[NETFILTER]: ctnetlink: use netlink policy
[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
19 #include <net/tcp.h>
20
21 #include <linux/netfilter.h>
22 #include <linux/netfilter_ipv4.h>
23 #include <linux/netfilter_ipv6.h>
24 #include <net/netfilter/nf_conntrack.h>
25 #include <net/netfilter/nf_conntrack_l4proto.h>
26 #include <net/netfilter/nf_conntrack_ecache.h>
27
28 /* Protects conntrack->proto.tcp */
29 static DEFINE_RWLOCK(tcp_lock);
30
31 /* "Be conservative in what you do,
32     be liberal in what you accept from others."
33     If it's non-zero, we mark only out of window RST segments as INVALID. */
34 static int nf_ct_tcp_be_liberal __read_mostly = 0;
35
36 /* If it is set to zero, we disable picking up already established
37    connections. */
38 static int nf_ct_tcp_loose __read_mostly = 1;
39
40 /* Max number of the retransmitted packets without receiving an (acceptable)
41    ACK from the destination. If this number is reached, a shorter timer
42    will be started. */
43 static int nf_ct_tcp_max_retrans __read_mostly = 3;
44
45   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
46      closely.  They're more complex. --RR */
47
48 static const char *tcp_conntrack_names[] = {
49         "NONE",
50         "SYN_SENT",
51         "SYN_RECV",
52         "ESTABLISHED",
53         "FIN_WAIT",
54         "CLOSE_WAIT",
55         "LAST_ACK",
56         "TIME_WAIT",
57         "CLOSE",
58         "LISTEN"
59 };
60
61 #define SECS * HZ
62 #define MINS * 60 SECS
63 #define HOURS * 60 MINS
64 #define DAYS * 24 HOURS
65
66 static unsigned int nf_ct_tcp_timeout_syn_sent __read_mostly =      2 MINS;
67 static unsigned int nf_ct_tcp_timeout_syn_recv __read_mostly =     60 SECS;
68 static unsigned int nf_ct_tcp_timeout_established __read_mostly =   5 DAYS;
69 static unsigned int nf_ct_tcp_timeout_fin_wait __read_mostly =      2 MINS;
70 static unsigned int nf_ct_tcp_timeout_close_wait __read_mostly =   60 SECS;
71 static unsigned int nf_ct_tcp_timeout_last_ack __read_mostly =     30 SECS;
72 static unsigned int nf_ct_tcp_timeout_time_wait __read_mostly =     2 MINS;
73 static unsigned int nf_ct_tcp_timeout_close __read_mostly =        10 SECS;
74
75 /* RFC1122 says the R2 limit should be at least 100 seconds.
76    Linux uses 15 packets as limit, which corresponds
77    to ~13-30min depending on RTO. */
78 static unsigned int nf_ct_tcp_timeout_max_retrans __read_mostly =   5 MINS;
79
80 static unsigned int * tcp_timeouts[] = {
81     NULL,                              /* TCP_CONNTRACK_NONE */
82     &nf_ct_tcp_timeout_syn_sent,       /* TCP_CONNTRACK_SYN_SENT, */
83     &nf_ct_tcp_timeout_syn_recv,       /* TCP_CONNTRACK_SYN_RECV, */
84     &nf_ct_tcp_timeout_established,    /* TCP_CONNTRACK_ESTABLISHED, */
85     &nf_ct_tcp_timeout_fin_wait,       /* TCP_CONNTRACK_FIN_WAIT, */
86     &nf_ct_tcp_timeout_close_wait,     /* TCP_CONNTRACK_CLOSE_WAIT, */
87     &nf_ct_tcp_timeout_last_ack,       /* TCP_CONNTRACK_LAST_ACK, */
88     &nf_ct_tcp_timeout_time_wait,      /* TCP_CONNTRACK_TIME_WAIT, */
89     &nf_ct_tcp_timeout_close,          /* TCP_CONNTRACK_CLOSE, */
90     NULL,                              /* TCP_CONNTRACK_LISTEN */
91  };
92
93 #define sNO TCP_CONNTRACK_NONE
94 #define sSS TCP_CONNTRACK_SYN_SENT
95 #define sSR TCP_CONNTRACK_SYN_RECV
96 #define sES TCP_CONNTRACK_ESTABLISHED
97 #define sFW TCP_CONNTRACK_FIN_WAIT
98 #define sCW TCP_CONNTRACK_CLOSE_WAIT
99 #define sLA TCP_CONNTRACK_LAST_ACK
100 #define sTW TCP_CONNTRACK_TIME_WAIT
101 #define sCL TCP_CONNTRACK_CLOSE
102 #define sLI TCP_CONNTRACK_LISTEN
103 #define sIV TCP_CONNTRACK_MAX
104 #define sIG TCP_CONNTRACK_IGNORE
105
106 /* What TCP flags are set from RST/SYN/FIN/ACK. */
107 enum tcp_bit_set {
108         TCP_SYN_SET,
109         TCP_SYNACK_SET,
110         TCP_FIN_SET,
111         TCP_ACK_SET,
112         TCP_RST_SET,
113         TCP_NONE_SET,
114 };
115
116 /*
117  * The TCP state transition table needs a few words...
118  *
119  * We are the man in the middle. All the packets go through us
120  * but might get lost in transit to the destination.
121  * It is assumed that the destinations can't receive segments
122  * we haven't seen.
123  *
124  * The checked segment is in window, but our windows are *not*
125  * equivalent with the ones of the sender/receiver. We always
126  * try to guess the state of the current sender.
127  *
128  * The meaning of the states are:
129  *
130  * NONE:        initial state
131  * SYN_SENT:    SYN-only packet seen
132  * SYN_RECV:    SYN-ACK packet seen
133  * ESTABLISHED: ACK packet seen
134  * FIN_WAIT:    FIN packet seen
135  * CLOSE_WAIT:  ACK seen (after FIN)
136  * LAST_ACK:    FIN seen (after FIN)
137  * TIME_WAIT:   last ACK seen
138  * CLOSE:       closed connection
139  *
140  * LISTEN state is not used.
141  *
142  * Packets marked as IGNORED (sIG):
143  *      if they may be either invalid or valid
144  *      and the receiver may send back a connection
145  *      closing RST or a SYN/ACK.
146  *
147  * Packets marked as INVALID (sIV):
148  *      if they are invalid
149  *      or we do not support the request (simultaneous open)
150  */
151 static enum tcp_conntrack tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
152         {
153 /* ORIGINAL */
154 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
155 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sIV },
156 /*
157  *      sNO -> sSS      Initialize a new connection
158  *      sSS -> sSS      Retransmitted SYN
159  *      sSR -> sIG      Late retransmitted SYN?
