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 =
345 state->flags = 0;
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 nf_ct_l4proto_log_invalid(skb, ct, hook_state,
675 "%s",
676 before(seq, sender->td_maxend + 1) ?
677 in_recv_win ?
678 before(sack, receiver->td_end + 1) ?
679 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
680 : "ACK is under the lower bound (possible overly delayed ACK)"
681 : "ACK is over the upper bound (ACKed data not seen yet)"
682 : "SEQ is under the lower bound (already ACKed data retransmitted)"
683 : "SEQ is over the upper bound (over the window of the receiver)");
684 }
685 }
686
687 pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
688 "receiver end=%u maxend=%u maxwin=%u\n",
689 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
690 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
691
692 return res;
693 }
694
695 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
696 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
697 TCPHDR_URG) + 1] =
698 {
699 [TCPHDR_SYN] = 1,
700 [TCPHDR_SYN|TCPHDR_URG] = 1,
701 [TCPHDR_SYN|TCPHDR_ACK] = 1,
702 [TCPHDR_RST] = 1,
703 [TCPHDR_RST|TCPHDR_ACK] = 1,
704 [TCPHDR_FIN|TCPHDR_ACK] = 1,
705 [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG] = 1,
706 [TCPHDR_ACK] = 1,
707 [TCPHDR_ACK|TCPHDR_URG] = 1,
708 };
709
tcp_error_log(const struct sk_buff * skb,const struct nf_hook_state * state,const char * msg)710 static void tcp_error_log(const struct sk_buff *skb,
711 const struct nf_hook_state *state,
712 const char *msg)
713 {
714 nf_l4proto_log_invalid(skb, state, IPPROTO_TCP, "%s", msg);
715 }
716
717 /* 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)718 static bool tcp_error(const struct tcphdr *th,
719 struct sk_buff *skb,
720 unsigned int dataoff,
721 const struct nf_hook_state *state)
722 {
723 unsigned int tcplen = skb->len - dataoff;
724 u8 tcpflags;
725
726 /* Not whole TCP header or malformed packet */
727 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
728 tcp_error_log(skb, state, "truncated packet");
729 return true;
730 }
731
732 /* Checksum invalid? Ignore.
733 * We skip checking packets on the outgoing path
734 * because the checksum is assumed to be correct.
735 */
736 /* FIXME: Source route IP option packets --RR */
737 if (state->net->ct.sysctl_checksum &&
738 state->hook == NF_INET_PRE_ROUTING &&
739 nf_checksum(skb, state->hook, dataoff, IPPROTO_TCP, state->pf)) {
740 tcp_error_log(skb, state, "bad checksum");
741 return true;
742 }
743
744 /* Check TCP flags. */
745 tcpflags = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
746 if (!tcp_valid_flags[tcpflags]) {
747 tcp_error_log(skb, state, "invalid tcp flag combination");
748 return true;
749 }
750
751 return false;
752 }
753
tcp_new(struct nf_conn * ct,const struct sk_buff * skb,unsigned int dataoff,const struct tcphdr * th)754 static noinline bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
755 unsigned int dataoff,
756 const struct tcphdr *th)
757 {
758 enum tcp_conntrack new_state;
759 struct net *net = nf_ct_net(ct);
760 const struct nf_tcp_net *tn = nf_tcp_pernet(net);
761 const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
762 const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
763
764 /* Don't need lock here: this conntrack not in circulation yet */
765 new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
766
767 /* Invalid: delete conntrack */
768 if (new_state >= TCP_CONNTRACK_MAX) {
769 pr_debug("nf_ct_tcp: invalid new deleting.\n");
770 return false;
771 }
772
773 if (new_state == TCP_CONNTRACK_SYN_SENT) {
774 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
775 /* SYN packet */
776 ct->proto.tcp.seen[0].td_end =
777 segment_seq_plus_len(ntohl(th->seq), skb->len,
778 dataoff, th);
779 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
780 if (ct->proto.tcp.seen[0].td_maxwin == 0)
781 ct->proto.tcp.seen[0].td_maxwin = 1;
782 ct->proto.tcp.seen[0].td_maxend =
783 ct->proto.tcp.seen[0].td_end;
784
785 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
786 } else if (tn->tcp_loose == 0) {
787 /* Don't try to pick up connections. */
788 return false;
789 } else {
790 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
791 /*
792 * We are in the middle of a connection,
793 * its history is lost for us.
