1 // SPDX-License-Identifier: GPL-2.0-only
2 /* (C) 1999-2001 Paul `Rusty' Russell
3 * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
4 * (C) 2002-2013 Jozsef Kadlecsik <kadlec@netfilter.org>
5 * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
6 */
7
8 #include <linux/types.h>
9 #include <linux/timer.h>
10 #include <linux/module.h>
11 #include <linux/in.h>
12 #include <linux/tcp.h>
13 #include <linux/spinlock.h>
14 #include <linux/skbuff.h>
15 #include <linux/ipv6.h>
16 #include <net/ip6_checksum.h>
17 #include <asm/unaligned.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 #include <net/netfilter/nf_conntrack_seqadj.h>
28 #include <net/netfilter/nf_conntrack_synproxy.h>
29 #include <net/netfilter/nf_conntrack_timeout.h>
30 #include <net/netfilter/nf_log.h>
31 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
32 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
33
34 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
35 closely. They're more complex. --RR */
36
37 static const char *const tcp_conntrack_names[] = {
38 "NONE",
39 "SYN_SENT",
40 "SYN_RECV",
41 "ESTABLISHED",
42 "FIN_WAIT",
43 "CLOSE_WAIT",
44 "LAST_ACK",
45 "TIME_WAIT",
46 "CLOSE",
47 "SYN_SENT2",
48 };
49
50 #define SECS * HZ
51 #define MINS * 60 SECS
52 #define HOURS * 60 MINS
53 #define DAYS * 24 HOURS
54
55 static const unsigned int tcp_timeouts[TCP_CONNTRACK_TIMEOUT_MAX] = {
56 [TCP_CONNTRACK_SYN_SENT] = 2 MINS,
57 [TCP_CONNTRACK_SYN_RECV] = 60 SECS,
58 [TCP_CONNTRACK_ESTABLISHED] = 5 DAYS,
59 [TCP_CONNTRACK_FIN_WAIT] = 2 MINS,
60 [TCP_CONNTRACK_CLOSE_WAIT] = 60 SECS,
61 [TCP_CONNTRACK_LAST_ACK] = 30 SECS,
62 [TCP_CONNTRACK_TIME_WAIT] = 2 MINS,
63 [TCP_CONNTRACK_CLOSE] = 10 SECS,
64 [TCP_CONNTRACK_SYN_SENT2] = 2 MINS,
65 /* RFC1122 says the R2 limit should be at least 100 seconds.
66 Linux uses 15 packets as limit, which corresponds
67 to ~13-30min depending on RTO. */
68 [TCP_CONNTRACK_RETRANS] = 5 MINS,
69 [TCP_CONNTRACK_UNACK] = 5 MINS,
70 };
71
72 #define sNO TCP_CONNTRACK_NONE
73 #define sSS TCP_CONNTRACK_SYN_SENT
74 #define sSR TCP_CONNTRACK_SYN_RECV
75 #define sES TCP_CONNTRACK_ESTABLISHED
76 #define sFW TCP_CONNTRACK_FIN_WAIT
77 #define sCW TCP_CONNTRACK_CLOSE_WAIT
78 #define sLA TCP_CONNTRACK_LAST_ACK
79 #define sTW TCP_CONNTRACK_TIME_WAIT
80 #define sCL TCP_CONNTRACK_CLOSE
81 #define sS2 TCP_CONNTRACK_SYN_SENT2
82 #define sIV TCP_CONNTRACK_MAX
83 #define sIG TCP_CONNTRACK_IGNORE
84
85 /* What TCP flags are set from RST/SYN/FIN/ACK. */
86 enum tcp_bit_set {
87 TCP_SYN_SET,
88 TCP_SYNACK_SET,
89 TCP_FIN_SET,
90 TCP_ACK_SET,
91 TCP_RST_SET,
92 TCP_NONE_SET,
93 };
94
95 /*
96 * The TCP state transition table needs a few words...
97 *
98 * We are the man in the middle. All the packets go through us
99 * but might get lost in transit to the destination.
100 * It is assumed that the destinations can't receive segments
101 * we haven't seen.
102 *
103 * The checked segment is in window, but our windows are *not*
104 * equivalent with the ones of the sender/receiver. We always
105 * try to guess the state of the current sender.
106 *
107 * The meaning of the states are:
108 *
109 * NONE: initial state
110 * SYN_SENT: SYN-only packet seen
111 * SYN_SENT2: SYN-only packet seen from reply dir, simultaneous open
112 * SYN_RECV: SYN-ACK packet seen
113 * ESTABLISHED: ACK packet seen
114 * FIN_WAIT: FIN packet seen
115 * CLOSE_WAIT: ACK seen (after FIN)
116 * LAST_ACK: FIN seen (after FIN)
117 * TIME_WAIT: last ACK seen
118 * CLOSE: closed connection (RST)
119 *
120 * Packets marked as IGNORED (sIG):
121 * if they may be either invalid or valid
122 * and the receiver may send back a connection
123 * closing RST or a SYN/ACK.
124 *
125 * Packets marked as INVALID (sIV):
126 * if we regard them as truly invalid packets
127 */
128 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
129 {
130 /* ORIGINAL */
131 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
132 /*syn*/ { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sS2 },
133 /*
134 * sNO -> sSS Initialize a new connection
135 * sSS -> sSS Retransmitted SYN
136 * sS2 -> sS2 Late retransmitted SYN
137 * sSR -> sIG
138 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
139 * are errors. Receiver will reply with RST
140 * and close the connection.
141 * Or we are not in sync and hold a dead connection.
142 * sFW -> sIG
143 * sCW -> sIG
144 * sLA -> sIG
145 * sTW -> sSS Reopened connection (RFC 1122).
146 * sCL -> sSS
147 */
148 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
149 /*synack*/ { sIV, sIV, sSR, sIV, sIV, sIV, sIV, sIV, sIV, sSR },
150 /*
151 * sNO -> sIV Too late and no reason to do anything
152 * sSS -> sIV Client can't send SYN and then SYN/ACK
153 * sS2 -> sSR SYN/ACK sent to SYN2 in simultaneous open
154 * sSR -> sSR Late retransmitted SYN/ACK in simultaneous open
155 * sES -> sIV Invalid SYN/ACK packets sent by the client
156 * sFW -> sIV
157 * sCW -> sIV
158 * sLA -> sIV
159 * sTW -> sIV
160 * sCL -> sIV
161 */
162 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
163 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
164 /*
165 * sNO -> sIV Too late and no reason to do anything...
166 * sSS -> sIV Client migth not send FIN in this state:
167 * we enforce waiting for a SYN/ACK reply first.
168 * sS2 -> sIV
169 * sSR -> sFW Close started.
170 * sES -> sFW
171 * sFW -> sLA FIN seen in both directions, waiting for
172 * the last ACK.
173 * Migth be a retransmitted FIN as well...
174 * sCW -> sLA
175 * sLA -> sLA Retransmitted FIN. Remain in the same state.
