ct->proto.tcp.retrans >= nf_ct_tcp_max_retrans &&
1018             tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans)
1019                 timeout = nf_ct_tcp_timeout_max_retrans;
1020         else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1021                  IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1022                  tcp_timeouts[new_state] > nf_ct_tcp_timeout_unacknowledged)
1023                 timeout = nf_ct_tcp_timeout_unacknowledged;
1024         else
1025                 timeout = tcp_timeouts[new_state];
1026         spin_unlock_bh(&ct->lock);
1027
1028         if (new_state != old_state)
1029                 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1030
1031         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1032                 /* If only reply is a RST, we can consider ourselves not to
1033                    have an established connection: this is a fairly common
1034                    problem case, so we can delete the conntrack
1035                    immediately.  --RR */
1036                 if (th->rst) {
1037                         nf_ct_kill_acct(ct, ctinfo, skb);
1038                         return NF_ACCEPT;
1039                 }
1040         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1041                    && (old_state == TCP_CONNTRACK_SYN_RECV
1042                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1043                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1044                 /* Set ASSURED if we see see valid ack in ESTABLISHED
1045                    after SYN_RECV or a valid answer for a picked up
1046                    connection. */
1047                 set_bit(IPS_ASSURED_BIT, &ct->status);
1048                 nf_conntrack_event_cache(IPCT_STATUS, ct);
1049         }
1050         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1051
1052         return NF_ACCEPT;
1053 }
1054
1055 /* Called when a new connection for this protocol found. */
1056 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
1057                     unsigned int dataoff)
1058 {
1059         enum tcp_conntrack new_state;
1060         const struct tcphdr *th;
1061         struct tcphdr _tcph;
1062         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
1063         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
1064
1065         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1066         BUG_ON(th == NULL);
1067
1068         /* Don't need lock here: this conntrack not in circulation yet */
1069         new_state
1070                 = tcp_conntracks[0][get_conntrack_index(th)]
1071                 [TCP_CONNTRACK_NONE];
1072
1073         /* Invalid: delete conntrack */
1074         if (new_state >= TCP_CONNTRACK_MAX) {
1075                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1076                 return false;
1077         }
1078
1079         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1080                 /* SYN packet */
1081                 ct->proto.tcp.seen[0].td_end =
1082                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1083                                              dataoff, th);
1084                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1085                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1086                         ct->proto.tcp.seen[0].td_maxwin = 1;
1087                 ct->proto.tcp.seen[0].td_maxend =
1088                         ct->proto.tcp.seen[0].td_end;
1089
1090                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1091                 ct->proto.tcp.seen[1].flags = 0;
1092         } else if (nf_ct_tcp_loose == 0) {
1093                 /* Don't try to pick up connections. */
1094                 return false;
1095         } else {
1096                 /*
1097                  * We are in the middle of a connection,
1098                  * its history is lost for us.
1099                  * Let's try to use the data from the packet.
1100                  */
1101                 ct->proto.tcp.seen[0].td_end =
1102                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1103                                              dataoff, th);
1104                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1105                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1106                         ct->proto.tcp.seen[0].td_maxwin = 1;
1107                 ct->proto.tcp.seen[0].td_maxend =
1108                         ct->proto.tcp.seen[0].td_end +
1109                         ct->proto.tcp.seen[0].td_maxwin;
1110                 ct->proto.tcp.seen[0].td_scale = 0;
1111
1112                 /* We assume SACK and liberal window checking to handle
1113                  * window scaling */
1114                 ct->proto.tcp.seen[0].flags =
1115                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1116                                               IP_CT_TCP_FLAG_BE_LIBERAL;
1117         }
1118
1119         ct->proto.tcp.seen[1].td_end = 0;
1120         ct->proto.tcp.seen[1].td_maxend = 0;
1121         ct->proto.tcp.seen[1].td_maxwin = 0;
1122         ct->proto.tcp.seen[1].td_scale = 0;
1123
1124         /* tcp_packet will set them */
1125         ct->proto.tcp.state = TCP_CONNTRACK_NONE;
1126         ct->proto.tcp.last_index = TCP_NONE_SET;
1127
1128         pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1129                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1130                  sender->td_end, sender->td_maxend, sender->td_maxwin,
1131                  sender->td_scale,