794 * Let's try to use the data from the packet.
795 */
796 ct->proto.tcp.seen[0].td_end =
797 segment_seq_plus_len(ntohl(th->seq), skb->len,
798 dataoff, th);
799 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
800 if (ct->proto.tcp.seen[0].td_maxwin == 0)
801 ct->proto.tcp.seen[0].td_maxwin = 1;
802 ct->proto.tcp.seen[0].td_maxend =
803 ct->proto.tcp.seen[0].td_end +
804 ct->proto.tcp.seen[0].td_maxwin;
805
806 /* We assume SACK and liberal window checking to handle
807 * window scaling */
808 ct->proto.tcp.seen[0].flags =
809 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
810 IP_CT_TCP_FLAG_BE_LIBERAL;
811 }
812
813 /* tcp_packet will set them */
814 ct->proto.tcp.last_index = TCP_NONE_SET;
815
816 pr_debug("%s: sender end=%u maxend=%u maxwin=%u scale=%i "
817 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
818 __func__,
819 sender->td_end, sender->td_maxend, sender->td_maxwin,
820 sender->td_scale,
821 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
822 receiver->td_scale);
823 return true;
824 }
825
tcp_can_early_drop(const struct nf_conn * ct)826 static bool tcp_can_early_drop(const struct nf_conn *ct)
827 {
828 switch (ct->proto.tcp.state) {
829 case TCP_CONNTRACK_FIN_WAIT:
830 case TCP_CONNTRACK_LAST_ACK:
831 case TCP_CONNTRACK_TIME_WAIT:
832 case TCP_CONNTRACK_CLOSE:
833 case TCP_CONNTRACK_CLOSE_WAIT:
834 return true;
835 default:
836 break;
837 }
838
839 return false;
840 }
841
842 /* 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)843 int nf_conntrack_tcp_packet(struct nf_conn *ct,
844 struct sk_buff *skb,
845 unsigned int dataoff,
846 enum ip_conntrack_info ctinfo,
847 const struct nf_hook_state *state)
848 {
849 struct net *net = nf_ct_net(ct);
850 struct nf_tcp_net *tn = nf_tcp_pernet(net);
851 struct nf_conntrack_tuple *tuple;
852 enum tcp_conntrack new_state, old_state;
853 unsigned int index, *timeouts;
854 enum ip_conntrack_dir dir;
855 const struct tcphdr *th;
856 struct tcphdr _tcph;
857 unsigned long timeout;
858
859 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
860 if (th == NULL)
861 return -NF_ACCEPT;
862
863 if (tcp_error(th, skb, dataoff, state))
864 return -NF_ACCEPT;
865
866 if (!nf_ct_is_confirmed(ct) && !tcp_new(ct, skb, dataoff, th))
867 return -NF_ACCEPT;
868
869 spin_lock_bh(&ct->lock);
870 old_state = ct->proto.tcp.state;
871 dir = CTINFO2DIR(ctinfo);
872 index = get_conntrack_index(th);
873 new_state = tcp_conntracks[dir][index][old_state];
874 tuple = &ct->tuplehash[dir].tuple;
875
876 switch (new_state) {
877 case TCP_CONNTRACK_SYN_SENT:
878 if (old_state < TCP_CONNTRACK_TIME_WAIT)
879 break;
880 /* RFC 1122: "When a connection is closed actively,
881 * it MUST linger in TIME-WAIT state for a time 2xMSL
882 * (Maximum Segment Lifetime). However, it MAY accept
883 * a new SYN from the remote TCP to reopen the connection
884 * directly from TIME-WAIT state, if..."
885 * We ignore the conditions because we are in the
886 * TIME-WAIT state anyway.
887 *
888 * Handle aborted connections: we and the server
889 * think there is an existing connection but the client
890 * aborts it and starts a new one.
891 */
892 if (((ct->proto.tcp.seen[dir].flags
893 | ct->proto.tcp.seen[!dir].flags)
894 & IP_CT_TCP_FLAG_CLOSE_INIT)
895 || (ct->proto.tcp.last_dir == dir
896 && ct->proto.tcp.last_index == TCP_RST_SET)) {
897 /* Attempt to reopen a closed/aborted connection.
898 * Delete this connection and look up again. */
899 spin_unlock_bh(&ct->lock);
900
901 /* Only repeat if we can actually remove the timer.
902 * Destruction may already be in progress in process
903 * context and we must give it a chance to terminate.