176 * sTW -> sTW
177 * sCL -> sCL
178 */
179 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
180 /*ack*/ { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
181 /*
182 * sNO -> sES Assumed.
183 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
184 * sS2 -> sIV
185 * sSR -> sES Established state is reached.
186 * sES -> sES :-)
187 * sFW -> sCW Normal close request answered by ACK.
188 * sCW -> sCW
189 * sLA -> sTW Last ACK detected (RFC5961 challenged)
190 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
191 * sCL -> sCL
192 */
193 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
194 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
195 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
196 },
197 {
198 /* REPLY */
199 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
200 /*syn*/ { sIV, sS2, sIV, sIV, sIV, sIV, sIV, sSS, sIV, sS2 },
201 /*
202 * sNO -> sIV Never reached.
203 * sSS -> sS2 Simultaneous open
204 * sS2 -> sS2 Retransmitted simultaneous SYN
205 * sSR -> sIV Invalid SYN packets sent by the server
206 * sES -> sIV
207 * sFW -> sIV
208 * sCW -> sIV
209 * sLA -> sIV
210 * sTW -> sSS Reopened connection, but server may have switched role
211 * sCL -> sIV
212 */
213 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
214 /*synack*/ { sIV, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
215 /*
216 * sSS -> sSR Standard open.
217 * sS2 -> sSR Simultaneous open
218 * sSR -> sIG Retransmitted SYN/ACK, ignore it.
219 * sES -> sIG Late retransmitted SYN/ACK?
220 * sFW -> sIG Might be SYN/ACK answering ignored SYN
221 * sCW -> sIG
222 * sLA -> sIG
223 * sTW -> sIG
224 * sCL -> sIG
225 */
226 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
227 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
228 /*
229 * sSS -> sIV Server might not send FIN in this state.
230 * sS2 -> sIV
231 * sSR -> sFW Close started.
232 * sES -> sFW
233 * sFW -> sLA FIN seen in both directions.
234 * sCW -> sLA
235 * sLA -> sLA Retransmitted FIN.
236 * sTW -> sTW
237 * sCL -> sCL
238 */
239 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
240 /*ack*/ { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIG },
241 /*
242 * sSS -> sIG Might be a half-open connection.
243 * sS2 -> sIG
244 * sSR -> sSR Might answer late resent SYN.
245 * sES -> sES :-)
246 * sFW -> sCW Normal close request answered by ACK.
247 * sCW -> sCW
248 * sLA -> sTW Last ACK detected (RFC5961 challenged)
249 * sTW -> sTW Retransmitted last ACK.
250 * sCL -> sCL
251 */
252 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
253 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
254 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
255 }
256 };
257
258 #ifdef CONFIG_NF_CONNTRACK_PROCFS
259 /* Print out the private part of the conntrack. */
tcp_print_conntrack(struct seq_file * s,struct nf_conn * ct)260 static void tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
261 {
262 if (test_bit(IPS_OFFLOAD_BIT, &ct->status))
263 return;
264
265 seq_printf(s, "%s ", tcp_conntrack_names[ct->proto.tcp.state]);
266 }
267 #endif
268
get_conntrack_index(const struct tcphdr * tcph)269 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
270 {
271 if (tcph->rst) return TCP_RST_SET;
272 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
273 else if (tcph->fin) return TCP_FIN_SET;
274 else if (tcph->ack) return TCP_ACK_SET;
275 else return TCP_NONE_SET;
276 }
277
278 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
279 in IP Filter' by Guido van Rooij.
280
281 http://www.sane.nl/events/sane2000/papers.html
282 http://www.darkart.com/mirrors/www.obfuscation.org/ipf/
283
284 The boundaries and the conditions are changed according to RFC793:
285 the packet must intersect the window (i.e. segments may be
286 after the right or before the left edge) and thus receivers may ACK
287 segments after the right edge of the window.
288
289 td_maxend = max(sack + max(win,1)) seen in reply packets
290 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
291 td_maxwin += seq + len - sender.td_maxend
292 if seq + len > sender.td_maxend
293 td_end = max(seq + len) seen in sent packets
294
295 I. Upper bound for valid data: seq <= sender.td_maxend
296 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
297 III. Upper bound for valid (s)ack: sack <= receiver.td_end
298 IV. Lower bound for valid (s)ack: sack >= receiver.td_end - MAXACKWINDOW
299
300 where sack is the highest right edge of sack block found in the packet
301 or ack in the case of packet without SACK option.
302
303 The upper bound limit for a valid (s)ack is not ignored -
304 we doesn't have to deal with fragments.
305 */
306
segment_seq_plus_len(__u32 seq,size_t len,unsigned int dataoff,const struct tcphdr * tcph)307 static inline __u32 segment_seq_plus_len(__u32 seq,
308 size_t len,
309 unsigned int dataoff,
310 const struct tcphdr *tcph)
311 {
312 /* XXX Should I use payload length field in IP/IPv6 header ?