160  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
161  *                      are errors. Receiver will reply with RST
162  *                      and close the connection.
163  *                      Or we are not in sync and hold a dead connection.
164  *      sFW -> sIG
165  *      sCW -> sIG
166  *      sLA -> sIG
167  *      sTW -> sSS      Reopened connection (RFC 1122).
168  *      sCL -> sSS
169  */
170 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
171 /*synack*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
172 /*
173  * A SYN/ACK from the client is always invalid:
174  *      - either it tries to set up a simultaneous open, which is
175  *        not supported;
176  *      - or the firewall has just been inserted between the two hosts
177  *        during the session set-up. The SYN will be retransmitted
178  *        by the true client (or it'll time out).
179  */
180 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
181 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
182 /*
183  *      sNO -> sIV      Too late and no reason to do anything...
184  *      sSS -> sIV      Client migth not send FIN in this state:
185  *                      we enforce waiting for a SYN/ACK reply first.
186  *      sSR -> sFW      Close started.
187  *      sES -> sFW
188  *      sFW -> sLA      FIN seen in both directions, waiting for
189  *                      the last ACK.
190  *                      Migth be a retransmitted FIN as well...
191  *      sCW -> sLA
192  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
193  *      sTW -> sTW
194  *      sCL -> sCL
195  */
196 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
197 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
198 /*
199  *      sNO -> sES      Assumed.
200  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
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, sLI   */
210 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
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, sLI   */
216 /*syn*/    { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV },
217 /*
218  *      sNO -> sIV      Never reached.
219  *      sSS -> sIV      Simultaneous open, not supported
220  *      sSR -> sIV      Simultaneous open, not supported.
221  *      sES -> sIV      Server may not initiate a connection.
222  *      sFW -> sIV
223  *      sCW -> sIV
224  *      sLA -> sIV
225  *      sTW -> sIV      Reopened connection, but server may not do it.
226  *      sCL -> sIV
227  */
228 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
229 /*synack*/ { sIV, sSR, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIV },
230 /*
231  *      sSS -> sSR      Standard open.
232  *      sSR -> sSR      Retransmitted SYN/ACK.
233  *      sES -> sIG      Late retransmitted SYN/ACK?
234  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
235  *      sCW -> sIG
236  *      sLA -> sIG
237  *      sTW -> sIG
238  *      sCL -> sIG
239  */
240 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
241 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
242 /*
243  *      sSS -> sIV      Server might not send FIN in this state.
244  *      sSR -> sFW      Close started.
245  *      sES -> sFW
246  *      sFW -> sLA      FIN seen in both directions.
247  *      sCW -> sLA
248  *      sLA -> sLA      Retransmitted FIN.
249  *      sTW -> sTW
250  *      sCL -> sCL
251  */
252 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
253 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIV },
254 /*
255  *      sSS -> sIG      Might be a half-open connection.
256  *      sSR -> sSR      Might answer late resent SYN.
257  *      sES -> sES      :-)
258  *      sFW -> sCW      Normal close request answered by ACK.
259  *      sCW -> sCW
260  *      sLA -> sTW      Last ACK detected.
261  *      sTW -> sTW      Retransmitted last ACK.
262  *      sCL -> sCL
263  */
264 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sLI   */
265 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sIV },
266 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
267         }
268 };
269
270 static int tcp_pkt_to_tuple(const struct sk_buff *skb,
271                             unsigned int dataoff,
272                             struct nf_conntrack_tuple *tuple)
273 {
274         struct tcphdr _hdr, *hp;
275
276         /* Actually only need first 8 bytes. */
277         hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
278         if (hp == NULL)
279                 return 0;
280
281         tuple->src.u.tcp.port = hp->source;
282         tuple->dst.u.tcp.port = hp->dest;
283
284         return 1;
285 }
286
287 static int tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
288                             const struct nf_conntrack_tuple *orig)
289 {
290         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
291         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
292         return 1;
293 }
294
295 /* Print out the per-protocol part of the tuple. */
296 static int tcp_print_tuple(struct seq_file *s,
297                            const struct nf_conntrack_tuple *tuple)
298 {
299         return seq_printf(s, "sport=%hu dport=%hu ",
300                           ntohs(tuple->src.u.tcp.port),
301                           ntohs(tuple->dst.u.tcp.port));
302 }
303
304 /* Print out the private part of the conntrack. */
305 static int tcp_print_conntrack(struct seq_file *s,
306                                const struct nf_conn *conntrack)
307 {
308         enum tcp_conntrack state;
309
310         read_lock_bh(&tcp_lock);
311         state = conntrack->proto.tcp.state;
312         read_unlock_bh(&tcp_lock);
313
314         return seq_printf(s, "%s ", tcp_conntrack_names[state]);
315 }
316
317 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
318 {
319         if (tcph->rst) return TCP_RST_SET;
320         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
321         else if (tcph->fin) return TCP_FIN_SET;
322         else if (tcph->ack) return TCP_ACK_SET;
323         else return TCP_NONE_SET;
324 }
325
326 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
327    in IP Filter' by Guido van Rooij.
328
329    http://www.nluug.nl/events/sane2000/papers.html
330    http://www.iae.nl/users/guido/papers/tcp_filtering.ps.gz
331
332    The boundaries and the conditions are changed according to RFC793:
333    the packet must intersect the window (i.e. segments may be
334    after the right or before the left edge) and thus receivers may ACK
335    segments after the right edge of the window.
336
337         td_maxend = max(sack + max(win,1)) seen in reply packets
338         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
339         td_maxwin += seq + len - sender.td_maxend
340                         if seq + len > sender.td_maxend
341         td_end    = max(seq + len) seen in sent packets
342
343    I.   Upper bound for valid data:     seq <= sender.td_maxend
344    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
345    III. Upper bound for valid ack:      sack <= receiver.td_end
346    IV.  Lower bound for valid ack:      ack >= receiver.td_end - MAXACKWINDOW
347
348    where sack is the highest right edge of sack block found in the packet.
349
350    The upper bound limit for a valid ack is not ignored -
351    we doesn't have to deal with fragments.
352 */
353
354 static inline __u32 segment_seq_plus_len(__u32 seq,
355                                          size_t len,
356                                          unsigned int dataoff,
357                                          struct tcphdr *tcph)
358 {
359         /* XXX Should I use payload length field in IP/IPv6 header ?