1132                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1133                  receiver->td_scale);
1134         return true;
1135 }
1136
1137 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1138
1139 #include <linux/netfilter/nfnetlink.h>
1140 #include <linux/netfilter/nfnetlink_conntrack.h>
1141
1142 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1143                          struct nf_conn *ct)
1144 {
1145         struct nlattr *nest_parms;
1146         struct nf_ct_tcp_flags tmp = {};
1147
1148         spin_lock_bh(&ct->lock);
1149         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1150         if (!nest_parms)
1151                 goto nla_put_failure;
1152
1153         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state);
1154
1155         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1156                    ct->proto.tcp.seen[0].td_scale);
1157
1158         NLA_PUT_U8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1159                    ct->proto.tcp.seen[1].td_scale);
1160
1161         tmp.flags = ct->proto.tcp.seen[0].flags;
1162         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1163                 sizeof(struct nf_ct_tcp_flags), &tmp);
1164
1165         tmp.flags = ct->proto.tcp.seen[1].flags;
1166         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1167                 sizeof(struct nf_ct_tcp_flags), &tmp);
1168         spin_unlock_bh(&ct->lock);
1169
1170         nla_nest_end(skb, nest_parms);
1171
1172         return 0;
1173
1174 nla_put_failure:
1175         spin_unlock_bh(&ct->lock);
1176         return -1;
1177 }
1178
1179 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1180         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1181         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1182         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1183         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1184         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len =  sizeof(struct nf_ct_tcp_flags) },
1185 };
1186
1187 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1188 {
1189         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1190         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1191         int err;
1192
1193         /* updates could not contain anything about the private
1194          * protocol info, in that case skip the parsing */
1195         if (!pattr)
1196                 return 0;
1197
1198         err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr, tcp_nla_policy);
1199         if (err < 0)
1200                 return err;
1201
1202         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1203             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1204                 return -EINVAL;
1205
1206         spin_lock_bh(&ct->lock);
1207         if (tb[CTA_PROTOINFO_TCP_STATE])
1208                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1209
1210         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1211                 struct nf_ct_tcp_flags *attr =
1212                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1213                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1214                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1215         }
1216
1217         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1218                 struct nf_ct_tcp_flags *attr =
1219                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1220                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1221                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1222         }
1223
1224         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1225             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1226             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1227             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1228                 ct->proto.tcp.seen[0].td_scale =
1229                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1230                 ct->proto.tcp.seen[1].td_scale =
1231                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1232         }
1233         spin_unlock_bh(&ct->lock);
1234
1235         return 0;
1236 }
1237
1238 static int tcp_nlattr_size(void)
1239 {
1240         return nla_total_size(0)           /* CTA_PROTOINFO_TCP */
1241                 + nla_policy_len(tcp_nla_policy, CTA_PROTOINFO_TCP_MAX + 1);
1242 }
1243
1244 static int tcp_nlattr_tuple_size(void)
1245 {
1246         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1247 }
1248 #endif
1249
1250 #ifdef CONFIG_SYSCTL
1251 static unsigned int tcp_sysctl_table_users;
1252 static struct ctl_table_header *tcp_sysctl_header;
1253 static struct ctl_table tcp_sysctl_table[] = {
1254         {
1255                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1256                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1257                 .maxlen         = sizeof(unsigned int),
1258                 .mode           = 0644,
1259                 .proc_handler   = proc_dointvec_jiffies,
1260         },
1261         {
1262                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1263                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1264                 .maxlen         = sizeof(unsigned int),
1265                 .mode           = 0644,
1266                 .proc_handler   = proc_dointvec_jiffies,
1267         },
1268         {
1269                 .procname       = "nf_conntrack_tcp_timeout_established",