313 * - YK */
314 return (seq + len - dataoff - tcph->doff*4
315 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
316 }
317
318 /* Fixme: what about big packets? */
319 #define MAXACKWINCONST 66000
320 #define MAXACKWINDOW(sender) \
321 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
322 : MAXACKWINCONST)
323
324 /*
325 * Simplified tcp_parse_options routine from tcp_input.c
326 */
tcp_options(const struct sk_buff * skb,unsigned int dataoff,const struct tcphdr * tcph,struct ip_ct_tcp_state * state)327 static void tcp_options(const struct sk_buff *skb,
328 unsigned int dataoff,
329 const struct tcphdr *tcph,
330 struct ip_ct_tcp_state *state)
331 {
332 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
333 const unsigned char *ptr;
334 int length = (tcph->doff*4) - sizeof(struct tcphdr);
335
336 if (!length)
337 return;
338
339 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
340 length, buff);
341 if (!ptr)
342 return;
343
344 state->td_scale = 0;
345 state->flags &= IP_CT_TCP_FLAG_BE_LIBERAL;
346
347 while (length > 0) {
348 int opcode=*ptr++;
349 int opsize;
350
351 switch (opcode) {
352 case TCPOPT_EOL:
353 return;
354 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
355 length--;
356 continue;
357 default:
358 if (length < 2)
359 return;
360 opsize=*ptr++;
361 if (opsize < 2) /* "silly options" */
362 return;
363 if (opsize > length)
364 return; /* don't parse partial options */
365
366 if (opcode == TCPOPT_SACK_PERM
367 && opsize == TCPOLEN_SACK_PERM)
368 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
369 else if (opcode == TCPOPT_WINDOW
370 && opsize == TCPOLEN_WINDOW) {
371 state->td_scale = *(u_int8_t *)ptr;
372
373 if (state->td_scale > TCP_MAX_WSCALE)
374 state->td_scale = TCP_MAX_WSCALE;
375
376 state->flags |=
377 IP_CT_TCP_FLAG_WINDOW_SCALE;
378 }
379 ptr += opsize - 2;
380 length -= opsize;
381 }
382 }
383 }
384
tcp_sack(const struct sk_buff * skb,unsigned int dataoff,const struct tcphdr * tcph,__u32 * sack)385 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
386 const struct tcphdr *tcph, __u32 *sack)
387 {
388 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
389 const unsigned char *ptr;
390 int length = (tcph->doff*4) - sizeof(struct tcphdr);
391 __u32 tmp;
392
393 if (!length)
394 return;
395
396 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
397 length, buff);
398 if (!ptr)
399 return;
400
401 /* Fast path for timestamp-only option */
402 if (length == TCPOLEN_TSTAMP_ALIGNED
403 && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
404 | (TCPOPT_NOP << 16)
405 | (TCPOPT_TIMESTAMP << 8)
406 | TCPOLEN_TIMESTAMP))
407 return;
408
409 while (length > 0) {
410 int opcode = *ptr++;
411 int opsize, i;
412
413 switch (opcode) {
414 case TCPOPT_EOL:
415 return;
416 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
417 length--;
418 continue;
419 default:
420 if (length < 2)
421 return;
422 opsize = *ptr++;
423 if (opsize < 2) /* "silly options" */
424 return;
425 if (opsize > length)
426 return; /* don't parse partial options */
427
428 if (opcode == TCPOPT_SACK
429 && opsize >= (TCPOLEN_SACK_BASE
430 + TCPOLEN_SACK_PERBLOCK)
431 && !((opsize - TCPOLEN_SACK_BASE)
432 % TCPOLEN_SACK_PERBLOCK)) {
433 for (i = 0;
434 i < (opsize - TCPOLEN_SACK_BASE);
435 i += TCPOLEN_SACK_PERBLOCK) {
436 tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
437
438 if (after(tmp, *sack))
439 *sack = tmp;
440 }
441 return;
442 }
443 ptr += opsize - 2;
444 length -= opsize;
445 }
446 }
447 }
448
tcp_in_window(struct nf_conn * ct,enum ip_conntrack_dir dir,unsigned int index,const struct sk_buff * skb,unsigned int dataoff,const struct tcphdr * tcph,const struct nf_hook_state * hook_state)449 static bool tcp_in_window(struct nf_conn *ct,
450 enum ip_conntrack_dir dir,
451 unsigned int index,
452 const struct sk_buff *skb,
453 unsigned int dataoff,
454 const struct tcphdr *tcph,
455 const struct nf_hook_state *hook_state)
456 {
457 struct ip_ct_tcp *state = &ct->proto.tcp;
458 struct net *net = nf_ct_net(ct);
459 struct nf_tcp_net *tn = nf_tcp_pernet(net);
460 struct ip_ct_tcp_state *sender = &state->seen[dir];
461 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
462 const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
463 __u32 seq, ack, sack, end, win, swin;
464 u16 win_raw;
465 s32 receiver_offset;
466 bool res, in_recv_win;
467
468 /*
469 * Get the required data from the packet.
470 */
471 seq = ntohl(tcph->seq);
472 ack = sack = ntohl(tcph->ack_seq);
473 win_raw = ntohs(tcph->window);
474 win = win_raw;
475 end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
476
477 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
478 tcp_sack(skb, dataoff, tcph, &sack);
479
480 /* Take into account NAT sequence number mangling */
481 receiver_offset = nf_ct_seq_offset(ct, !dir, ack - 1);
482 ack -= receiver_offset;
483 sack -= receiver_offset;
484
485 pr_debug("tcp_in_window: START\n");
486 pr_debug("tcp_in_window: ");
487 nf_ct_dump_tuple(tuple);
488 pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
489 seq, ack, receiver_offset, sack, receiver_offset, win, end);
490 pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
491 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
492 sender->td_end, sender->td_maxend, sender->td_maxwin,
493 sender->td_scale,
494 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
495 receiver->td_scale);
496
497 if (sender->td_maxwin == 0) {
498 /*
499 * Initialize sender data.
500 */
501 if (tcph->syn) {
502 /*
503 * SYN-ACK in reply to a SYN
504 * or SYN from reply direction in simultaneous open.
505 */
506 sender->td_end =
507 sender->td_maxend = end;
508 sender->td_maxwin = (win == 0 ? 1 : win);
509
510 tcp_options(skb, dataoff, tcph, sender);
511 /*
512 * RFC 1323:
513 * Both sides must send the Window Scale option
514 * to enable window scaling in either direction.
515 */
516 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
517 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
518 sender->td_scale =
519 receiver->td_scale = 0;
520 if (!tcph->ack)
521 /* Simultaneous open */
522 return true;
523 } else {
524 /*
525 * We are in the middle of a connection,
526 * its history is lost for us.
527 * Let's try to use the data from the packet.
528 */
529 sender->td_end = end;
530 swin = win << sender->td_scale;
531 sender->td_maxwin = (swin == 0 ? 1 : swin);
532 sender->td_maxend = end + sender->td_maxwin;
533 if (receiver->td_maxwin == 0) {
534 /* We haven't seen traffic in the other
535 * direction yet but we have to tweak window
536 * tracking to pass III and IV until that
537 * happens.
538 */
539 receiver->td_end = receiver->td_maxend = sack;
540 } else if (sack == receiver->td_end + 1) {
541 /* Likely a reply to a keepalive.
542 * Needed for III.
543 */
544 receiver->td_end++;
545 }
546
547 }
548 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
549 && dir == IP_CT_DIR_ORIGINAL)
550 || (state->state == TCP_CONNTRACK_SYN_RECV
551 && dir == IP_CT_DIR_REPLY))
552 && after(end, sender->td_end)) {
553 /*
554 * RFC 793: "if a TCP is reinitialized ... then it need
555 * not wait at all; it must only be sure to use sequence
556 * numbers larger than those recently used."
557 */
558 sender->td_end =
559 sender->td_maxend = end;
560 sender->td_maxwin = (win == 0 ? 1 : win);
561
562 tcp_options(skb, dataoff, tcph, sender);
563 }
564
565 if (!(tcph->ack)) {
566 /*
567 * If there is no ACK, just pretend it was set and OK.
568 */
569 ack = sack = receiver->td_end;
570 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
571 (TCP_FLAG_ACK|TCP_FLAG_RST))
572 && (ack == 0)) {
573 /*
574 * Broken TCP stacks, that set ACK in RST packets as well
575 * with zero ack value.
576 */
577 ack = sack = receiver->td_end;
578 }
579
580 if (tcph->rst && seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)
581 /*
582 * RST sent answering SYN.