360          * - YK */
361         return (seq + len - dataoff - tcph->doff*4
362                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
363 }
364
365 /* Fixme: what about big packets? */
366 #define MAXACKWINCONST                  66000
367 #define MAXACKWINDOW(sender)                                            \
368         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
369                                               : MAXACKWINCONST)
370
371 /*
372  * Simplified tcp_parse_options routine from tcp_input.c
373  */
374 static void tcp_options(const struct sk_buff *skb,
375                         unsigned int dataoff,
376                         struct tcphdr *tcph,
377                         struct ip_ct_tcp_state *state)
378 {
379         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
380         unsigned char *ptr;
381         int length = (tcph->doff*4) - sizeof(struct tcphdr);
382
383         if (!length)
384                 return;
385
386         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
387                                  length, buff);
388         BUG_ON(ptr == NULL);
389
390         state->td_scale =
391         state->flags = 0;
392
393         while (length > 0) {
394                 int opcode=*ptr++;
395                 int opsize;
396
397                 switch (opcode) {
398                 case TCPOPT_EOL:
399                         return;
400                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
401                         length--;
402                         continue;
403                 default:
404                         opsize=*ptr++;
405                         if (opsize < 2) /* "silly options" */
406                                 return;
407                         if (opsize > length)
408                                 break;  /* don't parse partial options */
409
410                         if (opcode == TCPOPT_SACK_PERM
411                             && opsize == TCPOLEN_SACK_PERM)
412                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
413                         else if (opcode == TCPOPT_WINDOW
414                                  && opsize == TCPOLEN_WINDOW) {
415                                 state->td_scale = *(u_int8_t *)ptr;
416
417                                 if (state->td_scale > 14) {
418                                         /* See RFC1323 */
419                                         state->td_scale = 14;
420                                 }
421                                 state->flags |=
422                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
423                         }
424                         ptr += opsize - 2;
425                         length -= opsize;
426                 }
427         }
428 }
429
430 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
431                      struct tcphdr *tcph, __u32 *sack)
432 {
433         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
434         unsigned char *ptr;
435         int length = (tcph->doff*4) - sizeof(struct tcphdr);
436         __u32 tmp;
437
438         if (!length)
439                 return;
440
441         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
442                                  length, buff);
443         BUG_ON(ptr == NULL);
444
445         /* Fast path for timestamp-only option */
446         if (length == TCPOLEN_TSTAMP_ALIGNED*4
447             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
448                                        | (TCPOPT_NOP << 16)
449                                        | (TCPOPT_TIMESTAMP << 8)
450                                        | TCPOLEN_TIMESTAMP))
451                 return;
452
453         while (length > 0) {
454                 int opcode = *ptr++;
455                 int opsize, i;
456
457                 switch (opcode) {
458                 case TCPOPT_EOL:
459                         return;
460                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
461                         length--;
462                         continue;
463                 default:
464                         opsize = *ptr++;
465                         if (opsize < 2) /* "silly options" */
466                                 return;
467                         if (opsize > length)
468                                 break;  /* don't parse partial options */
469
470                         if (opcode == TCPOPT_SACK
471                             && opsize >= (TCPOLEN_SACK_BASE
472                                           + TCPOLEN_SACK_PERBLOCK)
473                             && !((opsize - TCPOLEN_SACK_BASE)
474                                  % TCPOLEN_SACK_PERBLOCK)) {
475                                 for (i = 0;
476                                      i < (opsize - TCPOLEN_SACK_BASE);
477                                      i += TCPOLEN_SACK_PERBLOCK) {
478                                         tmp = ntohl(*((__be32 *)(ptr+i)+1));
479
480                                         if (after(tmp, *sack))
481                                                 *sack = tmp;
482                                 }
483                                 return;
484                         }
485                         ptr += opsize - 2;
486                         length -= opsize;
487                 }
488         }
489 }
490
491 static int tcp_in_window(struct nf_conn *ct,
492                          struct ip_ct_tcp *state,
493                          enum ip_conntrack_dir dir,
494                          unsigned int index,
495                          const struct sk_buff *skb,
496                          unsigned int dataoff,
497                          struct tcphdr *tcph,
498                          int pf)
499 {
500         struct ip_ct_tcp_state *sender = &state->seen[dir];
501         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
502         struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
503         __u32 seq, ack, sack, end, win, swin;
504         int res;
505
506         /*
507          * Get the required data from the packet.
508          */
509         seq = ntohl(tcph->seq);
510         ack = sack = ntohl(tcph->ack_seq);
511         win = ntohs(tcph->window);
512         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
513
514         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
515                 tcp_sack(skb, dataoff, tcph, &sack);
516
517         pr_debug("tcp_in_window: START\n");
518         pr_debug("tcp_in_window: ");
519         NF_CT_DUMP_TUPLE(tuple);
520         pr_debug("seq=%u ack=%u sack=%u win=%u end=%u\n",
521                  seq, ack, sack, win, end);
522         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
523                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
524                  sender->td_end, sender->td_maxend, sender->td_maxwin,
525                  sender->td_scale,
526                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
527                  receiver->td_scale);
528
529         if (sender->td_end == 0) {
530                 /*
531                  * Initialize sender data.
532                  */
533                 if (tcph->syn && tcph->ack) {
534                         /*
535                          * Outgoing SYN-ACK in reply to a SYN.
536                          */
537                         sender->td_end =
538                         sender->td_maxend = end;
539                         sender->td_maxwin = (win == 0 ? 1 : win);
540
541                         tcp_options(skb, dataoff, tcph, sender);
542                         /*
543                          * RFC 1323:
544                          * Both sides must send the Window Scale option
545                          * to enable window scaling in either direction.
546                          */
547                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
548                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
549                                 sender->td_scale =
550                                 receiver->td_scale = 0;
551                 } else {
552                         /*
553                          * We are in the middle of a connection,
554                          * its history is lost for us.
555                          * Let's try to use the data from the packet.
556                          */
557                         sender->td_end = end;
558                         sender->td_maxwin = (win == 0 ? 1 : win);
559                         sender->td_maxend = end + sender->td_maxwin;
560                 }
561         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
562                      && dir == IP_CT_DIR_ORIGINAL)
563                    || (state->state == TCP_CONNTRACK_SYN_RECV
564                      && dir == IP_CT_DIR_REPLY))
565                    && after(end, sender->td_end)) {
566                 /*
567                  * RFC 793: "if a TCP is reinitialized ... then it need
568                  * not wait at all; it must only be sure to use sequence
569                  * numbers larger than those recently used."
570                  */
571                 sender->td_end =
572                 sender->td_maxend = end;
573                 sender->td_maxwin = (win == 0 ? 1 : win);
574
575                 tcp_options(skb, dataoff, tcph, sender);
576         }
577
578         if (!(tcph->ack)) {
579                 /*
580                  * If there is no ACK, just pretend it was set and OK.