904 */
905 if (nf_ct_kill(ct))
906 return -NF_REPEAT;
907 return NF_DROP;
908 }
909 fallthrough;
910 case TCP_CONNTRACK_IGNORE:
911 /* Ignored packets:
912 *
913 * Our connection entry may be out of sync, so ignore
914 * packets which may signal the real connection between
915 * the client and the server.
916 *
917 * a) SYN in ORIGINAL
918 * b) SYN/ACK in REPLY
919 * c) ACK in reply direction after initial SYN in original.
920 *
921 * If the ignored packet is invalid, the receiver will send
922 * a RST we'll catch below.
923 */
924 if (index == TCP_SYNACK_SET
925 && ct->proto.tcp.last_index == TCP_SYN_SET
926 && ct->proto.tcp.last_dir != dir
927 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
928 /* b) This SYN/ACK acknowledges a SYN that we earlier
929 * ignored as invalid. This means that the client and
930 * the server are both in sync, while the firewall is
931 * not. We get in sync from the previously annotated
932 * values.
933 */
934 old_state = TCP_CONNTRACK_SYN_SENT;
935 new_state = TCP_CONNTRACK_SYN_RECV;
936 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
937 ct->proto.tcp.last_end;
938 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
939 ct->proto.tcp.last_end;
940 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
941 ct->proto.tcp.last_win == 0 ?
942 1 : ct->proto.tcp.last_win;
943 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
944 ct->proto.tcp.last_wscale;
945 ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
946 ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
947 ct->proto.tcp.last_flags;
948 memset(&ct->proto.tcp.seen[dir], 0,
949 sizeof(struct ip_ct_tcp_state));
950 break;
951 }
952 ct->proto.tcp.last_index = index;
953 ct->proto.tcp.last_dir = dir;
954 ct->proto.tcp.last_seq = ntohl(th->seq);
955 ct->proto.tcp.last_end =
956 segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
957 ct->proto.tcp.last_win = ntohs(th->window);
958
959 /* a) This is a SYN in ORIGINAL. The client and the server
960 * may be in sync but we are not. In that case, we annotate
961 * the TCP options and let the packet go through. If it is a
962 * valid SYN packet, the server will reply with a SYN/ACK, and
963 * then we'll get in sync. Otherwise, the server potentially
964 * responds with a challenge ACK if implementing RFC5961.
965 */
966 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
967 struct ip_ct_tcp_state seen = {};
968
969 ct->proto.tcp.last_flags =
970 ct->proto.tcp.last_wscale = 0;
971 tcp_options(skb, dataoff, th, &seen);
972 if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
973 ct->proto.tcp.last_flags |=
974 IP_CT_TCP_FLAG_WINDOW_SCALE;
975 ct->proto.tcp.last_wscale = seen.td_scale;
976 }
977 if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
978 ct->proto.tcp.last_flags |=
979 IP_CT_TCP_FLAG_SACK_PERM;
980 }
981 /* Mark the potential for RFC5961 challenge ACK,
982 * this pose a special problem for LAST_ACK state
983 * as ACK is intrepretated as ACKing last FIN.
984 */
985 if (old_state == TCP_CONNTRACK_LAST_ACK)
986 ct->proto.tcp.last_flags |=
987 IP_CT_EXP_CHALLENGE_ACK;
988 }
989 spin_unlock_bh(&ct->lock);
990 nf_ct_l4proto_log_invalid(skb, ct, state,
991 "packet (index %d) in dir %d ignored, state %s",
992 index, dir,
993 tcp_conntrack_names[old_state]);
994 return NF_ACCEPT;
995 case TCP_CONNTRACK_MAX:
996 /* Special case for SYN proxy: when the SYN to the server or
997 * the SYN/ACK from the server is lost, the client may transmit
998 * a keep-alive packet while in SYN_SENT state. This needs to
999 * be associated with the original conntrack entry in order to
1000 * generate a new SYN with the correct sequence number.
1001 */
1002 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
1003 index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
1004 ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
1005 ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
1006 pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
1007 spin_unlock_bh(&ct->lock);
1008 return NF_ACCEPT;
1009 }
1010
1011 /* Invalid packet */
1012 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
1013 dir, get_conntrack_index(th), old_state);
1014 spin_unlock_bh(&ct->lock);
1015 nf_ct_l4proto_log_invalid(skb, ct, state, "invalid state");
1016 return -NF_ACCEPT;
1017 case TCP_CONNTRACK_TIME_WAIT:
1018 /* RFC5961 compliance cause stack to send "challenge-ACK"
1019 * e.g. in response to spurious SYNs. Conntrack MUST
1020 * not believe this ACK is acking last FIN.