583 */
584 seq = end = sender->td_end;
585
586 pr_debug("tcp_in_window: ");
587 nf_ct_dump_tuple(tuple);
588 pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
589 seq, ack, receiver_offset, sack, receiver_offset, win, end);
590 pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
591 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
592 sender->td_end, sender->td_maxend, sender->td_maxwin,
593 sender->td_scale,
594 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
595 receiver->td_scale);
596
597 /* Is the ending sequence in the receive window (if available)? */
598 in_recv_win = !receiver->td_maxwin ||
599 after(end, sender->td_end - receiver->td_maxwin - 1);
600
601 pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
602 before(seq, sender->td_maxend + 1),
603 (in_recv_win ? 1 : 0),
604 before(sack, receiver->td_end + 1),
605 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
606
607 if (before(seq, sender->td_maxend + 1) &&
608 in_recv_win &&
609 before(sack, receiver->td_end + 1) &&
610 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
611 /*
612 * Take into account window scaling (RFC 1323).
613 */
614 if (!tcph->syn)
615 win <<= sender->td_scale;
616
617 /*
618 * Update sender data.
619 */
620 swin = win + (sack - ack);
621 if (sender->td_maxwin < swin)
622 sender->td_maxwin = swin;
623 if (after(end, sender->td_end)) {
624 sender->td_end = end;
625 sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
626 }
627 if (tcph->ack) {
628 if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
629 sender->td_maxack = ack;
630 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
631 } else if (after(ack, sender->td_maxack))
632 sender->td_maxack = ack;
633 }
634
635 /*
636 * Update receiver data.
637 */
638 if (receiver->td_maxwin != 0 && after(end, sender->td_maxend))
639 receiver->td_maxwin += end - sender->td_maxend;
640 if (after(sack + win, receiver->td_maxend - 1)) {
641 receiver->td_maxend = sack + win;
642 if (win == 0)
643 receiver->td_maxend++;
644 }
645 if (ack == receiver->td_end)
646 receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
647
648 /*
649 * Check retransmissions.
650 */
651 if (index == TCP_ACK_SET) {
652 if (state->last_dir == dir
653 && state->last_seq == seq
654 && state->last_ack == ack
655 && state->last_end == end
656 && state->last_win == win_raw)
657 state->retrans++;
658 else {
659 state->last_dir = dir;
660 state->last_seq = seq;
661 state->last_ack = ack;
662 state->last_end = end;
663 state->last_win = win_raw;
664 state->retrans = 0;
665 }
666 }
667 res = true;
668 } else {
669 res = false;
670 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
671 tn->tcp_be_liberal)
672 res = true;
673 if (!res) {
674 bool seq_ok = before(seq, sender->td_maxend + 1);
675
676 if (!seq_ok) {
677 u32 overshot = end - sender->td_maxend + 1;
678 bool ack_ok;
679
680 ack_ok = after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1);
681
682 if (in_recv_win &&
683 ack_ok &&
684 overshot <= receiver->td_maxwin &&
685 before(sack, receiver->td_end + 1)) {
686 /* Work around TCPs that send more bytes than allowed by
687 * the receive window.
688 *
689 * If the (marked as invalid) packet is allowed to pass by
690 * the ruleset and the peer acks this data, then its possible
691 * all future packets will trigger 'ACK is over upper bound' check.
692 *
693 * Thus if only the sequence check fails then do update td_end so
694 * possible ACK for this data can update internal state.
695 */
696 sender->td_end = end;
697 sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
698
699 nf_ct_l4proto_log_invalid(skb, ct, hook_state,
700 "%u bytes more than expected", overshot);
701 return res;
702 }
703 }
704
705 nf_ct_l4proto_log_invalid(skb, ct, hook_state,
706 "%s",
707 before(seq, sender->td_maxend + 1) ?
708 in_recv_win ?
709 before(sack, receiver->td_end + 1) ?
710 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
711 : "ACK is under the lower bound (possible overly delayed ACK)"
712 : "ACK is over the upper bound (ACKed data not seen yet)"
713 : "SEQ is under the lower bound (already ACKed data retransmitted)"
714 : "SEQ is over the upper bound (over the window of the receiver)");
715 }
716 }
717
718 pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
719 "receiver end=%u maxend=%u maxwin=%u\n",
720 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
721 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
722
723 return res;
724 }
725
726 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
727 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
728 TCPHDR_URG) + 1] =
729 {
730 [TCPHDR_SYN] = 1,
731 [TCPHDR_SYN|TCPHDR_URG] = 1,
732 [TCPHDR_SYN|TCPHDR_ACK] = 1,
733 [TCPHDR_RST] = 1,
734 [TCPHDR_RST|TCPHDR_ACK] = 1,
735 [TCPHDR_FIN|TCPHDR_ACK] = 1,
736 [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG] = 1,
737 [TCPHDR_ACK] = 1,
738 [TCPHDR_ACK|TCPHDR_URG] = 1,
739 };
740
tcp_error_log(const struct sk_buff * skb,const struct nf_hook_state * state,const char * msg)741 static void tcp_error_log(const struct sk_buff *skb,
742 const struct nf_hook_state *state,
743 const char *msg)
744 {
745 nf_l4proto_log_invalid(skb, state, IPPROTO_TCP, "%s", msg);
746 }
747
748 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
tcp_error(const struct tcphdr * th,struct sk_buff * skb,unsigned int dataoff,const struct nf_hook_state * state)749 static bool tcp_error(const struct tcphdr *th,
750 struct sk_buff *skb,
751 unsigned int dataoff,
752 const struct nf_hook_state *state)
753 {
754 unsigned int tcplen = skb->len - dataoff;
755 u8 tcpflags;
756
757 /* Not whole TCP header or malformed packet */
758 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
759 tcp_error_log(skb, state, "truncated packet");
760 return true;
761 }
762
763 /* Checksum invalid? Ignore.
764 * We skip checking packets on the outgoing path
765 * because the checksum is assumed to be correct.
766 */
767 /* FIXME: Source route IP option packets --RR */
768 if (state->net->ct.sysctl_checksum &&
769 state->hook == NF_INET_PRE_ROUTING &&
770 nf_checksum(skb, state->hook, dataoff, IPPROTO_TCP, state->pf)) {
771 tcp_error_log(skb, state, "bad checksum");
772 return true;
773 }
774
775 /* Check TCP flags. */
776 tcpflags = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
777 if (!tcp_valid_flags[tcpflags]) {
778 tcp_error_log(skb, state, "invalid tcp flag combination");
779 return true;
780 }
781
782 return false;
783 }
784
tcp_new(struct nf_conn * ct,const struct sk_buff * skb,unsigned int dataoff,const struct tcphdr * th)785 static noinline bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
786 unsigned int dataoff,
787 const struct tcphdr *th)
788 {
789 enum tcp_conntrack new_state;
790 struct net *net = nf_ct_net(ct);
791 const struct nf_tcp_net *tn = nf_tcp_pernet(net);
792 const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
793 const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
794
795 /* Don't need lock here: this conntrack not in circulation yet */
796 new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
797
798 /* Invalid: delete conntrack */
799 if (new_state >= TCP_CONNTRACK_MAX) {
800 pr_debug("nf_ct_tcp: invalid new deleting.\n");
801 return false;
802 }
803
804 if (new_state == TCP_CONNTRACK_SYN_SENT) {
805 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
806 /* SYN packet */
807 ct->proto.tcp.seen[0].td_end =
808 segment_seq_plus_len(ntohl(th->seq), skb->len,
809 dataoff, th);
810 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
811 if (ct->proto.tcp.seen[0].td_maxwin == 0)
812 ct->proto.tcp.seen[0].td_maxwin = 1;
813 ct->proto.tcp.seen[0].td_maxend =
814 ct->proto.tcp.seen[0].td_end;
815
816 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
817 } else if (tn->tcp_loose == 0) {
818 /* Don't try to pick up connections. */
819 return false;
820 } else {
821 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
822 /*
823 * We are in the middle of a connection,
824 * its history is lost for us.