581                  */
582                 ack = sack = receiver->td_end;
583         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
584                     (TCP_FLAG_ACK|TCP_FLAG_RST))
585                    && (ack == 0)) {
586                 /*
587                  * Broken TCP stacks, that set ACK in RST packets as well
588                  * with zero ack value.
589                  */
590                 ack = sack = receiver->td_end;
591         }
592
593         if (seq == end
594             && (!tcph->rst
595                 || (seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)))
596                 /*
597                  * Packets contains no data: we assume it is valid
598                  * and check the ack value only.
599                  * However RST segments are always validated by their
600                  * SEQ number, except when seq == 0 (reset sent answering
601                  * SYN.
602                  */
603                 seq = end = sender->td_end;
604
605         pr_debug("tcp_in_window: ");
606         NF_CT_DUMP_TUPLE(tuple);
607         pr_debug("seq=%u ack=%u sack =%u win=%u end=%u\n",
608                  seq, ack, sack, win, end);
609         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
610                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
611                  sender->td_end, sender->td_maxend, sender->td_maxwin,
612                  sender->td_scale,
613                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
614                  receiver->td_scale);
615
616         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
617                  before(seq, sender->td_maxend + 1),
618                  after(end, sender->td_end - receiver->td_maxwin - 1),
619                  before(sack, receiver->td_end + 1),
620                  after(ack, receiver->td_end - MAXACKWINDOW(sender)));
621
622         if (before(seq, sender->td_maxend + 1) &&
623             after(end, sender->td_end - receiver->td_maxwin - 1) &&
624             before(sack, receiver->td_end + 1) &&
625             after(ack, receiver->td_end - MAXACKWINDOW(sender))) {
626                 /*
627                  * Take into account window scaling (RFC 1323).
628                  */
629                 if (!tcph->syn)
630                         win <<= sender->td_scale;
631
632                 /*
633                  * Update sender data.
634                  */
635                 swin = win + (sack - ack);
636                 if (sender->td_maxwin < swin)
637                         sender->td_maxwin = swin;
638                 if (after(end, sender->td_end))
639                         sender->td_end = end;
640                 /*
641                  * Update receiver data.
642                  */
643                 if (after(end, sender->td_maxend))
644                         receiver->td_maxwin += end - sender->td_maxend;
645                 if (after(sack + win, receiver->td_maxend - 1)) {
646                         receiver->td_maxend = sack + win;
647                         if (win == 0)
648                                 receiver->td_maxend++;
649                 }
650
651                 /*
652                  * Check retransmissions.
653                  */
654                 if (index == TCP_ACK_SET) {
655                         if (state->last_dir == dir
656                             && state->last_seq == seq
657                             && state->last_ack == ack
658                             && state->last_end == end
659                             && state->last_win == win)
660                                 state->retrans++;
661                         else {
662                                 state->last_dir = dir;
663                                 state->last_seq = seq;
664                                 state->last_ack = ack;
665                                 state->last_end = end;
666                                 state->last_win = win;
667                                 state->retrans = 0;
668                         }
669                 }
670                 res = 1;
671         } else {
672                 res = 0;
673                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
674                     nf_ct_tcp_be_liberal)
675                         res = 1;
676                 if (!res && LOG_INVALID(IPPROTO_TCP))
677                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
678                         "nf_ct_tcp: %s ",
679                         before(seq, sender->td_maxend + 1) ?
680                         after(end, sender->td_end - receiver->td_maxwin - 1) ?
681                         before(sack, receiver->td_end + 1) ?
682                         after(ack, receiver->td_end - MAXACKWINDOW(sender)) ? "BUG"
683                         : "ACK is under the lower bound (possible overly delayed ACK)"
684                         : "ACK is over the upper bound (ACKed data not seen yet)"
685                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
686                         : "SEQ is over the upper bound (over the window of the receiver)");
687         }
688
689         pr_debug("tcp_in_window: res=%i sender end=%u maxend=%u maxwin=%u "
690                  "receiver end=%u maxend=%u maxwin=%u\n",
691                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
692                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
693
694         return res;
695 }
696
697 #ifdef CONFIG_NF_NAT_NEEDED
698 /* Update sender->td_end after NAT successfully mangled the packet */
699 /* Caller must linearize skb at tcp header. */
700 void nf_conntrack_tcp_update(struct sk_buff *skb,
701                              unsigned int dataoff,
702                              struct nf_conn *conntrack,
703                              int dir)
704 {
705         struct tcphdr *tcph = (void *)skb->data + dataoff;
706         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[dir];
707         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[!dir];
708         __u32 end;
709
710         end = segment_seq_plus_len(ntohl(tcph->seq), skb->len, dataoff, tcph);
711
712         write_lock_bh(&tcp_lock);
713         /*
714          * We have to worry for the ack in the reply packet only...
715          */
716         if (after(end, conntrack->proto.tcp.seen[dir].td_end))
717                 conntrack->proto.tcp.seen[dir].td_end = end;
718         conntrack->proto.tcp.last_end = end;
719         write_unlock_bh(&tcp_lock);
720         pr_debug("tcp_update: sender end=%u maxend=%u maxwin=%u scale=%i "
721                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
722                  sender->td_end, sender->td_maxend, sender->td_maxwin,
723                  sender->td_scale,
724                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
725                  receiver->td_scale);
726 }
727 EXPORT_SYMBOL_GPL(nf_conntrack_tcp_update);
728 #endif
729
730 #define TH_FIN  0x01
731 #define TH_SYN  0x02
732 #define TH_RST  0x04
733 #define TH_PUSH 0x08
734 #define TH_ACK  0x10
735 #define TH_URG  0x20
736 #define TH_ECE  0x40
737 #define TH_CWR  0x80
738
739 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
740 static u8 tcp_valid_flags[(TH_FIN|TH_SYN|TH_RST|TH_ACK|TH_URG) + 1] =
741 {
742         [TH_SYN]                        = 1,
743         [TH_SYN|TH_URG]                 = 1,
744         [TH_SYN|TH_ACK]                 = 1,
745         [TH_RST]                        = 1,
746         [TH_RST|TH_ACK]                 = 1,
747         [TH_FIN|TH_ACK]                 = 1,
748         [TH_FIN|TH_ACK|TH_URG]          = 1,
749         [TH_ACK]                        = 1,
750         [TH_ACK|TH_URG]                 = 1,
751 };
752
753 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
754 static int tcp_error(struct sk_buff *skb,
755                      unsigned int dataoff,
756                      enum ip_conntrack_info *ctinfo,
757                      int pf,
758                      unsigned int hooknum)
759 {
760         struct tcphdr _tcph, *th;
761         unsigned int tcplen = skb->len - dataoff;
762         u_int8_t tcpflags;
763
764         /* Smaller that minimal TCP header? */
765         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
766         if (th == NULL) {
767                 if (LOG_INVALID(IPPROTO_TCP))
768                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
769                                 "nf_ct_tcp: short packet ");
770                 return -NF_ACCEPT;
771         }
772
773         /* Not whole TCP header or malformed packet */
774         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
775                 if (LOG_INVALID(IPPROTO_TCP))
776                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
777                                 "nf_ct_tcp: truncated/malformed packet ");
778                 return -NF_ACCEPT;
779         }
780
781         /* Checksum invalid? Ignore.