1021 */
1022 if (old_state == TCP_CONNTRACK_LAST_ACK &&
1023 index == TCP_ACK_SET &&
1024 ct->proto.tcp.last_dir != dir &&
1025 ct->proto.tcp.last_index == TCP_SYN_SET &&
1026 (ct->proto.tcp.last_flags & IP_CT_EXP_CHALLENGE_ACK)) {
1027 /* Detected RFC5961 challenge ACK */
1028 ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
1029 spin_unlock_bh(&ct->lock);
1030 nf_ct_l4proto_log_invalid(skb, ct, state, "challenge-ack ignored");
1031 return NF_ACCEPT; /* Don't change state */
1032 }
1033 break;
1034 case TCP_CONNTRACK_SYN_SENT2:
1035 /* tcp_conntracks table is not smart enough to handle
1036 * simultaneous open.
1037 */
1038 ct->proto.tcp.last_flags |= IP_CT_TCP_SIMULTANEOUS_OPEN;
1039 break;
1040 case TCP_CONNTRACK_SYN_RECV:
1041 if (dir == IP_CT_DIR_REPLY && index == TCP_ACK_SET &&
1042 ct->proto.tcp.last_flags & IP_CT_TCP_SIMULTANEOUS_OPEN)
1043 new_state = TCP_CONNTRACK_ESTABLISHED;
1044 break;
1045 case TCP_CONNTRACK_CLOSE:
1046 if (index != TCP_RST_SET)
1047 break;
1048
1049 /* If we are closing, tuple might have been re-used already.
1050 * last_index, last_ack, and all other ct fields used for
1051 * sequence/window validation are outdated in that case.
1052 *
1053 * As the conntrack can already be expired by GC under pressure,
1054 * just skip validation checks.
1055 */
1056 if (tcp_can_early_drop(ct))
1057 goto in_window;
1058
1059 /* td_maxack might be outdated if we let a SYN through earlier */
1060 if ((ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET) &&
1061 ct->proto.tcp.last_index != TCP_SYN_SET) {
1062 u32 seq = ntohl(th->seq);
1063
1064 /* If we are not in established state and SEQ=0 this is most
1065 * likely an answer to a SYN we let go through above (last_index
1066 * can be updated due to out-of-order ACKs).
1067 */
1068 if (seq == 0 && !nf_conntrack_tcp_established(ct))
1069 break;
1070
1071 if (before(seq, ct->proto.tcp.seen[!dir].td_maxack) &&
1072 !tn->tcp_ignore_invalid_rst) {
1073 /* Invalid RST */
1074 spin_unlock_bh(&ct->lock);
1075 nf_ct_l4proto_log_invalid(skb, ct, state, "invalid rst");
1076 return -NF_ACCEPT;
1077 }
1078
1079 if (!nf_conntrack_tcp_established(ct) ||
1080 seq == ct->proto.tcp.seen[!dir].td_maxack)
1081 break;
1082
1083 /* Check if rst is part of train, such as
1084 * foo:80 > bar:4379: P, 235946583:235946602(19) ack 42
1085 * foo:80 > bar:4379: R, 235946602:235946602(0) ack 42
1086 */
1087 if (ct->proto.tcp.last_index == TCP_ACK_SET &&
1088 ct->proto.tcp.last_dir == dir &&
1089 seq == ct->proto.tcp.last_end)
1090 break;
1091
1092 /* ... RST sequence number doesn't match exactly, keep
1093 * established state to allow a possible challenge ACK.
1094 */
1095 new_state = old_state;
1096 }
1097 if (((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
1098 && ct->proto.tcp.last_index == TCP_SYN_SET)
1099 || (!test_bit(IPS_ASSURED_BIT, &ct->status)
1100 && ct->proto.tcp.last_index == TCP_ACK_SET))
1101 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
1102 /* RST sent to invalid SYN or ACK we had let through
1103 * at a) and c) above:
1104 *
1105 * a) SYN was in window then
1106 * c) we hold a half-open connection.
1107 *
1108 * Delete our connection entry.