825 * Let's try to use the data from the packet.
826 */
827 ct->proto.tcp.seen[0].td_end =
828 segment_seq_plus_len(ntohl(th->seq), skb->len,
829 dataoff, th);
830 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
831 if (ct->proto.tcp.seen[0].td_maxwin == 0)
832 ct->proto.tcp.seen[0].td_maxwin = 1;
833 ct->proto.tcp.seen[0].td_maxend =
834 ct->proto.tcp.seen[0].td_end +
835 ct->proto.tcp.seen[0].td_maxwin;
836
837 /* We assume SACK and liberal window checking to handle
838 * window scaling */
839 ct->proto.tcp.seen[0].flags =
840 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
841 IP_CT_TCP_FLAG_BE_LIBERAL;
842 }
843
844 /* tcp_packet will set them */
845 ct->proto.tcp.last_index = TCP_NONE_SET;
846
847 pr_debug("%s: sender end=%u maxend=%u maxwin=%u scale=%i "
848 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
849 __func__,
850 sender->td_end, sender->td_maxend, sender->td_maxwin,
851 sender->td_scale,
852 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
853 receiver->td_scale);
854 return true;
855 }
856
tcp_can_early_drop(const struct nf_conn * ct)857 static bool tcp_can_early_drop(const struct nf_conn *ct)
858 {
859 switch (ct->proto.tcp.state) {
860 case TCP_CONNTRACK_FIN_WAIT:
861 case TCP_CONNTRACK_LAST_ACK:
862 case TCP_CONNTRACK_TIME_WAIT:
863 case TCP_CONNTRACK_CLOSE:
864 case TCP_CONNTRACK_CLOSE_WAIT:
865 return true;
866 default:
867 break;
868 }
869
870 return false;
871 }
872
nf_ct_tcp_state_reset(struct ip_ct_tcp_state * state)873 static void nf_ct_tcp_state_reset(struct ip_ct_tcp_state *state)
874 {
875 state->td_end = 0;
876 state->td_maxend = 0;
877 state->td_maxwin = 0;
878 state->td_maxack = 0;
879 state->td_scale = 0;
880 state->flags &= IP_CT_TCP_FLAG_BE_LIBERAL;
881 }
882
883 /* Returns verdict for packet, or -1 for invalid. */
nf_conntrack_tcp_packet(struct nf_conn * ct,struct sk_buff * skb,unsigned int dataoff,enum ip_conntrack_info ctinfo,const struct nf_hook_state * state)884 int nf_conntrack_tcp_packet(struct nf_conn *ct,
885 struct sk_buff *skb,
886 unsigned int dataoff,
887 enum ip_conntrack_info ctinfo,
888 const struct nf_hook_state *state)
889 {
890 struct net *net = nf_ct_net(ct);
891 struct nf_tcp_net *tn = nf_tcp_pernet(net);
892 struct nf_conntrack_tuple *tuple;
893 enum tcp_conntrack new_state, old_state;
894 unsigned int index, *timeouts;
895 enum ip_conntrack_dir dir;
896 const struct tcphdr *th;
897 struct tcphdr _tcph;
898 unsigned long timeout;
899
900 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
901 if (th == NULL)
902 return -NF_ACCEPT;
903
904 if (tcp_error(th, skb, dataoff, state))
905 return -NF_ACCEPT;
906
907 if (!nf_ct_is_confirmed(ct) && !tcp_new(ct, skb, dataoff, th))
908 return -NF_ACCEPT;
909
910 spin_lock_bh(&ct->lock);
911 old_state = ct->proto.tcp.state;
912 dir = CTINFO2DIR(ctinfo);
913 index = get_conntrack_index(th);
914 new_state = tcp_conntracks[dir][index][old_state];
915 tuple = &ct->tuplehash[dir].tuple;
916
917 switch (new_state) {
918 case TCP_CONNTRACK_SYN_SENT:
919 if (old_state < TCP_CONNTRACK_TIME_WAIT)
920 break;
921 /* RFC 1122: "When a connection is closed actively,
922 * it MUST linger in TIME-WAIT state for a time 2xMSL
923 * (Maximum Segment Lifetime). However, it MAY accept
924 * a new SYN from the remote TCP to reopen the connection
925 * directly from TIME-WAIT state, if..."
926 * We ignore the conditions because we are in the
927 * TIME-WAIT state anyway.
928 *
929 * Handle aborted connections: we and the server
930 * think there is an existing connection but the client
931 * aborts it and starts a new one.
932 */
933 if (((ct->proto.tcp.seen[dir].flags
934 | ct->proto.tcp.seen[!dir].flags)
935 & IP_CT_TCP_FLAG_CLOSE_INIT)
936 || (ct->proto.tcp.last_dir == dir
937 && ct->proto.tcp.last_index == TCP_RST_SET)) {
938 /* Attempt to reopen a closed/aborted connection.
939 * Delete this connection and look up again. */
940 spin_unlock_bh(&ct->lock);
941
942 /* Only repeat if we can actually remove the timer.
943 * Destruction may already be in progress in process
944 * context and we must give it a chance to terminate.
945 */
946 if (nf_ct_kill(ct))
947 return -NF_REPEAT;
948 return NF_DROP;
949 }
950 fallthrough;
951 case TCP_CONNTRACK_IGNORE:
952 /* Ignored packets:
953 *
954 * Our connection entry may be out of sync, so ignore
955 * packets which may signal the real connection between
956 * the client and the server.
957 *
958 * a) SYN in ORIGINAL
959 * b) SYN/ACK in REPLY
960 * c) ACK in reply direction after initial SYN in original.
961 *
962 * If the ignored packet is invalid, the receiver will send
963 * a RST we'll catch below.
964 */
965 if (index == TCP_SYNACK_SET
966 && ct->proto.tcp.last_index == TCP_SYN_SET
967 && ct->proto.tcp.last_dir != dir
968 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
969 /* b) This SYN/ACK acknowledges a SYN that we earlier
970 * ignored as invalid. This means that the client and
971 * the server are both in sync, while the firewall is
972 * not. We get in sync from the previously annotated
973 * values.