782          * We skip checking packets on the outgoing path
783          * because the checksum is assumed to be correct.
784          */
785         /* FIXME: Source route IP option packets --RR */
786         if (nf_conntrack_checksum &&
787             ((pf == PF_INET && hooknum == NF_IP_PRE_ROUTING) ||
788              (pf == PF_INET6 && hooknum == NF_IP6_PRE_ROUTING)) &&
789             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
790                 if (LOG_INVALID(IPPROTO_TCP))
791                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
792                                   "nf_ct_tcp: bad TCP checksum ");
793                 return -NF_ACCEPT;
794         }
795
796         /* Check TCP flags. */
797         tcpflags = (((u_int8_t *)th)[13] & ~(TH_ECE|TH_CWR|TH_PUSH));
798         if (!tcp_valid_flags[tcpflags]) {
799                 if (LOG_INVALID(IPPROTO_TCP))
800                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
801                                   "nf_ct_tcp: invalid TCP flag combination ");
802                 return -NF_ACCEPT;
803         }
804
805         return NF_ACCEPT;
806 }
807
808 /* Returns verdict for packet, or -1 for invalid. */
809 static int tcp_packet(struct nf_conn *conntrack,
810                       const struct sk_buff *skb,
811                       unsigned int dataoff,
812                       enum ip_conntrack_info ctinfo,
813                       int pf,
814                       unsigned int hooknum)
815 {
816         struct nf_conntrack_tuple *tuple;
817         enum tcp_conntrack new_state, old_state;
818         enum ip_conntrack_dir dir;
819         struct tcphdr *th, _tcph;
820         unsigned long timeout;
821         unsigned int index;
822
823         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
824         BUG_ON(th == NULL);
825
826         write_lock_bh(&tcp_lock);
827         old_state = conntrack->proto.tcp.state;
828         dir = CTINFO2DIR(ctinfo);
829         index = get_conntrack_index(th);
830         new_state = tcp_conntracks[dir][index][old_state];
831         tuple = &conntrack->tuplehash[dir].tuple;
832
833         switch (new_state) {
834         case TCP_CONNTRACK_IGNORE:
835                 /* Ignored packets:
836                  *
837                  * a) SYN in ORIGINAL
838                  * b) SYN/ACK in REPLY
839                  * c) ACK in reply direction after initial SYN in original.
840                  */
841                 if (index == TCP_SYNACK_SET
842                     && conntrack->proto.tcp.last_index == TCP_SYN_SET
843                     && conntrack->proto.tcp.last_dir != dir
844                     && ntohl(th->ack_seq) ==
845                              conntrack->proto.tcp.last_end) {
846                         /* This SYN/ACK acknowledges a SYN that we earlier
847                          * ignored as invalid. This means that the client and
848                          * the server are both in sync, while the firewall is
849                          * not. We kill this session and block the SYN/ACK so
850                          * that the client cannot but retransmit its SYN and
851                          * thus initiate a clean new session.
852                          */
853                         write_unlock_bh(&tcp_lock);
854                         if (LOG_INVALID(IPPROTO_TCP))
855                                 nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
856                                           "nf_ct_tcp: killing out of sync session ");
857                         if (del_timer(&conntrack->timeout))
858                                 conntrack->timeout.function((unsigned long)
859                                                             conntrack);
860                         return -NF_DROP;
861                 }
862                 conntrack->proto.tcp.last_index = index;
863                 conntrack->proto.tcp.last_dir = dir;
864                 conntrack->proto.tcp.last_seq = ntohl(th->seq);
865                 conntrack->proto.tcp.last_end =
866                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
867
868                 write_unlock_bh(&tcp_lock);
869                 if (LOG_INVALID(IPPROTO_TCP))
870                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
871                                   "nf_ct_tcp: invalid packed ignored ");
872                 return NF_ACCEPT;
873         case TCP_CONNTRACK_MAX:
874                 /* Invalid packet */
875                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
876                          dir, get_conntrack_index(th), old_state);
877                 write_unlock_bh(&tcp_lock);
878                 if (LOG_INVALID(IPPROTO_TCP))
879                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
880                                   "nf_ct_tcp: invalid state ");
881                 return -NF_ACCEPT;
882         case TCP_CONNTRACK_SYN_SENT:
883                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
884                         break;
885                 if ((conntrack->proto.tcp.seen[dir].flags &
886                         IP_CT_TCP_FLAG_CLOSE_INIT)
887                     || after(ntohl(th->seq),
888                              conntrack->proto.tcp.seen[dir].td_end)) {
889                         /* Attempt to reopen a closed connection.
890                         * Delete this connection and look up again. */
891                         write_unlock_bh(&tcp_lock);
892                         if (del_timer(&conntrack->timeout))
893                                 conntrack->timeout.function((unsigned long)
894                                                             conntrack);
895                         return -NF_REPEAT;
896                 } else {
897                         write_unlock_bh(&tcp_lock);
898                         if (LOG_INVALID(IPPROTO_TCP))
899                                 nf_log_packet(pf, 0, skb, NULL, NULL,
900                                               NULL, "nf_ct_tcp: invalid SYN");
901                         return -NF_ACCEPT;
902                 }
903         case TCP_CONNTRACK_CLOSE:
904                 if (index == TCP_RST_SET
905                     && ((test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)
906                          && conntrack->proto.tcp.last_index == TCP_SYN_SET)
907                         || (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
908                             && conntrack->proto.tcp.last_index == TCP_ACK_SET))
909                     && ntohl(th->ack_seq) == conntrack->proto.tcp.last_end) {
910                         /* RST sent to invalid SYN or ACK we had let through
911                          * at a) and c) above:
912                          *
913                          * a) SYN was in window then
914                          * c) we hold a half-open connection.
915                          *
916                          * Delete our connection entry.