1109 * We skip window checking, because packet might ACK
1110 * segments we ignored. */
1111 goto in_window;
1112 }
1113 break;
1114 default:
1115 /* Keep compilers happy. */
1116 break;
1117 }
1118
1119 if (!tcp_in_window(ct, dir, index,
1120 skb, dataoff, th, state)) {
1121 spin_unlock_bh(&ct->lock);
1122 return -NF_ACCEPT;
1123 }
1124 in_window:
1125 /* From now on we have got in-window packets */
1126 ct->proto.tcp.last_index = index;
1127 ct->proto.tcp.last_dir = dir;
1128
1129 pr_debug("tcp_conntracks: ");
1130 nf_ct_dump_tuple(tuple);
1131 pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1132 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1133 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1134 old_state, new_state);
1135
1136 ct->proto.tcp.state = new_state;
1137 if (old_state != new_state
1138 && new_state == TCP_CONNTRACK_FIN_WAIT)
1139 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1140
1141 timeouts = nf_ct_timeout_lookup(ct);
1142 if (!timeouts)
1143 timeouts = tn->timeouts;
1144
1145 if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1146 timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1147 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1148 else if (unlikely(index == TCP_RST_SET))
1149 timeout = timeouts[TCP_CONNTRACK_CLOSE];
1150 else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1151 IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1152 timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1153 timeout = timeouts[TCP_CONNTRACK_UNACK];
1154 else if (ct->proto.tcp.last_win == 0 &&
1155 timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1156 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1157 else
1158 timeout = timeouts[new_state];
1159 spin_unlock_bh(&ct->lock);
1160
1161 if (new_state != old_state)
1162 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1163
1164 if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1165 /* If only reply is a RST, we can consider ourselves not to
1166 have an established connection: this is a fairly common
1167 problem case, so we can delete the conntrack
1168 immediately. --RR */
1169 if (th->rst) {
1170 nf_ct_kill_acct(ct, ctinfo, skb);
1171 return NF_ACCEPT;
1172 }
1173
1174 if (index == TCP_SYN_SET && old_state == TCP_CONNTRACK_SYN_SENT) {
1175 /* do not renew timeout on SYN retransmit.
1176 *
1177 * Else port reuse by client or NAT middlebox can keep
1178 * entry alive indefinitely (including nat info).
1179 */
1180 return NF_ACCEPT;
1181 }
1182
1183 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1184 * pickup with loose=1. Avoid large ESTABLISHED timeout.
1185 */
1186 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1187 timeout > timeouts[TCP_CONNTRACK_UNACK])
1188 timeout = timeouts[TCP_CONNTRACK_UNACK];
1189 } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1190 && (old_state == TCP_CONNTRACK_SYN_RECV
1191 || old_state == TCP_CONNTRACK_ESTABLISHED)
1192 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1193 /* Set ASSURED if we see valid ack in ESTABLISHED
1194 after SYN_RECV or a valid answer for a picked up
1195 connection. */
1196 set_bit(IPS_ASSURED_BIT, &ct->status);
1197 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1198 }
1199 nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1200
1201 return NF_ACCEPT;
1202 }
1203
1204 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1205
1206 #include <linux/netfilter/nfnetlink.h>
1207 #include <linux/netfilter/nfnetlink_conntrack.h>
1208
tcp_to_nlattr(struct sk_buff * skb,struct nlattr * nla,struct nf_conn * ct,bool destroy)1209 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1210 struct nf_conn *ct, bool destroy)
1211 {
1212 struct nlattr *nest_parms;
1213 struct nf_ct_tcp_flags tmp = {};
1214
1215 spin_lock_bh(&ct->lock);
1216 nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP);
1217 if (!nest_parms)
1218 goto nla_put_failure;
1219
1220 if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state))
1221 goto nla_put_failure;
1222
1223 if (destroy)
1224 goto skip_state;
1225
1226 if (nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1227 ct->proto.tcp.seen[0].td_scale) ||
1228 nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1229 ct->proto.tcp.seen[1].td_scale))
1230 goto nla_put_failure;
1231
1232 tmp.flags = ct->proto.tcp.seen[0].flags;
1233 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1234 sizeof(struct nf_ct_tcp_flags), &tmp))
1235 goto nla_put_failure;
1236
1237 tmp.flags = ct->proto.tcp.seen[1].flags;
1238 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1239 sizeof(struct nf_ct_tcp_flags), &tmp))
1240 goto nla_put_failure;
1241 skip_state:
1242 spin_unlock_bh(&ct->lock);
1243 nla_nest_end(skb, nest_parms);
1244
1245 return 0;
1246
1247 nla_put_failure:
1248 spin_unlock_bh(&ct->lock);
1249 return -1;
1250 }
1251
1252 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1253 [CTA_PROTOINFO_TCP_STATE] = { .type = NLA_U8 },