1270                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1271                 .maxlen         = sizeof(unsigned int),
1272                 .mode           = 0644,
1273                 .proc_handler   = proc_dointvec_jiffies,
1274         },
1275         {
1276                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1277                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1278                 .maxlen         = sizeof(unsigned int),
1279                 .mode           = 0644,
1280                 .proc_handler   = proc_dointvec_jiffies,
1281         },
1282         {
1283                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1284                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1285                 .maxlen         = sizeof(unsigned int),
1286                 .mode           = 0644,
1287                 .proc_handler   = proc_dointvec_jiffies,
1288         },
1289         {
1290                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1291                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1292                 .maxlen         = sizeof(unsigned int),
1293                 .mode           = 0644,
1294                 .proc_handler   = proc_dointvec_jiffies,
1295         },
1296         {
1297                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1298                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1299                 .maxlen         = sizeof(unsigned int),
1300                 .mode           = 0644,
1301                 .proc_handler   = proc_dointvec_jiffies,
1302         },
1303         {
1304                 .procname       = "nf_conntrack_tcp_timeout_close",
1305                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1306                 .maxlen         = sizeof(unsigned int),
1307                 .mode           = 0644,
1308                 .proc_handler   = proc_dointvec_jiffies,
1309         },
1310         {
1311                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1312                 .data           = &nf_ct_tcp_timeout_max_retrans,
1313                 .maxlen         = sizeof(unsigned int),
1314                 .mode           = 0644,
1315                 .proc_handler   = proc_dointvec_jiffies,
1316         },
1317         {
1318                 .procname       = "nf_conntrack_tcp_timeout_unacknowledged",
1319                 .data           = &nf_ct_tcp_timeout_unacknowledged,
1320                 .maxlen         = sizeof(unsigned int),
1321                 .mode           = 0644,
1322                 .proc_handler   = proc_dointvec_jiffies,
1323         },
1324         {
1325                 .procname       = "nf_conntrack_tcp_loose",
1326                 .data           = &nf_ct_tcp_loose,
1327                 .maxlen         = sizeof(unsigned int),
1328                 .mode           = 0644,
1329                 .proc_handler   = proc_dointvec,
1330         },
1331         {
1332                 .procname       = "nf_conntrack_tcp_be_liberal",
1333                 .data           = &nf_ct_tcp_be_liberal,
1334                 .maxlen         = sizeof(unsigned int),
1335                 .mode           = 0644,
1336                 .proc_handler   = proc_dointvec,
1337         },
1338         {
1339                 .procname       = "nf_conntrack_tcp_max_retrans",
1340                 .data           = &nf_ct_tcp_max_retrans,
1341                 .maxlen         = sizeof(unsigned int),
1342                 .mode           = 0644,
1343                 .proc_handler   = proc_dointvec,
1344         },
1345         { }
1346 };
1347
1348 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1349 static struct ctl_table tcp_compat_sysctl_table[] = {
1350         {
1351                 .procname       = "ip_conntrack_tcp_timeout_syn_sent",
1352                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT],
1353                 .maxlen         = sizeof(unsigned int),
1354                 .mode           = 0644,
1355                 .proc_handler   = proc_dointvec_jiffies,
1356         },
1357         {
1358                 .procname       = "ip_conntrack_tcp_timeout_syn_sent2",
1359                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_SENT2],
1360                 .maxlen         = sizeof(unsigned int),
1361                 .mode           = 0644,
1362                 .proc_handler   = proc_dointvec_jiffies,
1363         },
1364         {
1365                 .procname       = "ip_conntrack_tcp_timeout_syn_recv",
1366                 .data           = &tcp_timeouts[TCP_CONNTRACK_SYN_RECV],
1367                 .maxlen         = sizeof(unsigned int),
1368                 .mode           = 0644,
1369                 .proc_handler   = proc_dointvec_jiffies,
1370         },
1371         {
1372                 .procname       = "ip_conntrack_tcp_timeout_established",
1373                 .data           = &tcp_timeouts[TCP_CONNTRACK_ESTABLISHED],
1374                 .maxlen         = sizeof(unsigned int),
1375                 .mode           = 0644,
1376                 .proc_handler   = proc_dointvec_jiffies,
1377         },
1378         {
1379                 .procname       = "ip_conntrack_tcp_timeout_fin_wait",
1380                 .data           = &tcp_timeouts[TCP_CONNTRACK_FIN_WAIT],
1381                 .maxlen         = sizeof(unsigned int),
1382                 .mode           = 0644,
1383                 .proc_handler   = proc_dointvec_jiffies,
1384         },
1385         {
1386                 .procname       = "ip_conntrack_tcp_timeout_close_wait",