917                          * We skip window checking, because packet might ACK
918                          * segments we ignored. */
919                         goto in_window;
920                 }
921                 /* Just fall through */
922         default:
923                 /* Keep compilers happy. */
924                 break;
925         }
926
927         if (!tcp_in_window(conntrack, &conntrack->proto.tcp, dir, index,
928                            skb, dataoff, th, pf)) {
929                 write_unlock_bh(&tcp_lock);
930                 return -NF_ACCEPT;
931         }
932      in_window:
933         /* From now on we have got in-window packets */
934         conntrack->proto.tcp.last_index = index;
935
936         pr_debug("tcp_conntracks: ");
937         NF_CT_DUMP_TUPLE(tuple);
938         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
939                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
940                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
941                  old_state, new_state);
942
943         conntrack->proto.tcp.state = new_state;
944         if (old_state != new_state
945             && (new_state == TCP_CONNTRACK_FIN_WAIT
946                 || new_state == TCP_CONNTRACK_CLOSE))
947                 conntrack->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
948         timeout = conntrack->proto.tcp.retrans >= nf_ct_tcp_max_retrans
949                   && *tcp_timeouts[new_state] > nf_ct_tcp_timeout_max_retrans
950                   ? nf_ct_tcp_timeout_max_retrans : *tcp_timeouts[new_state];
951         write_unlock_bh(&tcp_lock);
952
953         nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
954         if (new_state != old_state)
955                 nf_conntrack_event_cache(IPCT_PROTOINFO, skb);
956
957         if (!test_bit(IPS_SEEN_REPLY_BIT, &conntrack->status)) {
958                 /* If only reply is a RST, we can consider ourselves not to
959                    have an established connection: this is a fairly common
960                    problem case, so we can delete the conntrack
961                    immediately.  --RR */
962                 if (th->rst) {
963                         if (del_timer(&conntrack->timeout))
964                                 conntrack->timeout.function((unsigned long)
965                                                             conntrack);
966                         return NF_ACCEPT;
967                 }
968         } else if (!test_bit(IPS_ASSURED_BIT, &conntrack->status)
969                    && (old_state == TCP_CONNTRACK_SYN_RECV
970                        || old_state == TCP_CONNTRACK_ESTABLISHED)
971                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
972                 /* Set ASSURED if we see see valid ack in ESTABLISHED
973                    after SYN_RECV or a valid answer for a picked up
974                    connection. */
975                 set_bit(IPS_ASSURED_BIT, &conntrack->status);
976                 nf_conntrack_event_cache(IPCT_STATUS, skb);
977         }
978         nf_ct_refresh_acct(conntrack, ctinfo, skb, timeout);
979
980         return NF_ACCEPT;
981 }
982
983 /* Called when a new connection for this protocol found. */
984 static int tcp_new(struct nf_conn *conntrack,
985                    const struct sk_buff *skb,
986                    unsigned int dataoff)
987 {
988         enum tcp_conntrack new_state;
989         struct tcphdr *th, _tcph;
990         struct ip_ct_tcp_state *sender = &conntrack->proto.tcp.seen[0];
991         struct ip_ct_tcp_state *receiver = &conntrack->proto.tcp.seen[1];
992
993         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
994         BUG_ON(th == NULL);
995
996         /* Don't need lock here: this conntrack not in circulation yet */
997         new_state
998                 = tcp_conntracks[0][get_conntrack_index(th)]
999                 [TCP_CONNTRACK_NONE];
1000
1001         /* Invalid: delete conntrack */
1002         if (new_state >= TCP_CONNTRACK_MAX) {
1003                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1004                 return 0;
1005         }
1006
1007         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1008                 /* SYN packet */
1009                 conntrack->proto.tcp.seen[0].td_end =
1010                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1011                                              dataoff, th);
1012                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1013                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1014                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1015                 conntrack->proto.tcp.seen[0].td_maxend =
1016                         conntrack->proto.tcp.seen[0].td_end;
1017
1018                 tcp_options(skb, dataoff, th, &conntrack->proto.tcp.seen[0]);
1019                 conntrack->proto.tcp.seen[1].flags = 0;
1020         } else if (nf_ct_tcp_loose == 0) {
1021                 /* Don't try to pick up connections. */
1022                 return 0;
1023         } else {
1024                 /*
1025                  * We are in the middle of a connection,
1026                  * its history is lost for us.
1027                  * Let's try to use the data from the packet.
1028                  */
1029                 conntrack->proto.tcp.seen[0].td_end =
1030                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1031                                              dataoff, th);
1032                 conntrack->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1033                 if (conntrack->proto.tcp.seen[0].td_maxwin == 0)
1034                         conntrack->proto.tcp.seen[0].td_maxwin = 1;
1035                 conntrack->proto.tcp.seen[0].td_maxend =
1036                         conntrack->proto.tcp.seen[0].td_end +
1037                         conntrack->proto.tcp.seen[0].td_maxwin;
1038                 conntrack->proto.tcp.seen[0].td_scale = 0;
1039
1040                 /* We assume SACK and liberal window checking to handle
1041                  * window scaling */
1042                 conntrack->proto.tcp.seen[0].flags =
1043                 conntrack->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1044                                                      IP_CT_TCP_FLAG_BE_LIBERAL;
1045         }
1046
1047         conntrack->proto.tcp.seen[1].td_end = 0;
1048         conntrack->proto.tcp.seen[1].td_maxend = 0;
1049         conntrack->proto.tcp.seen[1].td_maxwin = 1;
1050         conntrack->proto.tcp.seen[1].td_scale = 0;
1051
1052         /* tcp_packet will set them */
1053         conntrack->proto.tcp.state = TCP_CONNTRACK_NONE;
1054         conntrack->proto.tcp.last_index = TCP_NONE_SET;
1055
1056         pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1057                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1058                  sender->td_end, sender->td_maxend, sender->td_maxwin,
1059                  sender->td_scale,
1060                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1061                  receiver->td_scale);
1062         return 1;
1063 }
1064
1065 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1066
1067 #include <linux/netfilter/nfnetlink.h>
1068 #include <linux/netfilter/nfnetlink_conntrack.h>
1069
1070 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1071                          const struct nf_conn *ct)
1072 {
1073         struct nlattr *nest_parms;
1074         struct nf_ct_tcp_flags tmp = {};
1075
1076         read_lock_bh(&tcp_lock);
1077         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1078         if (!nest_parms)
1079                 goto nla_put_failure;
1080
1081         NLA_PUT(skb, CTA_PROTOINFO_TCP_STATE, sizeof(u_int8_t),
1082                 &ct->proto.tcp.state);
1083
1084         NLA_PUT(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL, sizeof(u_int8_t),
1085                 &ct->proto.tcp.seen[0].td_scale);
1086
1087         NLA_PUT(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY, sizeof(u_int8_t),
1088                 &ct->proto.tcp.seen[1].td_scale);
1089
1090         tmp.flags = ct->proto.tcp.seen[0].flags;