1254 [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1255 [CTA_PROTOINFO_TCP_WSCALE_REPLY] = { .type = NLA_U8 },
1256 [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL] = { .len = sizeof(struct nf_ct_tcp_flags) },
1257 [CTA_PROTOINFO_TCP_FLAGS_REPLY] = { .len = sizeof(struct nf_ct_tcp_flags) },
1258 };
1259
1260 #define TCP_NLATTR_SIZE ( \
1261 NLA_ALIGN(NLA_HDRLEN + 1) + \
1262 NLA_ALIGN(NLA_HDRLEN + 1) + \
1263 NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)) + \
1264 NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)))
1265
nlattr_to_tcp(struct nlattr * cda[],struct nf_conn * ct)1266 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1267 {
1268 struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1269 struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1270 int err;
1271
1272 /* updates could not contain anything about the private
1273 * protocol info, in that case skip the parsing */
1274 if (!pattr)
1275 return 0;
1276
1277 err = nla_parse_nested_deprecated(tb, CTA_PROTOINFO_TCP_MAX, pattr,
1278 tcp_nla_policy, NULL);
1279 if (err < 0)
1280 return err;
1281
1282 if (tb[CTA_PROTOINFO_TCP_STATE] &&
1283 nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1284 return -EINVAL;
1285
1286 spin_lock_bh(&ct->lock);
1287 if (tb[CTA_PROTOINFO_TCP_STATE])
1288 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1289
1290 if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1291 struct nf_ct_tcp_flags *attr =
1292 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1293 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1294 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1295 }
1296
1297 if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1298 struct nf_ct_tcp_flags *attr =
1299 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1300 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1301 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1302 }
1303
1304 if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1305 tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1306 ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1307 ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1308 ct->proto.tcp.seen[0].td_scale =
1309 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1310 ct->proto.tcp.seen[1].td_scale =
1311 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1312 }
1313 spin_unlock_bh(&ct->lock);
1314
1315 return 0;
1316 }
1317
tcp_nlattr_tuple_size(void)1318 static unsigned int tcp_nlattr_tuple_size(void)
1319 {
1320 static unsigned int size __read_mostly;
1321
1322 if (!size)
1323 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1324
1325 return size;
1326 }
1327 #endif
1328
1329 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1330
1331 #include <linux/netfilter/nfnetlink.h>
1332 #include <linux/netfilter/nfnetlink_cttimeout.h>
1333
tcp_timeout_nlattr_to_obj(struct nlattr * tb[],struct net * net,void * data)1334 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1335 struct net *net, void *data)
1336 {
1337 struct nf_tcp_net *tn = nf_tcp_pernet(net);
1338 unsigned int *timeouts = data;
1339 int i;
1340
1341 if (!timeouts)
1342 timeouts = tn->timeouts;
1343 /* set default TCP timeouts. */
1344 for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1345 timeouts[i] = tn->timeouts[i];
1346
1347 if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1348 timeouts[TCP_CONNTRACK_SYN_SENT] =
1349 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1350 }
1351
1352 if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1353 timeouts[TCP_CONNTRACK_SYN_RECV] =
1354 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1355 }
1356 if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1357 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1358 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1359 }
1360 if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1361 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1362 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1363 }
1364 if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1365 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1366 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1367 }
1368 if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1369 timeouts[TCP_CONNTRACK_LAST_ACK] =
1370 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1371 }
1372 if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1373 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1374 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1375 }
1376 if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1377 timeouts[TCP_CONNTRACK_CLOSE] =
1378 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1379 }
1380 if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1381 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1382 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1383 }
1384 if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1385 timeouts[TCP_CONNTRACK_RETRANS] =