974 */
975 old_state = TCP_CONNTRACK_SYN_SENT;
976 new_state = TCP_CONNTRACK_SYN_RECV;
977 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
978 ct->proto.tcp.last_end;
979 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
980 ct->proto.tcp.last_end;
981 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
982 ct->proto.tcp.last_win == 0 ?
983 1 : ct->proto.tcp.last_win;
984 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
985 ct->proto.tcp.last_wscale;
986 ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
987 ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
988 ct->proto.tcp.last_flags;
989 nf_ct_tcp_state_reset(&ct->proto.tcp.seen[dir]);
990 break;
991 }
992 ct->proto.tcp.last_index = index;
993 ct->proto.tcp.last_dir = dir;
994 ct->proto.tcp.last_seq = ntohl(th->seq);
995 ct->proto.tcp.last_end =
996 segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
997 ct->proto.tcp.last_win = ntohs(th->window);
998
999 /* a) This is a SYN in ORIGINAL. The client and the server
1000 * may be in sync but we are not. In that case, we annotate
1001 * the TCP options and let the packet go through. If it is a
1002 * valid SYN packet, the server will reply with a SYN/ACK, and
1003 * then we'll get in sync. Otherwise, the server potentially
1004 * responds with a challenge ACK if implementing RFC5961.
1005 */
1006 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
1007 struct ip_ct_tcp_state seen = {};
1008
1009 ct->proto.tcp.last_flags =
1010 ct->proto.tcp.last_wscale = 0;
1011 tcp_options(skb, dataoff, th, &seen);
1012 if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1013 ct->proto.tcp.last_flags |=
1014 IP_CT_TCP_FLAG_WINDOW_SCALE;
1015 ct->proto.tcp.last_wscale = seen.td_scale;
1016 }
1017 if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
1018 ct->proto.tcp.last_flags |=
1019 IP_CT_TCP_FLAG_SACK_PERM;
1020 }
1021 /* Mark the potential for RFC5961 challenge ACK,
1022 * this pose a special problem for LAST_ACK state
1023 * as ACK is intrepretated as ACKing last FIN.
1024 */
1025 if (old_state == TCP_CONNTRACK_LAST_ACK)
1026 ct->proto.tcp.last_flags |=
1027 IP_CT_EXP_CHALLENGE_ACK;
1028 }
1029 spin_unlock_bh(&ct->lock);
1030 nf_ct_l4proto_log_invalid(skb, ct, state,
1031 "packet (index %d) in dir %d ignored, state %s",
1032 index, dir,
1033 tcp_conntrack_names[old_state]);
1034 return NF_ACCEPT;
1035 case TCP_CONNTRACK_MAX:
1036 /* Special case for SYN proxy: when the SYN to the server or
1037 * the SYN/ACK from the server is lost, the client may transmit
1038 * a keep-alive packet while in SYN_SENT state. This needs to
1039 * be associated with the original conntrack entry in order to
1040 * generate a new SYN with the correct sequence number.
1041 */
1042 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
1043 index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
1044 ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
1045 ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
1046 pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
1047 spin_unlock_bh(&ct->lock);
1048 return NF_ACCEPT;
1049 }
1050
1051 /* Invalid packet */
1052 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
1053 dir, get_conntrack_index(th), old_state);
1054 spin_unlock_bh(&ct->lock);
1055 nf_ct_l4proto_log_invalid(skb, ct, state, "invalid state");
1056 return -NF_ACCEPT;
1057 case TCP_CONNTRACK_TIME_WAIT:
1058 /* RFC5961 compliance cause stack to send "challenge-ACK"
1059 * e.g. in response to spurious SYNs. Conntrack MUST
1060 * not believe this ACK is acking last FIN.
1061 */
1062 if (old_state == TCP_CONNTRACK_LAST_ACK &&
1063 index == TCP_ACK_SET &&
1064 ct->proto.tcp.last_dir != dir &&
1065 ct->proto.tcp.last_index == TCP_SYN_SET &&
1066 (ct->proto.tcp.last_flags & IP_CT_EXP_CHALLENGE_ACK)) {
1067 /* Detected RFC5961 challenge ACK */
1068 ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
1069 spin_unlock_bh(&ct->lock);
1070 nf_ct_l4proto_log_invalid(skb, ct, state, "challenge-ack ignored");
1071 return NF_ACCEPT; /* Don't change state */
1072 }
1073 break;
1074 case TCP_CONNTRACK_SYN_SENT2:
1075 /* tcp_conntracks table is not smart enough to handle
1076 * simultaneous open.
1077 */
1078 ct->proto.tcp.last_flags |= IP_CT_TCP_SIMULTANEOUS_OPEN;
1079 break;
1080 case TCP_CONNTRACK_SYN_RECV:
1081 if (dir == IP_CT_DIR_REPLY && index == TCP_ACK_SET &&
1082 ct->proto.tcp.last_flags & IP_CT_TCP_SIMULTANEOUS_OPEN)
1083 new_state = TCP_CONNTRACK_ESTABLISHED;
1084 break;
1085 case TCP_CONNTRACK_CLOSE:
1086 if (index != TCP_RST_SET)
1087 break;
1088
1089 /* If we are closing, tuple might have been re-used already.
1090 * last_index, last_ack, and all other ct fields used for
1091 * sequence/window validation are outdated in that case.
1092 *
1093 * As the conntrack can already be expired by GC under pressure,
1094 * just skip validation checks.
1095 */
1096 if (tcp_can_early_drop(ct))
1097 goto in_window;
1098
1099 /* td_maxack might be outdated if we let a SYN through earlier */
1100 if ((ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET) &&
1101 ct->proto.tcp.last_index != TCP_SYN_SET) {
1102 u32 seq = ntohl(th->seq);
1103
1104 /* If we are not in established state and SEQ=0 this is most
1105 * likely an answer to a SYN we let go through above (last_index
1106 * can be updated due to out-of-order ACKs).
1107 */
1108 if (seq == 0 && !nf_conntrack_tcp_established(ct))
1109 break;
1110
1111 if (before(seq, ct->proto.tcp.seen[!dir].td_maxack) &&
1112 !tn->tcp_ignore_invalid_rst) {
1113 /* Invalid RST */
1114 spin_unlock_bh(&ct->lock);
1115 nf_ct_l4proto_log_invalid(skb, ct, state, "invalid rst");
1116 return -NF_ACCEPT;
1117 }
1118
1119 if (!nf_conntrack_tcp_established(ct) ||
1120 seq == ct->proto.tcp.seen[!dir].td_maxack)
1121 break;
1122
1123 /* Check if rst is part of train, such as
1124 * foo:80 > bar:4379: P, 235946583:235946602(19) ack 42
1125 * foo:80 > bar:4379: R, 235946602:235946602(0) ack 42
1126 */
1127 if (ct->proto.tcp.last_index == TCP_ACK_SET &&
1128 ct->proto.tcp.last_dir == dir &&
1129 seq == ct->proto.tcp.last_end)
1130 break;
1131
1132 /* ... RST sequence number doesn't match exactly, keep
1133 * established state to allow a possible challenge ACK.