1091         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1092                 sizeof(struct nf_ct_tcp_flags), &tmp);
1093
1094         tmp.flags = ct->proto.tcp.seen[1].flags;
1095         NLA_PUT(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1096                 sizeof(struct nf_ct_tcp_flags), &tmp);
1097         read_unlock_bh(&tcp_lock);
1098
1099         nla_nest_end(skb, nest_parms);
1100
1101         return 0;
1102
1103 nla_put_failure:
1104         read_unlock_bh(&tcp_lock);
1105         return -1;
1106 }
1107
1108 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1109         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1110         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1111         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1112         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1113         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len =  sizeof(struct nf_ct_tcp_flags) },
1114 };
1115
1116 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1117 {
1118         struct nlattr *attr = cda[CTA_PROTOINFO_TCP];
1119         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1120         int err;
1121
1122         /* updates could not contain anything about the private
1123          * protocol info, in that case skip the parsing */
1124         if (!attr)
1125                 return 0;
1126
1127         err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, attr, tcp_nla_policy);
1128         if (err < 0)
1129                 return err;
1130
1131         if (!tb[CTA_PROTOINFO_TCP_STATE])
1132                 return -EINVAL;
1133
1134         write_lock_bh(&tcp_lock);
1135         ct->proto.tcp.state =
1136                 *(u_int8_t *)nla_data(tb[CTA_PROTOINFO_TCP_STATE]);
1137
1138         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1139                 struct nf_ct_tcp_flags *attr =
1140                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1141                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1142                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1143         }
1144
1145         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1146                 struct nf_ct_tcp_flags *attr =
1147                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1148                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1149                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1150         }
1151
1152         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1153             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1154             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1155             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1156                 ct->proto.tcp.seen[0].td_scale = *(u_int8_t *)
1157                         nla_data(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1158                 ct->proto.tcp.seen[1].td_scale = *(u_int8_t *)
1159                         nla_data(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1160         }
1161         write_unlock_bh(&tcp_lock);
1162
1163         return 0;
1164 }
1165 #endif
1166
1167 #ifdef CONFIG_SYSCTL
1168 static unsigned int tcp_sysctl_table_users;
1169 static struct ctl_table_header *tcp_sysctl_header;
1170 static struct ctl_table tcp_sysctl_table[] = {
1171         {
1172                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_SYN_SENT,
1173                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1174                 .data           = &nf_ct_tcp_timeout_syn_sent,
1175                 .maxlen         = sizeof(unsigned int),
1176                 .mode           = 0644,
1177                 .proc_handler   = &proc_dointvec_jiffies,
1178         },
1179         {
1180                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_SYN_RECV,
1181                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1182                 .data           = &nf_ct_tcp_timeout_syn_recv,
1183                 .maxlen         = sizeof(unsigned int),
1184                 .mode           = 0644,
1185                 .proc_handler   = &proc_dointvec_jiffies,
1186         },
1187         {
1188                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_ESTABLISHED,
1189                 .procname       = "nf_conntrack_tcp_timeout_established",
1190                 .data           = &nf_ct_tcp_timeout_established,
1191                 .maxlen         = sizeof(unsigned int),
1192                 .mode           = 0644,
1193                 .proc_handler   = &proc_dointvec_jiffies,
1194         },
1195         {
1196                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_FIN_WAIT,
1197                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1198                 .data           = &nf_ct_tcp_timeout_fin_wait,
1199                 .maxlen         = sizeof(unsigned int),
1200                 .mode           = 0644,
1201                 .proc_handler   = &proc_dointvec_jiffies,
1202         },
1203         {
1204                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_CLOSE_WAIT,
1205                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1206                 .data           = &nf_ct_tcp_timeout_close_wait,
1207                 .maxlen         = sizeof(unsigned int),
1208                 .mode           = 0644,
1209                 .proc_handler   = &proc_dointvec_jiffies,
1210         },
1211         {
1212                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_LAST_ACK,
1213                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1214                 .data           = &nf_ct_tcp_timeout_last_ack,
1215                 .maxlen         = sizeof(unsigned int),
1216                 .mode           = 0644,
1217                 .proc_handler   = &proc_dointvec_jiffies,
1218         },
1219         {
1220                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_TIME_WAIT,
1221                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1222                 .data           = &nf_ct_tcp_timeout_time_wait,
1223                 .maxlen         = sizeof(unsigned int),
1224                 .mode           = 0644,
1225                 .proc_handler   = &proc_dointvec_jiffies,
1226         },
1227         {
1228                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_CLOSE,
1229                 .procname       = "nf_conntrack_tcp_timeout_close",
1230                 .data           = &nf_ct_tcp_timeout_close,
1231                 .maxlen         = sizeof(unsigned int),
1232                 .mode           = 0644,
1233                 .proc_handler   = &proc_dointvec_jiffies,
1234         },
1235         {
1236                 .ctl_name       = NET_NF_CONNTRACK_TCP_TIMEOUT_MAX_RETRANS,
1237                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1238                 .data           = &nf_ct_tcp_timeout_max_retrans,
1239                 .maxlen         = sizeof(unsigned int),
1240                 .mode           = 0644,
1241                 .proc_handler   = &proc_dointvec_jiffies,
1242         },
1243         {
1244                 .ctl_name       = NET_NF_CONNTRACK_TCP_LOOSE,
1245                 .procname       = "nf_conntrack_tcp_loose",
1246                 .data           = &nf_ct_tcp_loose,
1247                 .maxlen         = sizeof(unsigned int),
1248                 .mode           = 0644,
1249                 .proc_handler   = &proc_dointvec,
1250         },
1251         {
1252                 .ctl_name       = NET_NF_CONNTRACK_TCP_BE_LIBERAL,
1253                 .procname       = "nf_conntrack_tcp_be_liberal",
1254                 .data           = &nf_ct_tcp_be_liberal,
1255                 .maxlen         = sizeof(unsigned int),
1256                 .mode           = 0644,
1257                 .proc_handler   = &proc_dointvec,
1258         },
1259         {
1260                 .ctl_name       = NET_NF_CONNTRACK_TCP_MAX_RETRANS,
1261                 .procname       = "nf_conntrack_tcp_max_retrans",
1262                 .data           = &nf_ct_tcp_max_retrans,