1386 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1387 }
1388 if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1389 timeouts[TCP_CONNTRACK_UNACK] =
1390 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1391 }
1392
1393 timeouts[CTA_TIMEOUT_TCP_UNSPEC] = timeouts[CTA_TIMEOUT_TCP_SYN_SENT];
1394 return 0;
1395 }
1396
1397 static int
tcp_timeout_obj_to_nlattr(struct sk_buff * skb,const void * data)1398 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1399 {
1400 const unsigned int *timeouts = data;
1401
1402 if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1403 htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1404 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1405 htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1406 nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1407 htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1408 nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1409 htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1410 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1411 htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1412 nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1413 htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1414 nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1415 htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1416 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1417 htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1418 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1419 htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1420 nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1421 htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1422 nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1423 htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1424 goto nla_put_failure;
1425 return 0;
1426
1427 nla_put_failure:
1428 return -ENOSPC;
1429 }
1430
1431 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1432 [CTA_TIMEOUT_TCP_SYN_SENT] = { .type = NLA_U32 },
1433 [CTA_TIMEOUT_TCP_SYN_RECV] = { .type = NLA_U32 },
1434 [CTA_TIMEOUT_TCP_ESTABLISHED] = { .type = NLA_U32 },
1435 [CTA_TIMEOUT_TCP_FIN_WAIT] = { .type = NLA_U32 },
1436 [CTA_TIMEOUT_TCP_CLOSE_WAIT] = { .type = NLA_U32 },
1437 [CTA_TIMEOUT_TCP_LAST_ACK] = { .type = NLA_U32 },
1438 [CTA_TIMEOUT_TCP_TIME_WAIT] = { .type = NLA_U32 },
1439 [CTA_TIMEOUT_TCP_CLOSE] = { .type = NLA_U32 },
1440 [CTA_TIMEOUT_TCP_SYN_SENT2] = { .type = NLA_U32 },
1441 [CTA_TIMEOUT_TCP_RETRANS] = { .type = NLA_U32 },
1442 [CTA_TIMEOUT_TCP_UNACK] = { .type = NLA_U32 },
1443 };
1444 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1445
nf_conntrack_tcp_init_net(struct net * net)1446 void nf_conntrack_tcp_init_net(struct net *net)
1447 {
1448 struct nf_tcp_net *tn = nf_tcp_pernet(net);
1449 int i;
1450
1451 for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1452 tn->timeouts[i] = tcp_timeouts[i];
1453
1454 /* timeouts[0] is unused, make it same as SYN_SENT so
1455 * ->timeouts[0] contains 'new' timeout, like udp or icmp.
1456 */
1457 tn->timeouts[0] = tcp_timeouts[TCP_CONNTRACK_SYN_SENT];
1458
1459 /* If it is set to zero, we disable picking up already established
1460 * connections.
1461 */
1462 tn->tcp_loose = 1;
1463
1464 /* "Be conservative in what you do,
1465 * be liberal in what you accept from others."
1466 * If it's non-zero, we mark only out of window RST segments as INVALID.
1467 */
1468 tn->tcp_be_liberal = 0;
1469
1470 /* If it's non-zero, we turn off RST sequence number check */
1471 tn->tcp_ignore_invalid_rst = 0;
1472
1473 /* Max number of the retransmitted packets without receiving an (acceptable)
1474 * ACK from the destination. If this number is reached, a shorter timer
1475 * will be started.
1476 */
1477 tn->tcp_max_retrans = 3;
1478
1479 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
1480 tn->offload_timeout = 30 * HZ;
1481 #endif
1482 }
1483
1484 const struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp =
1485 {
1486 .l4proto = IPPROTO_TCP,
1487 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1488 .print_conntrack = tcp_print_conntrack,
1489 #endif
1490 .can_early_drop = tcp_can_early_drop,
1491 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1492 .to_nlattr = tcp_to_nlattr,
1493 .from_nlattr = nlattr_to_tcp,
1494 .tuple_to_nlattr = nf_ct_port_tuple_to_nlattr,
1495 .nlattr_to_tuple = nf_ct_port_nlattr_to_tuple,
1496 .nlattr_tuple_size = tcp_nlattr_tuple_size,
1497 .nlattr_size = TCP_NLATTR_SIZE,
1498 .nla_policy = nf_ct_port_nla_policy,
1499 #endif
1500 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1501 .ctnl_timeout = {
1502 .nlattr_to_obj = tcp_timeout_nlattr_to_obj,
1503 .obj_to_nlattr = tcp_timeout_obj_to_nlattr,
1504 .nlattr_max = CTA_TIMEOUT_TCP_MAX,
1505 .obj_size = sizeof(unsigned int) *
1506 TCP_CONNTRACK_TIMEOUT_MAX,
1507 .nla_policy = tcp_timeout_nla_policy,
1508 },
1509 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1510 };
1511