1134 */
1135 new_state = old_state;
1136 }
1137 if (((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
1138 && ct->proto.tcp.last_index == TCP_SYN_SET)
1139 || (!test_bit(IPS_ASSURED_BIT, &ct->status)
1140 && ct->proto.tcp.last_index == TCP_ACK_SET))
1141 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
1142 /* RST sent to invalid SYN or ACK we had let through
1143 * at a) and c) above:
1144 *
1145 * a) SYN was in window then
1146 * c) we hold a half-open connection.
1147 *
1148 * Delete our connection entry.
1149 * We skip window checking, because packet might ACK
1150 * segments we ignored. */
1151 goto in_window;
1152 }
1153 break;
1154 default:
1155 /* Keep compilers happy. */
1156 break;
1157 }
1158
1159 if (!tcp_in_window(ct, dir, index,
1160 skb, dataoff, th, state)) {
1161 spin_unlock_bh(&ct->lock);
1162 return -NF_ACCEPT;
1163 }
1164 in_window:
1165 /* From now on we have got in-window packets */
1166 ct->proto.tcp.last_index = index;
1167 ct->proto.tcp.last_dir = dir;
1168
1169 pr_debug("tcp_conntracks: ");
1170 nf_ct_dump_tuple(tuple);
1171 pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1172 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1173 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1174 old_state, new_state);
1175
1176 ct->proto.tcp.state = new_state;
1177 if (old_state != new_state
1178 && new_state == TCP_CONNTRACK_FIN_WAIT)
1179 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1180
1181 timeouts = nf_ct_timeout_lookup(ct);
1182 if (!timeouts)
1183 timeouts = tn->timeouts;
1184
1185 if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1186 timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1187 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1188 else if (unlikely(index == TCP_RST_SET))
1189 timeout = timeouts[TCP_CONNTRACK_CLOSE];
1190 else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1191 IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1192 timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1193 timeout = timeouts[TCP_CONNTRACK_UNACK];
1194 else if (ct->proto.tcp.last_win == 0 &&
1195 timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1196 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1197 else
1198 timeout = timeouts[new_state];
1199 spin_unlock_bh(&ct->lock);
1200
1201 if (new_state != old_state)
1202 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1203
1204 if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1205 /* If only reply is a RST, we can consider ourselves not to
1206 have an established connection: this is a fairly common
1207 problem case, so we can delete the conntrack
1208 immediately. --RR */
1209 if (th->rst) {
1210 nf_ct_kill_acct(ct, ctinfo, skb);
1211 return NF_ACCEPT;
1212 }
1213
1214 if (index == TCP_SYN_SET && old_state == TCP_CONNTRACK_SYN_SENT) {
1215 /* do not renew timeout on SYN retransmit.
1216 *
1217 * Else port reuse by client or NAT middlebox can keep
1218 * entry alive indefinitely (including nat info).
1219 */
1220 return NF_ACCEPT;
1221 }
1222
1223 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1224 * pickup with loose=1. Avoid large ESTABLISHED timeout.
1225 */
1226 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1227 timeout > timeouts[TCP_CONNTRACK_UNACK])
1228 timeout = timeouts[TCP_CONNTRACK_UNACK];
1229 } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1230 && (old_state == TCP_CONNTRACK_SYN_RECV
1231 || old_state == TCP_CONNTRACK_ESTABLISHED)
1232 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1233 /* Set ASSURED if we see valid ack in ESTABLISHED
1234 after SYN_RECV or a valid answer for a picked up
1235 connection. */
1236 set_bit(IPS_ASSURED_BIT, &ct->status);
1237 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1238 }
1239 nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1240
1241 return NF_ACCEPT;
1242 }
1243
1244 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1245
1246 #include <linux/netfilter/nfnetlink.h>
1247 #include <linux/netfilter/nfnetlink_conntrack.h>
1248
tcp_to_nlattr(struct sk_buff * skb,struct nlattr * nla,struct nf_conn * ct,bool destroy)1249 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1250 struct nf_conn *ct, bool destroy)
1251 {
1252 struct nlattr *nest_parms;
1253 struct nf_ct_tcp_flags tmp = {};
1254
1255 spin_lock_bh(&ct->lock);
1256 nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP);
1257 if (!nest_parms)
1258 goto nla_put_failure;
1259
1260 if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state))
1261 goto nla_put_failure;
1262
1263 if (destroy)
1264 goto skip_state;
1265
1266 if (nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1267 ct->proto.tcp.seen[0].td_scale) ||
1268 nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1269 ct->proto.tcp.seen[1].td_scale))
1270 goto nla_put_failure;
1271
1272 tmp.flags = ct->proto.tcp.seen[0].flags;
1273 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1274 sizeof(struct nf_ct_tcp_flags), &tmp))
1275 goto nla_put_failure;
1276
1277 tmp.flags = ct->proto.tcp.seen[1].flags;
1278 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1279 sizeof(struct nf_ct_tcp_flags), &tmp))
1280 goto nla_put_failure;
1281 skip_state:
1282 spin_unlock_bh(&ct->lock);
1283 nla_nest_end(skb, nest_parms);
1284
1285 return 0;
1286
1287 nla_put_failure:
1288 spin_unlock_bh(&ct->lock);
1289 return -1;
1290 }
1291
1292 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1293 [CTA_PROTOINFO_TCP_STATE] = { .type = NLA_U8 },
1294 [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1295 [CTA_PROTOINFO_TCP_WSCALE_REPLY] = { .type = NLA_U8 },
1296 [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL] = { .len = sizeof(struct nf_ct_tcp_flags) },
1297 [CTA_PROTOINFO_TCP_FLAGS_REPLY] = { .len = sizeof(struct nf_ct_tcp_flags) },
1298 };
1299
1300 #define TCP_NLATTR_SIZE ( \
1301 NLA_ALIGN(NLA_HDRLEN + 1) + \
1302 NLA_ALIGN(NLA_HDRLEN + 1) + \
1303 NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)) + \
1304 NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)))
1305
nlattr_to_tcp(struct nlattr * cda[],struct nf_conn * ct)1306 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1307 {
1308 struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1309 struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1310 int err;
1311
1312 /* updates could not contain anything about the private
1313 * protocol info, in that case skip the parsing */
1314 if (!pattr)
1315 return 0;
1316
1317 err = nla_parse_nested_deprecated(tb, CTA_PROTOINFO_TCP_MAX, pattr,
1318 tcp_nla_policy, NULL);
1319 if (err < 0)
1320 return err;
1321
1322 if (tb[CTA_PROTOINFO_TCP_STATE] &&
1323 nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1324 return -EINVAL;
1325
1326 spin_lock_bh(&ct->lock);
1327 if (tb[CTA_PROTOINFO_TCP_STATE])
1328 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1329
1330 if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1331 struct nf_ct_tcp_flags *attr =
1332 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1333 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1334 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1335 }
1336
1337 if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1338 struct nf_ct_tcp_flags *attr =
1339 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1340 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1341 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1342 }
1343
1344 if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1345 tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1346 ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1347 ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1348 ct->proto.tcp.seen[0].td_scale =