1387                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE_WAIT],
1388                 .maxlen         = sizeof(unsigned int),
1389                 .mode           = 0644,
1390                 .proc_handler   = proc_dointvec_jiffies,
1391         },
1392         {
1393                 .procname       = "ip_conntrack_tcp_timeout_last_ack",
1394                 .data           = &tcp_timeouts[TCP_CONNTRACK_LAST_ACK],
1395                 .maxlen         = sizeof(unsigned int),
1396                 .mode           = 0644,
1397                 .proc_handler   = proc_dointvec_jiffies,
1398         },
1399         {
1400                 .procname       = "ip_conntrack_tcp_timeout_time_wait",
1401                 .data           = &tcp_timeouts[TCP_CONNTRACK_TIME_WAIT],
1402                 .maxlen         = sizeof(unsigned int),
1403                 .mode           = 0644,
1404                 .proc_handler   = proc_dointvec_jiffies,
1405         },
1406         {
1407                 .procname       = "ip_conntrack_tcp_timeout_close",
1408                 .data           = &tcp_timeouts[TCP_CONNTRACK_CLOSE],
1409                 .maxlen         = sizeof(unsigned int),
1410                 .mode           = 0644,
1411                 .proc_handler   = proc_dointvec_jiffies,
1412         },
1413         {
1414                 .procname       = "ip_conntrack_tcp_timeout_max_retrans",
1415                 .data           = &nf_ct_tcp_timeout_max_retrans,
1416                 .maxlen         = sizeof(unsigned int),
1417                 .mode           = 0644,
1418                 .proc_handler   = proc_dointvec_jiffies,
1419         },
1420         {
1421                 .procname       = "ip_conntrack_tcp_loose",
1422                 .data           = &nf_ct_tcp_loose,
1423                 .maxlen         = sizeof(unsigned int),
1424                 .mode           = 0644,
1425                 .proc_handler   = proc_dointvec,
1426         },
1427         {
1428                 .procname       = "ip_conntrack_tcp_be_liberal",
1429                 .data           = &nf_ct_tcp_be_liberal,
1430                 .maxlen         = sizeof(unsigned int),
1431                 .mode           = 0644,
1432                 .proc_handler   = proc_dointvec,
1433         },
1434         {
1435                 .procname       = "ip_conntrack_tcp_max_retrans",
1436                 .data           = &nf_ct_tcp_max_retrans,
1437                 .maxlen         = sizeof(unsigned int),
1438                 .mode           = 0644,
1439                 .proc_handler   = proc_dointvec,
1440         },
1441         { }
1442 };
1443 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
1444 #endif /* CONFIG_SYSCTL */
1445
1446 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1447 {
1448         .l3proto                = PF_INET,
1449         .l4proto                = IPPROTO_TCP,
1450         .name                   = "tcp",
1451         .pkt_to_tuple           = tcp_pkt_to_tuple,
1452         .invert_tuple           = tcp_invert_tuple,
1453         .print_tuple            = tcp_print_tuple,
1454         .print_conntrack        = tcp_print_conntrack,
1455         .packet                 = tcp_packet,
1456         .new                    = tcp_new,
1457         .error                  = tcp_error,
1458 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1459         .to_nlattr              = tcp_to_nlattr,
1460         .nlattr_size            = tcp_nlattr_size,
1461         .from_nlattr            = nlattr_to_tcp,
1462         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1463         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1464         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1465         .nla_policy             = nf_ct_port_nla_policy,
1466 #endif
1467 #ifdef CONFIG_SYSCTL
1468         .ctl_table_users        = &tcp_sysctl_table_users,
1469         .ctl_table_header       = &tcp_sysctl_header,
1470         .ctl_table              = tcp_sysctl_table,
1471 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1472         .ctl_compat_table       = tcp_compat_sysctl_table,
1473 #endif
1474 #endif
1475 };
1476 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1477
1478 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1479 {
1480         .l3proto                = PF_INET6,
1481         .l4proto                = IPPROTO_TCP,
1482         .name                   = "tcp",
1483         .pkt_to_tuple           = tcp_pkt_to_tuple,
1484         .invert_tuple           = tcp_invert_tuple,
1485         .print_tuple            = tcp_print_tuple,
1486         .print_conntrack        = tcp_print_conntrack,
1487         .packet                 = tcp_packet,
1488         .new                    = tcp_new,
1489         .error                  = tcp_error,
1490 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1491         .to_nlattr              = tcp_to_nlattr,
1492         .nlattr_size            = tcp_nlattr_size,
1493         .from_nlattr            = nlattr_to_tcp,
1494         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1495         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1496         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1497         .nla_policy             = nf_ct_port_nla_policy,
1498 #endif
1499 #ifdef CONFIG_SYSCTL
1500         .ctl_table_users        = &tcp_sysctl_table_users,
1501         .ctl_table_header       = &tcp_sysctl_header,
1502         .ctl_table              = tcp_sysctl_table,
1503 #endif
1504 };
1505 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);