1263                 .maxlen         = sizeof(unsigned int),
1264                 .mode           = 0644,
1265                 .proc_handler   = &proc_dointvec,
1266         },
1267         {
1268                 .ctl_name       = 0
1269         }
1270 };
1271
1272 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1273 static struct ctl_table tcp_compat_sysctl_table[] = {
1274         {
1275                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_SYN_SENT,
1276                 .procname       = "ip_conntrack_tcp_timeout_syn_sent",
1277                 .data           = &nf_ct_tcp_timeout_syn_sent,
1278                 .maxlen         = sizeof(unsigned int),
1279                 .mode           = 0644,
1280                 .proc_handler   = &proc_dointvec_jiffies,
1281         },
1282         {
1283                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_SYN_RECV,
1284                 .procname       = "ip_conntrack_tcp_timeout_syn_recv",
1285                 .data           = &nf_ct_tcp_timeout_syn_recv,
1286                 .maxlen         = sizeof(unsigned int),
1287                 .mode           = 0644,
1288                 .proc_handler   = &proc_dointvec_jiffies,
1289         },
1290         {
1291                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_ESTABLISHED,
1292                 .procname       = "ip_conntrack_tcp_timeout_established",
1293                 .data           = &nf_ct_tcp_timeout_established,
1294                 .maxlen         = sizeof(unsigned int),
1295                 .mode           = 0644,
1296                 .proc_handler   = &proc_dointvec_jiffies,
1297         },
1298         {
1299                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_FIN_WAIT,
1300                 .procname       = "ip_conntrack_tcp_timeout_fin_wait",
1301                 .data           = &nf_ct_tcp_timeout_fin_wait,
1302                 .maxlen         = sizeof(unsigned int),
1303                 .mode           = 0644,
1304                 .proc_handler   = &proc_dointvec_jiffies,
1305         },
1306         {
1307                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_CLOSE_WAIT,
1308                 .procname       = "ip_conntrack_tcp_timeout_close_wait",
1309                 .data           = &nf_ct_tcp_timeout_close_wait,
1310                 .maxlen         = sizeof(unsigned int),
1311                 .mode           = 0644,
1312                 .proc_handler   = &proc_dointvec_jiffies,
1313         },
1314         {
1315                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_LAST_ACK,
1316                 .procname       = "ip_conntrack_tcp_timeout_last_ack",
1317                 .data           = &nf_ct_tcp_timeout_last_ack,
1318                 .maxlen         = sizeof(unsigned int),
1319                 .mode           = 0644,
1320                 .proc_handler   = &proc_dointvec_jiffies,
1321         },
1322         {
1323                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_TIME_WAIT,
1324                 .procname       = "ip_conntrack_tcp_timeout_time_wait",
1325                 .data           = &nf_ct_tcp_timeout_time_wait,
1326                 .maxlen         = sizeof(unsigned int),
1327                 .mode           = 0644,
1328                 .proc_handler   = &proc_dointvec_jiffies,
1329         },
1330         {
1331                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_CLOSE,
1332                 .procname       = "ip_conntrack_tcp_timeout_close",
1333                 .data           = &nf_ct_tcp_timeout_close,
1334                 .maxlen         = sizeof(unsigned int),
1335                 .mode           = 0644,
1336                 .proc_handler   = &proc_dointvec_jiffies,
1337         },
1338         {
1339                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_TIMEOUT_MAX_RETRANS,
1340                 .procname       = "ip_conntrack_tcp_timeout_max_retrans",
1341                 .data           = &nf_ct_tcp_timeout_max_retrans,
1342                 .maxlen         = sizeof(unsigned int),
1343                 .mode           = 0644,
1344                 .proc_handler   = &proc_dointvec_jiffies,
1345         },
1346         {
1347                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_LOOSE,
1348                 .procname       = "ip_conntrack_tcp_loose",
1349                 .data           = &nf_ct_tcp_loose,
1350                 .maxlen         = sizeof(unsigned int),
1351                 .mode           = 0644,
1352                 .proc_handler   = &proc_dointvec,
1353         },
1354         {
1355                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_BE_LIBERAL,
1356                 .procname       = "ip_conntrack_tcp_be_liberal",
1357                 .data           = &nf_ct_tcp_be_liberal,
1358                 .maxlen         = sizeof(unsigned int),
1359                 .mode           = 0644,
1360                 .proc_handler   = &proc_dointvec,
1361         },
1362         {
1363                 .ctl_name       = NET_IPV4_NF_CONNTRACK_TCP_MAX_RETRANS,
1364                 .procname       = "ip_conntrack_tcp_max_retrans",
1365                 .data           = &nf_ct_tcp_max_retrans,
1366                 .maxlen         = sizeof(unsigned int),
1367                 .mode           = 0644,
1368                 .proc_handler   = &proc_dointvec,
1369         },
1370         {
1371                 .ctl_name       = 0
1372         }
1373 };
1374 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
1375 #endif /* CONFIG_SYSCTL */
1376
1377 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1378 {
1379         .l3proto                = PF_INET,
1380         .l4proto                = IPPROTO_TCP,
1381         .name                   = "tcp",
1382         .pkt_to_tuple           = tcp_pkt_to_tuple,
1383         .invert_tuple           = tcp_invert_tuple,
1384         .print_tuple            = tcp_print_tuple,
1385         .print_conntrack        = tcp_print_conntrack,
1386         .packet                 = tcp_packet,
1387         .new                    = tcp_new,
1388         .error                  = tcp_error,
1389 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1390         .to_nlattr              = tcp_to_nlattr,
1391         .from_nlattr            = nlattr_to_tcp,
1392         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1393         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1394         .nla_policy             = nf_ct_port_nla_policy,
1395 #endif
1396 #ifdef CONFIG_SYSCTL
1397         .ctl_table_users        = &tcp_sysctl_table_users,
1398         .ctl_table_header       = &tcp_sysctl_header,
1399         .ctl_table              = tcp_sysctl_table,
1400 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1401         .ctl_compat_table       = tcp_compat_sysctl_table,
1402 #endif
1403 #endif
1404 };
1405 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1406
1407 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1408 {
1409         .l3proto                = PF_INET6,
1410         .l4proto                = IPPROTO_TCP,
1411         .name                   = "tcp",
1412         .pkt_to_tuple           = tcp_pkt_to_tuple,
1413         .invert_tuple           = tcp_invert_tuple,
1414         .print_tuple            = tcp_print_tuple,
1415         .print_conntrack        = tcp_print_conntrack,
1416         .packet                 = tcp_packet,
1417         .new                    = tcp_new,
1418         .error                  = tcp_error,
1419 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
1420         .to_nlattr              = tcp_to_nlattr,
1421         .from_nlattr            = nlattr_to_tcp,
1422         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1423         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1424         .nla_policy             = nf_ct_port_nla_policy,
1425 #endif
1426 #ifdef CONFIG_SYSCTL
1427         .ctl_table_users        = &tcp_sysctl_table_users,
1428         .ctl_table_header       = &tcp_sysctl_header,
1429         .ctl_table              = tcp_sysctl_table,
1430 #endif
1431 };
1432 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);