1349 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1350 ct->proto.tcp.seen[1].td_scale =
1351 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1352 }
1353 spin_unlock_bh(&ct->lock);
1354
1355 return 0;
1356 }
1357
tcp_nlattr_tuple_size(void)1358 static unsigned int tcp_nlattr_tuple_size(void)
1359 {
1360 static unsigned int size __read_mostly;
1361
1362 if (!size)
1363 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1364
1365 return size;
1366 }
1367 #endif
1368
1369 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1370
1371 #include <linux/netfilter/nfnetlink.h>
1372 #include <linux/netfilter/nfnetlink_cttimeout.h>
1373
tcp_timeout_nlattr_to_obj(struct nlattr * tb[],struct net * net,void * data)1374 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1375 struct net *net, void *data)
1376 {
1377 struct nf_tcp_net *tn = nf_tcp_pernet(net);
1378 unsigned int *timeouts = data;
1379 int i;
1380
1381 if (!timeouts)
1382 timeouts = tn->timeouts;
1383 /* set default TCP timeouts. */
1384 for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1385 timeouts[i] = tn->timeouts[i];
1386
1387 if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1388 timeouts[TCP_CONNTRACK_SYN_SENT] =
1389 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1390 }
1391
1392 if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1393 timeouts[TCP_CONNTRACK_SYN_RECV] =
1394 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1395 }
1396 if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1397 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1398 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1399 }
1400 if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1401 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1402 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1403 }
1404 if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1405 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1406 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1407 }
1408 if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1409 timeouts[TCP_CONNTRACK_LAST_ACK] =
1410 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1411 }
1412 if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1413 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1414 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1415 }
1416 if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1417 timeouts[TCP_CONNTRACK_CLOSE] =
1418 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1419 }
1420 if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1421 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1422 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1423 }
1424 if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1425 timeouts[TCP_CONNTRACK_RETRANS] =
1426 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1427 }
1428 if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1429 timeouts[TCP_CONNTRACK_UNACK] =
1430 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1431 }
1432
1433 timeouts[CTA_TIMEOUT_TCP_UNSPEC] = timeouts[CTA_TIMEOUT_TCP_SYN_SENT];
1434 return 0;
1435 }
1436
1437 static int
tcp_timeout_obj_to_nlattr(struct sk_buff * skb,const void * data)1438 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1439 {
1440 const unsigned int *timeouts = data;
1441
1442 if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1443 htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1444 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1445 htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1446 nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1447 htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1448 nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1449 htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1450 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1451 htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1452 nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1453 htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1454 nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1455 htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1456 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1457 htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1458 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1459 htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1460 nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1461 htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1462 nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1463 htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1464 goto nla_put_failure;
1465 return 0;
1466
1467 nla_put_failure:
1468 return -ENOSPC;
1469 }
1470
1471 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1472 [CTA_TIMEOUT_TCP_SYN_SENT] = { .type = NLA_U32 },
1473 [CTA_TIMEOUT_TCP_SYN_RECV] = { .type = NLA_U32 },
1474 [CTA_TIMEOUT_TCP_ESTABLISHED] = { .type = NLA_U32 },
1475 [CTA_TIMEOUT_TCP_FIN_WAIT] = { .type = NLA_U32 },
1476 [CTA_TIMEOUT_TCP_CLOSE_WAIT] = { .type = NLA_U32 },
1477 [CTA_TIMEOUT_TCP_LAST_ACK] = { .type = NLA_U32 },
1478 [CTA_TIMEOUT_TCP_TIME_WAIT] = { .type = NLA_U32 },
1479 [CTA_TIMEOUT_TCP_CLOSE] = { .type = NLA_U32 },
1480 [CTA_TIMEOUT_TCP_SYN_SENT2] = { .type = NLA_U32 },
1481 [CTA_TIMEOUT_TCP_RETRANS] = { .type = NLA_U32 },
1482 [CTA_TIMEOUT_TCP_UNACK] = { .type = NLA_U32 },
1483 };
1484 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1485
nf_conntrack_tcp_init_net(struct net * net)1486 void nf_conntrack_tcp_init_net(struct net *net)
1487 {
1488 struct nf_tcp_net *tn = nf_tcp_pernet(net);
1489 int i;
1490
1491 for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1492 tn->timeouts[i] = tcp_timeouts[i];
1493
1494 /* timeouts[0] is unused, make it same as SYN_SENT so
1495 * ->timeouts[0] contains 'new' timeout, like udp or icmp.
1496 */
1497 tn->timeouts[0] = tcp_timeouts[TCP_CONNTRACK_SYN_SENT];
1498
1499 /* If it is set to zero, we disable picking up already established
1500 * connections.
1501 */
1502 tn->tcp_loose = 1;
1503
1504 /* "Be conservative in what you do,
1505 * be liberal in what you accept from others."
1506 * If it's non-zero, we mark only out of window RST segments as INVALID.
1507 */
1508 tn->tcp_be_liberal = 0;
1509
1510 /* If it's non-zero, we turn off RST sequence number check */
1511 tn->tcp_ignore_invalid_rst = 0;
1512
1513 /* Max number of the retransmitted packets without receiving an (acceptable)
1514 * ACK from the destination. If this number is reached, a shorter timer
1515 * will be started.
1516 */
1517 tn->tcp_max_retrans = 3;
1518
1519 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
1520 tn->offload_timeout = 30 * HZ;
1521 #endif
1522 }
1523
1524 const struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp =
1525 {
1526 .l4proto = IPPROTO_TCP,
1527 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1528 .print_conntrack = tcp_print_conntrack,
1529 #endif
1530 .can_early_drop = tcp_can_early_drop,
1531 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1532 .to_nlattr = tcp_to_nlattr,
1533 .from_nlattr = nlattr_to_tcp,
1534 .tuple_to_nlattr = nf_ct_port_tuple_to_nlattr,
1535 .nlattr_to_tuple = nf_ct_port_nlattr_to_tuple,
1536 .nlattr_tuple_size = tcp_nlattr_tuple_size,
1537 .nlattr_size = TCP_NLATTR_SIZE,
1538 .nla_policy = nf_ct_port_nla_policy,
1539 #endif
1540 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1541 .ctnl_timeout = {
1542 .nlattr_to_obj = tcp_timeout_nlattr_to_obj,
1543 .obj_to_nlattr = tcp_timeout_obj_to_nlattr,
1544 .nlattr_max = CTA_TIMEOUT_TCP_MAX,
1545 .obj_size = sizeof(unsigned int) *
1546 TCP_CONNTRACK_TIMEOUT_MAX,
1547 .nla_policy = tcp_timeout_nla_policy,
1548 },
1549 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1550 };
1551